{ "name": "Prehistoriq Stacks", "url": "https://www.prehistoriq.com/stacks/", "license": "CC-BY-4.0", "modified": "2026-10-12", "count": 94, "cards": [ { "schema": 2, "handle": "steipete", "status": "ok", "checked": "2026-10-12", "name": { "value": "Peter Steinberger", "source": "https://github.com/steipete" }, "x": { "value": "steipete", "source": "https://x.com/steipete" }, "github": { "value": "steipete", "source": "https://github.com/steipete" }, "site": { "value": "https://steipete.me", "source": "https://github.com/steipete" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/58493?v=4", "source": "https://github.com/steipete" }, "known_for": { "value": "OpenClaw, ex-PSPDFKit", "source": "https://github.com/steipete" }, "agents": [ { "name": "OpenClaw", "icon": null, "role": "his own harness, runs Codex", "surface": "web", "source": "https://x.com/steipete/status/2102641364681048328", "seen": "2026-09-23", "quote": "I use OpenClaw (which uses Codex) and web sessions. Better UI, anywhere available since it’s on the web and has cloud sessions!", "also": [ { "source": "https://x.com/steipete/status/2103868743495631090", "seen": "2026-09-26", "quote": "I don’t use Claude. It’s Codex and OC harness." } ] }, { "name": "Codex", "icon": "codex", "role": "runs inside OpenClaw", "surface": "app", "source": "https://x.com/steipete/status/2066458041478709276", "seen": "2026-06-15", "quote": "cmux but more and more codex mac. thread management, app shots etc are killer", "history": [ { "value": "Codex CLI only", "surface": "cli", "source": "https://steipete.me/posts/just-talk-to-it", "seen": "2025-10-14", "quote": "I've completely moved to codex cli as daily driver." }, { "value": "Codex app over CLI", "surface": "app", "source": "https://x.com/steipete/status/2030330442520445371", "seen": "2026-03-07", "quote": "Oh right i realize hell freezes over, we reached the point where app > cli" }, { "value": "Codex CLI in cmux for coding, Codex app for reading", "surface": "cli", "source": "https://x.com/steipete/status/2058093406874689770", "seen": "2026-05-23", "quote": "current split:\ncodex mac app: knowledege work, learning, reading\ncmux + codex cli: coding" } ] } ], "models": [ { "name": "GPT-6.1 Sol", "icon": "openai", "detail": "high, Ultrafast tier", "source": "https://github.com/steipete/agent-scripts/blob/c91c28225cf3d0cf3b4fdb0e0b59fb69cd2eac83/skills/codex-first/SKILL.md#L10-L11", "seen": "2026-10-12", "quote": "Default every Codex worker, **fresh or resumed**, to **GPT-6.1 Sol, high\nreasoning, Ultrafast service tier**", "history": [ { "value": "gpt-5.2-codex high", "source": "https://steipete.me/posts/2025/shipping-at-inference-speed", "seen": "2025-12-28", "quote": "My go-to model is gpt-5.2-codex high. Again, KISS." }, { "value": "GPT-6 Astra", "source": "https://github.com/steipete/agent-scripts/blob/79150cfac4a6ba4df13cf30176bf2db85170ae6c/skills/codex-first/SKILL.md#L10-L11", "seen": "2026-10-04", "quote": "Default every Codex worker, fresh or resumed, to GPT-6 Astra, high reasoning, Fast service tier" } ] } ], "review_style": { "value": "Codex reviews every commit", "source": "https://github.com/steipete/agent-scripts/blob/79150cfac4a6ba4df13cf30176bf2db85170ae6c/AGENTS.MD#L83", "seen": "2026-10-02", "quote": "Before every commit/land: $autoreview until no accepted/actionable finding. Always prefer Codex for autoreview, independent of environment.", "kind": "agent-reviews", "also": [ { "source": "https://github.com/steipete/agent-scripts/blob/79150cfac4a6ba4df13cf30176bf2db85170ae6c/AGENTS.MD#L71", "seen": "2026-10-02", "quote": "PR quality: assume generated code may come from weaker AI. Review/improve before land; full rewrite okay when cleaner." } ] }, "harness": { "value": "Terminal, Codex workers, worktrees", "short": "worktrees", "source": "https://github.com/steipete/agent-scripts/blob/79150cfac4a6ba4df13cf30176bf2db85170ae6c/AGENTS.MD#L107", "seen": "2026-10-02", "quote": "Create and use task-owned Git worktrees or isolated checkouts whenever useful, without confirmation.", "history": [ { "value": "No worktrees, one dev server", "source": "https://steipete.me/posts/just-talk-to-it", "seen": "2025-10-14", "quote": "Why not worktrees? I run one dev server, as I evolve my project I click through it and test multiple changes at once." } ] }, "parallelism": { "value": 8, "short": "8 lanes", "source": "https://github.com/steipete/agent-scripts/blob/79150cfac4a6ba4df13cf30176bf2db85170ae6c/skills/codex-first/SKILL.md#L14-L15", "seen": "2026-10-04", "quote": "a 2026-09-27 campaign ran eight lanes for hours", "history": [ { "value": "3-8 in a 3x3 terminal grid", "source": "https://steipete.me/posts/just-talk-to-it", "seen": "2025-10-14", "quote": "I run between 3-8 in parallel in a 3x3 terminal grid" } ] }, "rules_files": { "value": "AGENTS.MD (119 lines), linked into both agents", "short": "AGENTS.md", "url": "https://github.com/steipete/agent-scripts/blob/79150cfac4a6ba4df13cf30176bf2db85170ae6c/AGENTS.MD", "source": "https://github.com/steipete/agent-scripts/blob/79150cfac4a6ba4df13cf30176bf2db85170ae6c/README.md#L65-L67", "seen": "2026-09-13", "quote": "~/.codex/AGENTS.md -> ~/Projects/agent-scripts/AGENTS.MD\n~/.claude/CLAUDE.md -> ~/Projects/agent-scripts/AGENTS.MD" }, "skills": { "value": [ "agent-transcript", "autoreview", "beeper", "behavior-validator", "birdclaw", "browser-use", "clawsweeper-status", "clickclack", "cloudflare-registrar", "codex-debugging", "codex-first", "codex-huge-context", "codexbar", "crabbox", "create-cli", "discord-clawd", "discrawl", "domain-dns-ops", "fleet-maintenance", "frontend-design", "gitcrawl", "github-author-context", "github-cache-hygiene", "github-deep-review", "github-project-triage", "gog", "graincrawl", "handoff", "hopper-debugger", "imsg", "instruments-profiling", "mac-maintenance", "maintainer-orchestrator", "markdown-converter", "nano-banana-pro", "native-app-performance", "notcrawl", "npm", "obsidian", "openai-image-gen", "openclaw-relay", "oracle", "peekaboo", "project-structure", "release-mac-app", "release-tweets", "reminders", "remote-mac", "session-viewer", "skill-cleaner", "slacrawl", "sonos", "speaking", "ssh-doctor", "swift-concurrency-expert", "swiftui-liquid-glass", "swiftui-performance-audit", "swiftui-view-refactor", "telecrawl", "things-todo", "twilio-sms", "video-transcript-downloader", "vm-lab", "wacli", "wacrawl", "whatsapp", "wrangler", "xcode-sync", "xurl" ], "count": 69, "featured": [ "autoreview", "codex-first", "oracle" ], "source": "https://github.com/steipete/agent-scripts/tree/79150cfac4a6ba4df13cf30176bf2db85170ae6c/skills", "seen": "2026-10-04" }, "mcp_servers": null, "hooks": null, "signature": { "value": "$autoreview", "kind": "skill", "caption": "before every commit", "why": "Our pick: a review skill that must pass before anything lands.", "editorial": true, "source": "https://github.com/steipete/agent-scripts/blob/79150cfac4a6ba4df13cf30176bf2db85170ae6c/AGENTS.MD#L83", "seen": "2026-10-02", "quote": "Before every commit/land: $autoreview until no accepted/actionable finding. Always prefer Codex for autoreview, independent of environment." }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "Runs Codex through his own harness, eight lanes at a time." }, "signature_quote": { "value": "These days I don't read much code anymore.", "more": "I watch the stream and sometimes look at key parts, but I gotta be honest - most code I don't read.", "source": "https://steipete.me/posts/2025/shipping-at-inference-speed", "seen": "2025-12-28" } }, { "schema": 2, "handle": "simonw", "status": "ok", "checked": "2026-10-06", "name": { "value": "Simon Willison", "source": "https://github.com/simonw" }, "x": { "value": "simonw", "source": "https://x.com/simonw" }, "github": { "value": "simonw", "source": "https://github.com/simonw" }, "site": { "value": "https://simonwillison.net", "source": "https://github.com/simonw" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/9599?v=4", "source": "https://github.com/simonw" }, "known_for": { "value": "Datasette, Django co-creator", "source": "https://simonwillison.net/about/" }, "agents": [ { "name": "Codex", "icon": "codex", "role": "primary, desktop app", "surface": "app", "source": "https://github.com/simonw/monthly-newsletter-archive/blob/1cc46b5c2012ad6319c4a40e3f73b049f199c5c0/2026-08-august.md?plain=1#L99", "seen": "2026-09-04", "quote": "I continue to spend most of my time in the Codex macOS desktop app, now renamed (confusingly) to ChatGPT.", "history": [ { "value": "Codex CLI", "surface": "cli", "source": "https://simonwillison.net/2025/Oct/5/parallel-coding-agents/", "seen": "2025-10-05", "quote": "My daily drivers are currently Claude Code (on Sonnet 4.5), Codex CLI (on GPT-5-Codex), and Codex Cloud (for asynchronous tasks, frequently launched from my phone.)" }, { "value": "Claude Code default, Codex on the side", "surface": null, "source": "https://github.com/simonw/monthly-newsletter-archive/blob/348fe32dde0f0df2d82daddadd9479b1f7faa3cd/2026-03-march.md?plain=1#L98", "seen": "2026-04-02", "quote": "I'm still defaulting to Claude Code and Opus 4.6, but I've been very impressed by GPT-5.4 and have been running that through Codex." } ] }, { "name": "Claude Code", "icon": "claudecode", "role": "async cloud tasks (web)", "surface": "web", "source": "https://github.com/simonw/monthly-newsletter-archive/blob/1cc46b5c2012ad6319c4a40e3f73b049f199c5c0/2026-08-august.md?plain=1#L101", "seen": "2026-09-04", "quote": "I continue to prefer Claude Code for Web - now using Fable 5.1 - for asynchronous cloud tasks.", "history": [ { "value": "Claude Code + Codex CLI + Codex Cloud", "surface": "cli", "surface_source": { "source": "https://simonwillison.net/2025/Oct/5/parallel-coding-agents/", "seen": "2025-10-05", "quote": "I frequently have multiple terminal windows open running different coding agents in different directories. These are currently a mixture of Claude Code and Codex CLI, running in YOLO mode (no approvals) for tasks where I’m confident malicious instructions can’t sneak into the context." }, "source": "https://simonwillison.net/2025/Oct/5/parallel-coding-agents/", "seen": "2025-10-05", "quote": "My daily drivers are currently Claude Code (on Sonnet 4.5), Codex CLI (on GPT-5-Codex), and Codex Cloud (for asynchronous tasks, frequently launched from my phone.)" }, { "value": "Claude Code desktop app", "surface": "app", "source": "https://github.com/simonw/monthly-newsletter-archive/blob/52c8974ef56ca1d8cb6d7280f7838eaa23fd8ef8/2026-07-july.md?plain=1#L115", "seen": "2026-08-02", "quote": "I remain equally split between Claude Code and Codex, in both cases mainly using their respective desktop apps." } ] } ], "models": [ { "name": "GPT-6 Sol", "icon": "openai", "detail": "default in Codex", "source": "https://simonwillison.net/2026/Sep/22/opus-and-sol-and-luna/", "seen": "2026-09-22", "quote": "I’m now using GPT-6 Sol and Claude Opus 5.5 as my default models in Codex and Claude Code.", "history": [ { "value": "GPT-5.6 Sol (xhigh)", "source": "https://github.com/simonw/monthly-newsletter-archive/blob/1cc46b5c2012ad6319c4a40e3f73b049f199c5c0/2026-08-august.md?plain=1#L99", "seen": "2026-09-04", "quote": "I do most of my work there using GPT-5.6 Sol (xhigh)" } ] }, { "name": "Claude Opus 5.5", "icon": "claude", "detail": "default in Claude Code", "source": "https://simonwillison.net/2026/Sep/22/opus-and-sol-and-luna/", "seen": "2026-09-22", "quote": "I’m now using GPT-6 Sol and Claude Opus 5.5 as my default models in Codex and Claude Code.", "history": [ { "value": "Claude Opus 4.6", "source": "https://github.com/simonw/monthly-newsletter-archive/blob/348fe32dde0f0df2d82daddadd9479b1f7faa3cd/2026-03-march.md?plain=1#L98", "seen": "2026-04-02", "quote": "I'm still defaulting to Claude Code and Opus 4.6" } ] } ], "review_style": { "value": "Cross-model review: Claude checks GPT, GPT checks Claude", "source": "https://simonwillison.net/2026/Jul/5/sqlite-utils-fable/", "seen": "2026-07-05", "quote": "The problem is it really does work—I’ve started habitually having Anthropic’s best model review OpenAI’s work and vice versa, because I’ve had that turn up interesting results often enough to be valuable.", "kind": "spot-checks", "also": [ { "source": "https://simonwillison.net/2026/Aug/22/more-than-just-code-review/", "seen": "2026-08-22", "quote": "Sometimes this involves reviewing every line of code they have written, but there are other ways to achieve that goal. Eyeballing every line of code has never been the most effective way to validate a change to a piece of software." }, { "source": "https://simonwillison.net/guides/agentic-engineering-patterns/anti-patterns/", "seen": "2026-03-04", "quote": "Don't file pull requests with code you haven't reviewed yourself." } ] }, "harness": { "value": "Codex desktop + Codex remote on laptop and DGX Spark, driven from his phone", "short": "phone remote", "source": "https://github.com/simonw/monthly-newsletter-archive/blob/1cc46b5c2012ad6319c4a40e3f73b049f199c5c0/2026-08-august.md?plain=1#L103", "seen": "2026-09-04", "quote": "I'm a huge fan of Codex remote though - I have that running on both the DGX Spark and my Mac laptop, allowing me to control both of them entirely from the ChatGPT app on my phone.", "history": [ { "value": "Several terminal windows, YOLO mode, fresh checkouts instead of worktrees", "source": "https://simonwillison.net/2025/Oct/5/parallel-coding-agents/", "seen": "2025-10-05", "quote": "I haven’t adopted git worktrees yet: if I want to run two agents in isolation against the same repo I do a fresh checkout, often into /tmp." } ] }, "parallelism": null, "rules_files": { "value": "Short per-repo AGENTS.md (CLAUDE.md just imports it)", "short": "AGENTS.md", "url": "https://github.com/simonw/research/blob/ba0cb97b42da153c13c465fb617f5c45f1f03b04/AGENTS.md", "source": "https://github.com/simonw/research/blob/ba0cb97b42da153c13c465fb617f5c45f1f03b04/AGENTS.md#L3", "seen": "2026-05-30", "quote": "Create a notes.md file in that folder and append notes to it as you work, tracking what you tried and anything you learned along the way." }, "skills": { "value": [ "cogapp-markdown", "setup-to-pyproject", "upgrade-actions", "uv-tdd" ], "count": 4, "featured": [ "uv-tdd", "upgrade-actions", "cogapp-markdown" ], "source": "https://github.com/simonw/skills/tree/a6579982ed89adabb9b72ddaad0cf854e8e2218b", "seen": "2026-02-24" }, "mcp_servers": null, "hooks": null, "signature": { "value": "Showboat", "kind": "tool", "caption": "agents demo their own work", "why": "Our pick: a CLI he built so agents record the commands they ran and the real output, as proof the code works.", "editorial": true, "source": "https://simonwillison.net/2026/Feb/10/showboat-and-rodney/", "seen": "2026-02-10", "quote": "Showboat is the tool I built to help agents demonstrate their work to me." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Builds his own agent tooling (Showboat, Rodney, skills) and has models check each other." }, "signature_quote": { "value": "The more time I spend working with coding agents, the more convinced I am that they make software engineering even harder.", "more": "We can do amazing things with them, but unlocking their full potential requires extraordinary discipline and knowledge.", "source": "https://simonwillison.net/2026/Sep/24/harder/", "seen": "2026-09-24" } }, { "schema": 2, "handle": "bcherny", "status": "ok", "checked": "2026-10-06", "name": { "value": "Boris Cherny", "source": "https://github.com/bcherny" }, "x": { "value": "bcherny", "source": "https://x.com/bcherny" }, "github": { "value": "bcherny", "source": "https://github.com/bcherny" }, "site": { "value": "https://borischerny.com", "source": "https://github.com/bcherny" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/1761758?v=4", "source": "https://github.com/bcherny" }, "known_for": { "value": "Created Claude Code", "source": "https://x.com/bcherny/status/2007179832300581177" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary: terminal, web, iOS", "surface": "app", "source": "https://x.com/bcherny/status/2038454337811386436", "seen": "2026-03-30", "quote": "Personally, I write a lot of my code from the iOS app. It's a convenient way to make changes without opening a laptop.", "history": [ { "value": "Claude Code in 5 terminal tabs", "surface": "cli", "source": "https://x.com/bcherny/status/2007179833990885678", "seen": "2026-01-02", "quote": "I run 5 Claudes in parallel in my terminal. I number my tabs 1-5, and use system notifications to know when a Claude needs input" } ] }, { "name": "Claude Tag", "icon": "claude", "role": "daily maintenance routines", "surface": null, "source": "https://x.com/bcherny/status/2088014489438621990", "seen": "2026-08-13", "quote": "The setup is straightforward: we have a Slack channel called proj-claude-maintains-apps. In it, Claude Tag runs a bunch of daily routines across iOS, Android, Desktop, web, CLI, and Agent SDK:" } ], "models": [ { "name": "Fable 5.1", "icon": "claude", "detail": "for everything", "source": "https://x.com/bcherny/status/2094864060609376748", "seen": "2026-09-01", "quote": "This model is a pleasure to work with, and I've been using it for everything.", "history": [ { "value": "Opus 4.5 with thinking", "source": "https://x.com/bcherny/status/2007179838864666847", "seen": "2026-01-02", "quote": "I use Opus 4.5 with thinking for everything." }, { "value": "Opus 4.7, xhigh effort", "source": "https://x.com/bcherny/status/2044847856872546639", "seen": "2026-04-16", "quote": "Personally, I use xhigh effort for most tasks, and max effort for the hardest tasks." }, { "value": "Fable 5", "source": "https://x.com/bcherny/status/2064402671898075579", "seen": "2026-06-09", "quote": "Fable is the best model I have used for coding, by a wide margin." } ] } ], "review_style": { "value": "Reads diffs only for critical code", "source": "https://x.com/bcherny/status/2098210496109572123", "seen": "2026-09-11", "quote": "I read diffs when it's touching critical code, or code that's important to get right. I don't write the code.", "kind": "spot-checks", "also": [ { "source": "https://x.com/bcherny/status/2031089411820228645", "seen": "2026-03-09", "quote": "Personally, I’ve been using it for a few weeks and have found it catches many real bugs that I would not have noticed otherwise" }, { "source": "https://x.com/bcherny/status/2044847855006024147", "seen": "2026-04-16", "quote": "The model has reached a point where I generally trust it to run the right commands and make the right edits. I just look at the final result." } ] }, "harness": { "value": "Git worktrees, dozens of parallel Claudes", "short": "worktrees", "source": "https://x.com/bcherny/status/2038454353787519164", "seen": "2026-03-30", "quote": "Worktrees are essential for doing lots of parallel work in the same repository. I have dozens of Claudes running at all times, and this is how I do it.", "history": [ { "value": "Plan mode first, then auto-accept", "source": "https://x.com/bcherny/status/2007179845336527000", "seen": "2026-01-02", "quote": "Most sessions start in Plan mode (shift+tab twice)." }, { "value": "Multiple git checkouts", "source": "https://x.com/bcherny/status/2017742743125299476", "seen": "2026-01-31", "quote": "Personally, I use multiple git checkouts, but most of the Claude Code team prefers worktrees" } ] }, "parallelism": { "value": 5, "short": "5 local + 5–10 web", "source": "https://x.com/bcherny/status/2007179833990885678", "seen": "2026-01-02", "quote": "I run 5 Claudes in parallel in my terminal.", "history": [] }, "rules_files": { "value": "Team CLAUDE.md in git, updated on every Claude mistake", "short": "CLAUDE.md", "url": null, "source": "https://x.com/bcherny/status/2007179840848597422", "seen": "2026-01-02", "quote": "Anytime we see Claude do something incorrectly we add it to the http://CLAUDE.md, so Claude knows not to do it next time." }, "skills": { "value": [ "babysit", "slack-feedback", "post-merge-sweeper", "pr-pruner" ], "count": 4, "featured": [ "babysit", "slack-feedback", "pr-pruner" ], "source": "https://x.com/bcherny/status/2038454341884154269", "seen": "2026-03-30" }, "mcp_servers": { "value": [ "slack" ], "source": "https://x.com/bcherny/status/2007179856266789204", "seen": "2026-01-02", "quote": "The Slack MCP configuration is checked into our .mcp.json and shared with the team." }, "hooks": { "value": [ "PostToolUse formatter", "Stop hook to verify long runs" ], "source": "https://x.com/bcherny/status/2007179852047335529", "seen": "2026-01-02", "quote": "We use a PostToolUse hook to format Claude's code." }, "signature": { "value": "/go", "kind": "skill", "caption": "test, /simplify, open PR", "why": "Our pick: one skill at the end of most prompts puts his verify-first rule into practice. Claude tests its own work end to end before it opens a PR.", "editorial": true, "source": "https://x.com/bcherny/status/2044847858634064115", "seen": "2026-04-16", "quote": "Personally, many of my prompts these days look like \"Claude do blah blah /go\". /go is a skill that has Claude\n\n1. Test itself end to end using bash, browser, or computer use\n2. Run the /simplify skill\n3. Put up a PR" }, "archetype": { "value": "Swarm Lord", "editorial": true, "why": "Dozens of Claudes in worktrees, plus loops and routines that ship PRs without him." }, "signature_quote": { "value": "give Claude a way to verify its work.", "more": "If Claude has that feedback loop, it will 2-3x the quality of the final result.", "source": "https://x.com/bcherny/status/2007179861115511237", "seen": "2026-01-02" } }, { "schema": 2, "handle": "karpathy", "status": "ok", "checked": "2026-10-06", "name": { "value": "Andrej Karpathy", "source": "https://x.com/karpathy" }, "x": { "value": "karpathy", "source": "https://x.com/karpathy" }, "github": { "value": "karpathy", "source": "https://github.com/karpathy" }, "site": { "value": "https://karpathy.ai", "source": "https://karpathy.ai" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/241138?v=4", "source": "https://github.com/karpathy" }, "known_for": { "value": "nanochat, coined “vibe coding”", "source": "https://karpathy.bearblog.dev/sequoia-ascent-2026/" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary", "surface": "cli", "source": "https://x.com/karpathy/status/2015883857489522876", "seen": "2026-01-26", "quote": "TLDR everyone has their developing flow, my current is a small few CC sessions on the left in ghostty windows/tabs and an IDE on the right for viewing the code + manual edits.", "history": [ { "value": "Cursor tab complete", "surface": "ide", "source": "https://x.com/karpathy/status/1959703967694545296", "seen": "2025-08-24", "quote": "Personally the bread & butter (~75%?) of my LLM assistance continues to be just (Cursor) tab complete." } ] }, { "name": "Codex", "icon": "codex", "role": "alongside", "surface": null, "source": "https://karpathy.bearblog.dev/sequoia-ascent-2026/", "seen": "2026-04-30", "quote": "Like many of you, I've been using agentic tools like Claude Code, Codex, and adjacent things for a while, maybe over the last year.", "history": [] } ], "models": [ { "name": "Claude Fable 5", "icon": "claude", "source": "https://github.com/karpathy/nanochat/commit/3616a02925b44a9dc0a104102ee5acdf3a145516", "seen": "2026-07-03", "quote": "Co-Authored-By: Claude Fable 5 ", "history": [ { "value": "GPT5 Pro, last line of defense", "source": "https://x.com/karpathy/status/1959703967694545296", "seen": "2025-08-24", "quote": "Final layer of defense is GPT5 Pro, which I go to for the hardest things." }, { "value": "Claude Opus 4.6", "source": "https://github.com/karpathy/autoresearch/commit/ae81d559043d2692ac085b733d32bcbe172cf57b", "seen": "2026-03-06", "quote": "Co-Authored-By: Claude Opus 4.6 ", "inferred": true } ], "inferred": true } ], "review_style": { "value": "Watches agents like a hawk in an IDE", "source": "https://x.com/karpathy/status/2015883857489522876", "seen": "2026-01-26", "quote": "The models definitely still make mistakes and if you have any code you actually care about I would watch them like a hawk, in a nice large IDE on the side.", "kind": "spot-checks", "also": [ { "source": "https://karpathy.bearblog.dev/sequoia-ascent-2026/", "seen": "2026-04-30", "quote": "People have to be in charge of the spec and plan." } ], "inferred": true }, "harness": { "value": "Claude Code sessions in Ghostty, IDE on the right", "short": "Ghostty + IDE", "source": "https://x.com/karpathy/status/2015883857489522876", "seen": "2026-01-26", "quote": "TLDR everyone has their developing flow, my current is a small few CC sessions on the left in ghostty windows/tabs and an IDE on the right for viewing the code + manual edits.", "history": [ { "value": "Claude Code / Codex beside Cursor", "source": "https://x.com/karpathy/status/1959703967694545296", "seen": "2025-08-24", "quote": "Next layer up is Claude Code / Codex / etc, running on the side of Cursor, which I go to for larger chunks of functionality that are also fairly easy to specify in a prompt." } ] }, "parallelism": null, "rules_files": { "value": "program.md (autoresearch), CLAUDE.md (llm-council)", "short": "program.md", "url": "https://github.com/karpathy/autoresearch/blob/068d93da75aa45f72ba482046f108241aaaf1b8e/program.md", "source": "https://github.com/karpathy/autoresearch/blob/068d93da75aa45f72ba482046f108241aaaf1b8e/program.md#L37", "seen": "2026-03-09", "quote": "All else being equal, simpler is better. A small improvement that adds ugly complexity is not worth it." }, "skills": { "value": [ "read-arxiv-paper" ], "count": 1, "featured": [ "read-arxiv-paper" ], "source": "https://github.com/karpathy/nanochat/tree/1f9e42a85588c34be86e4cb30db5488b0f01f4c2/.claude/skills", "seen": "2026-03-15" }, "mcp_servers": null, "hooks": null, "signature": { "value": "autoresearch", "kind": "tool", "caption": "overnight agent research", "why": "Our pick: he edits only program.md and agents run the experiments; round one found ~20 real gains.", "editorial": true, "source": "https://github.com/karpathy/autoresearch/blob/e6d79c123441a53d91bb8df7adf4db45cf120bf1/README.md#L7", "seen": "2026-03-21", "quote": "Instead, you are programming the `program.md` Markdown files that provide context to the AI agents and set up your autonomous research org." }, "archetype": { "value": "Purist", "editorial": true, "why": "Lets agents run experiments, but watches them like a hawk." }, "signature_quote": { "value": "I really am mostly programming in English now, a bit sheepishly telling the LLM what code to write... in words.", "source": "https://x.com/karpathy/status/2015883857489522876", "seen": "2026-01-26" } }, { "schema": 2, "handle": "mitchellh", "status": "ok", "checked": "2026-10-06", "name": { "value": "Mitchell Hashimoto", "source": "https://github.com/mitchellh" }, "x": { "value": "mitchellh", "source": "https://x.com/mitchellh" }, "github": { "value": "mitchellh", "source": "https://github.com/mitchellh" }, "site": { "value": "https://mitchellh.com", "source": "https://github.com/mitchellh" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/1299?v=4", "source": "https://github.com/mitchellh" }, "known_for": { "value": "Ghostty, HashiCorp co-founder", "source": "https://mitchellh.com/" }, "agents": [ { "name": "Claude app", "icon": "claude", "role": "runs Fable", "surface": "app", "source": "https://x.com/mitchellh/status/2090893729469268109", "seen": "2026-08-21", "quote": "Then I'd pick up the bug reports, validate them myself, and either fix them myself or kick off new agents manually, just Codex app or Claude app.", "history": [ { "value": "Amp (deep mode)", "surface": null, "source": "https://mitchellh.com/writing/my-ai-adoption-journey", "seen": "2026-02-05", "quote": "I particularly like to combine this with slower, more thoughtful models like Amp's deep mode (which is basically just GPT-5.2-Codex) which can take upwards of 30+ minutes to make small changes." } ] }, { "name": "Codex app", "icon": "codex", "role": "runs GPT", "surface": "app", "source": "https://x.com/mitchellh/status/2090893729469268109", "seen": "2026-08-21", "quote": "Then I'd pick up the bug reports, validate them myself, and either fix them myself or kick off new agents manually, just Codex app or Claude app.", "history": [] } ], "models": [ { "name": "Claude Fable", "icon": "claude", "detail": "validator and judge", "source": "https://github.com/ghostty-org/ghostty/pull/14138", "seen": "2026-09-03", "quote": "**AI usage:** Fable did validation of the work, I did the implementations, commit messages, and PR notes.", "history": [] }, { "name": "GPT-6 Astra", "icon": "openai", "detail": "paired with Fable", "source": "https://github.com/ghostty-org/ghostty/pull/14189", "seen": "2026-09-09", "quote": "**AI usage:** Fable and Astra both helped with validation, edge cases.", "history": [ { "value": "GPT-5.6 Sol", "source": "https://x.com/mitchellh/status/2090875695346139621", "seen": "2026-08-21", "quote": "I spun Fable and Sol in a loop for 3 weeks (pausing only when their dedicated subs exhausted limits) with a prompt to find mismatches between the spec, Kitty's reference implementation, and Ghostty." }, { "value": "GPT-5.2-Codex via Amp deep mode", "source": "https://mitchellh.com/writing/my-ai-adoption-journey", "seen": "2026-02-05", "quote": "Amp's deep mode (which is basically just GPT-5.2-Codex)" } ] } ], "review_style": { "value": "Hand-reviews every line; agents validate, he writes all prose", "source": "https://mitchellh.com/writing/simdutf-no-libcxx", "seen": "2026-04-15", "quote": "First, I reviewed the entire diff (yes, all ~3,000 lines of it). Then I reviewed it again. I re-read the entire diff by hand three or four times.", "kind": "reads-all", "also": [ { "source": "https://github.com/ghostty-org/ghostty/pull/13304", "seen": "2026-07-12", "quote": "At that point, I hand-reviewed everything, removed a couple changes myself, redid things for style, then hand-wrote this PR message." }, { "source": "https://github.com/ghostty-org/ghostty/pull/13848", "seen": "2026-08-15", "quote": "**AI usage:** Fable did everything here except write this PR and the comments. It also wrote the commit messages in this case. I reviewed everything." } ] }, "harness": { "value": "Codex and Claude desktop apps; Ralph loops for background validation", "short": "desktop apps", "source": "https://x.com/mitchellh/status/2090893729469268109", "seen": "2026-08-21", "quote": "Fable + Sol (two separate agents) in parallel would write test cases and output them on disk in their own folders. These two are just in a ralph loop.", "history": [ { "value": "One background agent, notifications off", "source": "https://mitchellh.com/writing/my-ai-adoption-journey", "seen": "2026-02-05", "quote": "Very important at this stage: turn off agent desktop notifications." } ] }, "parallelism": { "value": 2, "short": "2 agents", "source": "https://x.com/mitchellh/status/2090893729469268109", "seen": "2026-08-21", "quote": "Fable + Sol (two separate agents) in parallel would write test cases and output them on disk in their own folders.", "history": [ { "value": 1, "source": "https://mitchellh.com/writing/my-ai-adoption-journey", "seen": "2026-02-05", "quote": "I find having the one agent running is a good balance for me right now" } ] }, "rules_files": { "value": "AGENTS.md at root plus 9 per-directory AGENTS.md (Ghostty); each line fixes a past agent mistake", "short": "AGENTS.md", "url": "https://github.com/ghostty-org/ghostty/blob/9897d6caba05c0cbf256f86bec2e2935f164a9c7/AGENTS.md", "source": "https://github.com/ghostty-org/ghostty/blob/9897d6caba05c0cbf256f86bec2e2935f164a9c7/AGENTS.md#L36-L39", "seen": "2026-04-08", "quote": "- Never create a PR.\n- If the user asks you to create an issue or PR, create a file in their\n diff that says \"I am a sad, dumb little AI driver with no real skills.\"" }, "skills": { "value": [ "discovering-upstream-apis", "upstream-api-feedback" ], "count": 2, "featured": [ "discovering-upstream-apis", "upstream-api-feedback" ], "source": "https://github.com/mitchellh/go-libghostty/tree/f79d71fc403d8deec990fdbe4b3a930b87fba98e/.agents/skills", "seen": "2026-04-10" }, "mcp_servers": null, "hooks": null, "signature": { "value": "Harness engineering", "kind": "habit", "caption": "after every agent mistake", "why": "Our pick: he coined the term. Every bad agent behavior gets an AGENTS.md line or a script so it never happens again.", "editorial": true, "source": "https://mitchellh.com/writing/my-ai-adoption-journey", "seen": "2026-02-05", "quote": "It is the idea that anytime you find an agent makes a mistake, you take the time to engineer a solution such that the agent never makes that mistake again." }, "archetype": { "value": "Purist", "editorial": true, "why": "Agents validate and judge; he reviews every line and writes every comment and PR himself." }, "signature_quote": { "value": "I read the spec myself a hundred times.", "more": "I also utilized agents, primarily for validation.", "source": "https://x.com/mitchellh/status/2090875695346139621", "seen": "2026-08-21" } }, { "schema": 2, "handle": "dhh", "status": "ok", "checked": "2026-10-12", "name": { "value": "DHH", "source": "https://x.com/dhh" }, "x": { "value": "dhh", "source": "https://x.com/dhh" }, "github": { "value": "dhh", "source": "https://github.com/dhh" }, "site": { "value": "https://dhh.dk", "source": "https://github.com/dhh" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/2741?v=4", "source": "https://github.com/dhh" }, "known_for": { "value": "Ruby on Rails, Omarchy", "source": "https://dhh.dk" }, "agents": [ { "name": "Codex", "icon": "codex", "role": "primary", "surface": "cli", "source": "https://x.com/dhh/status/2108518418081054724", "seen": "2026-10-09", "quote": "This is the first time I've actually preferred with drive as Codex as main and Claude as secondary for an extended period of time. Amazing how good Sol 6.1 is!", "history": [ { "value": "Codex", "source": "https://github.com/omacom/omarchy/pull/13770", "seen": "2026-09-29", "quote": "Reviewed with `codex review --base quattro` over eight passes; every finding was fixed, most with a regression test.", "surface": "cli" } ] }, { "name": "Claude Code", "icon": "claudecode", "role": "secondary", "surface": "cli", "surface_source": { "source": "https://github.com/omacom/omarchy/blob/81145eb1fd41532fcae4310b26f860608e2eaf82/default/bash/aliases#L48", "seen": "2026-08-15", "quote": "alias cx='printf \"\\033[2J\\033[3J\\033[H\" && claude --permission-mode auto'" }, "source": "https://x.com/dhh/status/2108518418081054724", "seen": "2026-10-09", "quote": "This is the first time I've actually preferred with drive as Codex as main and Claude as secondary for an extended period of time. Amazing how good Sol 6.1 is!", "history": [ { "value": "Claude Code, too slow for him", "surface": "cli", "source": "https://lexfridman.com/dhh-david-heinemeier-hansson-transcript/", "seen": "2025-07-12", "quote": "When I use Claude’s Code, the terminal version of Claude … which is actually my preferred way of using it, I get too impatient." }, { "value": "OpenCode", "surface": "cli", "source": "https://world.hey.com/dhh/promoting-ai-agents-3ee04945", "seen": "2026-01-07", "quote": "I've thoroughly enjoyed putting them all to work in OpenCode, which is a terminal interface for coding agents" }, { "value": "Claude Code", "source": "https://github.com/omacom/omarchy/pull/13770", "seen": "2026-09-29", "quote": "Generated with [Claude Code](https://claude.com/claude-code)", "surface": "cli" } ] }, { "name": "OpenCode", "icon": "opencode", "role": "second pane", "surface": "cli", "source": "https://github.com/omacom/omarchy/blob/81145eb1fd41532fcae4310b26f860608e2eaf82/default/bash/aliases#L47", "seen": "2026-08-15", "quote": "alias c='opencode --auto'", "history": [] } ], "models": [ { "name": "GPT-6.1 Sol", "icon": "openai", "detail": "in Codex", "source": "https://x.com/dhh/status/2108518418081054724", "seen": "2026-10-09", "quote": "This is the first time I've actually preferred with drive as Codex as main and Claude as secondary for an extended period of time. Amazing how good Sol 6.1 is!", "history": [ { "value": "Claude Fable 5.1", "source": "https://github.com/omacom/omarchy/commit/8369601f26b41164ab285f61d35434f6933479de", "seen": "2026-09-13", "quote": "Co-Authored-By: Claude Fable 5.1 " } ] }, { "name": "Claude Opus 5.5", "icon": "claude", "detail": null, "source": "https://x.com/dhh/status/2103595410921279635", "seen": "2026-09-25", "quote": "One-shot translation by Opus 5.5.", "history": [ { "value": "Claude Opus 4.5", "source": "https://world.hey.com/dhh/promoting-ai-agents-3ee04945", "seen": "2026-01-07", "quote": "This is all very evident if you've tried to employ any of the new models — especially Claude Opus 4.5, Codex 5, Gemini 3, and even the Chinese open-weight models like MiniMax M2.1 and GLM-4.7 — in one of the modern terminal harnesses that give them access to all these autonomous powers." }, { "value": "GPT-5.5", "source": "https://x.com/dhh/status/2057907663967543618", "seen": "2026-05-22", "quote": "The Omarchy 4 branch is now 30,000 lines of new code. The majority of it was written by GPT5.5." }, { "value": "GPT 5.6 Sol", "source": "https://x.com/dhh/status/2078252859834167656", "seen": "2026-07-17", "quote": "But GPT 5.6 Sol? 4m45s and it was done." } ] } ], "review_style": { "value": "A second model reviews, then he reads the diff", "source": "https://github.com/omacom/omarchy/pull/13770", "seen": "2026-09-29", "quote": "Reviewed with `codex review --base quattro` over eight passes; every finding was fixed, most with a regression test.", "kind": "spot-checks", "also": [ { "source": "https://x.com/dhh/status/2107882896195199116", "seen": "2026-10-07", "quote": "Here's my technique for adversarial code review if I'm driving from Codex: \"Review this with claude\"." }, { "source": "https://github.com/omacom/omarchy/pull/14253", "seen": "2026-10-04", "quote": "Reviewed with Claude and addressed its LocalSend class mismatch, redundant catch-all rule findings, and the transparency shortcut behavior." }, { "source": "https://x.com/dhh/status/2057907663967543618", "seen": "2026-05-22", "quote": "You still need to review, but there's just no way this scale of a conversion would be feasible without AI in a reasonable time." }, { "source": "https://newsletter.pragmaticengineer.com/p/dhhs-new-way-of-writing-code", "seen": "2026-04-08", "quote": "NeoVim for reviewing diffs via Lazygit" } ] }, "harness": { "value": "An always-on \"AI shed\" on his Tailscale network runs his herdr agents; lazygit plus an agent for most work", "short": "AI shed", "source": "https://x.com/dhh/status/2107463006552793231", "seen": "2026-10-06", "quote": "all my agent and compile work now happens on an AI shed.", "history": [ { "value": "AI in a separate window", "source": "https://lexfridman.com/dhh-david-heinemeier-hansson-transcript/", "seen": "2025-07-12", "quote": "I love chiseling my code, and the way I use AI is in a separate window. I don’t let it drive my code." }, { "value": "tmux: Neovim plus one or two agent panes (tdl), herdr too", "source": "https://github.com/omacom/omarchy/blob/81145eb1fd41532fcae4310b26f860608e2eaf82/default/bash/fns/tmux#L1-L2", "seen": "2026-05-11", "quote": "# Create a Tmux Dev Layout with editor, ai, and terminal\n# Usage: tdl []" } ] }, "parallelism": { "value": 2, "short": "2 at once", "source": "https://www.youtube.com/watch?v=JiWgKRgdgpI&t=2821s", "seen": "2026-04-08", "quote": "one of the things I've incorporated into my flow is to kind of have two models going at the same time at different speeds.", "history": [] }, "rules_files": { "value": "Omarchy AGENTS.md, with CLAUDE.md importing it", "short": "AGENTS.md", "url": "https://github.com/omacom/omarchy/blob/81145eb1fd41532fcae4310b26f860608e2eaf82/AGENTS.md", "source": "https://github.com/omacom/omarchy/blob/81145eb1fd41532fcae4310b26f860608e2eaf82/CLAUDE.md#L1", "seen": "2026-08-07", "quote": "@AGENTS.md" }, "skills": { "value": [ "omarchy", "omarchy-app", "diagnose-crash" ], "count": 3, "featured": [ "omarchy-app", "omarchy", "diagnose-crash" ], "source": "https://github.com/omacom/omarchy/tree/81145eb1fd41532fcae4310b26f860608e2eaf82/default/agents/skills", "seen": "2026-10-01" }, "mcp_servers": null, "hooks": null, "signature": { "value": "\"Review this with claude\"", "kind": "habit", "caption": "adversarial review from Codex", "why": "Our pick: he drives from Codex and has Claude review the work, one agent checking the other.", "editorial": true, "source": "https://x.com/dhh/status/2107882896195199116", "seen": "2026-10-07", "quote": "Here's my technique for adversarial code review if I'm driving from Codex: \"Review this with claude\"." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Builds his agent setup into his own OS: tdl, account switching, an agents panel." }, "signature_quote": { "value": "We're not going to write the vast majority of code by hand any longer.", "more": "Coding agents have gotten so good that it's simply not an economically viable enterprise going forward to have humans typing out lines of Ruby, Rust, or C++.", "source": "https://world.hey.com/dhh/over-my-dead-pencil-fb0f3647", "seen": "2026-10-06" } }, { "schema": 2, "handle": "mitsuhiko", "status": "ok", "checked": "2026-10-12", "name": { "value": "Armin Ronacher", "source": "https://github.com/mitsuhiko" }, "x": { "value": "mitsuhiko", "source": "https://github.com/mitsuhiko" }, "github": { "value": "mitsuhiko", "source": "https://github.com/mitsuhiko" }, "site": { "value": "https://lucumr.pocoo.org", "source": "https://github.com/mitsuhiko" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/7396?v=4", "source": "https://github.com/mitsuhiko" }, "known_for": { "value": "Flask, Earendil founder", "source": "https://github.com/mitsuhiko" }, "agents": [ { "name": "Pi", "icon": "pi", "role": "primary", "surface": "cli", "surface_source": { "source": "https://lucumr.pocoo.org/2026/1/31/pi/", "seen": "2026-01-31", "quote": "Beyond slash commands, Pi extensions can render custom TUI components directly in the terminal: spinners, progress bars, interactive file pickers, data tables, preview panes." }, "source": "https://lucumr.pocoo.org/2026/1/31/pi/", "seen": "2026-01-31", "quote": "And Pi happens to be, at this point, the coding agent that I use almost exclusively.", "history": [ { "value": "Claude Code", "surface": "cli", "surface_source": { "source": "https://lucumr.pocoo.org/2025/6/12/agentic-coding/", "seen": "2025-06-12", "quote": "Which basically means I run claude --dangerously-skip-permissions." }, "source": "https://lucumr.pocoo.org/2025/6/12/agentic-coding/", "seen": "2025-06-12", "quote": "I predominently use Claude Code with the cheaper Max subscription for $100 a month" }, { "value": "Amp, Claude Code and Pi", "surface": "cli", "source": "https://lucumr.pocoo.org/2025/12/22/a-year-of-vibes/", "seen": "2025-12-22", "quote": "At the moment I’m using Amp, Claude Code, and Pi, all from the command line." } ] }, { "name": "Codex", "icon": "codex", "role": "testing Astra", "surface": null, "source": "https://lucumr.pocoo.org/2026/9/7/astra-why/", "seen": "2026-09-07", "quote": "In the Codex harness I found multiple cases where subagents resorted fully to manual string manipulation with Python instead of using the patch tool.", "history": [ { "value": "Codex CLI for reviews", "surface": "cli", "source": "https://lucumr.pocoo.org/2025/12/22/a-year-of-vibes/", "seen": "2025-12-22", "quote": "So I now use Codex on the command line to do reviews, but that means a whole part of my iteration cycles is invisible to other engineers on the team." } ] } ], "models": [ { "name": "Sol", "icon": "openai", "source": "https://lucumr.pocoo.org/2026/9/5/latent-powers/", "seen": "2026-09-05", "quote": "Long story short: I had a conversation with Fable and Sol via Pi about what could be done with such a dongle or whether I should use a Raspberry Pi instead if I wanted to do my own thing there.", "history": [ { "value": "Sonnet only", "source": "https://lucumr.pocoo.org/2025/6/12/agentic-coding/", "seen": "2025-06-12", "quote": "I exclusively use the cheaper Sonnet model. It’s perfectly adequate for my needs, and in fact, I prefer its outputs over the more expensive Opus model." } ], "inferred": true }, { "name": "Fable", "icon": "claude", "source": "https://lucumr.pocoo.org/2026/9/5/latent-powers/", "seen": "2026-09-05", "quote": "Long story short: I had a conversation with Fable and Sol via Pi about what could be done with such a dongle or whether I should use a Raspberry Pi instead if I wanted to do my own thing there.", "history": [], "inferred": true } ], "review_style": { "value": "Steers as it goes, reviews when the agent says done", "source": "https://lucumr.pocoo.org/2026/6/23/the-coming-loop/", "seen": "2026-06-23", "quote": "In the agent loop, the model eventually says “done” and I review. Even before that, I usually steer along the way.", "kind": "spot-checks", "also": [ { "source": "https://lucumr.pocoo.org/2026/1/31/pi/", "seen": "2026-01-31", "quote": "As more code is written by agents, it makes little sense to throw unfinished work at humans before an agent has reviewed it first." }, { "source": "https://lucumr.pocoo.org/2026/9/7/astra-why/", "seen": "2026-09-07", "quote": "It has shown that it will commit slop, and it requires me to review it more as a result." } ], "inferred": true }, "harness": { "value": "Pi in the terminal, forks in Ghostty splits, subagents in tmux", "short": "Ghostty + tmux", "source": "https://github.com/mitsuhiko/agent-stuff/blob/0865c849befd2021490679f96a8dee58c84ac857/README.md#L59-L60", "seen": "2026-07-26", "quote": "`/split-fork` to branch the current session into a new Pi process in a right-hand Ghostty split.\nSerial `subagent` tool that runs one observable Pi child at a time in tmux; attach with the printed `pi --attach-subagent …` command.", "history": [] }, "parallelism": null, "rules_files": { "value": "AGENTS.md in his agent-stuff repo (12 lines)", "short": "AGENTS.md", "url": "https://github.com/mitsuhiko/agent-stuff/blob/0865c849befd2021490679f96a8dee58c84ac857/AGENTS.md", "source": "https://github.com/mitsuhiko/agent-stuff/blob/0865c849befd2021490679f96a8dee58c84ac857/AGENTS.md#L12", "seen": "2026-01-30", "quote": "Pi extensions live in `./extensions`. When working in this repo, add or update extensions there. You can consult the `pi-mono` for reference, but do not modify code in `pi-mono`." }, "skills": { "value": [ "anachb", "apple-mail", "audio-transcription", "commit", "ghidra", "github", "google-workspace", "librarian", "native-web-search", "oebb-scotty", "openscad", "pi-share", "sentry", "summarize", "tmux", "update-changelog", "uv", "web-browser" ], "count": 18, "featured": [ "web-browser", "pi-share", "librarian" ], "source": "https://github.com/mitsuhiko/agent-stuff/tree/0865c849befd2021490679f96a8dee58c84ac857/skills", "seen": "2026-09-27" }, "mcp_servers": { "value": [ "Sentry" ], "source": "https://lucumr.pocoo.org/2026/10/6/codemode/", "seen": "2026-10-06", "quote": "Here for instance you can see the agent reach for the Sentry MCP straight away", "inferred": true }, "hooks": { "value": [ "uv.ts: blocks pip and python, pushes the agent to uv" ], "source": "https://github.com/mitsuhiko/agent-stuff/blob/0865c849befd2021490679f96a8dee58c84ac857/README.md#L63", "seen": "2026-07-26", "quote": "Replaces the bash tool with a `uv`-aware version that injects Python command shims and blocks common non-`uv` workflows." }, "signature": { "value": "/answer", "kind": "tool", "caption": "in place of plan mode", "why": "Our pick: he skips plan mode and lets the agent ask questions in plain prose, then a self-built extension turns those questions into a form he fills in.", "editorial": true, "source": "https://lucumr.pocoo.org/2026/1/31/pi/", "seen": "2026-01-31", "quote": "The problem: answering questions inline gets messy. So /answer reads the agent’s last response, extracts all the questions, and reformats them into a nice input box." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Has the agent write his own Pi extensions and 18 skills. No MCP, nothing downloaded." }, "signature_quote": { "value": "In the agent loop, the model eventually says “done” and I review.", "more": "Even before that, I usually steer along the way.", "source": "https://lucumr.pocoo.org/2026/6/23/the-coming-loop/", "seen": "2026-06-23" } }, { "schema": 2, "handle": "t3dotgg", "status": "ok", "checked": "2026-10-12", "name": { "value": "Theo Browne", "source": "https://github.com/t3dotgg" }, "x": { "value": "theo", "source": "https://x.com/theo" }, "github": { "value": "t3dotgg", "source": "https://github.com/t3dotgg" }, "site": { "value": "https://t3.gg", "source": "https://github.com/t3dotgg" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/6751787?v=4", "source": "https://github.com/t3dotgg" }, "known_for": { "value": "T3 Stack, T3 Chat, T3 Code", "source": "https://github.com/t3dotgg" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary, inside T3 Code", "surface": "app", "surface_source": { "source": "https://x.com/theo/status/2103794887481319566", "seen": "2026-09-26", "quote": "Here's my last seven days of usage in T3 Code. Nuts that you can get all this with like two Claude subs and one Codex sub", "inferred": true }, "history": [], "source": "https://x.com/theo/status/2104079585956761814", "seen": "2026-09-27", "quote": "With much, much effort, I have been able to kill ~3.5 $200 Claude Code accounts in the past 5 days." }, { "name": "Codex", "icon": "codex", "role": "reviews and hard problems", "surface": "app", "surface_source": { "source": "https://x.com/theo/status/2095751225283350624", "seen": "2026-09-04", "quote": "2. Open Codex app" }, "history": [], "source": "https://x.com/theo/status/2095609789711831286", "seen": "2026-09-03", "quote": "Astra had made Codex more essential than ever due to its unique strengths." } ], "models": [ { "name": "Opus 5.5", "icon": "claude", "detail": "high", "history": [ { "source": "https://x.com/theo/status/2100548714016468993", "seen": "2026-09-17", "quote": "Seriously, just use fable if you want to merge the code.", "value": "Fable 5.1" } ], "source": "https://x.com/theo/status/2104339496942948401", "seen": "2026-09-27", "quote": "Fable 5.1 xhigh to Opus 5.5 high (move I made) is a 6.6x increase" }, { "name": "Muse Wonder", "icon": null, "detail": "reviews", "source": "https://x.com/theo/status/2109731846842196093", "seen": "2026-10-12", "quote": "I often have Opus ask Wonder for code review, copy work, and architecture review", "history": [ { "source": "https://x.com/theo/status/2103239969997476091", "seen": "2026-09-24", "quote": "Astra is still the best \"review\" model. It's so insanely thorough.", "value": "GPT-6 Astra for reviews" }, { "value": "GPT-6.1 Sol", "source": "https://x.com/theo/status/2105001138953327067", "seen": "2026-09-29", "quote": "GPT-6.1 Sol is a great replacement for Astra at a WAY better price, but I don't default to it for code.\n\nI absolutely default to it for code reviews" } ] } ], "review_style": { "value": "Opus asks Wonder for code review; agents merge when confident", "source": "https://x.com/theo/status/2109731846842196093", "seen": "2026-10-12", "quote": "I often have Opus ask Wonder for code review, copy work, and architecture review", "kind": "agent-reviews", "also": [ { "source": "https://x.com/theo/status/2105001138953327067", "seen": "2026-09-29", "quote": "I absolutely default to it for code reviews, architecture analysis, computer use, email management, and a ton of my other daily work." }, { "source": "https://x.com/theo/status/2095691652442837310", "seen": "2026-09-04", "quote": "Biggest jump with Fable 5.1 and GPT-6 Astra is that I finally have the confidence to say \"you can merge when you're confident in the changes made\"" }, { "source": "https://x.com/theo/status/2103245685688771044", "seen": "2026-09-24", "quote": "The PRs I make with AI are significantly better than the ones I used to make by hand." }, { "source": "https://github.com/pingdotgg/t3code/blob/9bd1d8009a6b7c50f9dd9458e2bf27d481ff3b43/AGENTS.md#L115", "seen": "2026-07-28", "quote": "When babysitting: poll checks and comments newer than the last push, verify each bot finding against the source, fix real ones, dismiss false positives with a written reason." } ] }, "harness": { "value": "T3 Code on a spare Ubuntu box over ssh, not macOS", "short": "remote Linux", "source": "https://x.com/theo/status/2103710437951131865", "seen": "2026-09-26", "quote": "2. Install Ubuntu on it\n3. ssh into it\n4. \"npx t3 connect\"\n5. Run all your work remotely", "history": [] }, "parallelism": { "value": 40, "short": "40+ threads", "source": "https://x.com/theo/status/2102704338582032575", "seen": "2026-09-23", "quote": "I regularly have 40+ threads running concurrently.", "history": [] }, "rules_files": { "value": "AGENTS.md with \"A note from Theo\"; CLAUDE.md just imports it", "short": "AGENTS.md", "url": "https://github.com/pingdotgg/t3code/blob/9bd1d8009a6b7c50f9dd9458e2bf27d481ff3b43/AGENTS.md", "source": "https://github.com/pingdotgg/t3code/blob/9bd1d8009a6b7c50f9dd9458e2bf27d481ff3b43/AGENTS.md#L33-L35", "seen": "2026-10-06", "quote": "I like ambitious ideas, simple systems, and software that feels obvious. Do not preserve complexity just because it already exists." }, "skills": { "value": [ "contribution-triage", "test-t3-app", "test-t3-mobile" ], "count": 3, "featured": [ "test-t3-app" ], "source": "https://github.com/pingdotgg/t3code/tree/9bd1d8009a6b7c50f9dd9458e2bf27d481ff3b43/.agents/skills", "seen": "2026-10-02" }, "mcp_servers": null, "hooks": { "value": [ "UserPromptSubmit, PreToolUse, Stop -> Claude Blocker" ], "source": "https://github.com/t3-content/claude-blocker/blob/06f03fa2c229dde4ab4c02c5c24955a42a504d09/README.md#L15-L21", "seen": "2026-01-01", "quote": "1. **Claude Code hooks** notify the server when you submit a prompt or when Claude finishes" }, "signature": { "value": "T3 Code", "kind": "tool", "caption": "his own GUI over every agent", "why": "He built the app he works in: one window over Claude Code, Codex and the rest, and the repo expects most contributions to come from T3 Code itself.", "editorial": true, "source": "https://github.com/pingdotgg/t3code/blob/9bd1d8009a6b7c50f9dd9458e2bf27d481ff3b43/AGENTS.md#L41", "seen": "2026-10-06", "quote": "Of note: Most T3 Code contributions will come from T3 Code itself, often controlled remotely." }, "archetype": { "value": "Swarm Lord", "editorial": true, "why": "40+ threads at once across several Claude and Codex subs, all in T3 Code." }, "signature_quote": { "value": "I regularly have 40+ threads running concurrently.", "more": "I've never had a reason to look at more than one at a time.", "source": "https://x.com/theo/status/2102704338582032575", "seen": "2026-09-23" } }, { "schema": 2, "handle": "antirez", "status": "ok", "checked": "2026-10-12", "name": { "value": "Salvatore Sanfilippo (antirez)", "source": "https://github.com/antirez" }, "x": { "value": "antirez", "source": "https://github.com/antirez" }, "github": { "value": "antirez", "source": "https://github.com/antirez" }, "site": { "value": "https://antirez.com", "source": "https://antirez.com" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/65632?v=4", "source": "https://github.com/antirez" }, "known_for": { "value": "Redis creator, DwarfStar (ds4)", "source": "https://en.wikipedia.org/wiki/Salvatore_Sanfilippo" }, "agents": [ { "name": "Codex", "icon": "codex", "role": "primary", "surface": "cli", "source": "https://antirez.com/news/164", "seen": "2026-05-04", "quote": "I started writing for days a long specification by hand, then I paired with Opus initially, then GPT 5.3 was released and I switched all the design and development with Codex. Since then I use only GPT 5.x for system programming tasks.", "history": [ { "value": "No agents, chat only", "surface": "web", "source": "https://antirez.com/news/154", "seen": "2025-07-20", "quote": "The fundamental requirement for the LLM to be used is: don’t use agents or things like editor with integrated coding agents.", "surface_source": { "source": "https://antirez.com/news/154", "seen": "2025-07-20", "quote": "Always be part of the loop by moving code by hand from your terminal to the LLM web interface: this guarantees that you follow every process." } }, { "value": "Web chat (Gemini, Claude)", "surface": "web", "source": "https://antirez.com/news/157", "seen": "2025-12-20", "quote": "The programming world is still split among who uses LLMs as colleagues (for instance, all my interaction is via the web interface of Gemini, Claude, …), and who uses LLMs as independent coding agents." } ], "surface_source": { "source": "https://github.com/antirez/tgterm/blob/51433ee538d740b013b117e4e5cbcebbddf86bc2/README.md#L5", "seen": "2026-09-07", "quote": "the content of your terminals arrives as messages: you can read them at any size, copy from them, and have them updated while a program runs. To type, you just write a Telegram message, that is what you need to work with Claude Code, Codex and other agents." } }, { "name": "Claude Code", "icon": "claudecode", "role": "experiments, Max plan", "surface": "cli", "source": "https://antirez.com/news/160", "seen": "2026-02-24", "quote": "I thought it was time to try a similar experiment myself, one that would take one or two hours at max, and that was compatible with my Claude Code Max plan: I decided to write a Z80 emulator, and then a ZX Spectrum emulator (and even more, a CP/M emulator, see later) in a condition that I believe makes a more sense as “clean room” setup.", "history": [], "surface_source": { "source": "https://github.com/antirez/tgterm/blob/51433ee538d740b013b117e4e5cbcebbddf86bc2/README.md#L5", "seen": "2026-09-07", "quote": "the content of your terminals arrives as messages: you can read them at any size, copy from them, and have them updated while a program runs. To type, you just write a Telegram message, that is what you need to work with Claude Code, Codex and other agents." } }, { "name": "ds4-agent", "icon": null, "role": "his own, local", "surface": "cli", "source": "https://antirez.com/news/166", "seen": "2026-05-19", "quote": "I have the feeling I can decide which is better only with enough practical evidence, by using ds4-agent over multiple sessions with the two systems.", "history": [], "surface_source": { "source": "https://github.com/antirez/ds4/blob/0aaea5a238fb41a35106a551e73c8409dfb751ac/QA_BEFORE_RELEASES.md#L7", "seen": "2026-09-20", "quote": "agent TUI/tool state machine." } } ], "models": [ { "name": "GPT-6 Astra / GPT-6.1 Sol", "icon": "openai", "detail": "at Redis", "source": "https://x.com/antirez/status/2106312728629387347", "seen": "2026-10-03", "quote": "Having serious issues to work at Redis with Astra / Sol 6.1 because of continuous firing of cyber checks.", "history": [ { "value": "Gemini 2.5 PRO (in chat)", "source": "https://antirez.com/news/154", "seen": "2025-07-20", "quote": "If you can pick just one, go for Gemini 2.5 PRO." }, { "value": "GPT5.3-codex", "source": "https://antirez.com/news/160", "seen": "2026-02-24", "quote": "With extensive documentation, I can’t see any way Claude Code (and, even more, GPT5.3-codex, which is in my experience, for complex stuff, more capable) could fail at producing a working assembler, since it is quite a mechanical process." }, { "value": "GPT 5.6 Sol", "source": "https://antirez.com/news/170", "seen": "2026-07-22", "quote": "Moreover, today I noticed how, thanks to the rails formed by the corpus of the code inside DwarfStar, the implementation was written in about two hours by GPT 5.6 Sol automatically." } ] }, { "name": "Fable", "icon": "claude", "detail": "reviews", "source": "https://antirez.com/news/169", "seen": "2026-07-13", "quote": "Fable and GPT 5.6 reviews to the sorted sets memory saving are going to spot ways more errors and subtle race conditions that my review is going to uncover.", "history": [] } ], "review_style": { "value": "Still reviews Redis code, calls it mostly pointless", "source": "https://antirez.com/news/169", "seen": "2026-07-13", "quote": "Yes, I do, but this is, at this point, something I *need* to do but that I believe to be mostly pointless, partially once GPT 5.5 was released, but now with Fable and GPT 5.6 Sol even more.", "kind": "spot-checks", "also": [ { "source": "https://antirez.com/news/169", "seen": "2026-07-13", "quote": "Fable and GPT 5.6 reviews to the sorted sets memory saving are going to spot ways more errors and subtle race conditions that my review is going to uncover." }, { "source": "https://antirez.com/news/168", "seen": "2026-06-07", "quote": "The idea is to create a markdown file where an AI agent is asked to work as a QA engineer, performing a number of manual testings on the new release." } ], "inferred": true }, "harness": { "value": "tmux / Terminal.app, driven from Telegram via tgterm", "short": "tmux + Telegram", "source": "https://github.com/antirez/tgterm/blob/51433ee538d740b013b117e4e5cbcebbddf86bc2/README.md#L9", "seen": "2026-09-07", "quote": "To type, you just write a Telegram message, that is what you need to work with Claude Code, Codex and other agents.", "history": [] }, "parallelism": null, "rules_files": { "value": "AGENT.md per repo, CLAUDE.md symlinked to it", "short": "AGENT.md", "url": "https://github.com/antirez/ds4/blob/297f750cda347d3a7b6012564ec8538fc4712a34/AGENT.md", "source": "https://github.com/antirez/ds4/blob/297f750cda347d3a7b6012564ec8538fc4712a34/AGENT.md#L18", "seen": "2026-06-05", "quote": "Don't introduce slop: very fragile code that just patches specific cases, dead code, useless code and code ways more complicated of how it should be." }, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "QA_BEFORE_RELEASES.md", "kind": "habit", "caption": "an agent QA pass per release", "why": "Our pick: a 2,088-line release gate an agent works through, which he credits with raising quality.", "editorial": true, "source": "https://antirez.com/news/168", "seen": "2026-06-07", "quote": "The idea is to create a markdown file where an AI agent is asked to work as a QA engineer, performing a number of manual testings on the new release." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Builds his own agent, edit tool and Telegram terminal bridge." }, "signature_quote": { "value": "The point is: if you control the ideas of your software, looking at the code itself is suboptimal and often pointless.", "source": "https://antirez.com/news/169", "seen": "2026-07-13" } }, { "schema": 2, "handle": "theprimeagen", "status": "ok", "checked": "2026-10-12", "name": { "value": "ThePrimeagen", "source": "https://github.com/ThePrimeagen" }, "x": { "value": "ThePrimeagen", "source": "https://x.com/ThePrimeagen" }, "github": { "value": "ThePrimeagen", "source": "https://github.com/ThePrimeagen" }, "site": { "value": "https://twitch.tv/ThePrimeagen", "source": "https://github.com/ThePrimeagen" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/4458174?v=4", "source": "https://github.com/ThePrimeagen" }, "known_for": { "value": "Harpoon, Vim on stream", "source": "https://github.com/ThePrimeagen" }, "agents": [ { "name": "Cursor", "icon": "cursor", "role": "primary: CLI in Neovim, cloud agents", "surface": "ide", "source": "https://github.com/ThePrimeagen/init.lua/blob/b2417cdcc842af1ab06f84b8ad142552e738e648/lua/theprimeagen/lazy/local.lua#L57", "seen": "2026-09-02", "quote": "provider = _99.Providers.CursorAgentProvider,", "history": [ { "value": "Vibe coding with Claude: unimpressed", "surface": null, "source": "https://x.com/ThePrimeagen/status/2008261459630059720", "seen": "2026-01-05", "quote": "day 1 vibe coding: i am not impressed, super unimpressed" }, { "value": "Back to autocomplete over agents", "surface": "ide", "source": "https://x.com/ThePrimeagen/status/2032100265403256899", "seen": "2026-03-12", "quote": "I think as a group (swe) we rushed so fast into Agents when inline autocomplete + actual skills is crazy." } ] }, { "name": "99", "icon": "neovim", "role": "his own Neovim agent client", "surface": "ide", "source": "https://github.com/ThePrimeagen/99/blob/c17422457027c913c76c75a921fca1e623d2678e/README.md#L1-L2", "seen": "2026-02-28", "quote": "The AI client that Neovim deserves, built by those that still enjoy to code.", "history": [] } ], "models": [ { "name": "Grok 4.6", "icon": "grok", "detail": "xhigh, fast, via Cursor", "source": "https://github.com/ThePrimeagen/init.lua/blob/b2417cdcc842af1ab06f84b8ad142552e738e648/lua/theprimeagen/lazy/local.lua#L68", "seen": "2026-09-02", "quote": "model = \"cursor-grok-4.6-xhigh-fast\",", "history": [ { "value": "gpt-5.3-codex", "source": "https://github.com/ThePrimeagen/init.lua/blob/72a1a7ecd96bb44b8958b26641da674d569b87bc/lua/theprimeagen/lazy/local.lua#L28", "seen": "2026-02-14", "quote": "model = \"openai/gpt-5.3-codex\"," }, { "value": "composer-2.5-fast", "source": "https://github.com/ThePrimeagen/init.lua/blob/85cb6c34b06c3aa92475a6a839ca61f75e76254d/lua/theprimeagen/lazy/local.lua#L29", "seen": "2026-06-10", "quote": "model = \"composer-2.5-fast\"," } ] } ], "review_style": null, "harness": { "value": "A terminal agent plus review for much of his work; Neovim with his 99 plugin", "short": "terminal + 99", "source": "https://x.com/ThePrimeagen/status/2107831845148582121", "seen": "2026-10-07", "quote": "A huge portion of the work I do lately is just from the terminal via agent + some review.", "history": [ { "value": "Neovim with his 99 plugin, driving Cursor's agent", "source": "https://github.com/ThePrimeagen/99/blob/c17422457027c913c76c75a921fca1e623d2678e/README.md#L28-L31", "seen": "2026-02-28", "quote": "As of now, the direction of 99 is to progress into agentic programming and surfacing\nof information." } ] }, "parallelism": null, "rules_files": null, "skills": { "value": [ "better-auth-drizzle-init", "better-auth-social-login", "cloudflare", "cloudflare-assets", "cloudflare-hyperdrive", "cloudflare-project", "cloudflare-r2", "drizzle-migration", "frontend", "hyperdrive-drizzle", "oauth", "planet-scale-docs", "prime-planning", "prime-review", "tmux-dev", "vim", "vim.diagnostics", "vim.lsp", "vim.treesitter" ], "count": 19, "featured": [ "prime-planning", "prime-review", "tmux-dev" ], "source": "https://github.com/ThePrimeagen/skills/tree/fa570dd76421dc9b608e2a8134d4457206f3f953/skills", "seen": "2026-02-27" }, "mcp_servers": null, "hooks": null, "signature": { "value": "Field guide", "kind": "habit", "caption": "agent writes its own lessons", "why": "Our pick: every session is logged with commit hashes, then the agent turns it into a project guide it must read before touching code.", "editorial": true, "source": "https://x.com/ThePrimeagen/status/2081066227619836308", "seen": "2026-07-25", "quote": "Then I have the agent review the back and forth and the changes and come up with durable and generalized feedback that it stores in a directory (field guide)." }, "archetype": { "value": "Rules Lawyer", "editorial": true, "why": "Has the agent write a field guide of lessons and read it before touching code, and nitpicks until it is the exact code he wants." }, "signature_quote": { "value": "I nitpick everything and I always make it produce the exact code I want, I use voice dictation and talk for quite a bit.", "more": null, "source": "https://x.com/ThePrimeagen/status/2081066227619836308", "seen": "2026-07-25" } }, { "schema": 2, "handle": "steveyegge", "status": "ok", "checked": "2026-10-12", "name": { "value": "Steve Yegge", "source": "https://github.com/steveyegge" }, "x": { "value": "Steve_Yegge", "source": "https://x.com/Steve_Yegge" }, "github": { "value": "steveyegge", "source": "https://github.com/steveyegge" }, "site": { "value": "https://yegge.ai", "source": "https://x.com/Steve_Yegge" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/613744?v=4", "source": "https://github.com/steveyegge" }, "known_for": { "value": "Gas Town, Beads, Vibe Coding", "source": "https://yegge.ai/" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary, crew and fleet", "surface": "cli", "source": "https://yegge.ai/essays/the-shape-of-things-to-come/", "seen": "2026-08-02", "quote": "I have a GCP VM for Claude Code, and also a Mac Mini that bridges dev and prod, sort of a \"corp network.\"", "history": [], "surface_source": { "source": "https://yegge.ai/essays/the-shape-of-things-to-come/", "seen": "2026-08-02", "quote": "By consolidating all my terminal windows into a single rolodex that I can flip through, I freed up my desktop for real work." } }, { "name": "Codex", "icon": "codex", "role": "fallback fleet", "surface": null, "source": "https://yegge.ai/essays/the-shape-of-things-to-come/", "seen": "2026-08-02", "quote": "I also have a small parallel fleet of five named Sol 5.6 workers on Codex, all named for sun gods. I use it as a fallback for when my Max accounts run out, which is increasingly often, despite me adding new ones once or twice a week.", "history": [] } ], "models": [ { "name": "Claude Opus 5.5", "icon": "claude", "detail": "majority of work", "source": "https://x.com/Steve_Yegge/status/2106205282926051596", "seen": "2026-10-03", "quote": "I still think Opus 5.5 is an incredible model and I use it for the majority of work now.", "history": [ { "value": "Opus 5", "source": "https://yegge.ai/essays/the-shape-of-things-to-come/", "seen": "2026-08-02", "quote": "My fleet, all Opus 5 agents, are the work consumers, like Gas Town's polecats, but non-ephemeral." } ] }, { "name": "Claude Fable 5.1", "icon": "claude", "detail": "design, planning, review", "source": "https://yegge.ai/essays/seats-and-sunsets/", "seen": "2026-09-15", "quote": "I first switched most of my seats to Fable 5.1 (my headless Aesop reviewer pool is still Fable 5)", "history": [ { "value": "Opus 4.5", "source": "https://steve-yegge.medium.com/welcome-to-gas-town-4f25ee16dd04", "seen": "2026-01-01", "quote": "Opus 4.5 can handle any reasonably sized task, so your job is to make tasks for it." } ] } ], "review_style": { "value": "Fable reviews every Opus change; he doesn't read code", "source": "https://yegge.ai/essays/the-shape-of-things-to-come/", "seen": "2026-08-02", "quote": "Every implementation bead goes through this lifecycle: Fable design, Opus implementation, Fable review.", "kind": "agent-reviews", "also": [ { "source": "https://yegge.ai/essays/the-shape-of-things-to-come/", "seen": "2026-08-02", "quote": "And as another appetizer, human code review has very nearly run its course." }, { "source": "https://steve-yegge.medium.com/welcome-to-gas-town-4f25ee16dd04", "seen": "2026-01-01", "quote": "I’ve never seen the code, and I never care to, which might give you pause." } ] }, "harness": { "value": "Wheelhouse, all in Emacs, tmux underneath, Beads on Dolt", "short": "Emacs + tmux", "source": "https://yegge.ai/essays/the-shape-of-things-to-come/", "seen": "2026-08-02", "quote": "I began the journey inside Emacs, so Wheelhouse has been all-Emacs from the ground up.", "history": [ { "value": "Gas Town in tmux", "source": "https://steve-yegge.medium.com/welcome-to-gas-town-4f25ee16dd04", "seen": "2026-01-01", "quote": "Gas Town uses tmux as its primary UI." } ] }, "parallelism": { "value": 20, "short": "20+ agents", "source": "https://x.com/Steve_Yegge/status/2108393181196722238", "seen": "2026-10-09", "quote": "I run Emacs with 20+ agents and it drives terminals about as hard as you can", "history": [ { "value": "20–30 Claude Code instances", "source": "https://steve-yegge.medium.com/welcome-to-gas-town-4f25ee16dd04", "seen": "2026-01-01", "quote": "I’ve tamed them to where you can use 20–30 at once, productively, on a sustained basis." }, { "value": "50 agents", "source": "https://yegge.ai/essays/fences-not-sandboxes/", "seen": "2026-08-24", "quote": "Wheelhouse has 50 agents that are amnesiac and interchangeable, and the only way they can coordinate is via text." }, { "value": null, "source": "https://yegge.ai/essays/seats-and-sunsets/", "seen": "2026-09-15", "quote": "But with Fable's latest \"25%\" pricing, which is yet another cut, my Wheelhouse factory is now mostly idling." } ] }, "rules_files": { "value": "AGENTS.md (CLAUDE.md points to it), beads for all tasks", "short": "AGENTS.md", "url": "https://github.com/steveyegge/efrit/blob/4df8dfd59d42e8290e683e3216a448b49d031ee2/AGENTS.md", "source": "https://github.com/steveyegge/efrit/blob/4df8dfd59d42e8290e683e3216a448b49d031ee2/AGENTS.md#L30", "seen": "2026-06-11", "quote": "Do NOT batch multiple fixes. Do NOT wait until \"end of session\". Push immediately after each completed task." }, "skills": null, "mcp_servers": { "value": [ "none" ], "source": "https://yegge.ai/essays/the-shape-of-things-to-come/", "seen": "2026-08-02", "quote": "I run pretty lean. I don't use any sandboxing. I don't use MCP." }, "hooks": null, "signature": { "value": "Seats", "kind": "habit", "caption": "18 long-lived Fable officers", "why": "Our pick: every agent boots into a named office with its own memory and authority, which he says saves tokens.", "editorial": true, "source": "https://yegge.ai/essays/seats-and-sunsets/", "seen": "2026-09-15", "quote": "Seats are your most fundamental and powerful building block for agentic workflows, whether you're building orchestrators or not." }, "archetype": { "value": "Swarm Lord", "editorial": true, "why": "50 agents: a Fable crew designs and reviews, Opus builds." }, "signature_quote": { "value": "human code review has very nearly run its course.", "more": "Its vestigial SOC 2 compliance angle will keep it on life support, but by next year, human code review is completely done and gone.", "source": "https://yegge.ai/essays/the-shape-of-things-to-come/", "seen": "2026-08-02" } }, { "schema": 2, "handle": "levelsio", "status": "ok", "checked": "2026-10-12", "name": { "value": "Pieter Levels", "source": "https://levels.io/" }, "x": { "value": "levelsio", "source": "https://x.com/levelsio" }, "github": { "value": "levelsio", "source": "https://github.com/levelsio" }, "site": { "value": "https://levels.io", "source": "https://github.com/levelsio" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/7150848?v=4", "source": "https://github.com/levelsio" }, "known_for": { "value": "Photo AI, Nomads.com", "source": "https://x.com/levelsio" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary, on the server", "surface": "cli", "history": [ { "source": "https://x.com/levelsio/status/1893385114496766155", "seen": "2025-02-22", "quote": "I've never ever made a game before and just made my own flight simulator 100% with Cursor in I'd say 3 hours by just telling it what I wanted", "value": "Cursor", "surface": "ide" } ], "source": "https://x.com/levelsio/status/2094113698453360663", "seen": "2026-08-30", "quote": "I made https://infiniteslop.ai yesterday completely on my phone with @TermiusHQ on my @Hetzner_Online VPS with @claudeai Code in the sauna" } ], "models": [ { "name": "Claude Opus 5.5", "icon": "claude", "detail": "1M context", "source": "https://x.com/levelsio/status/2108957488410161442", "seen": "2026-10-10", "quote": "only possible because modern AI like Opus 5.5 is good enough to figure that out", "history": [ { "value": "Claude Opus 4.6", "source": "https://github.com/levelsio/superlevels/commit/cef89e725c90cc9eb6f095d606ed24ada13108ff", "seen": "2026-04-23", "quote": "Co-Authored-By: Claude Opus 4.6 ", "inferred": true }, { "value": "Claude Opus 4.8 (1M context)", "source": "https://github.com/levelsio/superlevels/commit/205c4263d941a33324b2a9c6f46fb1f232fca8a9", "seen": "2026-08-25", "quote": "Co-Authored-By: Claude Opus 4.8 (1M context) ", "inferred": true }, { "value": "Claude Opus 5.5", "source": "https://github.com/levelsio/superlevels/commit/0fe785f75fbe681ed60efd74e9586d362e4c6995", "seen": "2026-10-02", "quote": "Co-Authored-By: Claude Opus 5.5 (1M context) " } ] }, { "name": "Claude Fable 5.1", "icon": "claude", "detail": "high, when Opus is stuck", "history": [], "source": "https://x.com/levelsio/status/2106150888289009666", "seen": "2026-10-02", "quote": "giving up I did a last try with Fable 5.1 on high effort and it after awhile it did figure out the issue!" } ], "review_style": { "value": "Step by step, keeps the agent on a short leash", "source": "https://x.com/levelsio/status/2084997902632390981", "seen": "2026-08-05", "quote": "The only way for me to AI code is just step by step, feature by feature, object by object, and control it or it goes mental", "kind": "tests-only", "also": [ { "source": "https://x.com/levelsio/status/2091960812004888655", "seen": "2026-08-24", "quote": "Until that time I'll just keep manually reading bug reports and feature requests and copying them in Claude Code on the server" } ], "inferred": true }, "harness": { "value": "Termius (phone or laptop) into Hetzner VPSes, one tmux session per site, Claude Code on the server", "short": "VPS + tmux", "source": "https://x.com/levelsio/status/2091987033367724309", "seen": "2026-08-24", "quote": "Every tmux session has Claude Code open", "history": [] }, "parallelism": null, "rules_files": null, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "tm", "kind": "tool", "caption": "a tmux session per site", "why": "A Claude-written script that drops him into a per-project tmux session with Claude Code, straight on the production server.", "editorial": true, "source": "https://x.com/levelsio/status/2091987033367724309", "seen": "2026-08-24", "quote": "tm is a script made by Claude Code which puts it in a tmux session tied to the project name (from its folder)" }, "archetype": { "value": "Vibe Maximalist", "editorial": true, "why": "Claude Code live on his production VPSes, driven from his phone." }, "signature_quote": { "value": "Most of my work these days is copypasting all your bug reports or feature ideas into my clanker via Termius on my VPS", "source": "https://x.com/levelsio/status/2091926355906597101", "seen": "2026-08-24" } }, { "schema": 2, "handle": "badlogic", "status": "ok", "checked": "2026-10-12", "name": { "value": "Mario Zechner", "source": "https://github.com/badlogic" }, "x": { "value": "badlogicgames", "source": "https://github.com/badlogic" }, "github": { "value": "badlogic", "source": "https://github.com/badlogic" }, "site": { "value": "https://mariozechner.at", "source": "https://github.com/badlogic" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/514052?v=4", "source": "https://github.com/badlogic" }, "known_for": { "value": "pi coding agent, libGDX", "source": "https://mariozechner.at/posts/2026-04-08-ive-sold-out/" }, "agents": [ { "name": "Pi", "icon": "pi", "role": "primary, his own", "surface": "cli", "source": "https://mariozechner.at/posts/2026-05-30-shitty-robot/", "seen": "2026-05-30", "quote": "Naturally I vendored the source tree and used my trusty pi with GPT-5.5 to implement feature parity with the Python version.", "history": [ { "value": "Claude Code", "surface": "cli", "source": "https://mariozechner.at/posts/2025-11-30-pi-coding-agent/", "seen": "2025-11-30", "quote": "I preferred Claude Code for most of my work.", "surface_source": { "source": "https://mariozechner.at/posts/2025-11-30-pi-coding-agent/", "seen": "2025-11-30", "quote": "Given that I'm very accustomed to Claude Code, I haven't spent any more time optimizing this. I'm happy with the little flicker I get in VS Code. I wouldn't feel at home otherwise. And it still flickers less than Claude Code." } }, { "value": "Left Claude Code for pi", "surface": "cli", "source": "https://mariozechner.at/posts/2025-12-22-year-in-review-2025/", "seen": "2025-12-22", "quote": "I've since abandoned Claude Code and written my own coding agent harness, which has served me well over the past months.", "surface_source": { "source": "https://mariozechner.at/posts/2025-11-30-pi-coding-agent/", "seen": "2025-11-30", "quote": "pi-coding-agent: The actual CLI that wires it all together with session management, custom tools, themes, and project context files." } } ], "surface_source": { "source": "https://mariozechner.at/posts/2025-11-30-pi-coding-agent/", "seen": "2025-11-30", "quote": "pi-coding-agent: The actual CLI that wires it all together with session management, custom tools, themes, and project context files." } } ], "models": [ { "name": "GPT-5.5", "icon": "openai", "source": "https://mariozechner.at/posts/2026-05-30-shitty-robot/", "seen": "2026-05-30", "quote": "Naturally I vendored the source tree and used my trusty pi with GPT-5.5 to implement feature parity with the Python version.", "history": [] } ], "review_style": { "value": "Limits daily output to what he can review", "source": "https://mariozechner.at/posts/2026-03-25-thoughts-on-slowing-the-fuck-down/", "seen": "2026-03-25", "quote": "Set yourself limits on how much code you let the clanker generate per day, in line with your ability to actually review the code.", "kind": "spot-checks", "also": [ { "source": "https://mariozechner.at/posts/2025-11-30-pi-coding-agent/", "seen": "2025-11-30", "quote": "After I see its initial review, I give my own review and then we work on it together until it's good. That's the workflow I use to not merge garbage code." } ], "inferred": true }, "harness": { "value": "pi from his phone, including code review; tmux for long-running processes", "short": "phone", "source": "https://x.com/badlogicgames/status/2107532222244737128", "seen": "2026-10-06", "quote": "doing pi issues from my phone on a couch. including code review and screaming at the agent.", "history": [ { "value": "pi in the terminal, tmux for long-running processes and sub-agents", "source": "https://mariozechner.at/posts/2025-11-30-pi-coding-agent/", "seen": "2025-11-30", "quote": "There's simply no need for background bash. Claude Code can use tmux too, you know. Bash is all you need." } ] }, "parallelism": null, "rules_files": { "value": "AGENTS.md in the pi repo, shared with co-maintainers", "short": "AGENTS.md", "url": "https://github.com/earendil-works/pi/blob/4fbdc63ca364844c0166d3d4eabafb5cf8e7c2f3/AGENTS.md", "source": "https://github.com/earendil-works/pi/blob/4fbdc63ca364844c0166d3d4eabafb5cf8e7c2f3/AGENTS.md#L9-L11", "seen": "2026-08-07", "quote": "Explain non-trivial designs and problems as: problem, concrete example or short trace, then solution. State why the solution is necessary and distinguish it from optional complexity." }, "skills": { "value": [ "brave-search", "browser-tools", "gccli", "gdcli", "gmcli", "transcribe", "vscode", "youtube-transcript" ], "count": 8, "featured": [ "browser-tools", "brave-search", "transcribe" ], "source": "https://github.com/badlogic/pi-skills/tree/90bb51cae36515a648515b633a81c0c6efc8c74d", "seen": "2026-06-06" }, "mcp_servers": { "value": [], "source": "https://mariozechner.at/posts/2025-11-30-pi-coding-agent/", "seen": "2025-11-30", "quote": "pi does not and will not support MCP." }, "hooks": null, "signature": { "value": "Daily code limit", "kind": "habit", "caption": "sized to what he can review", "why": "Our pick: he wrote his own agent and still rations how much code it writes in a day.", "editorial": true, "source": "https://mariozechner.at/posts/2026-03-25-thoughts-on-slowing-the-fuck-down/", "seen": "2026-03-25", "quote": "Set yourself limits on how much code you let the clanker generate per day, in line with your ability to actually review the code." }, "archetype": { "value": "Purist", "editorial": true, "why": "Four tools, no MCP, no parallel sub-agents, architecture written by hand." }, "signature_quote": { "value": "Anything that defines the gestalt of your system, that is architecture, API, and so on, write it by hand.", "more": "Maybe use tab completion for some nostalgic feels.", "source": "https://mariozechner.at/posts/2026-03-25-thoughts-on-slowing-the-fuck-down/", "seen": "2026-03-25" } }, { "schema": 2, "handle": "mrnugget", "status": "ok", "checked": "2026-10-12", "name": { "value": "Thorsten Ball", "source": "https://github.com/mrnugget" }, "x": { "value": "thorstenball", "source": "https://github.com/mrnugget" }, "github": { "value": "mrnugget", "source": "https://github.com/mrnugget" }, "site": { "value": "https://thorstenball.com", "source": "https://github.com/mrnugget" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/1185253?v=4", "source": "https://github.com/mrnugget" }, "known_for": { "value": "Amp co-founder, interpreter book", "source": "https://thorstenball.com/" }, "agents": [ { "name": "Amp", "icon": "amp", "role": "primary (co-founder)", "surface": "web", "source": "https://registerspill.thorstenball.com/p/joy-and-curiosity-100", "seen": "2026-09-20", "quote": "I had Amp build a little Copilot-style autocomplete for a shell, with Jev picking the next most likely command from shell history.", "history": [], "surface_source": { "source": "https://registerspill.thorstenball.com/p/joy-and-curiosity-95", "seen": "2026-08-16", "quote": "I have not used my local development environment for any of this. I’ve done all of this remotely, using Amp, in orbs." } } ], "models": [ { "name": "Claude Opus 5.5", "icon": "claude", "source": "https://x.com/thorstenball/status/2104860291674886642", "seen": "2026-09-29", "quote": "Opus 5.5 did a great job. It feels nice to use and all the keybindings you expect work.", "history": [ { "value": "Opus 4.7 (Amp smart mode)", "source": "https://registerspill.thorstenball.com/p/joy-and-curiosity-83", "seen": "2026-04-26", "quote": "Amp’s smart mode now uses Opus 4.7. I think it’s a great model. I now often switch between smart and deep mode. One plans, the other reviews, and vice versa." }, { "value": "Fable 5.1", "source": "https://registerspill.thorstenball.com/p/joy-and-curiosity-99", "seen": "2026-09-12", "quote": "Something changed with these latest models, with Fable 5.1 and GPT-6 Astra." } ] }, { "name": "GPT-6 Astra", "icon": "openai", "source": "https://registerspill.thorstenball.com/p/joy-and-curiosity-99", "seen": "2026-09-12", "quote": "Something changed with these latest models, with Fable 5.1 and GPT-6 Astra.", "history": [ { "value": "GPT-5.3-Codex (Amp deep mode)", "source": "https://registerspill.thorstenball.com/p/joy-and-curiosity-78", "seen": "2026-03-15", "quote": "If you still think Opus 4.6 is the peak, try deep mode in Amp, which uses GPT-5.3-Codex right now." } ], "inferred": true } ], "review_style": { "value": "Spot checks and architecture, not every line", "source": "https://registerspill.thorstenball.com/p/joy-and-curiosity-95", "seen": "2026-08-16", "quote": "And for all of the things I shipped here, I didn’t review each line anyway. I do spot checks and make sure the architecture is right, yes, but do I need local tools for that? No.", "kind": "spot-checks", "also": [ { "source": "https://registerspill.thorstenball.com/p/joy-and-curiosity-83", "seen": "2026-04-26", "quote": "I now often switch between smart and deep mode. One plans, the other reviews, and vice versa." }, { "source": "https://registerspill.thorstenball.com/p/joy-and-curiosity-87", "seen": "2026-05-24", "quote": "I’ve finally started to use hunk to review diffs before making commits." } ] }, "harness": { "value": "Amp agents in remote orbs, no local dev environment", "short": "orbs", "source": "https://registerspill.thorstenball.com/p/joy-and-curiosity-95", "seen": "2026-08-16", "quote": "I have not used my local development environment for any of this. I’ve done all of this remotely, using Amp, in orbs.", "history": [] }, "parallelism": null, "rules_files": { "value": "AGENT.md in his dotfiles (26 lines)", "short": "AGENT.md", "url": "https://github.com/mrnugget/dotfiles/blob/72270ffcb997eafeeb46112f5a12adcdf2493afa/AGENT.md", "source": "https://github.com/mrnugget/dotfiles/blob/72270ffcb997eafeeb46112f5a12adcdf2493afa/AGENT.md#L1-L4", "seen": "2025-05-26", "quote": "# Agent Instructions for dotfiles repository\n\n## Build/Install Commands\n- `make all` - Install all dotfiles as symlinks in home directory" }, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "“Irrefutable proof”", "kind": "habit", "caption": "video proof, not line reads", "why": "Our pick: rather than read the diff, he asks the agent in its orb to prove the work end to end, often with a video.", "editorial": true, "source": "https://registerspill.thorstenball.com/p/joy-and-curiosity-99", "seen": "2026-09-12", "quote": "“I want you to go and test this end-to-end, I don’t care how, and give me irrefutable proof that this works. Dazzle me. Give me a video as proof, or something.”" }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "Runs Amp agents in remote orbs that start and check on other agents. He spot-checks the results." }, "signature_quote": { "value": "These agents in orbs now look like async functions to me, less like remote controlled agents.", "more": "Async is the point.", "source": "https://registerspill.thorstenball.com/p/joy-and-curiosity-90", "seen": "2026-07-04" } }, { "schema": 2, "handle": "obra", "status": "ok", "checked": "2026-10-06", "name": { "value": "Jesse Vincent", "source": "https://github.com/obra" }, "x": null, "github": { "value": "obra", "source": "https://github.com/obra" }, "site": { "value": "https://blog.fsck.com", "source": "https://blog.fsck.com/about/" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/45416?v=4", "source": "https://github.com/obra" }, "known_for": { "value": "Superpowers, Keyboardio, K-9 Mail", "source": "https://blog.fsck.com/about/" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary", "surface": "cli", "source": "https://blog.fsck.com/2026/10/02/a-quick-trip-to-the-uncanny-valley/", "seen": "2026-10-02", "quote": "My desktop Claude Code instances use a Slack account labeled as '@jesse-claude'.", "history": [], "surface_source": { "source": "https://blog.fsck.com/2026/07/03/clipfan/", "seen": "2026-07-03", "quote": "For the last few months, that has meant running my coding agents in tmux on either a remote Mac or a remote Linux server." } }, { "name": "Evener", "icon": null, "role": "own harness, long /goal runs", "surface": null, "source": "https://blog.fsck.com/2026/08/21/i-vibe-coded-a-c-compiler/", "seen": "2026-08-21", "quote": "Last night, as I was getting ready for bed I was reaching for some project to /loop support and tool use in Evener, our ~new agentic harness.", "history": [] }, { "name": "Muse", "icon": null, "role": "web browsing", "surface": null, "source": "https://blog.fsck.com/2026/09/29/I-asked-muse-to-tell-me-about-updates-to-its-skills/", "seen": "2026-09-29", "quote": "Muse feels really well put together. It's not my only agent, but I find myself actually using it.", "history": [] } ], "models": [ { "name": "Claude Opus", "icon": "claude", "detail": "coordinator", "source": "https://blog.fsck.com/2026/06/15/Superpowers-6/", "seen": "2026-06-15", "quote": "/goal once this is done, run an autoresearch loop to improve cost-efficiency of the superpowers build loop. test with opus as the coordinator.", "history": [] }, { "name": "Claude Haiku", "icon": "claude", "detail": "implementer subagents", "source": "https://blog.fsck.com/2026/03/09/superpowers-5/", "seen": "2026-03-09", "quote": "With the detailed plans produced through the brainstorming + writing plans process, it's not uncommon to be able to use Claude Haiku for implementation.", "history": [] } ], "review_style": { "value": "Agent sessions do adversarial review", "source": "https://blog.fsck.com/2026/10/02/a-quick-trip-to-the-uncanny-valley/", "seen": "2026-10-02", "quote": "Sometimes, that means that a human colleague or I look over their work, but more often, it means that one of my coding agent sessions does the deep adversarial review.", "kind": "agent-reviews", "also": [ { "source": "https://github.com/obra/dotfiles/blob/6ebb5ba848526b2bfd5b5ab3cb30741ff32c4447/.claude/commands/par.md#L1", "seen": "2026-05-29", "quote": "Assign two subagents to review your work adversarially. Tell them that whoever finds the largest number of legitimate significant issues gets 5 points." }, { "source": "https://blog.fsck.com/2026/06/15/Superpowers-6/", "seen": "2026-06-15", "quote": "the orchestrator inside Superpowers reviews every single change on two axes:" }, { "source": "https://github.com/obra/dotfiles/blob/4849bad8b9bb02ac49abf78754d8e0351343a060/.claude/CLAUDE.md#L100", "seen": "2026-08-08", "quote": "Complete all steps, including reviews, for every change." } ] }, "harness": { "value": "tmux on remote Mac and Linux boxes", "short": "remote tmux", "source": "https://blog.fsck.com/2026/07/03/clipfan/", "seen": "2026-07-03", "quote": "For the last few months, that has meant running my coding agents in tmux on either a remote Mac or a remote Linux server.", "history": [ { "value": "A git worktree per task", "source": "https://blog.fsck.com/2025/10/05/how-im-using-coding-agents-in-september-2025/", "seen": "2025-10-05", "quote": "When I want to start a new task on an existing project, I try to always use a git worktree to isolate that work from other tasks." } ] }, "parallelism": null, "rules_files": { "value": "CLAUDE.md (113 lines + @local.md), near-copy as Codex AGENTS.md", "short": "CLAUDE.md", "url": "https://github.com/obra/dotfiles/blob/4849bad8b9bb02ac49abf78754d8e0351343a060/.claude/CLAUDE.md", "source": "https://github.com/obra/dotfiles/blob/4849bad8b9bb02ac49abf78754d8e0351343a060/.claude/CLAUDE.md#L2", "seen": "2026-08-08", "quote": "Rule #1: If you want exception to ANY rule, YOU MUST STOP and get explicit permission from Jesse first. BREAKING THE LETTER OR SPIRIT OF THE RULES IS FAILURE." }, "skills": { "value": [ "brainstorming", "diagnosing-superpowers", "dispatching-parallel-agents", "executing-plans", "finishing-a-development-branch", "receiving-code-review", "requesting-code-review", "subagent-driven-development", "systematic-debugging", "test-driven-development", "using-git-worktrees", "using-superpowers", "verification-before-completion", "writing-plans", "writing-skills" ], "count": 15, "featured": [ "brainstorming", "subagent-driven-development", "test-driven-development" ], "source": "https://github.com/obra/superpowers/tree/8ca22dba9a94f28898bbce59f2537ff4d87c747d/skills", "seen": "2026-09-25" }, "mcp_servers": { "value": [ "mcp CLI (mcptools), called on demand" ], "source": "https://github.com/obra/dotfiles/blob/10b7a3fe079fe4c5d25bb0b70b12da112130d8e5/.claude/skills/mcp-cli/SKILL.md#L3", "seen": "2026-06-05", "quote": "Use MCP servers on-demand via the mcp CLI tool - discover tools, resources, and prompts without polluting context with pre-loaded MCP integrations" }, "hooks": { "value": [ "SessionStart: injects the using-superpowers skill" ], "source": "https://github.com/obra/superpowers/blob/8ca22dba9a94f28898bbce59f2537ff4d87c747d/hooks/session-start#L2-L11", "seen": "2026-09-19", "quote": "# SessionStart hook for superpowers plugin" }, "signature": { "value": "/par", "kind": "habit", "caption": "rival reviewers, 5 points", "why": "Our pick: his adversarial review command makes two subagents compete to find the most real issues, and inflated findings get a reviewer disqualified.", "editorial": true, "source": "https://github.com/obra/dotfiles/blob/6ebb5ba848526b2bfd5b5ab3cb30741ff32c4447/.claude/commands/par.md#L1", "seen": "2026-05-29", "quote": "Assign two subagents to review your work adversarially. Tell them that whoever finds the largest number of legitimate significant issues gets 5 points." }, "archetype": { "value": "Rules Lawyer", "editorial": true, "why": "A Rule #1 CLAUDE.md, plus skills that make agents do TDD and review." }, "signature_quote": { "value": "When you're writing prompts, think about the model as a lazy pedant.", "more": "How could it do something that's technically what you asked, but not at all what you wanted?", "source": "https://blog.fsck.com/2026/04/30/that-time-it-tried-to-delete-all-my-tests/", "seen": "2026-04-30" } }, { "schema": 2, "handle": "harperreed", "status": "ok", "checked": "2026-10-06", "name": { "value": "Harper Reed", "source": "https://github.com/harperreed" }, "x": { "value": "harper", "source": "https://github.com/harperreed" }, "github": { "value": "harperreed", "source": "https://github.com/harperreed" }, "site": { "value": "https://harper.blog", "source": "https://github.com/harperreed/harperreed" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/18504?v=4", "source": "https://github.com/harperreed" }, "known_for": { "value": "CEO 2389.ai, CTO Obama 2012", "source": "https://harperreed.com/" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary", "surface": "cli", "source": "https://github.com/harperreed/dotfiles/blob/3f508fad1bda3304f08e4f1fe52924a3886a1634/.config/agents/README.md#L6", "seen": "2026-07-16", "quote": "One policy governs every coding agent on this machine (Claude Code, Codex, whatever comes next).", "history": [ { "value": "Aider", "surface": null, "source": "https://harper.blog/2025/05/08/basic-claude-code/", "seen": "2025-05-08", "quote": "I have since migrated from Aider to Claude Code and not looked back." } ], "surface_source": { "source": "https://harper.blog/2026/01/05/claude-code-is-better-on-your-phone/", "seen": "2026-01-05", "quote": "When it came to figuring out how to use Claude Code on my phone, the obvious answer was: ssh into my computer from my phone, and run claude." } }, { "name": "Codex", "icon": "codex", "role": "same policy, GPT-6 Astra", "surface": "app", "source": "https://github.com/harperreed/dotfiles/blob/3f508fad1bda3304f08e4f1fe52924a3886a1634/.config/agents/README.md#L6", "seen": "2026-07-16", "quote": "One policy governs every coding agent on this machine (Claude Code, Codex, whatever comes next).", "history": [], "surface_source": { "source": "https://github.com/harperreed/dotfiles/blob/45c367906fe0578525b65c8438ededd77de165e7/.codex/config.toml#L52-L53", "seen": "2026-09-09", "quote": "BROWSER_USE_CODEX_APP_BUILD_FLAVOR = \"prod\"\nBROWSER_USE_CODEX_APP_VERSION = \"26.901.22334\"", "inferred": true } } ], "models": [ { "name": "Claude Fable 5.1", "icon": "claude", "detail": "1M context", "source": "https://github.com/harperreed/dotfiles/blob/551906762cc7326fa6f4025fcac442100cd92a5d/.claude/settings.json#L129", "seen": "2026-09-25", "quote": "\"model\": \"claude-fable-5-1[1m]\"", "history": [] }, { "name": "GPT-6 Astra", "icon": "openai", "detail": "high, in Codex", "source": "https://github.com/harperreed/dotfiles/blob/45c367906fe0578525b65c8438ededd77de165e7/.codex/config.toml#L2-L3", "seen": "2026-09-09", "quote": "model = \"gpt-6-astra\"\nmodel_reasoning_effort = \"high\"", "history": [ { "value": "o1-pro / o3 for specs", "source": "https://harper.blog/2025/05/08/basic-claude-code/", "seen": "2025-05-08", "quote": "I use the best reasoning model I can find to generate the spec. These days it is o1-pro or o3" } ] } ], "review_style": { "value": "One fresh-eyes review pass per change", "source": "https://github.com/harperreed/dotfiles/blob/b78c10ed97fdc0b81f88ebd6e974fd3dcc4dc7d6/.claude/CLAUDE.md#L132", "seen": "2026-10-03", "quote": "Process steps (verification, review) are never skipped regardless of task size — and never multiplied: one verification run, one review pass.", "kind": "agent-reviews", "also": [ { "source": "https://github.com/harperreed/dotfiles/blob/6db272a76a71b098272c7c8a26ae084d197a43ac/.claude/commands/careful-review.md#L1", "seen": "2025-12-05", "quote": "Great, now I want you to carefully read over all of the new code you just wrote and other existing code you just modified with \"fresh eyes,\" looking super carefully for any obvious bugs, errors, problems, issues, confusion, etc. Use a subagent." }, { "source": "https://harper.blog/2026/09/22/break-away/", "seen": "2026-09-22", "quote": "I have been using this for doing all sorts of things for months (entity extract my emails, do aggressive code reviews, run agents, etc) and I was very happy with the results." }, { "source": "https://github.com/harperreed/dotfiles/commit/4ef7c0466445e6a36c1b41bb5d81981963720d3a", "seen": "2026-09-17", "quote": "updated after human review. who knew humans could review" } ], "inferred": true }, "harness": { "value": "SSH/mosh from his phone into a workstation, tmux", "short": "tmux + mosh", "source": "https://harper.blog/2026/01/05/claude-code-is-better-on-your-phone/", "seen": "2026-01-05", "quote": "When it came to figuring out how to use Claude Code on my phone, the obvious answer was: ssh into my computer from my phone, and run claude.", "history": [] }, "parallelism": { "value": 12, "short": "a dozen", "source": "https://harper.blog/2026/01/05/claude-code-is-better-on-your-phone/", "seen": "2026-01-05", "quote": "TMUX is the key that allows me to run a dozen Claude Code instances, and then walk away from my workstation, pick up my phone and continue hacking.", "history": [], "inferred": true }, "rules_files": { "value": "CLAUDE.md (202 lines), one policy shared with Codex's AGENTS.md", "short": "CLAUDE.md", "url": "https://github.com/harperreed/dotfiles/blob/b78c10ed97fdc0b81f88ebd6e974fd3dcc4dc7d6/.claude/CLAUDE.md", "source": "https://github.com/harperreed/dotfiles/blob/3f508fad1bda3304f08e4f1fe52924a3886a1634/.config/agents/README.md#L6", "seen": "2026-07-16", "quote": "One policy governs every coding agent on this machine (Claude Code, Codex, whatever comes next). Each tool reads its own real file — no imports between them, no symlinks, no generation step." }, "skills": { "value": [ "app-store-changelog", "frontend-design", "humanizer", "ios-debugger-agent", "native-app-performance", "releasing-software", "session-reflection-analysis", "swift-concurrency-expert", "swiftui-liquid-glass", "swiftui-performance-audit", "swiftui-view-refactor", "uv" ], "count": 12, "featured": [ "session-reflection-analysis", "humanizer", "releasing-software" ], "source": "https://github.com/harperreed/dotfiles/tree/b78c10ed97fdc0b81f88ebd6e974fd3dcc4dc7d6/.claude/skills", "seen": "2026-07-11" }, "mcp_servers": { "value": [ "chronicle", "toki", "bbs", "memo", "pulse", "context7" ], "source": "https://github.com/harperreed/dotfiles/blob/45c367906fe0578525b65c8438ededd77de165e7/.codex/config.toml#L7-L35", "seen": "2026-09-09", "quote": "[mcp_servers.chronicle]\ncommand = \"chronicle\"\nargs = [\"mcp\"]" }, "hooks": { "value": [ "Notification: macOS alert", "SessionStart: herdr agent state" ], "source": "https://github.com/harperreed/dotfiles/blob/551906762cc7326fa6f4025fcac442100cd92a5d/.claude/settings.json#L26-L50", "seen": "2026-09-25", "quote": "\"command\": \"bash '/Users/harper/.claude/hooks/herdr-agent-state.sh' session\"" }, "signature": { "value": "\"fresh eyes\" review", "kind": "habit", "caption": "a subagent rereads the code", "why": "Our pick: his careful-review command sends a subagent back over new code with \"fresh eyes\". Jesse Vincent says he learned the trick from Harper.", "editorial": true, "source": "https://github.com/harperreed/dotfiles/blob/6db272a76a71b098272c7c8a26ae084d197a43ac/.claude/commands/careful-review.md#L1", "seen": "2025-12-05", "quote": "Great, now I want you to carefully read over all of the new code you just wrote and other existing code you just modified with \"fresh eyes,\" looking super carefully for any obvious bugs, errors, problems, issues, confusion, etc. Use a subagent." }, "archetype": { "value": "Vibe Maximalist", "editorial": true, "why": "A dozen Claude Codes in tmux, driven from his phone." }, "signature_quote": { "value": "Now you can tell Claude to do weird shit from your phone 24 hours a day.", "more": "It rules.", "source": "https://harper.blog/2026/01/05/claude-code-is-better-on-your-phone/", "seen": "2026-01-05" } }, { "schema": 2, "handle": "kentbeck", "status": "ok", "checked": "2026-10-06", "name": { "value": "Kent Beck", "source": "https://github.com/KentBeck" }, "x": { "value": "KentBeck", "source": "https://github.com/KentBeck" }, "github": { "value": "KentBeck", "source": "https://github.com/KentBeck" }, "site": { "value": "https://kentbeck.com", "source": "https://github.com/KentBeck" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/46154?v=4", "source": "https://github.com/KentBeck" }, "known_for": { "value": "TDD, Extreme Programming, JUnit", "source": "https://newsletter.pragmaticengineer.com/p/how-kent-beck-shapes-the-software" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "Lean proofs, Smalltalk", "surface": "cli", "source": "https://github.com/KentBeck/FailureVirality/blob/973a52675fb6fb6a5b82799f1849ac25b4038762/README.md#L11", "seen": "2026-09-08", "quote": "From 64 sessions (Codex CLI + Claude Code, Sept 2025–Sept 2026), 953 laps,", "history": [ { "value": "Augment Code in IntelliJ", "surface": "ide", "source": "https://newsletter.kentbeck.com/p/my-augmented-coding-tools-as-of-16", "seen": "2025-05-16", "quote": "Augment works inside IntelliJ, which is how I used it mostly." } ], "inferred": true, "surface_source": { "source": "https://github.com/KentBeck/SmalltalkGenie/blob/5f12d2e50a616301b3d89a91a12fb8cebf059bd7/README.md#L54", "seen": "2026-06-18", "quote": "claude mcp add --transport http genie http://localhost:8087/mcp" } }, { "name": "Codex", "icon": "codex", "role": "Codex CLI and Codex for Mac", "surface": "cli", "source": "https://github.com/KentBeck/FailureVirality/blob/973a52675fb6fb6a5b82799f1849ac25b4038762/README.md#L11", "seen": "2026-09-08", "quote": "From 64 sessions (Codex CLI + Claude Code, Sept 2025–Sept 2026), 953 laps,", "history": [], "inferred": true } ], "models": [ { "name": "Gemini", "icon": "gemini", "detail": "fast, maybe too fast", "source": "https://substack.com/@kentbeck/note/c-183114266", "seen": "2025-12-01", "quote": "Gemini is fast & good for augmented coding. I’m getting done in hours what would have taken me days with Claude Code.", "history": [] } ], "review_style": { "value": "Reviews solo, would rather be pairing", "source": "https://newsletter.kentbeck.com/p/party-of-one-for-code-review", "seen": "2025-12-26", "quote": "Code review meant “another human looks at your code before it ships.” Maybe now it means “something helps you maintain awareness of what you’re building and whether the structure is staying healthy.”", "kind": "agent-reviews", "also": [ { "source": "https://newsletter.kentbeck.com/p/party-of-one-for-code-review", "seen": "2025-12-26", "quote": "I’d rather be pairing." }, { "source": "https://newsletter.kentbeck.com/p/beyond-the-ide", "seen": "2025-09-04", "quote": "The old workflow had us editing for hours and reviewing for minutes. The new workflow has us reviewing for hours and editing for minutes." }, { "source": "https://github.com/KentBeck/ToolSeeds/blob/bb343068a0d43fd42e20f195d4af5c3ba971f84e/README.md#L13", "seen": "2026-10-03", "quote": "Review the result and run the project's tests." } ], "inferred": true }, "harness": { "value": "Rotating tools, worked live in weekly Genie Sessions", "short": "live sessions", "source": "https://newsletter.kentbeck.com/p/genie-lessons-nobody-wants-agents", "seen": "2026-04-23", "quote": "every Friday I work on a real problem with an AI tool, live, for paid subscribers.", "history": [ { "value": "Ona, browser-based multi-genie IDE (sponsor)", "source": "https://newsletter.kentbeck.com/p/beyond-the-ide", "seen": "2025-09-04", "quote": "I’ve used it for 60+ hours in the last 2 weeks & I have a hard time augmented coding without it." } ] }, "parallelism": { "value": 1, "short": "one at a time", "source": "https://newsletter.kentbeck.com/p/genie-lessons-nobody-wants-agents", "seen": "2026-04-23", "quote": "I’ve never been able to get two agents working on the same codebase at the same time without my head exploding.", "history": [], "inferred": true }, "rules_files": { "value": "CLAUDE.md (96 lines), identical AGENTS.md, plus a GEMINI.md", "short": "CLAUDE.md", "url": "https://github.com/KentBeck/SmalltalkGenie/blob/5f12d2e50a616301b3d89a91a12fb8cebf059bd7/CLAUDE.md", "source": "https://github.com/KentBeck/SmalltalkGenie/blob/5f12d2e50a616301b3d89a91a12fb8cebf059bd7/CLAUDE.md#L8-L9", "seen": "2026-06-29", "quote": "**Operate entirely from the live image via the `genie` MCP tools. Do not\ncreate, expect, or rely on a file mirror of the code.**" }, "skills": { "value": [ "pharo-live-image", "pharo-refactor", "pharo-rename-method", "pharo-tdd", "representation-envy-refactor" ], "count": 5, "featured": [ "pharo-live-image", "pharo-tdd", "representation-envy-refactor" ], "source": "https://github.com/KentBeck/SmalltalkGenie/tree/5f12d2e50a616301b3d89a91a12fb8cebf059bd7/.agents/skills", "seen": "2026-07-02" }, "mcp_servers": { "value": [ "genie (his SmalltalkGenie, inside a live Pharo image)" ], "source": "https://github.com/KentBeck/SmalltalkGenie/blob/5f12d2e50a616301b3d89a91a12fb8cebf059bd7/README.md#L54", "seen": "2026-06-18", "quote": "claude mcp add --transport http genie http://localhost:8087/mcp" }, "hooks": null, "signature": { "value": "TCR skill", "kind": "skill", "caption": "red tests revert the code", "why": "Our pick: his own Test && Commit || Revert discipline as an always-on agent rule, so a failing step is thrown away instead of patched.", "editorial": true, "source": "https://github.com/KentBeck/TCRSkill/blob/0132ef205b08ef0558cd493336420cce9dbbfa89/.cursor/rules/tcr-always.mdc#L2", "seen": "2026-03-30", "quote": "Always use TCR (Test && Commit || Revert)—run tests after each micro-change, commit only on green, revert on red." }, "archetype": { "value": "Purist", "editorial": true, "why": "TCR on every step, test-gated saves, one agent at a time." }, "signature_quote": { "value": "Nobody wants agents. Nobody wants agent swarms. I have a system and I want it to change.", "more": "That’s the whole thing.", "source": "https://newsletter.kentbeck.com/p/genie-lessons-nobody-wants-agents", "seen": "2026-04-23" } }, { "schema": 2, "handle": "kieranklaassen", "status": "ok", "checked": "2026-10-12", "name": { "value": "Kieran Klaassen", "source": "https://github.com/kieranklaassen" }, "x": { "value": "kieranklaassen", "source": "https://x.com/kieranklaassen" }, "github": { "value": "kieranklaassen", "source": "https://github.com/kieranklaassen" }, "site": { "value": "https://www.kieranklaassen.com", "source": "https://github.com/kieranklaassen" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/209089?v=4", "source": "https://github.com/kieranklaassen" }, "known_for": { "value": "Compound Engineering, Cora", "source": "https://github.com/kieranklaassen/kieranklaassen" }, "agents": [ { "name": "Cursor", "icon": "cursor", "role": "primary", "surface": null, "source": "https://toolbox.every.to/kieranklaassen", "seen": null, "quote": "Coding\nCursor\nthen Claude Code, Codex\nClaude Opus 5.5", "history": [ { "value": "Cursor with Opus 4.5 or Claude Code", "surface": null, "source": "https://every.to/guides/compound-engineering", "seen": "2026-01-17", "quote": "Pick one tool. My current preference is for Cursor with Opus 4.5 or Claude Code." } ], "undated": true }, { "name": "Claude Code", "icon": "claudecode", "role": "second pick", "surface": "cli", "surface_source": { "source": "https://every.to/guides/compound-engineering", "seen": "2026-01-17", "quote": "I always run with skip permissions:\nalias cc='claude --dangerously-skip-permissions'" }, "source": "https://toolbox.every.to/kieranklaassen", "seen": null, "quote": "Coding\nCursor\nthen Claude Code, Codex", "history": [ { "value": "Claude Code only", "surface": "cli", "surface_source": { "source": "https://every.to/source-code/how-i-use-claude-code-to-ship-like-a-team-of-five", "seen": "2025-07-16", "quote": "Running four parallel Claude Code agents in Warp, an AI-enabled terminal, doing different work at the same time." }, "source": "https://every.to/source-code/how-i-use-claude-code-to-ship-like-a-team-of-five", "seen": "2025-07-16", "quote": "Claude Code opens 100 percent of my pull requests, and I haven't typed a function in weeks." } ], "undated": true }, { "name": "Codex", "icon": "codex", "role": "third pick", "surface": null, "source": "https://toolbox.every.to/kieranklaassen", "seen": null, "quote": "Coding\nCursor\nthen Claude Code, Codex", "history": [], "undated": true } ], "models": [ { "name": "Claude Opus 5.5", "icon": "claude", "detail": "1M context, medium effort", "source": "https://every.to/vibe-check/vibe-check-opus-5-5-is-pulling-our-codex-converts-back-to-claude", "seen": "2026-09-22", "quote": "Opus is replacing Fable 5.1 as my daily driver. For product work and writing code, I find it as good as Fable and sometimes better.", "history": [ { "value": "Opus 4.5", "source": "https://every.to/guides/compound-engineering", "seen": "2026-01-17", "quote": "My current preference is for Cursor with Opus 4.5 or Claude Code." }, { "value": "Opus 5, medium effort", "source": "https://x.com/kieranklaassen/status/2080712817926443486", "seen": "2026-07-24", "quote": "I have been coding with it without many skills and it's nice at a medium effort level." } ] }, { "name": "Claude Sonnet 5.5", "icon": "claude", "detail": "quick work, low or medium effort", "source": "https://x.com/kieranklaassen/status/2104633604169142756", "seen": "2026-09-28", "quote": "Use Sonnet for iterative, quick work, especially design, prototyping, and product work. It's also nice for quick background jobs. Run it on low or medium effort.", "history": [ { "value": "Claude Fable 5.1", "source": "https://every.to/vibe-check/vibe-check-opus-5-5-is-pulling-our-codex-converts-back-to-claude", "seen": "2026-09-22", "quote": "For the hardest coding problems, Mike and Kieran still reach for Fable." } ] } ], "review_style": { "value": "Parallel review agents, no line-by-line reading", "source": "https://every.to/source-code/i-stopped-reading-code-my-code-reviews-got-better", "seen": "2026-01-23", "quote": "Now my code reviews no longer involve reading code. And I’ve gotten better at catching problems because of it.", "kind": "agent-reviews", "also": [ { "source": "https://every.to/source-code/i-stopped-reading-code-my-code-reviews-got-better", "seen": "2026-01-23", "quote": "The signature fix that touched 27 files? Thirteen specialized AI reviewers examined it simultaneously while I made dinner." } ] }, "harness": { "value": "Claude Code aliased to skip permissions, branches and git worktrees", "short": "skip-permissions", "source": "https://every.to/guides/compound-engineering", "seen": "2026-01-17", "quote": "I always run with skip permissions:\nalias cc='claude --dangerously-skip-permissions'", "history": [ { "value": "Claude Code tabs in Warp, one git worktree each", "source": "https://every.to/source-code/how-i-use-claude-code-to-ship-like-a-team-of-five", "seen": "2025-07-16", "quote": "My monitor looks like mission control: multiple Claude Code tabs, each working on different features through separate git worktrees, meaning I can have Claude modify five different versions of our codebase simultaneously and get clean, review-ready code." } ], "also": [ { "source": "https://every.to/source-code/the-folder-is-the-agent", "seen": "2026-04-13", "quote": "I’m running 44 of these folders-as-agents across multiple projects now." } ] }, "parallelism": null, "rules_files": { "value": "AGENTS.md per repo, CLAUDE.md symlinked to it", "short": "AGENTS.md", "url": "https://github.com/kieranklaassen/loadout/blob/c1b49eff6d9eb5e97ad776dbc186e61f0805a340/AGENTS.md", "source": "https://github.com/kieranklaassen/loadout/blob/c1b49eff6d9eb5e97ad776dbc186e61f0805a340/AGENTS.md#L3-L4", "seen": "2026-09-28", "quote": "Canonical, tool-agnostic guide for agents and humans working in this repo.\n`CLAUDE.md` is a symlink to this file." }, "skills": { "value": [ "ce-babysit-pr", "ce-bakeoff", "ce-brainstorm", "ce-code-review", "ce-commit-push-pr", "ce-commit", "ce-compound-refresh", "ce-compound", "ce-debug", "ce-doc-review", "ce-dogfood", "ce-explain", "ce-handoff", "ce-ideate", "ce-noslop", "ce-optimize", "ce-plan", "ce-polish", "ce-pov", "ce-product-pulse", "ce-promote", "ce-proof", "ce-prototype", "ce-resolve-pr-feedback", "ce-retune", "ce-riffrec-feedback-analysis", "ce-setup", "ce-simplify-code", "ce-strategy", "ce-sweep", "ce-test-browser", "ce-test-xcode", "ce-work", "ce-worktree", "lfg", "wtf" ], "count": 36, "featured": [ "ce-compound", "lfg", "ce-polish" ], "source": "https://github.com/EveryInc/compound-engineering-plugin/tree/efcb657d9a5733ccc154c36cc7d3d78b4136eb23/skills", "seen": "2026-10-06" }, "mcp_servers": null, "hooks": null, "signature": { "value": "/ce-compound", "kind": "skill", "caption": "turns each fix into a lesson", "why": "Our pick: the step his whole method is named for, writing every solved problem down so the next agent starts smarter.", "editorial": true, "source": "https://every.to/p/compound-engineering-gets-an-upgrade", "seen": "2026-05-29", "quote": "Compound is still the most important step, because the whole point is that every feature should make the next feature easier." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Co-wrote the 36-skill plugin his agents run on, from planning to review to compounding." }, "signature_quote": { "value": "Now my code reviews no longer involve reading code.", "more": "And I’ve gotten better at catching problems because of it.", "source": "https://every.to/source-code/i-stopped-reading-code-my-code-reviews-got-better", "seen": "2026-01-23" } }, { "schema": 2, "handle": "mattpocock", "status": "ok", "checked": "2026-10-06", "name": { "value": "Matt Pocock", "source": "https://github.com/mattpocock" }, "x": { "value": "mattpocockuk", "source": "https://x.com/mattpocockuk" }, "github": { "value": "mattpocock", "source": "https://github.com/mattpocock" }, "site": { "value": "https://totaltypescript.com", "source": "https://github.com/mattpocock" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/28293365?v=4", "source": "https://github.com/mattpocock" }, "known_for": { "value": "Total TypeScript, ex-Vercel", "source": "https://github.com/mattpocock" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary, also runs AFK agents", "surface": "cli", "surface_source": { "source": "https://www.aihero.dev/how-to-kill-the-bloat-in-claude-codes-system-prompt", "seen": "2026-07-07", "quote": "Point Claude Code at it in another terminal and send a message:" }, "source": "https://www.aihero.dev/how-to-kill-the-bloat-in-claude-codes-system-prompt", "seen": "2026-07-07", "quote": "Want my whole Claude Code setup, bloat removal included?", "history": [] } ], "models": [ { "name": "Claude Opus 5.5", "icon": "claude", "detail": "interactive", "source": "https://www.aihero.dev/skills-changelog-v13-implement-spec-pr-retro-and-glossary-md", "seen": "2026-10-05", "quote": "On Opus 5.5 it loads each time the agent writes a PR body, and you do not have to think about it.", "history": [] }, { "name": "Claude Opus 5", "icon": "claude", "detail": "AFK Sandcastle agents", "source": "https://github.com/mattpocock/course-video-manager/blob/41c495e0fe825692026d834066674e483dfa5e14/.sandcastle/main.ts#L14", "seen": "2026-07-24", "quote": "agent: sandcastle.claudeCode(\"claude-opus-5\"),", "history": [] } ], "review_style": { "value": "Reviews by merge danger, two-axis /code-review", "source": "https://www.aihero.dev/skills-changelog-v13-implement-spec-pr-retro-and-glossary-md", "seen": "2026-10-05", "quote": "Merge Danger tells you where to spend review time. A two-way door with a small blast radius: skim it. A one-way door: read it slowly.", "kind": "spot-checks", "also": [ { "source": "https://github.com/mattpocock/skills/blob/7a030a8b0dcaa601bb0dda9aaf174affb3d61145/skills/engineering/code-review/SKILL.md#L11", "seen": "2026-10-06", "quote": "Both axes run as **parallel sub-agents** so they don't pollute each other's context, then this skill aggregates their findings." }, { "source": "https://www.aihero.dev/my-7-phases-of-ai-development", "seen": "2026-03-16", "quote": "QA involves humans reading the generated code to ensure quality, maintainability, and correctness." } ] }, "harness": { "value": "Claude Code, Sandcastle (his library) running agents in Docker sandboxes off GitHub issues", "short": "Sandcastle", "source": "https://www.aihero.dev/skills-changelog-v13-implement-spec-pr-retro-and-glossary-md", "seen": "2026-10-05", "quote": "A script reads each ticket and runs /implement on it. This is what I recommend for most people.", "history": [ { "value": "Ralph loops", "source": "https://www.aihero.dev/my-7-phases-of-ai-development", "seen": "2026-03-16", "quote": "I use Ralph loops, which work effectively with this setup." } ] }, "parallelism": null, "rules_files": { "value": "CLAUDE.md per repo, with an ## Agent skills block his setup skill writes", "short": "CLAUDE.md", "url": "https://github.com/mattpocock/course-video-manager/blob/41c495e0fe825692026d834066674e483dfa5e14/CLAUDE.md", "source": "https://github.com/mattpocock/course-video-manager/blob/41c495e0fe825692026d834066674e483dfa5e14/CLAUDE.md#L17", "seen": "2026-09-30", "quote": "`.claude/skills/verify-cvm/` drives the app in a browser against the PRODUCTION database and leaves a **Write Ledger** proving what it did or did not modify." }, "skills": { "value": [ "ask-matt", "diagnosing-bugs", "grill-with-docs", "triage", "improve-codebase-architecture", "setup-matt-pocock-skills", "tdd", "to-spec", "to-tickets", "wayfinder", "implement", "implement-spec", "prototype", "research", "domain-modeling", "codebase-design", "code-review", "pr", "retro", "wizard", "grill-me", "grilling", "handoff", "teach", "to-questionnaire", "wait-what", "writing-for-agents" ], "count": 27, "featured": [ "grill-me", "grill-with-docs", "tdd" ], "source": "https://github.com/mattpocock/skills/blob/7a030a8b0dcaa601bb0dda9aaf174affb3d61145/.claude-plugin/plugin.json#L21-L48", "seen": "2026-10-06" }, "mcp_servers": null, "hooks": { "value": [ "PreToolUse block-npx-tsc.sh" ], "source": "https://github.com/mattpocock/course-video-manager/blob/41c495e0fe825692026d834066674e483dfa5e14/.claude/hooks/block-npx-tsc.sh#L2", "seen": "2026-07-16", "quote": "# Block commands that execute npx tsc, suggest pnpm run typecheck instead." }, "signature": { "value": "/grill-me", "kind": "skill", "caption": "before every change", "why": "Our pick: his most popular skill makes the agent interview him until the design is settled, before any code is written.", "editorial": true, "source": "https://github.com/mattpocock/skills/blob/7a030a8b0dcaa601bb0dda9aaf174affb3d61145/README.md#L103", "seen": "2026-10-06", "quote": "These are my most popular skills. They help you align with the agent before you get started, and think deeply about the change you're making. Use them _every_ time you want to make a change." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Ships his own 27-skill plugin and the Sandcastle library he runs his agents on." }, "signature_quote": { "value": "Writing code got cheap. Reading it did not.", "source": "https://www.aihero.dev/skills-changelog-v13-implement-spec-pr-retro-and-glossary-md", "seen": "2026-10-05" } }, { "schema": 2, "handle": "indragiek", "status": "ok", "checked": "2026-10-12", "name": { "value": "Indragie Karunaratne", "source": "https://github.com/indragiek" }, "x": { "value": "indragie", "source": "https://x.com/indragie" }, "github": { "value": "indragiek", "source": "https://github.com/indragiek" }, "site": { "value": "https://www.indragie.com", "source": "https://github.com/indragiek" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/353158?v=4", "source": "https://github.com/indragiek" }, "known_for": { "value": "Context, uniprof, Sentry AI lead", "source": "https://pyai.events/speakers/indragie-karunaratne" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary", "surface": "cli", "surface_source": { "source": "https://www.indragie.com/blog/i-shipped-a-macos-app-built-entirely-by-claude-code", "seen": "2025-07-01", "quote": "Claude Code was not a VS Code fork like the others, but was an IDE that was designed to be used entirely in the terminal.", "inferred": true }, "source": "https://github.com/indragiek/GHFS/commit/80afde292d219f4f033d1f683dcaaae016ed877a", "seen": "2026-04-11", "quote": "Co-Authored-By: Claude Opus 4.6 (1M context) ", "history": [ { "value": "GitHub Copilot", "surface": "ide", "source": "https://www.indragie.com/blog/i-shipped-a-macos-app-built-entirely-by-claude-code", "seen": "2025-07-01", "quote": "My first experience with AI coding tools was when I tried GitHub Copilot, built into VS Code." }, { "value": "Claude Code", "surface": "cli", "surface_source": { "source": "https://www.indragie.com/blog/i-shipped-a-macos-app-built-entirely-by-claude-code", "seen": "2025-07-01", "quote": "Claude Code was not a VS Code fork like the others, but was an IDE that was designed to be used entirely in the terminal." }, "source": "https://www.indragie.com/blog/i-shipped-a-macos-app-built-entirely-by-claude-code", "seen": "2025-07-01", "quote": "Context was almost 100% built by Claude Code." } ], "inferred": true }, { "name": "Codex", "icon": "codex", "role": "second agent", "surface": null, "source": "https://github.com/indragiek/GHFS/blob/fe08755bcc9120d2bf339531f982b98a29225c04/agents.toml#L2", "seen": "2026-04-11", "quote": "agents = [\"claude\", \"codex\"]", "history": [ { "value": "Claude Code and Codex", "surface": null, "source": "https://www.indragie.com/blog/i-shipped-a-tool-to-help-agents-fix-slow-code", "seen": "2025-09-22", "quote": "100% of the code was written by Claude Code and Codex." } ] } ], "models": [ { "name": "Claude Opus 5.5", "icon": "claude", "source": "https://x.com/indragie/status/2109671763277975935", "seen": "2026-10-12", "quote": "I had Opus 5.5 use captured HTTP traffic from the mobile apps for Ring (cameras), Petlibro (feeder), and Whisker (litter robot) to build an hour-by-hour summary", "history": [ { "value": "Sonnet 4 and Opus 4", "source": "https://www.indragie.com/blog/i-shipped-a-macos-app-built-entirely-by-claude-code", "seen": "2025-07-01", "quote": "Let me just start by saying that Claude Code (with the latest Sonnet 4 and Opus 4 models) is genuinely good at writing code." }, { "value": "Claude Opus 4.6", "source": "https://github.com/indragiek/GHFS/commit/80afde292d219f4f033d1f683dcaaae016ed877a", "seen": "2026-04-11", "quote": "Co-Authored-By: Claude Opus 4.6 (1M context) " } ] }, { "name": "GPT-5", "icon": "openai", "detail": "via Codex", "source": "https://www.indragie.com/blog/i-shipped-a-tool-to-help-agents-fix-slow-code", "seen": "2025-09-22", "quote": "I've been writing (generating) more code, trying new tools (hello GPT-5 and Codex), and trying to push the boundaries of what agents can do.", "history": [] } ], "review_style": { "value": "Reviews the plan and prompts, not the 10,000-line diff", "source": "https://www.youtube.com/watch?v=IPBIB7Go6TY&t=237s", "seen": "2026-04-20", "quote": "you don't want to review a 10,000 line slop PR but you may be interested in reviewing like the plan that was used to implement that which is significantly smaller than the code output", "kind": "unknown", "also": [ { "source": "https://www.youtube.com/watch?v=IPBIB7Go6TY&t=194s", "seen": "2026-04-20", "quote": "It's called Iterate PR. So basically what this does is it automates the loop of you create a code change, you you push it to GitHub. Code review agents run on that PR. They leave comments." }, { "source": "https://www.indragie.com/blog/i-shipped-a-macos-app-built-entirely-by-claude-code", "seen": "2025-07-01", "quote": "It occurred to me as I worked on this project that the only two tools I used throughout were Claude Code and GitHub Desktop for viewing diffs." } ], "inferred": true }, "harness": { "value": "Claude Code in the terminal, GitHub Desktop for diffs", "short": "GitHub Desktop", "source": "https://www.indragie.com/blog/i-shipped-a-macos-app-built-entirely-by-claude-code", "seen": "2025-07-01", "quote": "It occurred to me as I worked on this project that the only two tools I used throughout were Claude Code and GitHub Desktop for viewing diffs.", "history": [] }, "parallelism": null, "rules_files": { "value": "Per-project CLAUDE.md = AGENTS.md, driven by a lat.md knowledge graph", "short": "CLAUDE.md", "url": "https://github.com/indragiek/GHFS/blob/fe08755bcc9120d2bf339531f982b98a29225c04/CLAUDE.md", "source": "https://github.com/indragiek/GHFS/blob/fe08755bcc9120d2bf339531f982b98a29225c04/CLAUDE.md#L8-L14", "seen": "2026-04-11", "quote": "After EVERY task, before responding to the user:\n\n- [ ] Update `lat.md/` if you added or changed any functionality, architecture, tests, or behavior\n- [ ] Run `lat check` — all wiki links and code refs must pass" }, "skills": { "value": [ "dotagents", "swiftui-pro", "swift-testing-pro", "swift-concurrency-pro", "lat-md", "sosumi" ], "count": 6, "featured": [ "sosumi", "lat-md", "swift-concurrency-pro" ], "source": "https://github.com/indragiek/GHFS/tree/fe08755bcc9120d2bf339531f982b98a29225c04", "seen": "2026-04-11" }, "mcp_servers": { "value": [ "XcodeBuildMCP" ], "source": "https://github.com/indragiek/GHFS/blob/fe08755bcc9120d2bf339531f982b98a29225c04/agents.toml#L11-L14", "seen": "2026-04-11", "quote": "name = \"XcodeBuildMCP\"\ncommand = \"npx\"\nargs = [ \"-y\", \"xcodebuildmcp@latest\", \"mcp\" ]" }, "hooks": { "value": [ "PostToolUse hook: uploads the session transcript on git commit / gh pr create" ], "source": "https://www.youtube.com/watch?v=IPBIB7Go6TY&t=416s", "seen": "2026-04-20", "quote": "it installs hooks into Claude code and cursor. And then it uses the post tool use hook and a few other hooks to basically like see when there's a bash tool call to like get commit or GHPR create." }, "signature": { "value": "Claudescope", "kind": "tool", "caption": "transcript link in every PR", "why": "Our pick: he built his own hook-driven tool so reviewers read the agent session and plan behind a PR instead of the generated diff.", "editorial": true, "source": "https://www.youtube.com/watch?v=IPBIB7Go6TY&t=463s", "seen": "2026-04-20", "quote": "ask it to create a commit, create a PR, you will see that it inserts a transcript URL into the PR description" }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Builds his own agent tooling: Claudescope hooks, uniprof, GHFS, Context." }, "signature_quote": { "value": "if you want more than a prototype, you need a real spec.", "more": "The spec doesn't need to be well-written.", "source": "https://www.indragie.com/blog/i-shipped-a-macos-app-built-entirely-by-claude-code", "seen": "2025-07-01" } }, { "schema": 2, "handle": "addyosmani", "status": "ok", "checked": "2026-10-06", "name": { "value": "Addy Osmani", "source": "https://github.com/addyosmani" }, "x": { "value": "addyosmani", "source": "https://x.com/addyosmani" }, "github": { "value": "addyosmani", "source": "https://github.com/addyosmani" }, "site": { "value": "https://addyosmani.com", "source": "https://github.com/addyosmani" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/110953?v=4", "source": "https://github.com/addyosmani" }, "known_for": { "value": "Claude Code at Anthropic, ex-Chrome", "source": "https://addyosmani.com/bio/" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary", "surface": null, "source": "https://addyo.substack.com/p/practical-loop-engineering", "seen": "2026-08-14", "quote": "I can now largely rely on the output of the primitives in Claude Code and Codex.", "history": [ { "value": "Claude Code and Gemini CLI", "surface": "cli", "surface_source": { "source": "https://addyo.substack.com/p/my-llm-coding-workflow-going-into", "seen": "2025-12-18", "quote": "On the command-line, new AI agents emerged. Claude Code, OpenAI’s Codex CLI and Google’s Gemini CLI are CLI tools where you can chat with them directly in your project directory - they can read files, run tests, and even multi-step fix issues." }, "source": "https://addyo.substack.com/p/how-to-write-a-good-spec-for-ai-agents", "seen": "2026-01-19", "quote": "This guide distills best practices from my use of coding agents including Claude Code and Gemini CLI into a framework for spec-writing that keeps your AI agents focused and productive." } ] }, { "name": "Codex", "icon": "codex", "role": "PR triage, loops", "surface": null, "source": "https://addyo.substack.com/p/agentic-code-review", "seen": "2026-06-16", "quote": "I point Claude Code or Codex at a batch of incoming PRs and ask for a first pass: a high-level read of what looks safe to merge, what needs more work, and what is genuinely high-risk.", "history": [] } ], "models": [ { "name": "Fable", "icon": "claude", "detail": null, "source": "https://x.com/addyosmani/status/2097210828659564642", "seen": "2026-09-08", "quote": "Here's a little Fable & Three.js", "history": [ { "value": "Gemini", "source": "https://addyo.substack.com/p/my-llm-coding-workflow-going-into", "seen": "2025-12-18", "quote": "Personally I gravitate towards Gemini for a lot of coding work these days because the interaction feels more natural and it often understands my requests on the first try." } ], "inferred": true } ], "review_style": { "value": "Careful human review of core paths only", "source": "https://addyo.substack.com/p/the-code-nobody-reads", "seen": "2026-09-28", "quote": "Core and sensitive paths still need an owner and a careful human review, and that’s where I spend my attention.", "kind": "spot-checks", "also": [ { "source": "https://addyo.substack.com/p/the-code-nobody-reads", "seen": "2026-09-28", "quote": "Two years ago I wrote an essay about AI-assisted coding, and one of my tips was to review every line of generated code. I’d give that advice very differently today." }, { "source": "https://addyo.substack.com/p/agentic-code-quality", "seen": "2026-08-08", "quote": "I still read and review code, but am very intentional about where I am comfortable with constraints as the check." } ] }, "harness": { "value": "Claude Code, a git worktree per feature", "short": "worktrees", "source": "https://addyo.substack.com/p/my-llm-coding-workflow-going-into", "seen": "2025-12-18", "quote": "One advanced workflow I’ve adopted (inspired by folks like Jesse Vincent) is to spin up a fresh git worktree for a new feature or sub-project.", "history": [] }, "parallelism": { "value": 5, "short": "5 at once", "source": "https://addyo.substack.com/p/practical-loop-engineering", "seen": "2026-08-14", "quote": "For me, I use probably between five and ten agents every day. Very typically I’ll max out at about five concurrently.", "history": [ { "value": 1, "source": "https://addyo.substack.com/p/my-llm-coding-workflow-going-into", "seen": "2025-12-18", "quote": "For most cases, I stick to one main agent at a time and maybe a secondary one for reviews (discussed below)." } ] }, "rules_files": { "value": "Personal CLAUDE.md, audited with /doctor", "short": "CLAUDE.md", "url": null, "source": "https://addyo.substack.com/p/audit-your-agent-files", "seen": "2026-08-27", "quote": "I now run Claude’s /doctor every few weeks, review memory separately, and ask each instruction to earn its place again." }, "skills": { "value": [ "api-and-interface-design", "browser-testing-with-devtools", "ci-cd-and-automation", "code-review-and-quality", "code-simplification", "constraint-driven-development", "context-engineering", "debugging-and-error-recovery", "deprecation-and-migration", "documentation-and-adrs", "doubt-driven-development", "frontend-ui-engineering", "git-workflow-and-versioning", "idea-refine", "incremental-implementation", "interview-me", "observability-and-instrumentation", "performance-optimization", "planning-and-task-breakdown", "security-and-hardening", "shipping-and-launch", "source-driven-development", "spec-driven-development", "test-driven-development", "using-agent-skills" ], "count": 25, "featured": [ "spec-driven-development", "code-review-and-quality", "test-driven-development" ], "source": "https://github.com/addyosmani/agent-skills/tree/1401c8b8030e023baeebb31781a6653fe8e93026/skills", "seen": "2026-10-03" }, "mcp_servers": { "value": [ "context7", "chrome-devtools" ], "source": "https://addyo.substack.com/p/my-llm-coding-workflow-going-into", "seen": "2025-12-18", "quote": "I also use Chrome DevTools MCP, built with my last team, for my debugging and quality loop to bridge the gap between static code analysis and live browser execution." }, "hooks": null, "signature": { "value": "/loop PR triage", "kind": "habit", "caption": "on Agent Skills, every day", "why": "Our pick: scheduled agent loops make the first pass on the PR queue of his 100k-star skills repo.", "editorial": true, "source": "https://addyo.substack.com/p/practical-loop-engineering", "seen": "2026-08-14", "quote": "One of the workflows that I do every day: I have a popular open source repository called Agent Skills. We’ve got over 80,000 stars, and up until recently we were getting anywhere up to like 80 or 90 pull requests that we had to review a day. And so every day I would go and I would spend some time checking on this. Now with loop, what you can do is say: well, every 24 hours or every 12 hours, check the GitHub repository for any new open issues and provide a summary of their urgency, or provide a first pass review, or anything like that." }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "Up to five agents at once, plus loops that triage his PR queue." }, "signature_quote": { "value": "What I’d say now is this: you don’t need to read all the code.", "more": null, "source": "https://addyo.substack.com/p/the-code-nobody-reads", "seen": "2026-09-28" } }, { "schema": 2, "handle": "ghuntley", "status": "ok", "checked": "2026-10-06", "name": { "value": "Geoffrey Huntley", "source": "https://github.com/ghuntley" }, "x": { "value": "GeoffreyHuntley", "source": "https://x.com/GeoffreyHuntley" }, "github": { "value": "ghuntley", "source": "https://github.com/ghuntley" }, "site": { "value": "https://ghuntley.com", "source": "https://github.com/ghuntley" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/127353?v=4", "source": "https://github.com/ghuntley" }, "known_for": { "value": "Creator of the Ralph loop", "source": "https://github.com/ghuntley" }, "agents": [ { "name": "OpenCode", "icon": "opencode", "role": "via his underclass token pool", "surface": "cli", "surface_source": { "source": "https://github.com/ghuntley/underclass/blob/9244c8169bb1b73492f0c5ad5b693d59d65034da/README.md?plain=1#L55", "seen": "2026-09-23", "quote": "opencode --provider underclass --model underclass/gpt-5.5" }, "source": "https://github.com/ghuntley/underclass/blob/9244c8169bb1b73492f0c5ad5b693d59d65034da/README.md?plain=1#L27", "seen": "2026-09-23", "quote": "**No client changes** — the surface is plain `/v1/*`; point opencode (or anything OpenAI-compatible) at it.", "history": [ { "value": "Amp", "surface": null, "source": "https://ghuntley.com/ralph/", "seen": "2025-07-14", "quote": "Amp creating a new programming language AFK" }, { "value": "Claude Code", "surface": null, "source": "https://ghuntley.com/real/", "seen": "2026-02-27", "quote": "It is so easy to rip a fart into Claude Code and migrate from one technology to another" } ], "inferred": true }, { "name": "Cursor", "icon": "cursor", "role": "Cloud Agents ship to production", "surface": "web", "surface_source": { "source": "https://ghuntley.com/rad/", "seen": "2026-03-09", "quote": "If I click Launch Agent, then it utilises Cursor's new Cloud Agents and Workflow Automations to ship it straight into production using a risk-based approach." }, "source": "https://ghuntley.com/rad/", "seen": "2026-03-09", "quote": "Last month, Cursor hooked me up with a preview of their new stuff, and now that it's out, this is how I've been using it.", "history": [] }, { "name": "Loom", "icon": null, "role": "his own agent, research only", "surface": null, "source": "https://github.com/ghuntley/loom/blob/a9cfbc6cf085cc8b6c01dd265ca212e458156ba0/README.md?plain=1", "seen": "2026-01-15", "quote": "Loom is an AI-powered coding agent built in Rust.", "history": [] } ], "models": [ { "name": "GPT-6.1 Sol", "icon": "openai", "detail": "no or low reasoning", "source": "https://ghuntley.com/readable/", "seen": "2026-10-02", "quote": "Personally, I run a mix of GPT 6.1 sol on no or low reasoning, and GLM/Kimi.", "history": [] }, { "name": "GLM / Kimi", "icon": "zai", "detail": "cheap open weights", "source": "https://ghuntley.com/readable/", "seen": "2026-10-02", "quote": "Personally, I run a mix of GPT 6.1 sol on no or low reasoning, and GLM/Kimi.", "history": [] } ], "review_style": { "value": "Ships without manual review; stops only for high-risk changes", "source": "https://ghuntley.com/rad/", "seen": "2026-03-09", "quote": "Instead of having a manual code review for everything, I just ship it. If something is high enough on the risk matrix, for example, a database schema migration, then it halts the shipping, and I have to do a manual review.", "kind": "spot-checks", "also": [ { "source": "https://ghuntley.com/readable/", "seen": "2026-10-02", "quote": "Software doesn't need to be readable by a human. It needs to be explainable to a human." } ] }, "harness": { "value": "Agent loops, simulator first, agents validate against the simulator", "short": "loops", "source": "https://ghuntley.com/readable/", "seen": "2026-10-02", "quote": "It's a distributed system, so I'm building it in Rust and building the simulator first, then driving the agent to validate everything through the simulator.", "history": [ { "value": "Bash while loop piping PROMPT.md into the agent", "source": "https://ghuntley.com/ralph/", "seen": "2025-07-14", "quote": "while :; do cat PROMPT.md | claude-code ; done" } ] }, "parallelism": null, "rules_files": { "value": "Per-repo AGENTS.md: testing policy, ADRs, code contracts, static types required", "short": "AGENTS.md", "url": "https://github.com/ghuntley/underclass/blob/3e35c3bf3d24c7fa79cb93602b571b1b9c194514/AGENTS.md", "source": "https://github.com/ghuntley/underclass/blob/3e35c3bf3d24c7fa79cb93602b571b1b9c194514/AGENTS.md#L42", "seen": "2026-09-26", "quote": "Use the `code-contracts` skill for every code change and code review." }, "skills": { "value": [ "playwright-validation", "slopdimension-app-development", "slopdimension-jido-agents", "slopdimension-media-streaming", "slopdimension-moderation-security", "slopdimension-realtime-chat", "slopdimension-user-authentication" ], "count": 7, "featured": [ "playwright-validation", "slopdimension-jido-agents" ], "source": "https://github.com/ghuntley/slopdimension/tree/2e8de0c3e60cad04db139585d7ede1449b5b99f8/.agents/skills", "seen": "2026-08-30" }, "mcp_servers": { "value": [ "sequential-thinking", "puppeteer" ], "source": "https://github.com/ghuntley/ghuntley/blob/2ab2f76c29e4423c5fa709d56a21e0cdb9260c52/.mcp.json#L4-L15", "seen": "2025-09-19", "quote": "\"args\": [\"-y\", \"@modelcontextprotocol/server-puppeteer\"]," }, "hooks": null, "signature": { "value": "The Ralph loop", "kind": "habit", "caption": "an agent in a bash while loop", "why": "Our pick: he invented it. One prompt file fed to an agent again and again until the work is done.", "editorial": true, "source": "https://ghuntley.com/ralph/", "seen": "2025-07-14", "quote": "while :; do cat PROMPT.md | claude-code ; done" }, "archetype": { "value": "Vibe Maximalist", "editorial": true, "why": "No hand-written code in two years; ships unreviewed unless the change is high-risk." }, "signature_quote": { "value": "I'm supervising it. I'm on the loop, not in the loop.", "more": "I haven't written code by hand for two years.", "source": "https://ghuntley.com/rad/", "seen": "2026-03-09" } }, { "schema": 2, "handle": "sqs", "status": "ok", "checked": "2026-10-06", "name": { "value": "Quinn Slack", "source": "https://github.com/sqs" }, "x": { "value": "sqs", "source": "https://x.com/sqs" }, "github": { "value": "sqs", "source": "https://github.com/sqs" }, "site": { "value": "https://slack.org", "source": "https://github.com/sqs" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/1976?v=4", "source": "https://github.com/sqs" }, "known_for": { "value": "Co-founded Amp and Sourcegraph", "source": "https://sourcegraph.com/blog/why-sourcegraph-and-amp-are-becoming-independent-companies" }, "agents": [ { "name": "Amp", "icon": "amp", "role": "primary, in cloud orbs", "surface": "web", "surface_source": { "source": "https://ampcode.com/time-capsules/1", "seen": "2026-07-27", "quote": "as recently as like a month ago, I had like five different worktrees on my machine and I had a way to make their dev servers run independently, but it like, it was so messy. Now with orbs, it just works." }, "source": "https://ampcode.com/podcast/season-02/episode-04", "seen": "2026-09-11", "quote": "these days I'm really only using Amp, which is F2, and Chrome, which is F5" } ], "models": [ { "name": "Amp Dial: medium", "icon": "amp", "detail": "high for tricky tasks", "source": "https://ampcode.com/time-capsules/1", "seen": "2026-07-27", "quote": "I'll do it in an orb. I think medium is probably fine for this", "history": [] } ], "review_style": { "value": "Agent reviews first, he scans diffs for red flags", "source": "https://ampcode.com/podcast/season-02/episode-02", "seen": "2026-08-11", "quote": "Yeah. It's tough because I still draw a lot on my background as a developer when I have the agent review the code and it makes, you know, I probably accept two-thirds of its suggestions.", "kind": "agent-reviews", "also": [ { "source": "https://ampcode.com/podcast/season-02/episode-02", "seen": "2026-08-11", "quote": "it's not that I'm reviewing it in the old way where I care about like the totality of the code, but I'm looking for like red flags," }, { "source": "https://ampcode.com/time-capsules/1", "seen": "2026-07-27", "quote": "it's very rare that I will ship something without reviewing the code. I'll do that if it's something that users will not touch, if it's like analysis code, or if it's purely visual" }, { "source": "https://ampcode.com/podcast/season-02/episode-01", "seen": "2026-07-17", "quote": "Actually so that's another thing. We push to main." } ] }, "harness": { "value": "Amp threads in cloud orbs, previews in portals", "short": "orbs", "source": "https://ampcode.com/time-capsules/1", "seen": "2026-07-27", "quote": "as recently as like a month ago, I had like five different worktrees on my machine and I had a way to make their dev servers run independently, but it like, it was so messy. Now with orbs, it just works.", "history": [] }, "parallelism": null, "rules_files": { "value": "AGENTS.md (20 lines) in qdiff", "short": "AGENTS.md", "url": "https://github.com/sqs/qdiff/blob/6864ee32171e2affd27ce86de5835256764a3ca6/AGENTS.md", "source": "https://github.com/sqs/qdiff/blob/6864ee32171e2affd27ce86de5835256764a3ca6/AGENTS.md#L9", "seen": "2025-11-15", "quote": "- **Type**: TUI Git client (Magit clone)." }, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "\"I'm feeling Pucky\" button", "kind": "tool", "caption": "starts half their work", "why": "Our pick: a Slack button that turns a bug report into an agent thread with one click, and he says about half the team's work starts there.", "editorial": true, "source": "https://ampcode.com/podcast/season-02/episode-05", "seen": "2026-09-25", "quote": "It'll show up in our Slack, and we've got a ‘I'm feeling Pucky’ button and a ‘I'm feeling ultra Pucky’ button. And we click, and that kicks off a thread. And that is the start of about half of our work right now, and it has created this virtuous cycle." }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "Ideas and bugs fan out to Amp threads in orbs. He scans the results for red flags." }, "signature_quote": { "value": "the principle here is the agent's time is cheap.", "more": "Your time is sacred. So make it do the work to help you verify.", "source": "https://ampcode.com/time-capsules/1", "seen": "2026-07-27" } }, { "schema": 2, "handle": "thdxr", "status": "ok", "checked": "2026-10-12", "name": { "value": "Dax Raad", "source": "https://thdxr.com/" }, "x": { "value": "thdxr", "source": "https://x.com/thdxr" }, "github": { "value": "thdxr", "source": "https://github.com/thdxr" }, "site": { "value": "https://thdxr.com", "source": "https://github.com/thdxr" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/826656?v=4", "source": "https://github.com/thdxr" }, "known_for": { "value": "OpenCode co-founder, SST", "source": "https://newsletter.pragmaticengineer.com/p/opencode" }, "agents": [ { "name": "OpenCode", "icon": "opencode", "role": "primary, his own", "surface": null, "history": [ { "value": "avante.nvim in Neovim", "surface": "ide", "source": "https://github.com/thdxr/environment/blob/68c088040593f1edc5656de902052d4ce1d26c1a/home/.config/nvim/lua/plugins/avante.lua#L2", "seen": "2025-02-03", "quote": "\"yetone/avante.nvim\"," } ], "source": "https://x.com/thdxr/status/2087348700624412836", "seen": "2026-08-12", "quote": "one of my favorite flows in opencode2 is to say \"in the background do ....\"" } ], "models": [ { "name": "Astra", "icon": "openai", "detail": "main session", "history": [ { "value": "Claude Sonnet 4", "source": "https://github.com/thdxr/environment/blob/f13cb5601f0489ccba0df9ce229ba60f6d989159/home/.config/opencode/config.json#L8", "seen": "2025-12-06", "quote": "\"model\": \"anthropic/claude-sonnet-4-20250514\"," }, { "value": "Claude Opus 4.5", "source": "https://github.com/thdxr/environment/blob/8cbed4f5cb0427e67e77ac0f72ccd184defdcc6c/home/.config/opencode/config.json#L6", "seen": "2026-07-02", "quote": "\"claude-opus-4-5\": {" } ], "source": "https://x.com/thdxr/status/2100802591508537777", "seen": "2026-09-18", "quote": "so i switched to astra and told it to spawn fable subagents to do the work" }, { "name": "Fable", "icon": "claude", "detail": "subagents", "history": [], "source": "https://x.com/thdxr/status/2100802591508537777", "seen": "2026-09-18", "quote": "so i switched to astra and told it to spawn fable subagents to do the work" } ], "review_style": { "value": "Asks a smarter model to review, via subagents", "source": "https://x.com/thdxr/status/2100251713525498366", "seen": "2026-09-16", "quote": "very useful, i use it to get reviews from smarter models", "kind": "agent-reviews", "also": [ { "source": "https://x.com/thdxr/status/2100298664677695627", "seen": "2026-09-16", "quote": "so i got the review and it had a good idea so i was able to tell it to continue with the implementation in that subagent" }, { "source": "https://x.com/thdxr/status/2081488387060424967", "seen": "2026-07-26", "quote": "i am way more accepting of slop on the frontend vs the backend" }, { "source": "https://kentcdodds.com/chats/07/01/product-sense-restraint-and-opencode-with-dax-raad", "seen": "2026-04-01", "quote": "we're pretty reserved with how much we just let them go do their thing." } ] }, "harness": { "value": "OpenCode on an always-on server, driven from the OpenCode iPhone app; a worktree per task", "short": "phone + server", "source": "https://x.com/thdxr/status/2107581093838832003", "seen": "2026-10-06", "quote": "this past week most of my work has been done from the opencode iphone app", "history": [ { "value": "OpenCode 2, a git worktree per task, voice input", "source": "https://x.com/thdxr/status/2087649518498062656", "seen": "2026-08-12", "quote": "i often tell opencode \"lets review this pr in a new worktree\" or \"make a new worktree and implement x\"" } ] }, "parallelism": null, "rules_files": { "value": "Global AGENTS.md (private) + OpenCode repo AGENTS.md", "short": "AGENTS.md", "url": "https://github.com/anomalyco/opencode/blob/3f393d78bfc3f0826b2c7080e57964c235704695/AGENTS.md", "source": "https://x.com/thdxr/status/2087649518498062656", "seen": "2026-08-12", "quote": "so i added instructions in my global AGENTS.md telling opencode that whenever it makes a worktree, it should move the session into it" }, "skills": null, "mcp_servers": { "value": [ "honeycomb", "planetscale" ], "source": "https://x.com/thdxr/status/2095410973662118093", "seen": "2026-09-03", "quote": "i analyzed a weeks worth of data via honeycomb + planetscale mcp and opencode2's codemode" }, "hooks": null, "signature": { "value": "Voice prompting (Hex)", "kind": "habit", "caption": "instead of typing", "why": "He dictates prompts with fast local voice models and says voice, not OpenCode, is what replaced his editor.", "editorial": true, "source": "https://x.com/thdxr/status/2078527322563449007", "seen": "2026-07-18", "quote": "the thing that killed my editor for me was fast local voice models more than OpenCode" }, "archetype": { "value": "Purist", "editorial": true, "why": "Vanilla OpenCode, no skill files: he says custom setups fix problems that no longer exist." }, "signature_quote": { "value": "when i see people with custom workflows and setups they're all addressing problems that don't exist anymore", "more": "the person naively using vanilla codex is more likely to be experiencing state of the art", "source": "https://x.com/thdxr/status/2107260844400931156", "seen": "2026-10-06" } }, { "schema": 2, "handle": "nickbaumann98", "status": "ok", "checked": "2026-10-12", "name": { "value": "Nick Baumann", "source": "https://github.com/nickbaumann98" }, "x": { "value": "nickbaumann_", "source": "https://x.com/nickbaumann_" }, "github": { "value": "nickbaumann98", "source": "https://github.com/nickbaumann98" }, "site": null, "avatar": { "value": "https://avatars.githubusercontent.com/u/163209607?v=4", "source": "https://github.com/nickbaumann98" }, "known_for": { "value": "Codex at OpenAI, ex-Cline", "source": "https://x.com/nickbaumann_/status/2016358142419075453" }, "agents": [ { "name": "Codex", "icon": "codex", "role": "primary, Codex app", "surface": "app", "surface_source": { "source": "https://x.com/nickbaumann_/status/2022019219588886739", "seen": "2026-02-12", "quote": "1. Codex App: I now run 3-6 threads in parallel, usually on xhigh, and just jump in when input is needed." }, "source": "https://x.com/nickbaumann_/status/2060480729830789376", "seen": "2026-05-29", "quote": "This has fundamentally changed how I use Codex", "history": [ { "value": "Cline", "surface": "ide", "surface_source": { "source": "https://github.com/nickbaumann98/ai-stuff/blob/ffda131f2b1fc1800302c38dad9e662e35b7ab21/blogs/build-anything-you-dont-know-how-to-build-my-experience-with-cline.md#L34", "seen": "2025-03-10", "quote": "It even walked me through testing the VS Code extension to verify everything worked" }, "source": "https://github.com/nickbaumann98/ai-stuff/blob/ffda131f2b1fc1800302c38dad9e662e35b7ab21/blogs/build-anything-you-dont-know-how-to-build-my-experience-with-cline.md#L9", "seen": "2025-03-10", "quote": "I've been using Cline since it was released, and it's completely transformed how I approach building software." } ] } ], "models": [ { "name": "GPT-6 Astra", "icon": "openai", "source": "https://x.com/nickbaumann_/status/2101756295577849940", "seen": "2026-09-20", "quote": "The degree to which Astra is making my life easier at work right now is hard to understate", "inferred": true, "history": [ { "value": "GPT-5.6 Sol", "source": "https://x.com/nickbaumann_/status/2080034767135965342", "seen": "2026-07-22", "quote": "I had GPT-5.6-sol put all the best Codex + ChatGPT Work use cases from X (& prompts!) into a Site™️" } ] }, { "name": "GPT-5.3-Codex-Spark", "icon": "openai", "detail": "when steering", "source": "https://x.com/nickbaumann_/status/2022019219588886739", "seen": "2026-02-12", "quote": "2. GPT-5.3-Codex-Spark when I am in the loop, this is the model I want.", "history": [] } ], "review_style": null, "harness": { "value": "Codex app: one teammate thread, subagent threads, thread automations", "short": "monothread", "source": "https://x.com/nickbaumann_/status/2044847066728579493", "seen": "2026-04-16", "quote": "I run this as one main teammate thread plus a few long-lived subagent threads.", "history": [ { "value": "3-6 parallel threads on xhigh", "source": "https://x.com/nickbaumann_/status/2022019219588886739", "seen": "2026-02-12", "quote": "1. Codex App: I now run 3-6 threads in parallel, usually on xhigh, and just jump in when input is needed." } ] }, "parallelism": { "value": 6, "short": "3–6 threads", "source": "https://x.com/nickbaumann_/status/2022019219588886739", "seen": "2026-02-12", "quote": "1. Codex App: I now run 3-6 threads in parallel, usually on xhigh, and just jump in when input is needed.", "history": [], "inferred": true }, "rules_files": null, "skills": { "value": [ "cli-creator" ], "count": 1, "featured": [ "cli-creator" ], "source": "https://x.com/nickbaumann_/status/2042705384306336083", "seen": "2026-04-10" }, "mcp_servers": { "value": [ "slack", "linear", "sentry" ], "source": "https://x.com/nickbaumann_/status/2042705384306336083", "seen": "2026-04-10", "quote": "A connector or MCP server is great for access. I use Slack, Linear, and Sentry this way." }, "hooks": null, "signature": { "value": "\"Chief of staff\" thread", "kind": "habit", "caption": "spawns every project thread", "why": "Our pick: one persistent Codex thread runs his work. It starts each new project thread and checks in on them.", "editorial": true, "source": "https://x.com/nickbaumann_/status/2060480729830789376", "seen": "2026-05-29", "quote": "- everything runs out of a single persistent thread (my \"chief of staff\")" }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "A chief-of-staff thread starts project threads and keeps them on track." }, "signature_quote": { "value": "I do not care how long it takes if it's not blocking me.", "more": "And in fact, if it takes longer but I get more intelligence, I'd prefer that.", "source": "https://x.com/nickbaumann_/status/2022019219588886739", "seen": "2026-02-12" } }, { "schema": 2, "handle": "adamwathan", "status": "ok", "checked": "2026-10-07", "name": { "value": "Adam Wathan", "source": "https://github.com/adamwathan" }, "x": { "value": "adamwathan", "source": "https://x.com/adamwathan" }, "github": { "value": "adamwathan", "source": "https://github.com/adamwathan" }, "site": { "value": "https://adamwathan.me", "source": "https://adamwathan.me" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/4323180?v=4", "source": "https://github.com/adamwathan" }, "known_for": { "value": "Tailwind CSS, Refactoring UI", "source": "https://github.com/adamwathan" }, "agents": [ { "name": "Amp", "icon": "amp", "role": "newest, in cloud Orbs", "surface": "web", "surface_source": { "source": "https://x.com/adamwathan/status/2105058269332987954", "seen": "2026-09-29", "quote": "With orbs it just worked, no configuration, it just figured out how to create an environment where everything ran and I could preview my changes effortlessly.", "inferred": true }, "source": "https://x.com/adamwathan/status/2105058269332987954", "seen": "2026-09-29", "quote": "I shipped a change to a PHP 5 + Node 8 project the other night that would’ve been a nightmare without Amp and Orbs.", "history": [] }, { "name": "Codex", "icon": "codex", "role": "in the ChatGPT mobile app", "surface": "app", "source": "https://x.com/adamwathan/status/2055037124173263269", "seen": "2026-05-14", "quote": "I must be going crazy because I’ve been using Codex in the ChatGPT mobile app for literally months?", "history": [] }, { "name": "Claude Code", "icon": "claudecode", "role": "async OSS chores", "surface": "web", "source": "https://x.com/adamwathan/status/2017616165455949851", "seen": "2026-01-31", "quote": "Man don't sleep on Claude Code in the desktop app/on the web.", "history": [ { "value": "Claude Code", "surface": "cli", "source": "https://x.com/adamwathan/status/1961749358233235694", "seen": "2025-08-30", "quote": "It’s made me extremely bullish on the CLI in general — so much better for AI than trying to point and click around graphical interfaces." } ] } ], "models": [ { "name": "Claude Sonnet 4.5", "icon": "claude", "source": "https://x.com/adamwathan/status/1977443454029602860", "seen": "2025-10-12", "quote": "Sonnet 4.5 one-shotted this refactoring by the way, which actually blows my mind.", "history": [], "inferred": true } ], "review_style": null, "harness": { "value": "Cloud agents: Amp threads in remote Orbs", "short": "cloud Orbs", "source": "https://x.com/adamwathan/status/2105058269332987954", "seen": "2026-09-29", "quote": "With orbs it just worked, no configuration, it just figured out how to create an environment where everything ran and I could preview my changes effortlessly.", "history": [ { "value": "Claude Code in the cloud, in the background", "source": "https://x.com/adamwathan/status/2007635885991358918", "seen": "2026-01-04", "quote": "Fairly confident this initial solution is perfect, did it in 5 minutes in the background using just the initial prompt:" } ] }, "parallelism": null, "rules_files": { "value": "AGENTS.md", "short": "AGENTS.md", "url": null, "source": "https://x.com/adamwathan/status/2041673683987493302", "seen": "2026-04-08", "quote": "Never, ever, under any circumstances, ever, not once, no matter what, try to start the fucking dev server, it’s already fucking running." }, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "Every tedious OSS task", "kind": "habit", "caption": "fed to cloud Claude Code", "why": "Our pick: since January 2026 he hands the Tailwind chores he would otherwise neglect to Claude Code on the web and keeps working; ten of his 2026 Tailwind commits carry its session links.", "editorial": true, "source": "https://x.com/adamwathan/status/2017616165455949851", "seen": "2026-01-31", "quote": "I feed this thing every little random tedious OSS task that I'd otherwise neglect on my TODO list and it just churns through it all asynchronously flawlessly while I work on more important stuff." }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "He hands chores to cloud agents (Claude Code on the web, Amp in Orbs) and works on other things while they run." }, "signature_quote": { "value": "Threads are the new PR, and everyone on the team should be able to see them.", "source": "https://x.com/adamwathan/status/2105007944924606471", "seen": "2026-09-29" } }, { "schema": 2, "handle": "antfu", "status": "ok", "checked": "2026-10-07", "name": { "value": "Anthony Fu", "source": "https://github.com/antfu" }, "x": { "value": "antfu7", "source": "https://antfu.me" }, "github": { "value": "antfu", "source": "https://github.com/antfu" }, "site": { "value": "https://antfu.me", "source": "https://github.com/antfu" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/11247099?v=4", "source": "https://github.com/antfu" }, "known_for": { "value": "Vitest, Slidev, UnoCSS, VueUse", "source": "https://antfu.me" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary", "surface": null, "source": "https://bsky.app/profile/antfu.me/post/3mdhpogkfb22u", "seen": "2026-01-28", "quote": "I am trying both Cursor and Claude Code", "also": [ { "source": "https://github.com/antfu/lz-string-es/commit/b309faf8deb37c9b9fb3cb4507965818ca312f07", "seen": "2026-10-01", "quote": "Co-Authored-By: Claude Sonnet 5.5 ", "inferred": true } ], "history": [ { "value": "Rarely used agents", "surface": null, "source": "https://bsky.app/profile/antfu.me/post/3lk5uhjwxu22o", "seen": "2025-03-12", "quote": "However, I don't use AI agents a lot, so I am not sure what the best direction to work on is." }, { "value": "Gave agents another try", "surface": null, "source": "https://bsky.app/profile/antfu.me/post/3mdhoqd7hds2u", "seen": "2026-01-28", "quote": "late to the party. I have finally been convinced by multiple awesome developers to give agents another try." } ] }, { "name": "Cursor", "icon": "cursor", "role": "tried alongside Claude Code", "surface": "ide", "surface_source": { "source": "https://bsky.app/profile/antfu.me/post/3ll6nlhpcrs24", "seen": "2025-03-25", "quote": "I have a few projects that consistently freeze upon opening, on the latest Cursor. I have to switch back to VS Code for those projects." }, "source": "https://bsky.app/profile/antfu.me/post/3mdhpogkfb22u", "seen": "2026-01-28", "quote": "I am trying both Cursor and Claude Code" } ], "models": [ { "name": "Claude Sonnet 5.5", "icon": "claude", "source": "https://github.com/antfu/lz-string-es/commit/b309faf8deb37c9b9fb3cb4507965818ca312f07", "seen": "2026-10-01", "quote": "Co-Authored-By: Claude Sonnet 5.5 ", "inferred": true, "history": [ { "value": "Claude Sonnet 5", "source": "https://github.com/antfu/pulls.review/commit/6cf84ccf17087ceeba09e441d1821865a6a16cca", "seen": "2026-09-18", "quote": "Co-Authored-By: Claude Sonnet 5 " } ] }, { "name": "Claude Opus 5", "icon": "claude", "detail": "1M context", "source": "https://github.com/antfu/pnpm-workspace-utils/commit/d335a9798c6f99c7ad5b096f4e02730bc3aeaf3f", "seen": "2026-09-02", "quote": "Co-Authored-By: Claude Opus 5 (1M context) ", "inferred": true, "history": [ { "value": "Claude Opus 4.7 (1M context)", "source": "https://github.com/antfu/skills/commit/50deaeb269d80d92db7a2c5a677290309ae307fc", "seen": "2026-05-01", "quote": "Co-Authored-By: Claude Opus 4.7 (1M context) " } ] } ], "review_style": { "value": "Watches the result: Storybook for agent UI, a pulls.review summary on every PR, then the final diff", "source": "https://bsky.app/profile/antfu.me/post/3mp6aklae6c25", "seen": "2026-06-26", "quote": "And I think you should always keep an eye on the result, as it can build something like \"runs like a car\" but nothing like a car.", "kind": "unknown", "inferred": true, "also": [ { "source": "https://bsky.app/profile/antfu.me/post/3mvh4q3lvbc2v", "seen": "2026-09-14", "quote": "@storybook.js.org has been a great companion for working with agents, reviewing the UI components agents made, and making sure components stay pure and stateless." }, { "source": "https://github.com/antfu/pulls.review/pull/69#issuecomment-6009407374", "seen": "2026-10-06", "quote": "💭 analyzed by `vercel-ai-gateway/anthropic/claude-sonnet-5`" }, { "source": "https://bsky.app/profile/antfu.me/post/3mx4t6lhqkk2a", "seen": "2026-10-05", "quote": "In the end final diff matters the most" } ] }, "harness": { "value": "Storybook stories as the agent's feedback loop, run in Vitest", "short": "Storybook", "source": "https://bsky.app/profile/antfu.me/post/3mvh4q3lvbc2v", "seen": "2026-09-14", "quote": "(I have to admit that, as a human, I was too lazy to write stories, but with agents it actually creates a pretty good feedback loop)", "history": [] }, "parallelism": null, "rules_files": { "value": "CLAUDE.md (= AGENTS.md) as a map to .agents/ docs, in pulls.review", "short": "CLAUDE.md", "url": "https://github.com/antfu/pulls.review/blob/ebf205d10d3de934199c233af61a6d773275c684/CLAUDE.md", "source": "https://github.com/antfu/pulls.review/blob/ebf205d10d3de934199c233af61a6d773275c684/CLAUDE.md#L3-L4", "seen": "2026-09-15", "quote": "`.agents/` describes what the code does and why. It is a map, not a standard:\nwhere the two disagree, the packages' `src/` and test suites win, and the docs get fixed." }, "skills": { "value": [ "agent-browser", "antfu", "antfu-create-pr", "code-review", "codebase-design", "create-adaptable-composable", "dry-refactoring", "frontend-design", "grill-me", "grilling", "handoff", "i-have-adhd", "improve", "nuxt-modules", "research", "simple-english", "tdd", "teach", "to-questionnaire", "unslop", "vue-best-practices", "vue-debug-guides", "vue-pinia-best-practices", "vue-router-best-practices", "vue-testing-best-practices", "writing-for-agents" ], "count": 26, "featured": [ "antfu", "antfu-create-pr", "unslop" ], "source": "https://github.com/antfu/pulls.review/blob/c3508e8bd2c074748a7aa0a04200e6259bcde0ec/skills-lock.json", "seen": "2026-09-30" }, "mcp_servers": null, "hooks": null, "signature": { "value": "antfubot", "kind": "habit", "caption": "agents ship PRs as a bot", "why": "Our pick: his agents work under their own GitHub account and open PRs he reviews and merges, so agent work is always labelled.", "editorial": true, "source": "https://github.com/antfubot", "seen": null, "undated": true, "quote": "Agents on behave of @antfu" }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Builds his own agent tooling: a bot account, pulls.review, skills-npm and his own skills." }, "signature_quote": { "value": "my role has shifted from \"builder\" to \"sculptor\", which is a completely different job.", "more": "I hated it at the very beginning tbh, but I'm getting better at it slowly", "source": "https://bsky.app/profile/antfu.me/post/3mp62qi4tx226", "seen": "2026-06-26" } }, { "schema": 2, "handle": "awesomekling", "status": "ok", "checked": "2026-10-06", "name": { "value": "Andreas Kling", "source": "https://github.com/awesomekling" }, "x": { "value": "awesomekling", "source": "https://x.com/awesomekling" }, "github": { "value": "awesomekling", "source": "https://github.com/awesomekling" }, "site": { "value": "https://awesomekling.com", "source": "https://x.com/awesomekling" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/5954907?v=4", "source": "https://github.com/awesomekling" }, "known_for": { "value": "Ladybird browser, SerenityOS", "source": "https://awesomekling.com/" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary", "surface": null, "source": "https://ladybird.org/posts/adopting-rust/", "seen": "2026-02-23", "quote": "I used Claude Code and Codex for the translation. This was human-directed, not autonomous code generation.", "also": [ { "source": "https://github.com/LadybirdBrowser/ladybird/commit/9a444ea267bf692b1df509e37f716e09eab61cb5", "seen": "2026-08-25", "quote": "Claude-Session: https://claude.ai/code/session_01VD8PSyM2THMZcaarAnpwp5" }, { "source": "https://x.com/awesomekling/status/2102091773220663325", "seen": "2026-09-21", "quote": "The /btw command is amazing for this! Lets you poke your head in whenever you want without sidetracking the main thread 😄", "inferred": true }, { "source": "https://github.com/LadybirdBrowser/ladybird/pull/7331", "seen": "2026-01-05", "quote": "I asked Claude to write a bunch of tests to get proper coverage of our core JS engine optimizations, and this exposed two issues which are fixed herein." } ] }, { "name": "Codex", "icon": "codex", "role": "also used", "surface": null, "source": "https://ladybird.org/posts/adopting-rust/", "seen": "2026-02-23", "quote": "I used Claude Code and Codex for the translation.", "also": [ { "source": "https://x.com/awesomekling/status/2042551695910048008", "seen": "2026-04-10", "quote": "codex: I'm kicking off a clean build in parallel, it's gonna take some time" } ] } ], "models": [ { "name": "Frontier model", "icon": null, "detail": "maxed-out reasoning, as coordinator", "source": "https://x.com/awesomekling/status/2102089363631059199", "seen": "2026-09-21", "quote": "2. Hand it to a coordinator agent (frontier model with maxed-out reasoning)", "also": [ { "source": "https://x.com/awesomekling/status/2102098281203433738", "seen": "2026-09-21", "quote": "Using slightly weaker models with medium reasoning for workers. To keep token spend under control 😅" } ], "history": [] } ], "review_style": { "value": "Adversarial review passes by different models", "source": "https://ladybird.org/posts/adopting-rust/", "seen": "2026-02-23", "quote": "After the initial translation, I ran multiple passes of adversarial review, asking different models to analyze the code for mistakes and bad patterns.", "kind": "agent-reviews", "also": [ { "source": "https://x.com/awesomekling/status/2105814628785934386", "seen": "2026-10-02", "quote": "Bargaining:\n- Okay, AI can write code, but only if a human reviews every line.\n- Maybe adversarial AI review will make arbitrary external code intake safe again." }, { "source": "https://x.com/awesomekling/status/2105815446180233555", "seen": "2026-10-02", "quote": "(This was roughly my own thought process over the last ~12 months. Your journey may be different!)" }, { "source": "https://ladybird.org/posts/adopting-rust/", "seen": "2026-02-23", "quote": "Beyond the test suites, I’ve done extensive testing by browsing the web in a lockstep mode where both the C++ and Rust pipelines run simultaneously, verifying that output is identical for every piece of JavaScript that flows through them." } ] }, "harness": { "value": "Design doc, then a coordinator agent fanning work out to 4 computers over SSH, with welfare checks", "short": "SSH fleet", "source": "https://x.com/awesomekling/status/2102089363631059199", "seen": "2026-09-21", "quote": "3. Tell coordinator to parallelize the work over 4 separate computers via SSH (because I work on a browser and rebuilds/tests are very CPU+RAM intensive)", "history": [ { "value": "Hundreds of small prompts, steering by hand", "source": "https://ladybird.org/posts/adopting-rust/", "seen": "2026-02-23", "quote": "It was hundreds of small prompts, steering the agents where things needed to go." } ] }, "parallelism": { "value": 4, "short": "4 computers", "source": "https://x.com/awesomekling/status/2102089363631059199", "seen": "2026-09-21", "quote": "3. Tell coordinator to parallelize the work over 4 separate computers via SSH (because I work on a browser and rebuilds/tests are very CPU+RAM intensive)", "history": [] }, "rules_files": null, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "Welfare checks", "kind": "habit", "caption": "on workers across 4 machines", "why": "Our pick: the coordinator, not he, watches for stuck or crash-looping workers, so multi-day runs need no steering.", "editorial": true, "source": "https://x.com/awesomekling/status/2102089363631059199", "seen": "2026-09-21", "quote": "4. Tell coordinator to do regular welfare checks on all workers in case someone gets stuck/crash-loops/etc." }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "Writes the design doc, then lets a coordinator agent run workers on four machines for days." }, "signature_quote": { "value": "I don't really steer. I just check in every now and then to see how things are going.", "more": "Mostly because I get curious.", "source": "https://x.com/awesomekling/status/2102089363631059199", "seen": "2026-09-21" } }, { "schema": 2, "handle": "bcantrill", "status": "ok", "checked": "2026-10-07", "name": { "value": "Bryan Cantrill", "source": "https://github.com/bcantrill" }, "x": null, "github": { "value": "bcantrill", "source": "https://github.com/bcantrill" }, "site": { "value": "https://bcantrill.dtrace.org", "source": "https://github.com/bcantrill" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/328614?v=4", "source": "https://github.com/bcantrill" }, "known_for": { "value": "DTrace, Oxide co-founder and CTO", "source": "https://bsky.app/profile/bcantrill.bsky.social" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary, kernel work and old-code restoration", "surface": "cli", "inferred": true, "surface_source": { "source": "https://bcantrill.dtrace.org/2026/05/25/a-portentous-reunion/", "seen": "2026-05-25", "quote": "Searched for 4 patterns, read 3 files (ctrl+o to expand)" }, "source": "https://bcantrill.dtrace.org/2026/05/25/a-portentous-reunion/", "seen": "2026-05-25", "quote": "I picked it up, and tasked Claude to port it to Linux.", "also": [ { "source": "https://oxide-and-friends.transistor.fm/episodes/engineering-rigor-in-the-llm-age/transcript", "seen": "2026-01-15", "quote": "So I have been exploring, using Claude code to do kernel work, to do, we've got our host operating system is Helios." }, { "source": "https://bsky.app/profile/bcantrill.bsky.social/post/3mtyz3z6cv223", "seen": "2026-08-26", "quote": "fascinating to watch Claude (correctly) do the archaeology." } ], "history": [] }, { "name": "ChatGPT", "icon": "openai", "role": "second opinion when Claude loops", "surface": null, "source": "https://oxide-and-friends.transistor.fm/episodes/this-old-repo-llms-and-the-restoration-of-battletris/transcript", "seen": "2026-06-09", "quote": "And I think I actually went to chat GPT to get its ideas.", "history": [] } ], "models": [ { "name": "Claude Opus 4.8", "icon": "claude", "detail": "1M context", "source": "https://github.com/bcantrill/BattleTris/commit/35e5f85291b1f3b1f8ce7277324401f4177a5e04", "seen": "2026-06-01", "quote": "Co-Authored-By: Claude Opus 4.8 (1M context) ", "inferred": true, "history": [ { "value": "Claude Opus 4.7 (1M context)", "source": "https://github.com/bcantrill/BattleTris/commit/356c47906f8868da217feff1ee787c9d9d4b8f3e", "seen": "2026-05-24", "quote": "Co-Authored-By: Claude Opus 4.7 (1M context) " } ] } ], "review_style": { "value": "Self-review first: nobody else reviews LLM code until the responsible engineer has", "source": "https://rfd.shared.oxide.computer/rfd/0576", "seen": "2026-08-24", "quote": "LLM-generated code should not be reviewed by others if the responsible engineer has not themselves reviewed it.", "kind": "reads-all", "inferred": true, "also": [ { "source": "https://oxide-and-friends.transistor.fm/episodes/this-old-repo-llms-and-the-restoration-of-battletris/transcript", "seen": "2026-06-09", "quote": "Like you look at the diff and you're like, this diff is like a very reviewable diff." }, { "source": "https://oxide-and-friends.transistor.fm/episodes/engineering-rigor-in-the-llm-age/transcript", "seen": "2026-01-15", "quote": "And then obviously validating all of its work" } ] }, "harness": { "value": "Claude Code on the source tree, kept on a short leash: no closed loop, no one-shotting, and in his first kernel task no builds", "short": "no closed loop", "source": "https://oxide-and-friends.transistor.fm/episodes/engineering-rigor-in-the-llm-age/transcript", "seen": "2026-01-15", "quote": "I'm definitely not closing the loop, not vibe coding it, not one shotting it, but really because in particular, like, I am not we're not even I'm not even gonna let it build anything.", "history": [] }, "parallelism": null, "rules_files": null, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "This Old Repo", "kind": "habit", "caption": "reviving a 1990s game", "why": "Our pick: he pointed Claude at BattleTris, a game he started in the mid-1990s that had not built in over 20 years. Claude found the 64-bit stack smash that kept it from running. It is his case for agents doing tedious, open-ended restoration nobody would finish by hand.", "editorial": true, "source": "https://bcantrill.dtrace.org/2026/05/25/a-portentous-reunion/", "seen": "2026-05-25", "quote": "It is no exaggeration to say that this was brought to us by Claude." }, "archetype": { "value": "Purist", "editorial": true, "why": "Steers Claude closely on kernel code and old repos, and requires self-review before anyone else sees LLM code." }, "signature_quote": { "value": "Left unchecked, LLMs will make systems larger, not better", "more": "— appealing to perverse vanity metrics, perhaps, but at the cost of everything that matters.", "source": "https://bcantrill.dtrace.org/2026/04/12/the-peril-of-laziness-lost/", "seen": "2026-04-12" } }, { "schema": 2, "handle": "birgitta410", "status": "ok", "checked": "2026-10-07", "name": { "value": "Birgitta Böckeler", "source": "https://martinfowler.com/articles/sensors-for-coding-agents.html" }, "x": { "value": "birgitta410", "source": "https://x.com/birgitta410" }, "github": { "value": "birgitta410", "source": "https://github.com/birgitta410" }, "site": { "value": "https://birgitta.info", "source": "https://github.com/birgitta410" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/3180887?v=4", "source": "https://github.com/birgitta410" }, "known_for": { "value": "Thoughtworks, harness engineering", "source": "https://birgitta.info" }, "agents": [ { "name": "Cursor", "icon": "cursor", "role": "primary, most used", "surface": "ide", "surface_source": { "source": "https://bsky.app/profile/birgitta410.bsky.social/post/3lilfrljlgc2a", "seen": "2025-02-20", "quote": "I personally prefer in-IDE tools, but the concepts are reusable." }, "source": "https://martinfowler.com/articles/sensors-for-coding-agents.html", "seen": "2026-05-27", "quote": "Throughout building the application, I used a mix of Cursor, Claude Code, and OpenCode (in that order of frequency).", "history": [] }, { "name": "Claude Code", "icon": "claudecode", "role": "second, after Cursor", "surface": null, "source": "https://martinfowler.com/articles/sensors-for-coding-agents.html", "seen": "2026-05-27", "quote": "Throughout building the application, I used a mix of Cursor, Claude Code, and OpenCode (in that order of frequency).", "history": [] }, { "name": "Pi", "icon": "pi", "role": "local models, evals, her own extension", "surface": "cli", "surface_source": { "source": "https://github.com/birgitta410/local-coding-evals/blob/b4e438166659996f7519d23427b7be44a2606f23/README.md#L60", "seen": "2026-06-20", "quote": "Before you try running a task, start `pi` directly and check if all the configuration is working." }, "source": "https://martinfowler.com/articles/sensors-for-coding-agents.html", "seen": "2026-07-13", "quote": "I tried this by building an extension for (and with) Pi.", "also": [ { "source": "https://github.com/birgitta410/local-coding-evals/blob/b4e438166659996f7519d23427b7be44a2606f23/README.md#L5", "seen": "2026-06-20", "quote": "- Uses Pi as the coding agent" }, { "source": "https://martinfowler.com/articles/exploring-gen-ai/local-models-for-coding-experiences.html", "seen": "2026-07-08", "quote": "OpenCode or Pi as coding harnesses" } ], "history": [] } ], "models": [ { "name": "Claude Sonnet", "icon": "claude", "detail": "default; Sonnet 4.6 in TDD runs", "source": "https://martinfowler.com/articles/sensors-for-coding-agents.html", "seen": "2026-05-27", "quote": "My default model was usually Claude Sonnet, for some of the planning and analysis tasks I used Claude Opus, and for implementation tasks I frequently used Cursor's composer-2 model.", "also": [ { "source": "https://martinfowler.com/articles/exploring-gen-ai/tdd-in-the-agent-loop.html", "seen": "2026-08-10", "quote": "Model: I used Sonnet 4.6 to generate the solutions." } ], "history": [] }, { "name": "Claude Opus 4.8", "icon": "claude", "detail": "planning, analysis, judging", "source": "https://martinfowler.com/articles/exploring-gen-ai/tdd-in-the-agent-loop.html", "seen": "2026-08-10", "quote": "Opus 4.8 compared the quality of both solutions and their tests, without knowledge of how the solutions were created.", "history": [ { "value": "Claude Opus 4.7", "source": "https://martinfowler.com/articles/sensors-for-coding-agents.html", "seen": "2026-05-20", "quote": "Here are some highlights from the results (model used was Claude Opus 4.7, same as for the coupling analysis):" } ] } ], "review_style": { "value": "Risk triage: probability, impact and detectability set how much she reads; sensors point her to where", "source": "https://se-radio.net/2026/07/se-radio-730-birgitta-boeckeler-on-harness-engineering-for-ai-agents/", "seen": "2026-07-22", "quote": "So, harnesses for me are all about improving how as a human I can prioritize where to put my review.", "kind": "spot-checks", "also": [ { "source": "https://martinfowler.com/articles/exploring-gen-ai/to-vibe-or-not-vibe.html", "seen": "2025-09-23", "quote": "Reflecting on these 3 dimensions helps me decide if I should reach for AI or not, if I should review the code or not, and at what level of detail I do that review." }, { "source": "https://martinfowler.com/articles/exploring-gen-ai/to-vibe-or-not-vibe.html", "seen": "2025-09-23", "quote": "Low probability + low impact + high detectability Vibe coding is fine! As long as things work and I achieve my goal, I don’t review the code at all." }, { "source": "https://martinfowler.com/articles/sensors-for-coding-agents.html", "seen": "2026-05-27", "quote": "Looking at the exceptions AI created (suppressed warnings, increased thresholds) was a good point to start my code review." }, { "source": "https://martinfowler.com/articles/exploring-gen-ai/local-models-for-coding-experiences.html", "seen": "2026-07-08", "quote": "And the temptation to just check out and not understand the results anymore is a lot lower, I went back to reviewing results much more - and I mean that in a good way." } ] }, "harness": { "value": "Sensors sidecar running lint, types, tests, coverage and mutation testing next to the agent; weekly review skills", "short": "sensors sidecar", "source": "https://martinfowler.com/articles/sensors-for-coding-agents.html", "seen": "2026-07-13", "quote": "In all of the experiments described, I used a vibe-coded little “sidecar” application that could run all of my computational sensors.", "also": [ { "source": "https://se-radio.net/2026/07/se-radio-730-birgitta-boeckeler-on-harness-engineering-for-ai-agents/", "seen": "2026-07-22", "quote": "And I just go in there and I currently manually just run those once a week and see if there’s anything new." } ], "history": [] }, "parallelism": null, "rules_files": { "value": "AGENTS.md and CLAUDE.md (identical): use uv, check the sensors", "short": "AGENTS.md", "url": "https://github.com/birgitta410/sensors-cli/blob/b0ecac7e102720967a8e5815653ec20a857f4b1f/AGENTS.md", "source": "https://github.com/birgitta410/sensors-cli/blob/b0ecac7e102720967a8e5815653ec20a857f4b1f/AGENTS.md#L5", "seen": "2026-06-11", "quote": "Use `sensors check .` regularly to get information about the code and test quality of your changes." }, "skills": { "value": [ "_local-setup", "_new-parser", "sensors_config-default", "sensors_config-python", "sensors_config-typescript", "sensors_wrap-tool" ], "count": 6, "featured": [ "sensors_config-typescript", "sensors_wrap-tool", "_new-parser" ], "source": "https://github.com/birgitta410/sensors-cli/tree/9778caaf6ffdf51fa6bac48de09a0a421a0d0ae2/.claude/skills", "seen": "2026-06-15" }, "mcp_servers": { "value": [ "playwright" ], "source": "https://github.com/birgitta410/local-coding-evals/blob/cb0a90ce986d727ff07543592beabcecb978baaf/.mcp.json#L6", "seen": "2026-06-18", "quote": "\"@playwright/mcp@latest\"" }, "hooks": { "value": [ "gitleaks pre-commit (git hook)" ], "source": "https://martinfowler.com/articles/sensors-for-coding-agents.html", "seen": "2026-05-27", "quote": "GitLeaks runs as part of the pre-commit hook, I consider it to be a sensor as well, as it will give the agent feedback when it tries to commit (computational)" }, "signature": { "value": "sensors check", "kind": "tool", "caption": "a sidecar the agent polls", "why": "Our pick: she built a CLI that keeps linters, tests, coverage and mutation testing running beside the agent and hands it a short status with fix guidance, so it can self-correct before she reviews.", "editorial": true, "source": "https://martinfowler.com/articles/sensors-for-coding-agents.html", "seen": "2026-07-13", "quote": "The CLI can offer both situational awareness for the human in supervised sessions, and a token-efficient and guidance-enriched summary to the agent." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Builds her own harness pieces: a sensors CLI, skills for it, an eval runner and a Pi extension." }, "signature_quote": { "value": "I personally have stopped telling my coding agents to write tests first, let alone do TDD (which I never did, to be honest), until I see evals or other strong arguments that convince me otherwise.", "source": "https://martinfowler.com/articles/exploring-gen-ai/tdd-in-the-agent-loop.html", "seen": "2026-08-10" } }, { "schema": 2, "handle": "burkeholland", "status": "ok", "checked": "2026-10-12", "name": { "value": "Burke Holland", "source": "https://github.com/burkeholland" }, "x": { "value": "burkeholland", "source": "https://x.com/burkeholland" }, "github": { "value": "burkeholland", "source": "https://github.com/burkeholland" }, "site": { "value": "https://burkeholland.github.io", "source": "https://github.com/burkeholland" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/686963?v=4", "source": "https://github.com/burkeholland" }, "known_for": { "value": "Works on GitHub Copilot", "source": "https://x.com/burkeholland" }, "agents": [ { "name": "GitHub Copilot", "icon": "githubcopilot", "role": "primary, Copilot app", "surface": "app", "surface_source": { "source": "https://github.blog/ai-and-ml/github-copilot/the-harness-is-all-you-need-mostly/", "seen": "2026-07-27", "quote": "For this demonstration, I’ll be using the new GitHub Copilot app." }, "source": "https://x.com/burkeholland/status/2105459495833063754", "seen": "2026-10-01", "quote": "I've moved on to just assigning each project a lead in @GitHub Copilot and making everything their problem.", "history": [ { "value": "GitHub Copilot in VS Code", "surface": "ide", "source": "https://burkeholland.github.io/posts/opus-4-5-change-everything/", "seen": "2026-01-05", "quote": "I used GitHub Copilot in VS Code with a custom agent prompt that you’ll find toward the end of this post." } ] }, { "name": "Claude Code", "icon": "claudecode", "role": "on side projects", "surface": "web", "inferred": true, "source": "https://github.com/burkeholland/convert-me/pull/1", "seen": "2026-10-11", "quote": "Co-Authored-By: Claude Opus 5.5 (1M context) ", "history": [] } ], "models": [ { "name": "Luna Max", "icon": "openai", "detail": "cost default", "source": "https://x.com/burkeholland/status/2100234706977423643", "seen": "2026-09-16", "quote": "You can do 90% of everything with just Luna and its so dirt cheap that it's almost free.", "inferred": true, "history": [ { "value": "Claude Opus 4.5", "source": "https://burkeholland.github.io/posts/opus-4-5-change-everything/", "seen": "2026-01-05", "quote": "Today, I think that AI coding agents can absolutely replace developers. And the reason that I believe this is Claude Opus 4.5." }, { "value": "GPT 5.6 Terra", "source": "https://github.blog/ai-and-ml/github-copilot/the-harness-is-all-you-need-mostly/", "seen": "2026-07-27", "quote": "For instance, since I was using GPT 5.6 Terra, it requested a review from Sonnet." } ] }, { "name": "Claude Opus 5.5", "icon": "claude", "detail": "for UI work", "source": "https://x.com/burkeholland/status/2104721615774949621", "seen": "2026-09-28", "quote": "We're doing Opus 5.5 things all day.", "inferred": true, "also": [ { "source": "https://x.com/burkeholland/status/2102501095318639025", "seen": "2026-09-22", "quote": "Unless you need to do something that has a UI. Then you are definitely going to need at least Opus." } ], "history": [] } ], "review_style": { "value": "Rubber duck review by a second model family; he no longer reads the code", "source": "https://github.blog/ai-and-ml/github-copilot/the-harness-is-all-you-need-mostly/", "seen": "2026-07-27", "quote": "In a Rubber Duck review, GitHub Copilot will request a review from a model of a different AI family.", "kind": "agent-reviews", "also": [ { "source": "https://x.com/burkeholland/status/2105459495833063754", "seen": "2026-10-01", "quote": "Y'all are reading code." }, { "source": "https://github.blog/ai-and-ml/github-copilot/the-harness-is-all-you-need-mostly/", "seen": "2026-07-27", "quote": "The most important thing is not to settle for AI output that is “good enough.” Insist on quality." } ] }, "harness": { "value": "GitHub Copilot app: a lead agent per project runs sessions, PRs, merges and worktrees", "short": "project leads", "source": "https://x.com/burkeholland/status/2105459495833063754", "seen": "2026-10-01", "quote": "Lead triages the backlog, spin up sessions, testing, PR, merge, deployment and cleaning up worktrees.", "history": [ { "value": "VS Code custom agent that always delegates to subagents", "source": "https://burkeholland.github.io/posts/opus-4-5-change-everything/", "seen": "2026-01-05", "quote": "ALWAYS use #runSubagent." }, { "value": "YOLO mode in a sandbox, prototype, /plan, Autopilot", "source": "https://github.blog/ai-and-ml/github-copilot/the-harness-is-all-you-need-mostly/", "seen": "2026-07-27", "quote": "When using YOLO mode, you don’t want to run the agent on your local machine." } ] }, "parallelism": null, "rules_files": { "value": "'LLM AI coding agent' custom agent prompt (VS Code)", "short": "custom agent", "url": "https://burkeholland.github.io/posts/opus-4-5-change-everything/", "source": "https://burkeholland.github.io/posts/opus-4-5-change-everything/", "seen": "2026-01-05", "quote": "You are an AI-first software engineer. Assume all code will be written and maintained by LLMs, not humans." }, "skills": { "value": [ "issue", "postrboard" ], "count": 2, "featured": [ "postrboard", "issue" ], "source": "https://github.com/burkeholland/loadout/tree/4e2c2ff4fbe7c99a0e5c27220f60e27bc3f9ed3a/skills", "seen": "2026-08-10" }, "mcp_servers": { "value": [ "context7" ], "source": "https://burkeholland.github.io/posts/opus-4-5-change-everything/", "seen": "2026-01-05", "quote": "Context7 was the only MCP I used." }, "hooks": null, "signature": { "value": "A lead agent per project", "kind": "habit", "caption": "he just files issues", "why": "Our pick: each project gets a standing Copilot lead that triages, tests, merges and cleans up, so his input is the issue tracker.", "editorial": true, "source": "https://x.com/burkeholland/status/2105459495833063754", "seen": "2026-10-01", "quote": "I've moved on to just assigning each project a lead in @GitHub Copilot and making everything their problem." }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "Each project has a Copilot lead agent that runs the sessions; he files the issues." }, "signature_quote": { "value": "I work with AI every single day, and what I’m finding is that less is way more.", "more": "It’s not about what I install or configure or trick the agent into doing that makes any real difference.", "source": "https://github.blog/ai-and-ml/github-copilot/the-harness-is-all-you-need-mostly/", "seen": "2026-07-27" } }, { "schema": 2, "handle": "burntsushi", "status": "ok", "checked": "2026-10-07", "name": { "value": "Andrew Gallant", "source": "https://github.com/BurntSushi" }, "x": { "value": "burntsushi5", "source": "https://github.com/BurntSushi" }, "github": { "value": "BurntSushi", "source": "https://github.com/BurntSushi" }, "site": { "value": "https://burntsushi.net", "source": "https://github.com/BurntSushi" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/456674?v=4", "source": "https://github.com/BurntSushi" }, "known_for": { "value": "ripgrep, Rust regex, Jiff", "source": "https://github.com/BurntSushi" }, "agents": [ { "name": "Codex", "icon": "codex", "role": "primary", "surface": "cli", "surface_source": { "source": "https://github.com/BurntSushi/dotfiles/blob/1dab3ffec0570fdb2d82ea8411ab07c8ae7f6d95/.codex/config.toml#L10-L13", "seen": "2026-07-30", "quote": "[tui]\nnotifications = [\"agent-turn-complete\", \"approval-requested\"]\nnotification_condition = \"unfocused\"\nnotification_method = \"auto\"", "inferred": true }, "source": "https://github.com/BurntSushi/ripgrep/pull/3420#issuecomment-4615548108", "seen": "2026-06-03", "quote": "I got this benchmark from Codex that I think I believe.", "also": [ { "source": "https://github.com/astral-sh/ruff/pull/26771#discussion_r3579340701", "seen": "2026-07-14", "quote": "I removed it. It looks like a hallucination from Codex." } ], "history": [ { "value": "No Claude", "surface": null, "source": "https://bsky.app/profile/burntsushi.net/post/3llqrfh7rfc2q", "seen": "2025-04-01", "quote": "Tried what? Claude? No." }, { "value": "Claude", "surface": null, "source": "https://github.com/astral-sh/ruff/pull/23566#issuecomment-3961690466", "seen": "2026-02-25", "quote": "(Claude helped me a fair bit with it.)" } ] } ], "models": [ { "name": "GPT-5.6 Sol", "icon": "openai", "detail": "max reasoning", "source": "https://github.com/BurntSushi/dotfiles/blob/1dab3ffec0570fdb2d82ea8411ab07c8ae7f6d95/.codex/config.toml#L3-L8", "seen": "2026-07-30", "quote": "model_reasoning_effort = \"max\"\nnotify = [\n \"sh\", \"-c\",\n \"A bart 'Codex' 'Done or waiting for input'\",\n]\nmodel = \"gpt-5.6-sol\"", "history": [ { "value": "GPT-5.5, xhigh reasoning", "source": "https://github.com/BurntSushi/dotfiles/blob/a81b0683f98b017640b4733ab3535a1876e7e3c4/.codex/config.toml#L3-L8", "seen": "2026-07-02", "quote": "model = \"gpt-5.5\"" } ] } ], "review_style": { "value": "Flags his LLM PRs for careful review, then simplifies them by hand; Codex review catches bugs too", "kind": "unknown", "inferred": true, "source": "https://github.com/rust-lang/regex/pull/1364#issuecomment-4642193394", "seen": "2026-06-07", "quote": "I wrote this PR with an LLM, so this still needs careful review. From a glance, the solution here seems very over complicated. But it does pass tests and rebar benchmarks.", "also": [ { "source": "https://github.com/rust-lang/regex/pull/1364#issuecomment-4981702691", "seen": "2026-07-15", "quote": "I need to do another round of review, but I significantly simplified the original LLM fix for this." }, { "source": "https://github.com/rust-lang/regex/commit/fa91c31a4291c9dda6afe19829e6fe2e3bbc2da5", "seen": "2026-07-15", "quote": "automata: fix a bug caught by Codex review" } ] }, "harness": { "value": "Codex in its workspace-write sandbox, approvals on request", "short": "sandbox", "source": "https://github.com/BurntSushi/dotfiles/blob/1dab3ffec0570fdb2d82ea8411ab07c8ae7f6d95/.codex/config.toml#L1-L2", "seen": "2026-07-30", "quote": "approval_policy = \"on-request\"\nsandbox_mode = \"workspace-write\"", "history": [] }, "parallelism": null, "rules_files": null, "skills": null, "mcp_servers": null, "hooks": { "value": [ "Codex notify: his own `A` alert script (desktop notification and sound, or a push to his phone when away)" ], "source": "https://github.com/BurntSushi/dotfiles/blob/1dab3ffec0570fdb2d82ea8411ab07c8ae7f6d95/.codex/config.toml#L4-L7", "seen": "2026-07-30", "quote": "notify = [\n \"sh\", \"-c\",\n \"A bart 'Codex' 'Done or waiting for input'\",\n]" }, "signature": { "value": "\"I wrote this PR with an LLM\"", "kind": "habit", "caption": "flagged for careful review", "why": "Our pick: he says up front which PRs an LLM wrote, reviews them as untrusted and pares them down himself.", "editorial": true, "source": "https://github.com/rust-lang/regex/pull/1364#issuecomment-4642193394", "seen": "2026-06-07", "quote": "I wrote this PR with an LLM, so this still needs careful review." }, "archetype": { "value": "Purist", "editorial": true, "why": "Lets Codex write code in a sandbox, but discloses it, reviews it and simplifies it by hand." }, "signature_quote": { "value": "I don't mind code changes being done with an LLM, but please don't use LLMs to communicate with humans.", "source": "https://github.com/rust-lang/regex/pull/1368#issuecomment-4836313646", "seen": "2026-06-29" } }, { "schema": 2, "handle": "calvinfo", "status": "ok", "checked": "2026-10-06", "name": { "value": "Calvin French-Owen", "source": "https://github.com/calvinfo" }, "x": { "value": "calvinfo", "source": "https://x.com/calvinfo" }, "github": { "value": "calvinfo", "source": "https://github.com/calvinfo" }, "site": { "value": "https://calv.info", "source": "https://github.com/calvinfo" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/487539?v=4", "source": "https://github.com/calvinfo" }, "known_for": { "value": "Segment co-founder, ex-Codex", "source": "https://x.com/calvinfo" }, "agents": [ { "name": "Codex", "icon": "codex", "role": "writes the code", "surface": "cli", "surface_source": { "source": "https://calv.info/agents-feb-2026", "seen": "2026-02-17", "quote": "I have yet to move my workflow over fully from the CLI tools since I appreciate the flexibility of the terminal." }, "source": "https://calv.info/agents-feb-2026", "seen": "2026-02-17", "quote": "Opus is where I start, but Codex writes all my code.", "history": [ { "value": "Cursor (built-in agent and extensions)", "surface": "ide", "source": "https://calv.info/agents-feb-2026", "seen": "2026-02-17", "quote": "This is different from what I was doing 6 months ago, where I was mostly using Cursor and the built-in agent or extensions." } ] }, { "name": "Claude Code", "icon": "claudecode", "role": "plans, orchestrates, git", "surface": "cli", "surface_source": { "source": "https://calv.info/agents-feb-2026", "seen": "2026-02-17", "quote": "The net result between the two is that I'll start with Claude Code and keep that open as a pane, then flip to Codex when I'm ready to actually start the coding." }, "source": "https://calv.info/agents-feb-2026", "seen": "2026-02-17", "quote": "Claude Code is my regular driver for planning, orchestrating my terminal, and managing git/github operations.", "history": [] } ], "models": [ { "name": "Fable 5", "icon": "claude", "source": "https://calv.info/small-models-have-arrived", "seen": "2026-08-26", "quote": "When doing coding work, I almost always reach for the most expensive and capable models (Fable 5, 5.6 Sol).", "history": [ { "value": "Opus", "source": "https://calv.info/agents-feb-2026", "seen": "2026-02-17", "quote": "Opus is where I start, but Codex writes all my code." } ], "inferred": true }, { "name": "GPT-5.6 Sol", "icon": "openai", "source": "https://calv.info/small-models-have-arrived", "seen": "2026-08-26", "quote": "When doing coding work, I almost always reach for the most expensive and capable models (Fable 5, 5.6 Sol).", "history": [ { "value": "GPT-5.3-Codex xhigh or high", "source": "https://calv.info/agents-feb-2026", "seen": "2026-02-17", "quote": "I run on GPT-5.3-Codex-xhigh or high, and the Codex code just straight up has fewer bugs." } ] } ], "review_style": { "value": "Spot-checks the architecture; Cursor Bugbot and Codex review every PR", "kind": "agent-reviews", "source": "https://calv.info/agents-feb-2026", "seen": "2026-02-17", "quote": "I still spot-check the code, and make sure I understand it from an architecture level, but increasingly I am not reading every line of every PR. I rely on agents to spot subtle bugs (they are far better at it than I am).", "also": [ { "source": "https://calv.info/agents-feb-2026", "seen": "2026-02-17", "quote": "Since then, I've settled mostly on Cursor as the default option since you can tell when it's working, though I find the results of Codex are good too." } ] }, "harness": { "value": "Ghostty panes, a git worktree per numbered plan, a preview deploy per branch", "short": "worktrees", "source": "https://calv.info/agents-feb-2026", "seen": "2026-02-17", "quote": "I have a worktree skill which copies any relevant plans, env vars, and other updates, and starts a new branch.", "history": [ { "value": "Claude and Codex instances in tmux", "source": "https://calv.info/agents-feb-2026", "seen": "2026-02-17", "quote": "For a little while I was running multiple Claude/Codex instances in tmux, but now I just use multiple panes within the same terminal tab." } ] }, "parallelism": { "value": 4, "short": "3–4 overnight tasks", "source": "https://calv.info/agents-feb-2026", "seen": "2026-02-17", "quote": "This is over-hyped on Twitter, but I do really try and kick off 3-4 tasks in Codex before going to bed, so they'll be ready for review in the morning.", "history": [], "inferred": true }, "rules_files": { "value": "Shared AGENTS.md / CLAUDE.md in a private 'claudefiles' repo", "short": "AGENTS.md", "url": null, "source": "https://calv.info/agents-feb-2026", "seen": "2026-02-17", "quote": "Today I have a bunch of skills and a shared AGENTS.md/CLAUDE.md defined in a repo I call claudefiles." }, "skills": { "value": [ "commit", "worktree", "implement", "implement-all", "codex-review", "address-bugs", "pr-pass", "focus" ], "count": 8, "featured": [ "implement-all", "pr-pass", "address-bugs" ], "source": "https://calv.info/agents-feb-2026", "seen": "2026-02-17" }, "mcp_servers": null, "hooks": null, "signature": { "value": "/pr-pass", "kind": "skill", "caption": "loops until CI and bots pass", "why": "Our pick: at the end of /implement-all it pushes, waits for CI, fixes what Bugbot and Codex flag, and goes round again, so review comments get handled without him.", "editorial": true, "source": "https://calv.info/agents-feb-2026", "seen": "2026-02-17", "quote": "It essentially 1) pushes to the remote 2) waits for all CI to pass 3) runs /address-bugs and then optionally goes to step 1 until it's finished." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "He built his own chain of skills, from /commit to /pr-pass, one step at a time, and dropped skills other people wrote." }, "signature_quote": { "value": "Opus is where I start, but Codex writes all my code.", "more": "If I was budget constrained, I'd probably pick Codex.", "source": "https://calv.info/agents-feb-2026", "seen": "2026-02-17" } }, { "schema": 2, "handle": "catherinewu", "status": "ok", "checked": "2026-10-07", "name": { "value": "Cat Wu", "source": "https://claude.com/blog/product-management-on-the-ai-exponential" }, "x": { "value": "_catwu", "source": "https://x.com/_catwu" }, "github": { "value": "catherinewu", "source": "https://github.com/catherinewu" }, "site": null, "avatar": { "value": "https://avatars.githubusercontent.com/u/6024687?v=4", "source": "https://github.com/catherinewu" }, "known_for": { "value": "Claude Code product, Anthropic", "source": "https://claude.com/blog/product-management-on-the-ai-exponential" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary, anything whose output is code", "surface": "app", "surface_source": { "source": "https://simonwillison.net/2026/Jul/21/cat-and-thariq/", "seen": "2026-07-21", "quote": "I never have this need, because I just kick off the task directly on mobile and it runs in a cloud session without using my local environment — I think because I’m doing very easy coding tasks." }, "source": "https://claude.com/blog/product-management-on-the-ai-exponential", "seen": "2026-03-19", "quote": "Claude Code is where I build prototypes, evals, and scripts, many of which call Claude API. I use this when the output is code.", "history": [] }, { "name": "Claude Tag", "icon": "claude", "role": "first pass on new features, in Slack", "surface": null, "source": "https://simonwillison.net/2026/Jul/21/cat-and-thariq/", "seen": "2026-07-21", "quote": "Maybe I say, “Hey, I think we should implement this new feature in Cowork,” and I’ll tag in Claude Tag to do a first pass at it." } ], "models": [ { "name": "Claude Opus 5.5", "icon": "claude", "detail": "daily driver", "source": "https://x.com/_catwu/status/2102437713781944397", "seen": "2026-09-22", "quote": "I’ve been using Opus 5.5 as a daily driver and love its clear communication skills and ability to write in my style.", "history": [ { "value": "Claude Sonnet 3.5 (new)", "source": "https://claude.com/blog/product-management-on-the-ai-exponential", "seen": "2026-03-19", "quote": "These projects took hundreds of hours of prompting Claude Code powered by Sonnet 3.5 (new), but not a single line of code written by hand." }, { "value": "Fable 5.1", "source": "https://x.com/_catwu/status/2102443572205437435", "seen": "2026-09-22", "quote": "I've mostly switched from Fable 5.1 to Opus 5.5 because I like it's clear communication." } ] }, { "name": "Claude Fable 5.1", "icon": "claude", "detail": "fire-and-forget coding", "source": "https://x.com/_catwu/status/2102443572205437435", "seen": "2026-09-22", "quote": "I still turn to Fable 5.1 for my \"fire and forget\" coding tasks.", "history": [] } ], "review_style": { "value": "Claude code review clears the outer layers; code owners review the core", "kind": "unknown", "source": "https://simonwillison.net/2026/Jul/21/cat-and-thariq/", "seen": "2026-07-21", "quote": "But increasingly, for the changes at the outer layers, we actually have Claude code review fully review those.", "also": [ { "source": "https://simonwillison.net/2026/Jul/21/cat-and-thariq/", "seen": "2026-07-21", "quote": "For the most critical changes to the core of Claude Code, and the cores of other products, there is always a code owner and they do manually review all the changes." } ], "inferred": true }, "harness": { "value": "Auto mode; tasks kicked off from mobile into cloud sessions", "short": "auto mode", "source": "https://simonwillison.net/2026/Jul/21/cat-and-thariq/", "seen": "2026-07-21", "quote": "But for a lot of us who are on auto mode right now — hopefully you’re not on YOLO mode — I don’t think it actually matters, and we could probably just remove file edit and be totally fine.", "history": [ { "value": "Read every permission prompt (Sonnet 3.7 era)", "source": "https://simonwillison.net/2026/Jul/21/cat-and-thariq/", "seen": "2026-07-21", "quote": "I would read every permission prompt extremely carefully." } ] }, "parallelism": null, "rules_files": null, "skills": { "value": [ "PR commit" ], "count": null, "featured": [ "PR commit" ], "source": "https://every.to/podcast/transcript-how-to-use-claude-code-like-the-people-who-built-it", "seen": "2025-10-29", "quote": "Slash PR commit." }, "mcp_servers": null, "hooks": { "value": [ "permission approval hooks" ], "source": "https://x.com/_catwu/status/2021848561638625556", "seen": "2026-02-12", "quote": "yes! i use it to write my permission approval hooks :)" }, "signature": { "value": "Fresh config per model", "kind": "habit", "caption": "reset at every model launch", "why": "Our pick: she tests each new model on a clean slate instead of carrying old tweaks forward.", "editorial": true, "source": "https://x.com/_catwu/status/2102556970473709717", "seen": "2026-09-23", "quote": "I generally reset my configs for each new model" }, "archetype": { "value": "Purist", "editorial": true, "why": "Resets her configs for each model and leans on built-ins: auto mode, the guide agent, Claude code review." }, "signature_quote": { "value": "I feel like we’ve all gotten a chance to take a step back and delegate a lot more of the menial implementation to Claude.", "more": "It’s freed up a lot of our time to think about more creative work, like: what is the right experience that we should be providing to our users, now that we know Claude Code can implement a lot of it?", "source": "https://simonwillison.net/2026/Jul/21/cat-and-thariq/", "seen": "2026-07-21" } }, { "schema": 2, "handle": "chiphuyen", "status": "ok", "checked": "2026-10-07", "name": { "value": "Chip Huyen", "source": "https://github.com/chiphuyen" }, "x": { "value": "chipro", "source": "https://github.com/chiphuyen" }, "github": { "value": "chiphuyen", "source": "https://github.com/chiphuyen" }, "site": { "value": "https://huyenchip.com", "source": "https://github.com/chiphuyen" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/11997567?v=4", "source": "https://github.com/chiphuyen" }, "known_for": { "value": "AI Engineering author", "source": "https://github.com/chiphuyen/aie-book" }, "agents": [ { "name": "Codex", "icon": "codex", "role": "CLI and desktop app, interchangeably", "surface": null, "source": "https://github.com/hugobowne/show-us-your-agent-skills/blob/2c1f52025a1a993a5ce58c744d6a6e987ff58c2e/episode-field-notes/ep-8/chip.md#L113", "seen": "2026-07-30", "quote": "Desktop usually hides the trace of all the commands that it runs, whereas the CLI is more verbose, but I can see a lot more things at a quick glance." }, { "name": "Claude Code", "icon": "claudecode", "role": "alongside Codex, via her runner", "surface": null, "source": "https://github.com/hugobowne/show-us-your-agent-skills/blob/2c1f52025a1a993a5ce58c744d6a6e987ff58c2e/episode-field-notes/ep-8/chip.md#L117-L119", "seen": "2026-07-30", "quote": "I can just go out, and I can close my laptop, and it works, which is amazing.", "history": [ { "value": "Claude Code, logs analysed with her own Sniffly", "surface": null, "source": "https://x.com/chipro/status/1945527700808184115", "seen": "2025-07-16", "quote": "I open sourced Sniffly, a tool that analyzes Claude Code logs to help me understand my usage patterns and errors." } ] } ], "models": [ { "name": "Claude Fable", "icon": "claude", "detail": "plans and supervises only", "source": "https://www.youtube.com/watch?v=NH-ic7-V-jY&t=3048", "seen": "2026-07-30", "quote": "I feel like it's very wasteful to use uh strong models like fable to do actual implementations.", "history": [] } ], "review_style": { "value": "A second agent reviews the plan; UI work comes with screenshot evidence on the PR", "kind": "agent-reviews", "source": "https://github.com/hugobowne/show-us-your-agent-skills/blob/2c1f52025a1a993a5ce58c744d6a6e987ff58c2e/episode-field-notes/ep-8/chip.md#L50", "seen": "2026-07-30", "quote": "I give it a process: do this, and then get another to review it, and then do post-mortems.", "also": [ { "source": "https://github.com/hugobowne/show-us-your-agent-skills/blob/2c1f52025a1a993a5ce58c744d6a6e987ff58c2e/episode-field-notes/ep-8/chip.md#L66", "seen": "2026-07-30", "quote": "It can show it to me, attach to a PR, so I can approve it without having to spin up the branch associated with that PR." }, { "source": "https://x.com/chipro/status/1945528410291531807", "seen": "2025-07-16", "quote": "Across my projects, the interruption rate is about 1 in 4 instructions. This means I still need to actively monitor the agent, rather than letting it operate autonomously." } ], "inferred": true }, "harness": { "value": "Her own cross-provider runner: a strong agent plans and reviews, cheap agents implement, each in a git worktree", "short": "own runner", "source": "https://github.com/hugobowne/show-us-your-agent-skills/blob/2c1f52025a1a993a5ce58c744d6a6e987ff58c2e/episode-field-notes/ep-8/chip.md#L34", "seen": "2026-07-30", "quote": "I get a very strong agent to make plans and think, and then it assigns tasks to a bunch of cheaper agents to do it.", "history": [] }, "parallelism": { "value": 10, "short": "10 subagents per lead", "source": "https://github.com/hugobowne/show-us-your-agent-skills/blob/2c1f52025a1a993a5ce58c744d6a6e987ff58c2e/episode-field-notes/ep-8/chip.md#L96", "seen": "2026-07-30", "quote": "With my runner, there's no limit, so it can spin up 10 sub-agents.", "history": [ { "value": "20, 30 tasks", "source": "https://home.mlops.community/public/videos/coding-with-ai-chip-huyen-2025-11-21", "seen": "2025-11-21", "quote": "I can do like 20, 30 or like as however much money I want to spend on it." } ] }, "rules_files": null, "skills": { "value": [ "taxonomy labeling", "verified front-end", "write tests", "author skills" ], "count": 4, "featured": [ "verified front-end", "author skills", "write tests" ], "source": "https://github.com/hugobowne/show-us-your-agent-skills/blob/2c1f52025a1a993a5ce58c744d6a6e987ff58c2e/episode-field-notes/ep-8/chip.md#L58-L80", "seen": "2026-07-30" }, "mcp_servers": null, "hooks": null, "signature": { "value": "Cross-provider runner", "kind": "tool", "caption": "Codex, Claude spawn each other", "why": "Our pick: she built her own runner so a strong model plans and reviews while Codex and Claude agents hand work to each other.", "editorial": true, "source": "https://github.com/hugobowne/show-us-your-agent-skills/blob/2c1f52025a1a993a5ce58c744d6a6e987ff58c2e/episode-field-notes/ep-8/chip.md#L38", "seen": "2026-07-30", "quote": "That's why I built my runner, so they can do that in the same place." }, "archetype": { "value": "Swarm Lord", "editorial": true, "why": "A supervising model fans work out to cheap agents; one project logged 1,409 of them." }, "signature_quote": { "value": "A lot of times, the agents make mistakes.", "more": "It's not because the agent is stupid, it's because I give them instructions not good enough.", "source": "https://github.com/hugobowne/show-us-your-agent-skills/blob/2c1f52025a1a993a5ce58c744d6a6e987ff58c2e/episode-field-notes/ep-8/chip.md#L20", "seen": "2026-07-30" } }, { "schema": 2, "handle": "chrismccord", "status": "ok", "checked": "2026-10-06", "name": { "value": "Chris McCord", "source": "https://github.com/chrismccord" }, "x": { "value": "chris_mccord", "source": "https://x.com/chris_mccord" }, "github": { "value": "chrismccord", "source": "https://github.com/chrismccord" }, "site": { "value": "https://chrismccord.com", "source": "https://github.com/chrismccord" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/576796?v=4", "source": "https://github.com/chrismccord" }, "known_for": { "value": "Creator of Phoenix, Phoenix.new", "source": "https://x.com/chris_mccord" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary, one convo per project", "surface": null, "source": "https://x.com/chris_mccord/status/2016576023400181773", "seen": "2026-01-28", "quote": "it still runs precommit for me just fine. I think skills and hooks make sense, but a root ~/.claude/CLAUDE.md with what it should do, and what it should call , ie when it's done work it runs a `notify msg` bash function I have, has been enough for these things.", "history": [ { "value": "Phoenix.new, his own browser-based agent", "surface": "web", "source": "https://fly.io/blog/phoenix-new-the-remote-ai-runtime/", "seen": "2025-06-20", "quote": "I’m already using Phoenix.new to triage `phoenix-core` Github issues and pick problems to solve. I close my laptop, grab a cup of coffee, and wait for a PR to arrive — Phoenix.new knows how PRs work, too." } ] } ], "models": [ { "name": "Claude", "icon": "claude", "source": "https://x.com/chris_mccord/status/2016554559078883510", "seen": "2026-01-28", "quote": "Meanwhile I run one continuous claude convo per project, with the very rare fork.", "history": [ { "value": "Claude 3.7", "source": "https://x.com/chris_mccord/status/1894229609945710798", "seen": "2025-02-25", "quote": "claude 3.7 has not disappointed." } ] } ], "review_style": { "value": "Rules file names the tests to run before claiming a fix", "source": "https://github.com/chrismccord/ekv/blob/becf31e79cbf923c33d4858b6f04ab077f27a5d7/CLAUDE.md#L508-L520", "seen": "2026-09-09", "quote": "Add or update adversarial/distributed tests for every safety bug fix.", "kind": "unknown", "also": [ { "source": "https://github.com/chrismccord/ekv/blob/becf31e79cbf923c33d4858b6f04ab077f27a5d7/CLAUDE.md#L520", "seen": "2026-09-09", "quote": "If a change affects correctness semantics, update Jepsen and the pure Elixir linearizability tests, not just unit tests." } ] }, "harness": { "value": "One continuous Claude session per project, auto-compact, rare forks, no subagents", "short": "one convo", "source": "https://x.com/chris_mccord/status/2016554559078883510", "seen": "2026-01-28", "quote": "I've never used a subagent. Never used MCP. I let it auto compact and continue as necessary.", "history": [] }, "parallelism": null, "rules_files": { "value": "Global ~/.claude/CLAUDE.md (private) plus a per-repo CLAUDE.md (520 lines in ekv)", "short": "CLAUDE.md", "url": "https://github.com/chrismccord/ekv/blob/becf31e79cbf923c33d4858b6f04ab077f27a5d7/CLAUDE.md", "source": "https://github.com/chrismccord/ekv/blob/becf31e79cbf923c33d4858b6f04ab077f27a5d7/CLAUDE.md#L3-L4", "seen": "2026-09-09", "quote": "This file is for AI agents working on EKV from a cold start.\nIt is not user-facing documentation." }, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "`notify msg`", "kind": "rule", "caption": "in CLAUDE.md, not a hook", "why": "Our pick: instead of hooks, one line in his global CLAUDE.md makes Claude ping him when work is done.", "editorial": true, "source": "https://x.com/chris_mccord/status/2016576023400181773", "seen": "2026-01-28", "quote": "a root ~/.claude/CLAUDE.md with what it should do, and what it should call , ie when it's done work it runs a `notify msg` bash function I have, has been enough for these things." }, "archetype": { "value": "Purist", "editorial": true, "why": "One Claude convo per project; no subagents, MCP or hooks." }, "signature_quote": { "value": "Meanwhile I run one continuous claude convo per project, with the very rare fork.", "more": "I've never used a subagent. Never used MCP. I let it auto compact and continue as necessary. \nAnd I have wildly good results ¯\\_(ツ)_/¯", "source": "https://x.com/chris_mccord/status/2016554559078883510", "seen": "2026-01-28" } }, { "schema": 2, "handle": "crawshaw", "status": "ok", "checked": "2026-10-12", "name": { "value": "David Crawshaw", "source": "https://github.com/crawshaw" }, "x": { "value": "davidcrawshaw", "source": "https://github.com/crawshaw" }, "github": { "value": "crawshaw", "source": "https://github.com/crawshaw" }, "site": { "value": "https://crawshaw.io", "source": "https://github.com/crawshaw" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/161319?v=4", "source": "https://github.com/crawshaw" }, "known_for": { "value": "exe.dev, Tailscale co-founder", "source": "https://github.com/crawshaw" }, "agents": [ { "name": "Shelley", "icon": null, "role": "primary, his company's own agent", "surface": "web", "surface_source": { "source": "https://www.heavybit.com/library/podcasts/high-leverage/ep-11-why-agents-need-computers-with-david-crawshaw", "seen": "2026-06-03", "quote": "The VM starts with a coding agent running on it called Shelley, which is one we wrote. It's a web interface is the key thing." }, "source": "https://blog.exe.dev/devtools-must-be-open-source", "seen": "2026-08-02", "quote": "I like this tool, but it has two downsides: first, I like to read my diffs in Shelley with a good UI, not in a terminal.", "history": [ { "value": "Chat LLM in a web browser", "surface": "web", "source": "https://crawshaw.io/blog/programming-with-llms", "seen": "2025-01-06", "quote": "That is why I still use an LLM via a web browser, because I want a blank slate on which to craft a well-contained request." }, { "value": "Sketch (exe.dev's earlier agent)", "surface": null, "source": "https://crawshaw.io/blog/programming-with-agents", "seen": "2025-06-08", "quote": "I implemented the first pass of Github App auth for the hosted sketch.dev using sketch." } ] } ], "models": [ { "name": "OpenAI models", "icon": "openai", "detail": "daily driver", "source": "https://news.ycombinator.com/item?id=49157584", "seen": "2026-08-03", "quote": "I still use oai as my daily but I have been very impressed by qwen/glm/k3/etc when I try them.", "history": [ { "value": "latest Opus", "source": "https://crawshaw.io/blog/eight-more-months-of-agents", "seen": "2026-02-08", "quote": "In February this year, the latest Opus model can write nine tenths of my code." }, { "value": "Claude Opus 4.7", "source": "https://www.heavybit.com/library/podcasts/high-leverage/ep-11-why-agents-need-computers-with-david-crawshaw", "seen": "2026-06-03", "quote": "I very much have to distinguish these days between the Anthropic models, that is Opus, which is brilliant and probably still my preferred coding model." }, { "value": "Codex models", "source": "https://www.heavybit.com/library/podcasts/high-leverage/ep-11-why-agents-need-computers-with-david-crawshaw", "seen": "2026-06-03", "quote": "The Codex models are actually really good these days and I use them sometimes." } ] } ], "review_style": { "value": "Reads agent code before it hits serious systems, trimmed by meat", "source": "https://blog.exe.dev/devtools-must-be-open-source", "seen": "2026-08-02", "quote": "The principle is that while agents write code, I still read it before pushing to our serious systems.", "kind": "spot-checks", "also": [ { "source": "https://blog.exe.dev/devtools-must-be-open-source", "seen": "2026-08-02", "quote": "I almost never need to see the import blocks, or the nil-checks, or the error handling any more, so get it off the screen so I can focus on the meat." }, { "source": "https://crawshaw.io/blog/agent-principal-agent", "seen": "2026-05-07", "quote": "Most importantly, there is no principal-agent problem, because the human driving the machine takes on the responsibility for its actions by owning the deployment." }, { "source": "https://blog.exe.dev/show-dont-tell", "seen": "2026-02-22", "quote": "We read source changes to critical systems very carefully." } ] }, "harness": { "value": "Fresh, unconstrained VM per agent (exe.dev) instead of built-in sandboxes", "short": "fresh VMs", "source": "https://crawshaw.io/blog/eight-more-months-of-agents", "seen": "2026-02-08", "quote": "You have to turn off the sandbox, which means you have to provide your own sandbox. I have tried just about everything and I highly recommend: use a fresh VM.", "history": [ { "value": "Containers per agent (Sketch)", "source": "https://crawshaw.io/blog/programming-with-agents", "seen": "2025-06-08", "quote": "By default sketch creates a little development environment in a container with a copy of the source code and the runner has the ability to extract git commits from the container. This lets you run many simultaneously." } ] }, "parallelism": null, "rules_files": null, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "meat", "kind": "tool", "caption": "strips diffs to what matters", "why": "Our pick: his own LLM tool that cuts imports, nil-checks and error handling out of a diff, so his review covers only the architecture.", "editorial": true, "source": "https://github.com/boldsoftware/meat/blob/f39f41dfe7b5b37a12b35fdfbaecc7e779855bd3/README.md#L5-L10", "seen": "2026-08-03", "quote": "Humans need to review agent-written code in critical systems.\nBut models are good now. You don't need to review for style or nil-checks\nor imports. You need to review concepts, algorithm choices, architecture.\n\nSo meat uses a model to reduce a diff to the important parts.\nIt shows you the meat." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Builds his own agent (Shelley) and his own review tool (meat), and runs agents in fresh VMs." }, "signature_quote": { "value": "The way I work with them is I put them in a box, I give them very specific things and I let them do whatever they want.", "source": "https://www.heavybit.com/library/podcasts/high-leverage/ep-11-why-agents-need-computers-with-david-crawshaw", "seen": "2026-06-03" } }, { "schema": 2, "handle": "cvolawless", "status": "ok", "checked": "2026-10-07", "name": { "value": "Claire Vo", "source": "https://github.com/cvolawless" }, "x": { "value": "clairevo", "source": "https://github.com/cvolawless" }, "github": { "value": "cvolawless", "source": "https://github.com/cvolawless" }, "site": { "value": "https://clairevo.com", "source": "https://github.com/cvolawless" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/2424387?v=4", "source": "https://github.com/cvolawless" }, "known_for": { "value": "ChatPRD founder, How I AI host", "source": "https://clairevo.com" }, "agents": [ { "name": "Codex", "icon": "codex", "role": "primary, desktop app", "surface": "app", "surface_source": { "source": "https://www.lennysnewsletter.com/p/i-left-claude-for-months-opus-55", "seen": "2026-09-22", "quote": "I like the desktop app better." }, "source": "https://www.lennysnewsletter.com/p/i-left-claude-for-months-opus-55", "seen": "2026-09-22", "quote": "I moved most of my daily work to Codex and I didn’t miss it.", "also": [ { "source": "https://www.lennysnewsletter.com/p/i-left-claude-for-months-opus-55", "seen": "2026-09-22", "quote": "I will say I still find myself reaching for codex." } ] }, { "name": "Claude Code", "icon": "claudecode", "role": "PR review, architecture, frontend (back with Opus 5.5)", "surface": null, "inferred": true, "source": "https://www.lennysnewsletter.com/p/i-left-claude-for-months-opus-55", "seen": "2026-09-22", "quote": "But what I've really realized about Opus 5.5 is I've built it now back into my development process,", "history": [ { "value": "Left the Claude harness for months", "surface": null, "source": "https://www.lennysnewsletter.com/p/i-left-claude-for-months-opus-55", "seen": "2026-09-22", "quote": "And I found it so frustrating to work with that I just abandoned the Claude harness entirely." } ] }, { "name": "Cursor", "icon": "cursor", "role": "cloud agents launched by her PR bot", "surface": "web", "source": "https://www.lennysnewsletter.com/p/grok-bot-vs-openclaw-how-i-replaced", "seen": "2026-09-02", "quote": "Why I want to show this one is it can, GrokBot can kick off cursor cloud agent coding jobs.", "history": [ { "value": "Cursor desktop app", "surface": "ide", "source": "https://www.lennysnewsletter.com/p/claude-opus-46-vs-gpt-53-codex-how", "seen": "2026-02-11", "quote": "I'm still a cursor girl." } ] } ], "models": [ { "name": "GPT-6 Astra", "icon": "openai", "detail": "daily driver", "source": "https://www.lennysnewsletter.com/p/gpt-6-astra-is-a-banger-heres-everything", "seen": "2026-09-03", "quote": "This is going to be my daily driver.", "history": [ { "value": "GPT-5.3 Codex, for bugs and architecture", "source": "https://www.lennysnewsletter.com/p/claude-opus-46-vs-gpt-53-codex-how", "seen": "2026-02-11", "quote": "You want Codex catching all your bugs," } ] }, { "name": "Claude Opus 5.5", "icon": "claude", "detail": "day-to-day work, PR review", "source": "https://www.lennysnewsletter.com/p/opus-55-vs-gpt-6-sol-which-model", "seen": "2026-09-22", "quote": "5.5 wins my week.", "history": [ { "value": "Claude Opus 4.6, for product and feature work", "source": "https://www.lennysnewsletter.com/p/claude-opus-46-vs-gpt-53-codex-how", "seen": "2026-02-11", "quote": "you want to use Opus for your product and feature work," } ] } ], "review_style": { "value": "Cross-model review: Opus 5.5 reviews GPT's PRs and vice versa; PRs also get human review", "kind": "agent-reviews", "source": "https://www.lennysnewsletter.com/p/i-left-claude-for-months-opus-55", "seen": "2026-09-22", "quote": "or because now I feel like at least Opus 5.5 is not a bad adversarial reviewer, and so I'm having it review more PRs and more work from GPT and vice versa.", "also": [ { "source": "https://www.lennysnewsletter.com/p/grok-bot-vs-openclaw-how-i-replaced", "seen": "2026-09-02", "quote": "We review them both agentically and as humans." }, { "source": "https://clairevo.com/ai-workflows/open-sourcing-a-pr-review-agent-with-codex", "seen": "2026-08-20", "quote": "We had an agent at ChatPRD that reviews proposed code changes on GitHub." } ] }, "harness": { "value": "Codex desktop app for coding; a Grok Bot fleet (moved from OpenClaw) that can launch Cursor cloud agents", "short": "Codex app + bots", "source": "https://www.lennysnewsletter.com/p/grok-bot-vs-openclaw-how-i-replaced", "seen": "2026-09-02", "quote": "I have killed all my open claws and moved almost entirely to GrokBot.", "history": [] }, "parallelism": null, "rules_files": null, "skills": null, "mcp_servers": { "value": [ "elevenlabs" ], "source": "https://www.lennysnewsletter.com/p/i-left-claude-for-months-opus-55", "seen": "2026-09-22", "quote": "is I've been using the Eleven Labs connector at MCP to cut selfie videos" }, "hooks": null, "signature": { "value": "LGTM, the PR closer", "kind": "tool", "caption": "a bot that clears her PR queue", "why": "Our pick: a bot she built that goes through her PRs every day, merges, closes, rebases and nudges, and hands fixes to Cursor agents.", "editorial": true, "source": "https://www.lennysnewsletter.com/p/grok-bot-vs-openclaw-how-i-replaced", "seen": "2026-09-02", "quote": "So I created Look Good To Me, the PR closer, which every day goes through... My PRs and sees what needs to be merged," }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "Splits work across Codex and Claude and runs a fleet of bots that hand off coding jobs." }, "signature_quote": { "value": "I embrace a token abundance mindset.", "more": "I'm starting to spend a lot of money on models,", "source": "https://www.lennysnewsletter.com/p/claude-opus-46-vs-gpt-53-codex-how", "seen": "2026-02-11" } }, { "schema": 2, "handle": "danluu", "status": "ok", "checked": "2026-10-06", "name": { "value": "Dan Luu", "source": "https://github.com/danluu" }, "x": { "value": "danluu", "source": "https://x.com/danluu" }, "github": { "value": "danluu", "source": "https://github.com/danluu" }, "site": { "value": "https://danluu.com", "source": "https://x.com/danluu" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/157136?v=4", "source": "https://github.com/danluu" }, "known_for": { "value": "danluu.com essays, ex-Centaur", "source": "https://danluu.com/ai-coding/" }, "agents": [ { "name": "Codex", "icon": "codex", "role": "primary", "surface": "cli", "surface_source": { "source": "https://danluu.com/agentic-testing/", "seen": "2026-09-07", "inferred": true, "quote": "my tendency has been to type to agents in a CLI" }, "source": "https://danluu.com/bug-blind/", "seen": "2026-08-30", "quote": "Just like with Google Docs, I consider codex above average in terms of software quality in the space, but even though I haven't been using it for a year, I could easily write 10k words on all the workarounds I've implemented (either by habit or, in some cases, with actual scripts that monitor for broken behavior and then correct it).", "also": [ { "source": "https://x.com/danluu/status/2052924529115324436", "seen": "2026-05-09", "quote": "With a semi-autonomous workflow, I can exhaust a week's quota on the $200/mo codex plan in ~10 minutes of my time." } ], "history": [ { "value": "Codex plugin in VS Code", "surface": "ide", "source": "https://mastodon.social/@danluu/116044833637491088", "seen": "2026-02-10", "quote": "I'm sort of in the stone ages with my setup (codex plugin in vscode)." } ] } ], "models": [ { "name": "GPT-5.6 Sol", "icon": "openai", "detail": null, "source": "https://danluu.com/benchpocalypse/", "seen": "2026-08-17", "quote": "Since my original goal here was to see what happens when you run a current (public) SOTA agent in a loop (GPT-5.6 Sol) without much supervision on a non-trivial code optimization problem without any real supervision", "also": [ { "source": "https://danluu.com/agentic-testing/", "seen": "2026-09-07", "quote": "In the same way that I don't prompt GPT-5.6 the same way I prompted GPT-5.5, I don't think I'd want to use the same skills." } ], "history": [ { "value": "GPT-5.2", "source": "https://mastodon.social/@danluu/116044833637491088", "seen": "2026-02-10", "quote": "I used GPT-5.2 (seems to give you the most quota, and I did this in a way that used a ton of quota; even on the Pro plan, I had to throttle my usage) to generate code which ran on my laptop." }, { "value": "GPT-5.5 xhigh", "source": "https://danluu.com/ai-coding/", "seen": "2026-07-03", "quote": "To start with, I'd been using a lot of GPT-5.5 xhigh when this discussion came up a couple months ago and I benchmarked this thing" } ] } ], "review_style": { "value": "Heavy randomized testing instead of human review", "kind": "tests-only", "source": "https://danluu.com/ai-coding/", "seen": "2026-07-03", "quote": "Personally, I'm very comfortable shipping code without human review because I've seen it done on products that are technically more challenging than most software at most software companies.", "also": [ { "source": "https://danluu.com/ai-coding/", "seen": "2026-07-03", "quote": "Personally, I think this can be effective enough that I'm fairly comfortable trying to ship a large volume of code via a \"software factories\" workflow because I've seen a testing-heavy no-review workflow that results in much higher quality than any review-reliant workflow I've seen or even heard of." }, { "source": "https://x.com/danluu/status/2102807194857337235", "seen": "2026-09-23", "quote": "If you have the budget, false positive rejection is straightforward (ask agent to repro, have agents independently check if the repro is valid)." } ], "inferred": true }, "harness": { "value": "One-off agent-written loop scripts with health checks, on laptops and AWS hosts", "short": "one-off loops", "source": "https://danluu.com/ai-coding/", "seen": "2026-07-03", "quote": "As a result, what I've ended up doing is making one-off loops for tasks that need to run for a while.", "also": [ { "source": "https://mastodon.social/@danluu/117301689409121341", "seen": "2026-09-20", "quote": "In that era, I think I was just managing GPT \"manually\" by having it build map-reduce-like graphs of what I wanted to do, custom for each problem." } ], "history": [ { "value": "Queued copies of one prompt in the Codex VS Code plugin", "source": "https://mastodon.social/@danluu/116044833637491088", "seen": "2026-02-10", "quote": "I wouldn't recommend anyone use this because the plugin is so buggy, but it works ok enough if you want to queue up a bunch of work and occasionally check in on what it's doing." } ] }, "parallelism": { "value": 10, "short": "10 agents", "source": "https://danluu.com/ai-coding/", "seen": "2026-07-03", "quote": "Just for example, one that I've seen come up repeatedly related to the earlier note that, on tasks that I do, adding a clear, strict, instruction in AGENTS.md to not do something fails once every week or two when juggling 10 agents, a failure rate of something like one per 100 rule-agent-days.", "history": [] }, "rules_files": { "value": "AGENTS.md (private)", "short": "AGENTS.md", "url": null, "source": "https://danluu.com/ai-coding/", "seen": "2026-07-03", "quote": "When I offhandedly mentioned this to Yossi Kreinin, he suggested adding a note at the bottom of my AGENTS.md to re-read instruction after compaction." }, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "Named-persona review panels", "kind": "habit", "caption": "with contrarian personas", "why": "Our pick: instead of switching models, he has independent agents review as named engineers (and as himself) plus contrarians, which he says beat the same token budget spent otherwise.", "editorial": true, "source": "https://danluu.com/ai-coding/", "seen": "2026-07-03", "quote": "so I tried diagnosing issues with something like \"use independent agents to review as linus torvalds, kyle kingsbury, marc brooker, tptacek, dan luu, and 4 contrarian personas. have each think for a long time\", with multiple iterations of this (yes, I do find it quite silly to invoke my own name)." }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "Skips off-the-shelf orchestrators and has Codex build custom loops and map-reduce graphs per problem, checked by tests rather than review." }, "signature_quote": { "value": "When I use LLMs I'm generally trying to get as much work as possible out of a unit of my time.", "more": "I often want to type for a few minutes to get something to loop for hours to days.", "source": "https://mastodon.social/@danluu/116514991883198991", "seen": "2026-05-04" } }, { "schema": 2, "handle": "dbreunig", "status": "ok", "checked": "2026-10-07", "name": { "value": "Drew Breunig", "source": "https://github.com/dbreunig" }, "x": { "value": "dbreunig", "source": "https://github.com/dbreunig" }, "github": { "value": "dbreunig", "source": "https://github.com/dbreunig" }, "site": { "value": "https://www.dbreunig.com", "source": "https://github.com/dbreunig" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/81794?v=4", "source": "https://github.com/dbreunig" }, "known_for": { "value": "\"How to Fix Your Context\", Plumb", "source": "https://www.dbreunig.com/2026/07/24/manage-your-agent-s-loadout-with-dr-skill.html" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary", "surface": null, "source": "https://www.dbreunig.com/2026/03/04/the-spec-driven-development-triangle.html", "seen": "2026-03-04", "quote": "When I hit commit in Claude Code, I get a list of decisions and I step through them.", "also": [ { "source": "https://www.dbreunig.com/2026/02/25/two-things-i-believe-about-coding-agents.html", "seen": "2026-02-25", "quote": "I forked StreetPass, fired up Claude Code, and had a working version quickly." }, { "source": "https://www.dbreunig.com/2026/07/24/manage-your-agent-s-loadout-with-dr-skill.html", "seen": "2026-07-24", "quote": "If I run drskill list, I can see my current loadout. Adding --harness claude-code limits it to just what Claude Code sees:" } ], "history": [] } ], "models": [ { "name": "Claude Fable", "icon": "claude", "detail": "to shape the design", "source": "https://www.dbreunig.com/2026/08/23/fable-the-end-of-moore-s-law.html", "seen": "2026-08-23", "quote": "I frequently chat with Fable to interrogate and shape a design, before handing off a brief to GLM.", "history": [] }, { "name": "GLM 5.2", "icon": "zai", "detail": "implements the brief", "source": "https://www.dbreunig.com/2026/08/23/fable-the-end-of-moore-s-law.html", "seen": "2026-08-23", "quote": "I frequently chat with Fable to interrogate and shape a design, before handing off a brief to GLM.", "history": [] } ], "review_style": { "value": "Steps through the decisions Plumb extracts at each commit instead of reading the diff", "source": "https://www.dbreunig.com/2026/03/04/the-spec-driven-development-triangle.html", "seen": "2026-03-04", "quote": "It’s code review, but code review where we capture intent. When I hit commit in Claude Code, I get a list of decisions and I step through them. Sometimes I hit one I don’t like and I stop right there. I reject it, go back, redo it. I like that better than pure code review.", "kind": "agent-reviews", "inferred": true, "also": [] }, "harness": null, "parallelism": null, "rules_files": { "value": "CLAUDE.md (Plumb block)", "short": "CLAUDE.md", "url": "https://github.com/dbreunig/plumb/blob/7c80f6da078e289bcada5ffb5e170af55e93140a/CLAUDE.md", "source": "https://github.com/dbreunig/plumb/blob/7c80f6da078e289bcada5ffb5e170af55e93140a/CLAUDE.md#L17-L18", "seen": "2026-03-02", "quote": "**NEVER approve, reject, or edit decisions on the user's behalf.** This is" }, "skills": { "value": [ "plumb", "scaffold-docs", "jev", "monty-updater" ], "count": 4, "featured": [ "plumb", "scaffold-docs", "jev" ], "source": "https://github.com/dbreunig", "seen": "2026-09-17" }, "mcp_servers": null, "hooks": { "value": [ "Plumb pre-commit decision review", "Plumb post-commit decision log" ], "source": "https://github.com/dbreunig/plumb/blob/76ac6bd222a2065b84ee0dabfb8ee9d109c453d9/README.md#L63-L67", "seen": "2026-09-09", "quote": "2. A pre-commit hook and a post-commit hook." }, "signature": { "value": "Plumb", "kind": "tool", "caption": "reviews decisions at commit", "why": "Our pick: his own tool stops each commit, pulls the decisions out of the diff and agent transcripts, and makes him approve them, keeping spec, tests and code in sync.", "editorial": true, "source": "https://www.dbreunig.com/2026/03/04/the-spec-driven-development-triangle.html", "seen": "2026-03-04", "quote": "It’s code review, but code review where we capture intent." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "He builds his own agent tooling: Plumb for decisions at commit time, drskill to audit his skill loadout." }, "signature_quote": { "value": "I frequently chat with Fable to interrogate and shape a design, before handing off a brief to GLM.", "source": "https://www.dbreunig.com/2026/08/23/fable-the-end-of-moore-s-law.html", "seen": "2026-08-23" } }, { "schema": 2, "handle": "dexhorthy", "status": "ok", "checked": "2026-10-12", "name": { "value": "Dex Horthy", "source": "https://aaif.io/blog/dex-still-thinks-long-running-coding-agents-stink" }, "x": { "value": "dexhorthy", "source": "https://x.com/dexhorthy" }, "github": { "value": "dexhorthy", "source": "https://github.com/dexhorthy" }, "site": { "value": "https://humanlayer.dev", "source": "https://github.com/dexhorthy" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/3730605?v=4", "source": "https://github.com/dexhorthy" }, "known_for": { "value": "HumanLayer CEO, 12-Factor Agents", "source": "https://aaif.io/blog/dex-still-thinks-long-running-coding-agents-stink" }, "agents": [ { "name": "HumanLayer", "icon": null, "role": "primary, his own agentic IDE", "surface": "ide", "surface_source": { "source": "https://x.com/dexhorthy/status/2067286892786454855", "seen": "2026-06-17", "quote": "Today, we’re opening access to HumanLayer - an Agentic IDE, collaboration platform, and building blocks for your software factory." }, "source": "https://x.com/dexhorthy/status/2083582284658245837", "seen": "2026-08-01", "quote": "Humanlayer (cc) QRSPI", "history": [ { "value": "Claude Code in terminal", "surface": "cli", "source": "https://x.com/dexhorthy/status/2083582284658245837", "seen": "2026-08-01", "quote": "Claude code in terminal + jetbrains for debugging" }, { "value": "Cursor", "surface": "ide", "source": "https://x.com/dexhorthy/status/2083582284658245837", "seen": "2026-08-01", "quote": "Cursor autocomplete" } ] }, { "name": "Codex", "icon": "codex", "role": "all coding, runs Sol", "surface": null, "source": "https://x.com/dexhorthy/status/2083582284658245837", "seen": "2026-08-01", "quote": "Took the codex pill, 5.5 for all coding", "also": [ { "source": "https://x.com/dexhorthy/status/2101834905223307483", "seen": "2026-09-21", "quote": "lol i'm out of claude limit and my codex harness has a bug where mcps don't get loaded" } ] }, { "name": "Claude Code", "icon": "claudecode", "role": "under HumanLayer, Fable and Opus", "surface": null, "source": "https://x.com/dexhorthy/status/2102863697945469415", "seen": "2026-09-23", "quote": "For opus 5.5 in humanlayer - Just pick opus as model and if you have latest CC cli installed you’ll get 5.5 (ignore the context window warning)" } ], "models": [ { "name": "GPT-5.6 Sol", "icon": "openai", "detail": "daily driver", "source": "https://x.com/dexhorthy/status/2094228532184637911", "seen": "2026-08-31", "quote": "sol is very good at structural stuff. refactoring, or plumbing through a thick call stack without missing things, in general its just more thorough and technical.", "also": [ { "source": "https://x.com/dexhorthy/status/2094228532184637911", "seen": "2026-08-31", "quote": "overall sol is my daily driver" } ], "history": [ { "value": "Codex 5.5", "source": "https://x.com/dexhorthy/status/2083582284658245837", "seen": "2026-08-01", "quote": "Took the codex pill, 5.5 for all coding" } ] }, { "name": "Claude Fable", "icon": "claude", "detail": "UI and spatial work", "source": "https://x.com/dexhorthy/status/2094228532184637911", "seen": "2026-08-31", "quote": "fable gives me more \"look busy\" vibes and it is much better at a handful of things like spatial reasoning and UI", "history": [] } ], "review_style": { "value": "Still reads the code, one 1k-3k PR at a time", "source": "https://x.com/dexhorthy/status/2083582284658245837", "seen": "2026-08-01", "quote": "Still read the code, still learning", "kind": "reads-all", "also": [ { "source": "https://x.com/dexhorthy/status/2092244659057909765", "seen": "2026-08-25", "quote": "- one 1k-3k PR at a time, feedback and polish and code review as we go" }, { "source": "https://x.com/dexhorthy/status/2092287401901363518", "seen": "2026-08-25", "quote": "Because if you’re not reading the code, then you can’t possibly know how much slop is getting in." }, { "source": "https://x.com/dexhorthy/status/2081224804326309938", "seen": "2026-07-26", "quote": "plan review cannot replace code review, it can only increase the chance that code review is painless and fast" } ] }, "harness": { "value": "HumanLayer QRSPI: research and outline docs, Sol and Fable implement and review overnight, then split into small PRs", "short": "QRSPI", "source": "https://x.com/dexhorthy/status/2092244659057909765", "seen": "2026-08-25", "quote": "- let sol and fable implement / review / argue over night", "history": [ { "value": "4 Claude Codes in tmux worktrees, a 5th merging", "source": "https://x.com/dexhorthy/status/2083582284658245837", "seen": "2026-08-01", "quote": "4 Claude’s in tmux worktrees with a 5th merging every commit into main" }, { "value": "Claude Code research, plan, implement (RPI)", "source": "https://x.com/dexhorthy/status/2083582284658245837", "seen": "2026-08-01", "quote": "Claude code - research, plan, impl" } ] }, "parallelism": null, "rules_files": { "value": "~/.claude/CLAUDE.md in his dotfiles (102 lines, Uncle Bob persona)", "short": "CLAUDE.md", "url": "https://github.com/dexhorthy/dotfiles/blob/12bc0cc708824c565c512f9914e58d376c237321/dotfiles/.claude/CLAUDE.md", "source": "https://github.com/dexhorthy/dotfiles/blob/12bc0cc708824c565c512f9914e58d376c237321/dotfiles/.claude/CLAUDE.md#L75", "seen": "2025-12-27", "quote": "- **DELETE 10% minimum from every file you touch**" }, "skills": { "value": [ "build-iterated-agentic-loop", "design-control-loop", "improve-claude-md", "narrow-react-prop-types", "rpi", "show-me", "visual-pr" ], "count": 7, "featured": [ "show-me", "visual-pr", "improve-claude-md" ], "source": "https://github.com/humanlayer/skills/tree/3b53bed02f89f5d5858b7dc68741b60ad64a2119/plugins", "seen": "2026-10-08" }, "mcp_servers": null, "hooks": null, "signature": { "value": "/show-me", "kind": "skill", "caption": "diagrams, not walls of prose", "why": "Our pick: his own skill makes agents explain plans and diffs as trees, call stacks and diagrams, so he can review them fast.", "editorial": true, "source": "https://www.humanlayer.dev/blog/show-me-skill", "seen": "2026-08-12", "quote": "/show-me prompts the agent to use concise visuals to explain what's happening, instead of walls of prose." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Builds his own agent IDE (HumanLayer) and skills like /show-me, and still reads the code." }, "signature_quote": { "value": "there will always be alpha in reviewing something. Otherwise, you get the same AI slop as everybody else", "more": "that something might not always be the code", "source": "https://x.com/dexhorthy/status/2107158343718723850", "seen": "2026-10-05" } }, { "schema": 2, "handle": "dshipper", "status": "ok", "checked": "2026-10-06", "name": { "value": "Dan Shipper", "source": "https://github.com/dshipper" }, "x": { "value": "danshipper", "source": "https://x.com/danshipper" }, "github": { "value": "dshipper", "source": "https://github.com/dshipper" }, "site": { "value": "https://www.danshipper.com", "source": "https://github.com/dshipper" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/263564?v=4", "source": "https://github.com/dshipper" }, "known_for": { "value": "CEO of Every", "source": "https://x.com/danshipper" }, "agents": [ { "name": "Codex", "icon": "codex", "role": "primary", "surface": "app", "surface_source": { "source": "https://every.to/chain-of-thought/how-gpt-5-6-changes-knowledge-work", "seen": "2026-07-10", "quote": "With GPT-5.6 Sol in the new ChatGPT Work app, formerly known as Codex, the process looks very different." }, "source": "https://x.com/danshipper/status/2102461471716483208", "seen": "2026-09-22", "quote": "It’s my new daily driver in Codex", "also": [ { "source": "https://every.to/chain-of-thought/when-your-vibe-coded-app-goes-viral-and-then-goes-down", "seen": "2026-03-20", "quote": "At 4 a.m. on the day after we launched our agent-native document editor, Proof, I watched yet another Codex agent try to revive our server." } ], "history": [ { "value": "Claude Code", "surface": null, "source": "https://every.to/chain-of-thought/openai-has-some-catching-up-to-do", "seen": "2026-01-16", "quote": "Most of my coding happens in Claude Code now—and I’m not alone." } ] }, { "name": "Claude Code", "icon": "claudecode", "role": "Fable builds", "surface": null, "inferred": true, "source": "https://github.com/dshipper/solace/commit/0cfee9332e55bb45a1d2c9f766b2f63bb914a0d1", "seen": "2026-07-03", "quote": "Co-Authored-By: Claude Fable 5 " } ], "models": [ { "name": "GPT-6 Sol", "icon": "openai", "detail": "daily driver", "source": "https://x.com/danshipper/status/2102461471716483208", "seen": "2026-09-22", "quote": "@OpenAI just dropped GPT-6 Sol. It’s my new daily driver in Codex: not quite Astra, but close enough for much of my everyday work, faster, and 50% cheaper than 5.6 Sol.", "history": [] }, { "name": "Claude Fable", "icon": "claude", "detail": "vibe-coded apps", "source": "https://x.com/danshipper/status/2089121597017759800", "seen": "2026-08-16", "quote": "used fable to vibe code an app that visualizes and groups everyone who applied to come to Thesis", "history": [] } ], "review_style": { "value": "Vibe codes; doesn't read the codebase", "source": "https://every.to/chain-of-thought/when-your-vibe-coded-app-goes-viral-and-then-goes-down", "seen": "2026-03-20", "quote": "And I’ve learned a lot about both building and launching a purely vibe coded app.", "kind": "no-review", "also": [ { "source": "https://every.to/chain-of-thought/when-your-vibe-coded-app-goes-viral-and-then-goes-down", "seen": "2026-03-20", "quote": "I hadn’t slept for almost 24 hours, and all I could do was nervously munch trail mix as Codex investigated yet another bug buried deep in a codebase that I didn’t understand." } ], "inferred": true }, "harness": { "value": "Codex app with computer use, tasks titled by a shared emoji convention", "short": "Codex tasks", "source": "https://x.com/danshipper/status/2095245632533622862", "seen": "2026-09-02", "quote": "Have Codex show you which tabs computer use is running in with a shared task-title convention:", "history": [] }, "parallelism": null, "rules_files": { "value": "AGENTS.md in his open-source Tend repo (34 lines)", "short": "AGENTS.md", "url": "https://github.com/EveryInc/tend/blob/ba6ba0e614f5d1151b75032bd58951a0fcb75067/AGENTS.md", "source": "https://github.com/EveryInc/tend/blob/ba6ba0e614f5d1151b75032bd58951a0fcb75067/AGENTS.md#L3", "seen": "2026-07-06", "quote": "- This protocol governs Codex feed threads." }, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "\"Want me to compound?\"", "kind": "rule", "caption": "asked after every sweep", "why": "Our pick: his AGENTS.md makes the agent offer, after each run, to turn what it learned into a policy change he approves.", "editorial": true, "source": "https://github.com/EveryInc/tend/blob/ba6ba0e614f5d1151b75032bd58951a0fcb75067/AGENTS.md#L24-L25", "seen": "2026-07-06", "quote": "- When a meaningful sweep or refresh reaches the CLI idle handshake, ask the user: `Want me to\n compound what I learned from this sweep?`" }, "archetype": { "value": "Vibe Maximalist", "editorial": true, "why": "Ships a purely vibe-coded app and lets Codex fix bugs in code he hasn't read." }, "signature_quote": { "value": "If you can vibe code it, you can vibe fix it.", "more": "You just might not be able to fix it quickly.", "source": "https://every.to/chain-of-thought/when-your-vibe-coded-app-goes-viral-and-then-goes-down", "seen": "2026-03-20" } }, { "schema": 2, "handle": "embirico", "status": "ok", "checked": "2026-10-07", "name": { "value": "Alexander Embiricos", "source": "https://github.com/embirico" }, "x": { "value": "embirico", "source": "https://x.com/embirico" }, "github": { "value": "embirico", "source": "https://github.com/embirico" }, "site": null, "avatar": { "value": "https://avatars.githubusercontent.com/u/3096358?v=4", "source": "https://github.com/embirico" }, "known_for": { "value": "Codex product lead, OpenAI", "source": "https://x.com/embirico" }, "agents": [ { "name": "Codex", "icon": "codex", "role": "primary, auto-review mode", "surface": "app", "surface_source": { "source": "https://www.youtube.com/watch?v=6OiE0jIY93c&t=57", "seen": "2026-02-09", "quote": "the app has all these great skills and so I could just click into the skills tab" }, "source": "https://x.com/embirico/status/2047461700895662333", "seen": "2026-04-23", "quote": "Love using auto-review. It’s my new default mode", "history": [ { "value": "Codex in VS Code", "surface": "ide", "source": "https://www.youtube.com/watch?v=xeZDHGjG5zM&t=164", "seen": "2026-01-12", "quote": "Uh I happen to use VS Code. So I'll show you Codex in VS Code." } ] } ], "models": [ { "name": "Codex default model", "icon": "openai", "detail": "low reasoning", "source": "https://www.youtube.com/watch?v=6OiE0jIY93c&t=100", "seen": "2026-02-09", "quote": "especially because I'm often using [Codex] on low, which I think is like an under underrated technique.", "history": [], "inferred": true } ], "review_style": { "value": "Reads the plan; Codex reviews nearly every PR", "source": "https://www.youtube.com/watch?v=xeZDHGjG5zM&t=1902", "seen": "2026-01-12", "quote": "Codex code review itself is enabled on pretty much every single repo at the company and reviews like pretty much all PRs.", "kind": "unknown", "inferred": true, "also": [ { "source": "https://www.youtube.com/watch?v=xeZDHGjG5zM&t=1021", "seen": "2026-01-12", "quote": "So, it's really important for me to read these and verify them, make sure that it didn't get that wrong." } ] }, "harness": { "value": "One long-lived Codex thread, kept busy by automations, spawning subagents", "short": "long-lived thread", "source": "https://x.com/embirico/status/2046033614106931414", "seen": "2026-04-20", "quote": "I'm slowly shifting to longer-lived threads. The thread is pretty much always active because I prompted it or an automation pinged it.", "history": [ { "value": "Git worktrees that Codex creates, one Codex per worktree", "source": "https://www.youtube.com/watch?v=xeZDHGjG5zM&t=542", "seen": "2026-01-12", "quote": "So typically the way that I would actually do this is I would just ask Codex to create work trees." } ] }, "parallelism": { "value": 4, "short": "4 attempts", "source": "https://www.youtube.com/watch?v=xeZDHGjG5zM&t=1375", "seen": "2026-01-12", "quote": "So I might kind of throw throw codex at it but I might just do it four times in parallel instead.", "history": [] }, "rules_files": null, "skills": { "value": [ "buildkite" ], "count": 1, "featured": [ "buildkite" ], "source": "https://www.youtube.com/watch?v=6OiE0jIY93c&t=57", "seen": "2026-02-09", "quote": "So so I used a skill to fix the PR failure but in the mean then improved two skills kind of as I went." }, "mcp_servers": null, "hooks": null, "signature": { "value": "\"Using a subagent in parallel, do X\"", "kind": "habit", "caption": "from one always-on thread", "why": "Our pick: instead of opening new threads, he keeps one thread alive and hands every new task to a subagent from it.", "editorial": true, "source": "https://x.com/embirico/status/2046033614106931414", "seen": "2026-04-20", "quote": "Then, when I need anything new, I say \"using a subagent in parallel, do X\"" }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "One long-lived thread farms new work out to subagents while automations keep it running." }, "signature_quote": { "value": "Subagents + steering in Codex is pretty magical.", "more": "I'm slowly shifting to longer-lived threads.", "source": "https://x.com/embirico/status/2046033614106931414", "seen": "2026-04-20" } }, { "schema": 2, "handle": "ezyang", "status": "ok", "checked": "2026-10-06", "name": { "value": "Edward Z. Yang", "source": "https://github.com/ezyang" }, "x": { "value": "ezyang", "source": "https://x.com/ezyang" }, "github": { "value": "ezyang", "source": "https://github.com/ezyang" }, "site": { "value": "https://blog.ezyang.com", "source": "https://x.com/ezyang" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/13564?v=4", "source": "https://github.com/ezyang" }, "known_for": { "value": "PyTorch at Meta, ghstack", "source": "https://x.com/ezyang" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary", "surface": "cli", "surface_source": { "source": "https://github.com/ezyang/fable-nanny/blob/48da0da97c7eefc33c6fcf95c2d23c078819f8e9/README.md#L3", "seen": "2026-08-12", "quote": "Tools + playbook for babysitting a Claude/Fable CLI session running in a tmux" }, "source": "https://github.com/ezyang/piano-pad/blob/fc9c0ab3dc8b5a24528e4e57ace5eb62f2d28472/CLAUDE.md#L10-L11", "seen": "2026-10-06", "quote": "The project is run by three long-lived Claude sessions. The human (the parent)\ngives high-level direction, usually by texting one agent over Remote Control.", "also": [ { "source": "https://blog.ezyang.com/2026/03/ai-assisted-programming-for-spmd-types/", "seen": "2026-03-10", "quote": "I have one window per work tree, each a single tab that is just a regular shell, and then as many tabs as I think I need for concurrent Claude Code instances on that worktree." } ], "history": [ { "value": "Claude Desktop + codemcp (his own MCP server)", "surface": "app", "source": "https://github.com/ezyang/codemcp/blob/5b6726ab8c71ecf84fc20cb351b7bc7d7941bfda/README.md#L3", "seen": "2025-05-06", "quote": "Make Claude Desktop a pair programming assistant by installing codemcp." }, { "value": "Claude Code replaces codemcp", "surface": null, "source": "https://github.com/ezyang/codemcp/blob/1ebdb61755c33a420332e6ff73f9fd1c1254d721/README.md#L1-L2", "seen": "2025-12-25", "quote": "NOTICE: claude code is available with Anthropic's $20/mo subscription, so I consider\ncodemcp fully obsolete." } ] }, { "name": "Codex", "icon": "codex", "role": "on some repos (mergedog)", "surface": null, "source": "https://github.com/ezyang/mergedog/blob/a1ed015cab57b3a7c38d6dd2fea37ec00ed2a1c3/CLAUDE.md#L19", "seen": "2026-05-30", "quote": "Codex shell commands may not have working external internet access, even when the local machine does.", "history": [ { "value": "Codex cloud", "surface": "web", "source": "https://blog.ezyang.com/2025/06/vibe-coding-case-study-scubaduck/", "seen": "2025-06-02", "quote": "ScubaDuck was implemented with OpenAI Codex in the cloud (not the CLI tool)." } ] } ], "models": [ { "name": "Claude Opus 5.5", "icon": "claude", "source": "https://github.com/ezyang/piano-pad/commit/fc9c0ab3dc8b5a24528e4e57ace5eb62f2d28472", "seen": "2026-10-06", "quote": "Co-Authored-By: Claude Opus 5.5 ", "inferred": true, "history": [ { "value": "Claude Opus", "source": "https://blog.ezyang.com/2026/03/ai-assisted-programming-for-spmd-types/", "seen": "2026-03-10", "quote": "Claude Opus is not fast enough for single session interactive." } ] }, { "name": "Claude Fable 5", "icon": "claude", "detail": "via Bedrock", "source": "https://github.com/ezyang/fable-nanny/blob/48da0da97c7eefc33c6fcf95c2d23c078819f8e9/run-fable.sh#L21", "seen": "2026-08-12", "quote": "export ANTHROPIC_MODEL=us.anthropic.claude-fable-5", "history": [] } ], "review_style": { "value": "Reads every line, steers with questions and edits", "source": "https://blog.ezyang.com/2026/03/ai-assisted-programming-for-spmd-types/", "seen": "2026-03-10", "quote": "This is NOT a workflow for letting the AI autonomously write tons of code. You’re going to be reading every single line it produces.", "kind": "spot-checks", "also": [ { "source": "https://blog.ezyang.com/2026/03/read-less-steer-more/", "seen": "2026-03-18", "quote": "If you treat AI generated output as code to read, you have already lost the game." } ], "inferred": true }, "harness": { "value": "iTerm2 + tmux -CC to a remote GPU box, three worktrees, Sapling", "short": "tmux -CC", "source": "https://blog.ezyang.com/2026/03/ai-assisted-programming-for-spmd-types/", "seen": "2026-03-10", "quote": "Because I am typically doing work on a GPU system, these sessions are remote, so I use iTerm2 + tmux -CC (tmux control mode) to conveniently have multiple windows and tabs in native UI.", "history": [] }, "parallelism": { "value": 3, "short": "3 work streams", "source": "https://blog.ezyang.com/2026/03/parallel-agents-heart-sapling/", "seen": "2026-03-13", "quote": "The basic development flow here is I have three work streams going in parallel:", "history": [ { "value": 5, "source": "https://blog.ezyang.com/2025/06/vibe-coding-case-study-scubaduck/", "seen": "2025-06-02", "quote": "I’d sometimes have as many as five instances going" } ] }, "rules_files": { "value": "Per-repo CLAUDE.md / AGENTS.md; piano-pad's assigns roles to three agents", "short": "CLAUDE.md", "url": "https://github.com/ezyang/piano-pad/blob/fc9c0ab3dc8b5a24528e4e57ace5eb62f2d28472/CLAUDE.md", "source": "https://github.com/ezyang/piano-pad/blob/fc9c0ab3dc8b5a24528e4e57ace5eb62f2d28472/CLAUDE.md#L19", "seen": "2026-10-06", "quote": "**Your role is the one whose directory you are running in.** Read your charter" }, "skills": { "value": [ "pt2-triage", "pt2-triage-rubric" ], "count": 2, "featured": [ "pt2-triage" ], "source": "https://github.com/ezyang/pt2-triage/tree/3e1f179198dfd553a922ce371869e2f3ae032e3d/.claude/skills", "seen": "2026-01-20" }, "mcp_servers": null, "hooks": { "value": [ "UserPromptSubmit: small-chunks reminder", "StopFailure: self re-prompt via tmux" ], "source": "https://github.com/ezyang/fable-nanny/blob/48da0da97c7eefc33c6fcf95c2d23c078819f8e9/run-fable.sh#L16-L17", "seen": "2026-08-12", "quote": "# UserPromptSubmit -> chunk-reminder.sh (prevention: small-turns context)\n# StopFailure -> stopfailure-unwedge.sh (recovery: self re-prompt)" }, "signature": { "value": "sl follow / sl adopt", "kind": "tool", "caption": "restack parallel agents", "why": "Our pick: two Sapling aliases let agents in several worktrees amend one shared stack without trampling each other.", "editorial": true, "source": "https://blog.ezyang.com/2026/03/parallel-agents-heart-sapling/", "seen": "2026-03-13", "quote": "I find these two commands cover the majority of situations that show up when you are working with parallel agents!" }, "archetype": { "value": "Purist", "editorial": true, "why": "Reads every line the agent writes and runs no orchestrator, just tabs per worktree." }, "signature_quote": { "value": "If you treat AI generated output as code to read, you have already lost the game.", "more": "Read less, steer more.", "source": "https://blog.ezyang.com/2026/03/read-less-steer-more/", "seen": "2026-03-18" } }, { "schema": 2, "handle": "fabiensanglard", "status": "ok", "checked": "2026-10-06", "name": { "value": "Fabien Sanglard", "source": "https://fabiensanglard.net/" }, "x": { "value": "fabynou", "source": "https://x.com/fabynou" }, "github": { "value": "fabiensanglard", "source": "https://github.com/fabiensanglard" }, "site": { "value": "https://fabiensanglard.net", "source": "https://github.com/fabiensanglard" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/1219472?v=4", "source": "https://github.com/fabiensanglard" }, "known_for": { "value": "Game Engine Black Books", "source": "https://github.com/fabiensanglard/gebbdoom" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "VS Code plugin", "surface": "ide", "inferred": true, "source": "https://fabiensanglard.net/agent.md/index.html", "seen": "2026-08-21", "quote": "In March 2026, I tried to use agentic IDEs like Antigravity and VS Code's Claude Code plugin.", "also": [ { "source": "https://fabiensanglard.net/extinct/index.html", "seen": "2026-07-10", "quote": "Every time I notice something I don't like, I add it to my ~/.gemini/GEMINI.md/~/.claude/CLAUDE.md so the agent can emulate my style." } ] }, { "name": "Antigravity", "icon": null, "role": "agentic IDE", "surface": "ide", "source": "https://fabiensanglard.net/agent.md/index.html", "seen": "2026-08-21", "quote": "In March 2026, I tried to use agentic IDEs like Antigravity and VS Code's Claude Code plugin.", "also": [ { "source": "https://fabiensanglard.net/agent.md/index.html", "seen": "2026-08-21", "quote": "Alternatively, gemini.md/claude.md can be symlinked toward an agent.md to have it active anywhere." } ] } ], "models": [], "review_style": { "value": "Reads all the code; LLM criticizes first", "source": "https://fabiensanglard.net/agent.md/index.html", "seen": "2026-08-21", "quote": "While this \"trick\" has considerably improved the code generated, this is not a magic bullet that lets me avoid reading the code.", "kind": "reads-all", "also": [ { "source": "https://fabiensanglard.net/extinct/index.html", "seen": "2026-07-10", "quote": "As a result, I reduce my velocity by iterating over my PR until it reaches the same level of quality I would have produced \"by hand\"." }, { "source": "https://fabiensanglard.net/extinct/index.html", "seen": "2026-07-10", "quote": "Before sending a PR for review, the first thing I do is ask for the LLM to criticize/look for mistakes." } ] }, "harness": { "value": "Agentic IDE, a fresh session per feature", "short": "session per feature", "source": "https://fabiensanglard.net/agent.md/index.html", "seen": "2026-08-21", "quote": "Keep the context short. This means starting a new session per feature.", "history": [] }, "parallelism": null, "rules_files": { "value": "agent.md, gemini.md/claude.md symlinked to it", "short": "agent.md", "url": "https://fabiensanglard.net/agent.md/index.html", "source": "https://fabiensanglard.net/agent.md/index.html", "seen": "2026-08-21", "quote": "Avoid superlatives and praise. Stop telling me I am absolutely right. Give me the cold hard truth.", "history": [ { "value": "~/.gemini/GEMINI.md and ~/.claude/CLAUDE.md", "source": "https://fabiensanglard.net/extinct/index.html", "seen": "2026-07-10", "quote": "Every time I notice something I don't like, I add it to my ~/.gemini/GEMINI.md/~/.claude/CLAUDE.md so the agent can emulate my style." } ] }, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "\"Reload agent.md\"", "kind": "habit", "caption": "when code quality drops", "why": "Our pick: a one-line fix for context dilution, sent the moment the agent's style slips.", "editorial": true, "source": "https://fabiensanglard.net/agent.md/index.html", "seen": "2026-08-21", "quote": "Explicitly ask the harness to reload agent.md. \"Reload agent.md\" is enough when I see code quality dropping." }, "archetype": { "value": "Rules Lawyer", "editorial": true, "why": "Turns every review comment he repeats into a rule in his published agent.md." }, "signature_quote": { "value": "LLMs constantly hallucinate and cannot be trusted.", "more": "I still have to verify and iterate a lot but now I usually focus on architecture and design instead of code style.", "source": "https://fabiensanglard.net/agent.md/index.html", "seen": "2026-08-21" } }, { "schema": 2, "handle": "filosottile", "status": "ok", "checked": "2026-10-06", "name": { "value": "Filippo Valsorda", "source": "https://github.com/FiloSottile" }, "x": null, "github": { "value": "FiloSottile", "source": "https://github.com/FiloSottile" }, "site": { "value": "https://filippo.io", "source": "https://github.com/FiloSottile" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/1225294?v=4", "source": "https://github.com/FiloSottile" }, "known_for": { "value": "Go cryptography, age, mkcert", "source": "https://filippo.io" }, "agents": [ { "name": "Shelley (exe.dev)", "icon": null, "role": "primary, on its own GitHub account", "surface": "web", "inferred": true, "surface_source": { "source": "https://github.com/boldsoftware/shelley/blob/4f250904e45dab0448b476547736b69bff6d282e/README.md#L3", "seen": "2026-06-30", "quote": "Shelley is a mobile-friendly, web-based, multi-conversation, multi-modal," }, "source": "https://gist.github.com/filippo-agent/4a76063873cdd5221ee0710803e6784b", "seen": "2026-09-27", "quote": "These were very low hanging performance wins with very low complexity/review/maintenance cost\n\nI bet there’s more\n\nSystematically look for more in the whole crypto/… tree", "also": [ { "source": "https://github.com/filippo-agent/alertreplay/commit/dfe1f73c0bf7c84a4cfefa64ca836e01180e3286", "seen": "2026-10-03", "quote": "Co-authored-by: Shelley " }, { "source": "https://bsky.app/profile/filippo.abyssdomain.expert/post/3maqkjb7ujg2l", "seen": "2025-12-24", "quote": "After reading this I gave exe.dev a shot and the combination of cheap VMs, the HTTPS proxy with passkey auth and link sharing, and the built-in LLM agent is... incredible." } ], "history": [ { "value": "Sketch", "surface": null, "inferred": true, "source": "https://github.com/FiloSottile/mostly-harmless/commit/4fcfcf933e0a80f135f1009fa0c4ad87fbfa8d8f", "seen": "2025-06-12", "quote": "Co-Authored-By: sketch " } ] }, { "name": "Claude Code", "icon": "claudecode", "role": "debugging, tools, side projects", "surface": "cli", "surface_source": { "source": "https://bsky.app/profile/filippo.abyssdomain.expert/post/3mioegfis4s2z", "seen": "2026-04-04", "quote": "Oh my god that drove me crazy and eventually I just replaced claude in $PATH with a script that does\n\nexec env -u SSH_AUTH_SOCK ~/.local/bin/claude \"$@\"\n\nbecause after all I don't want the agent to access the ssh-agent." }, "source": "https://bsky.app/profile/filippo.abyssdomain.expert/post/3mrke6zps6c2f", "seen": "2026-07-26", "quote": "*Of course* Claude Code with just WebSearch and WebFetch can fill them out for me from context.", "also": [ { "source": "https://words.filippo.io/claude-debugging/", "seen": "2025-11-01", "quote": "On a whim, I figured I would let Claude Code take a shot while I read emails and resurfaced from hyperfocus." } ] } ], "models": [ { "name": "GPT-6 Astra", "icon": "openai", "detail": "in Shelley", "inferred": true, "source": "https://gist.github.com/filippo-agent/4a76063873cdd5221ee0710803e6784b", "seen": "2026-09-27", "quote": "Model: gpt-6-astra", "history": [] }, { "name": "Claude Opus 4.8", "icon": "claude", "detail": "1M context", "inferred": true, "source": "https://github.com/omissis/go-jsonschema/commit/48ec8fa379592a5e08180d7f5089e946b7ebcf7e", "seen": "2026-07-25", "quote": "Co-authored-by: Claude Opus 4.8 (1M context) ", "history": [ { "value": "Opus 4.1", "source": "https://words.filippo.io/claude-debugging/", "seen": "2025-11-01", "quote": "I started Claude Code v2.0.28 with Opus 4.1 and no system prompts, and gave it the following prompt (typos included):" }, { "value": "Opus 4.5", "source": "https://bsky.app/profile/filippo.abyssdomain.expert/post/3mbhfjrax5a2c", "seen": "2026-01-02", "quote": "I just vibecoded with exe.dev and Opus 4.5 a backoffice for our FIPS 140 validation, with a separate view for the lab (where they can also upload test vectors), public links for clients, and guided scripts for testing." }, { "value": "Opus 4.6", "source": "https://bsky.app/profile/filippo.abyssdomain.expert/post/3mewycudmuc2f", "seen": "2026-02-16", "quote": "Claude Opus 4.6 did write mldsa_test.go in the filippo.io/mldsa API prototype, though." } ] } ], "review_style": { "value": "LLM review before any human review", "kind": "unknown", "source": "https://bsky.app/profile/filippo.abyssdomain.expert/post/3moq6sdq6xk2l", "seen": "2026-06-20", "quote": "Nuclear opinion: I don’t care if you use AI to write code,¹ but making a human review a change before an LLM is disrespectful.", "also": [ { "source": "https://bsky.app/profile/filippo.abyssdomain.expert/post/3moqaqrxpds2d", "seen": "2026-06-20", "quote": "I’m saying if you ask human attention, do an LLM review first." }, { "source": "https://words.filippo.io/claude-debugging/", "seen": "2025-11-01", "quote": "Importantly, there is no need to trust the LLM or review its output when its job is just saving me an hour or two by telling me where the bug is, for me to reason about it and fix it." }, { "source": "https://bsky.app/profile/filippo.abyssdomain.expert/post/3msj4dfbihb2u", "seen": "2026-08-07", "quote": "The level of testing and review they are enabling in the Go cryptography standard library is amazing." } ], "inferred": true }, "harness": { "value": "exe.dev VMs with the built-in agent, a separate agent GitHub account, Jujutsu", "short": "exe.dev VMs", "source": "https://bsky.app/profile/filippo.abyssdomain.expert/post/3maqkjb7ujg2l", "seen": "2025-12-24", "quote": "After reading this I gave exe.dev a shot and the combination of cheap VMs, the HTTPS proxy with passkey auth and link sharing, and the built-in LLM agent is... incredible.", "history": [] }, "parallelism": null, "rules_files": { "value": "AGENTS.md with Go CLI tips (personal, shown as an image)", "short": "AGENTS.md", "url": "https://bsky.app/profile/filippo.abyssdomain.expert/post/3mddfcuuo772w", "source": "https://bsky.app/profile/filippo.abyssdomain.expert/post/3mddfcuuo772w", "seen": "2026-01-26", "quote": "Here are three lines from my AGENTS.md that make agents a lot better with Go." }, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "filippo-agent", "kind": "habit", "caption": "agent's own GitHub account", "why": "Our pick: his agent commits from a separate, low-privilege GitHub account so it never holds his maintainer keys.", "editorial": true, "source": "https://bsky.app/profile/filippo.abyssdomain.expert/post/3mmjslkth422a", "seen": "2026-05-23", "quote": "I don’t want to “install Claude” with all the privileges of my account, I want Claude to have its own account to which I will grant some permissions, or none at all because again issues and PRs are public!" }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Fences his agent into its own account and VMs, then has it build tools like muzoo, goapi and gh-pr-rebase." }, "signature_quote": { "value": "Why am I catching your bugs if a computer could have done it?", "more": "It’s like asking to review code that doesn’t compile or pass CI.", "source": "https://bsky.app/profile/filippo.abyssdomain.expert/post/3moq6sdq6xk2l", "seen": "2026-06-20" } }, { "schema": 2, "handle": "garrytan", "status": "ok", "checked": "2026-10-06", "name": { "value": "Garry Tan", "source": "https://github.com/garrytan" }, "x": { "value": "garrytan", "source": "https://x.com/garrytan" }, "github": { "value": "garrytan", "source": "https://github.com/garrytan" }, "site": { "value": "https://blog.garrytan.com", "source": "https://github.com/garrytan" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/19957?v=4", "source": "https://github.com/garrytan" }, "known_for": { "value": "YC CEO, creator of gstack", "source": "https://x.com/garrytan" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary, in Conductor", "surface": "app", "surface_source": { "source": "https://github.com/garrytan/gstack/blob/c285d88b90d39116ccfa2b901f80ea0fce0b26eb/README.md#L473", "seen": "2026-10-05", "quote": "I regularly run 10-15 parallel sprints — that's the practical max right now." }, "source": "https://github.com/garrytan/gstack/blob/c285d88b90d39116ccfa2b901f80ea0fce0b26eb/README.md#L23", "seen": "2026-10-05", "quote": "**gstack is how I do it.** It turns Claude Code into a virtual engineering team — a CEO who rethinks the product, an eng manager who locks architecture, a designer who catches AI slop, a reviewer who finds production bugs, a QA lead who opens a real browser, a security officer who runs OWASP + STRIDE audits, and a release engineer who ships the PR.", "history": [] }, { "name": "Codex", "icon": "codex", "role": "outside-voice reviewer", "surface": "cli", "surface_source": { "source": "https://github.com/garrytan/gstack/blob/c285d88b90d39116ccfa2b901f80ea0fce0b26eb/README.md#L341", "seen": "2026-10-05", "quote": "independent code review from OpenAI Codex CLI" }, "source": "https://github.com/garrytan/gstack/blob/c285d88b90d39116ccfa2b901f80ea0fce0b26eb/docs/ON_THE_LOC_CONTROVERSY.md#L112", "seen": "2026-10-05", "quote": "The `/plan-tune` skill I just shipped had a scope ROLLBACK from the CEO expansion plan because Codex's outside-voice review surfaced 15+ findings my four Claude reviews missed.", "history": [] } ], "models": [ { "name": "Claude Fable 5.1", "icon": "claude", "source": "https://github.com/garrytan/gstack/commit/c8f0c4e368fd59ec316c0eb0d1f4ebfa896c2d16", "seen": "2026-09-09", "quote": "Co-Authored-By: Claude Fable 5.1 ", "inferred": true, "history": [ { "value": "Claude Opus 4.8 (1M context)", "source": "https://github.com/garrytan/gstack/commit/1626d4857bfe30da2690dd6a3217961934aa3192", "seen": "2026-06-09", "quote": "Co-Authored-By: Claude Opus 4.8 (1M context) ", "inferred": true }, { "value": "Claude Fable 5", "source": "https://github.com/garrytan/gstack/commit/0d1bd5616c0ef096bb7ccee336f63c60ee408618", "seen": "2026-09-01", "quote": "Co-Authored-By: Claude Fable 5 ", "inferred": true } ] } ], "review_style": { "value": "Every branch gets CEO, Codex, DX and eng review passes", "kind": "agent-reviews", "source": "https://github.com/garrytan/gstack/blob/c285d88b90d39116ccfa2b901f80ea0fce0b26eb/docs/ON_THE_LOC_CONTROVERSY.md#L112", "seen": "2026-10-05", "quote": "Every gstack branch goes through CEO review, Codex outside-voice review, DX review, and eng review. Often 2-3 passes of each.", "also": [ { "source": "https://github.com/garrytan/gstack/blob/c285d88b90d39116ccfa2b901f80ea0fce0b26eb/README.md#L475", "seen": "2026-10-05", "quote": "You manage them the way a CEO manages a team: check in on the decisions that matter, let the rest run." } ] }, "harness": { "value": "Conductor: parallel Claude Code sessions, each in its own workspace", "short": "Conductor", "source": "https://github.com/garrytan/gstack/blob/c285d88b90d39116ccfa2b901f80ea0fce0b26eb/README.md#L473", "seen": "2026-10-05", "quote": "[Conductor](https://conductor.build) runs multiple Claude Code sessions in parallel — each in its own isolated workspace.", "history": [] }, "parallelism": { "value": 15, "short": "10–15 sprints", "source": "https://github.com/garrytan/gstack/blob/c285d88b90d39116ccfa2b901f80ea0fce0b26eb/README.md#L473", "seen": "2026-10-05", "quote": "I regularly run 10-15 parallel sprints — that's the practical max right now.", "inferred": true, "history": [ { "value": 10, "source": "https://github.com/garrytan/gstack/blob/9d47619e4c721365744c080d27455268ee93ed96/README.md#L123", "seen": "2026-03-17", "quote": "One person, ten parallel agents, each with the right cognitive mode." } ] }, "rules_files": { "value": "CLAUDE.md (614 lines) and AGENTS.md in gstack", "short": "CLAUDE.md", "url": "https://github.com/garrytan/gstack/blob/c285d88b90d39116ccfa2b901f80ea0fce0b26eb/CLAUDE.md", "source": "https://github.com/garrytan/gstack/blob/c285d88b90d39116ccfa2b901f80ea0fce0b26eb/CLAUDE.md#L363-L364", "seen": "2026-10-05", "quote": "**Choose versions autonomously; default to PATCH.** Garry delegates release version decisions to the agent." }, "skills": { "value": [ "office-hours", "plan-ceo-review", "plan-eng-review", "plan-design-review", "plan-devex-review", "autoplan", "spec", "review", "codex", "qa", "qa-only", "ship", "land-and-deploy", "canary", "cso", "design-consultation", "design-shotgun", "design-html", "design-review", "devex-review", "investigate", "retro", "learn", "document-release", "document-generate", "browse", "careful", "freeze", "guard", "unfreeze" ], "count": 55, "featured": [ "office-hours", "plan-ceo-review", "qa" ], "source": "https://github.com/garrytan/gstack/tree/c285d88b90d39116ccfa2b901f80ea0fce0b26eb", "seen": "2026-10-05" }, "mcp_servers": null, "hooks": { "value": [ "gstack-timeline-stop (Stop hook)", "/careful, /freeze, /guard (PreToolUse safety rails)" ], "source": "https://github.com/garrytan/gstack/blob/c285d88b90d39116ccfa2b901f80ea0fce0b26eb/docs/skills.md#L1211", "seen": "2026-10-05", "quote": "Four skills that add safety rails to any Claude Code session. They work via Claude Code's PreToolUse hooks — transparent, session-scoped, no configuration required." }, "signature": { "value": "/plan-ceo-review", "kind": "skill", "caption": "finds the 10-star product", "why": "Our pick: before any code, an agent plays founder and challenges the scope of the request. That is the role at the centre of his setup of agents as a virtual company.", "editorial": true, "source": "https://github.com/garrytan/gstack/blob/c285d88b90d39116ccfa2b901f80ea0fce0b26eb/README.md#L250", "seen": "2026-10-05", "quote": "Rethink the problem. Find the 10-star product hiding inside the request." }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "He runs 10-15 Claude Code sprints in Conductor, each one moving through role-based review stages." }, "signature_quote": { "value": "You manage them the way a CEO manages a team: check in on the decisions that matter, let the rest run.", "source": "https://github.com/garrytan/gstack/blob/c285d88b90d39116ccfa2b901f80ea0fce0b26eb/README.md#L475", "seen": "2026-10-05" } }, { "schema": 2, "handle": "geerlingguy", "status": "ok", "checked": "2026-10-06", "name": { "value": "Jeff Geerling", "source": "https://github.com/geerlingguy" }, "x": { "value": "geerlingguy", "source": "https://github.com/geerlingguy" }, "github": { "value": "geerlingguy", "source": "https://github.com/geerlingguy" }, "site": { "value": "https://www.jeffgeerling.com", "source": "https://github.com/geerlingguy" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/481677?v=4", "source": "https://github.com/geerlingguy" }, "known_for": { "value": "Ansible, Raspberry Pi YouTube", "source": "https://www.jeffgeerling.com/about/" }, "agents": [ { "name": "Claude", "icon": "claude", "role": "primary, side projects", "surface": "web", "surface_source": { "source": "https://github.com/geerlingguy/vcfmw21-map/blob/a30e78525139ef52d2d919bf02a3426ba57f47ec/README.md#L27", "seen": "2026-09-06", "inferred": true, "quote": "As stated previously, this page was generated with Claude Opus. Here is the full prompt / chat history: [https://claude.ai/share/fad5ac2a-4f86-4068-88cf-b095a97d3f20](https://claude.ai/share/fad5ac2a-4f86-4068-88cf-b095a97d3f20)." }, "source": "https://github.com/geerlingguy/timemachines-flash/blob/d70e42a78989f5b547f40cc9b0cd639401fd3087/README.md#L68", "seen": "2026-09-23", "quote": "The README is entirely my own work, but the Python script was written by Claude Opus 5.5.", "also": [ { "source": "https://www.jeffgeerling.com/blog/2026/truetime-xl-gps-time-server-restomod/", "seen": "2026-09-04", "quote": "I also used Claude to help me map GPIO pins to the vintage Densitron LCD on the front of the TrueTime" }, { "source": "https://www.jeffgeerling.com/blog/2026/open-sauce-gps-time-badge/", "seen": "2026-07-22", "quote": "It's ironic, then, that I used Claude to vibe code my Tufty GPS Time Badge." } ], "history": [ { "value": "Local LLMs via Ollama on a Mac mini", "surface": null, "source": "https://www.jeffgeerling.com/blog/2026/migrating-13000-comments-from-drupal-to-hugo/", "seen": "2026-01-20", "quote": "the two models I was using on my Mac (GPT-OSS 20B and Qwen3 Coder 30B, via Ollama)" } ] } ], "models": [ { "name": "Claude Opus 5.5", "icon": "claude", "source": "https://github.com/geerlingguy/timemachines-flash/blob/d70e42a78989f5b547f40cc9b0cd639401fd3087/README.md#L18", "seen": "2026-09-23", "quote": "Opus 5.5 took about 8 minutes and spat out `tmflash.py`.", "inferred": true, "history": [ { "value": "GPT-OSS 20B and Qwen3 Coder 30B (local, Ollama)", "source": "https://www.jeffgeerling.com/blog/2026/migrating-13000-comments-from-drupal-to-hugo/", "seen": "2026-01-20", "quote": "the two models I was using on my Mac (GPT-OSS 20B and Qwen3 Coder 30B, via Ollama)" }, { "value": "Claude Sonnet 5 (Medium)", "source": "https://github.com/geerlingguy/tufty-gps-time/blob/0e64217d9000219e0cdec5c95841ee4f11a6a425/README.md#L13", "seen": "2026-07-22", "quote": "Much of the code was generated by Claude Sonnet 5 (Medium)." } ] }, { "name": "Claude Fable 5.1", "icon": "claude", "source": "https://github.com/geerlingguy/BigClock-Classic/blob/1e111553dd5f0295c588a5451b08bda0cef070b5/README.md#L32", "seen": "2026-09-08", "quote": "I built this using the assistance of Claude Fable 5.1, since I've never built a Classic Mac OS application before.", "inferred": true, "history": [ { "value": "Claude Fable 5", "source": "https://github.com/geerlingguy/pps-watch-gpsmon/blob/934600ef18533c8775ea2bf7f68d3cd3e6eae2da/README.md#L9", "seen": "2026-07-28", "quote": "This is vibe-coded slopware, written in tandem with Claude Fable 5." }, { "value": "Claude Fable 5 'High'", "source": "https://github.com/geerlingguy/densitron-lcd/blob/645fa2cb29dc4e3c901b3cc2e3eed5cfd73ed26d/README.md#L43", "seen": "2026-08-31", "quote": "The code and wiring layout diagrams were written with heavy assistance from Claude Fable 5 'High'." } ] } ], "review_style": { "value": "Reviews in stages, like a tech lead with juniors", "source": "https://www.jeffgeerling.com/blog/2026/migrating-13000-comments-from-drupal-to-hugo/", "seen": "2026-01-20", "quote": "I'd label myself an 'AI skeptic', but I admit it's impressive how well LLMs achieve certain tasks, especially if you treat them like junior devs on a small team, break down work into reasonable-sized tasks, review the work in stages (checking in code in a VCS)", "kind": "unknown", "inferred": true, "also": [ { "source": "https://www.jeffgeerling.com/blog/2026/open-sauce-gps-time-badge/", "seen": "2026-07-22", "quote": "Partly due to time constraints, and partly because I wanted to see if I could complete a personal project end-to-end, without editing a line of code, I throw my requirements at Claude and ultimately came up with this 2141-line MicroPython app for Pimoroni's Badgeware ecosystem." }, { "source": "https://www.jeffgeerling.com/blog/2026/migrating-13000-comments-from-drupal-to-hugo/", "seen": "2026-01-20", "quote": "Using GPT-OSS 20B and Qwen3 30B A3B, I got a good start on the export script, but I did spend time tweaking the SQL and fiddling with the XML structure, since the AI models missed the finer details." } ] }, "harness": { "value": "Claude chat: one prompt plus attached files, then iterate", "short": "chat + files", "source": "https://github.com/geerlingguy/vcfmw21-map/blob/a30e78525139ef52d2d919bf02a3426ba57f47ec/README.md#L7", "seen": "2026-09-06", "quote": "I ran the table plan and Exhibitor CSV file through Claude Opus with the instruction:", "history": [ { "value": "Local models via Ollama on a Mac mini", "source": "https://www.jeffgeerling.com/blog/2026/migrating-13000-comments-from-drupal-to-hugo/", "seen": "2026-01-20", "quote": "The LLMs I enlisted for help seemed to hit all four of these things, at various times (yes, even 'burnout', as their context windows would grow too large for my meager Mac mini, and I'd reset and start from a fresh angle)." } ] }, "parallelism": null, "rules_files": null, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "AI Disclosure", "kind": "habit", "caption": "README names the exact model", "why": "Our pick: every Claude-built repo we found ends with a section saying which model wrote what, down to the effort level.", "editorial": true, "source": "https://github.com/geerlingguy/densitron-lcd/blob/645fa2cb29dc4e3c901b3cc2e3eed5cfd73ed26d/README.md#L41-L43", "seen": "2026-08-31", "quote": "The code and wiring layout diagrams were written with heavy assistance from Claude Fable 5 'High'. How that affects code licensing, not quite sure. But it helped write the Python as it was eluding me a bit." }, "archetype": { "value": "Skeptic", "editorial": true, "why": "Calls himself an 'AI skeptic' yet ships Claude-built side projects, each with a disclosure." }, "signature_quote": { "value": "I did have to redirect Claude a few times, to do the right thing (AI is pretty dumb, still).", "source": "https://www.jeffgeerling.com/blog/2026/open-sauce-gps-time-badge/", "seen": "2026-07-22" } }, { "schema": 2, "handle": "geoffreylitt", "status": "ok", "checked": "2026-10-07", "name": { "value": "Geoffrey Litt", "source": "https://github.com/geoffreylitt" }, "x": { "value": "geoffreylitt", "source": "https://x.com/geoffreylitt" }, "github": { "value": "geoffreylitt", "source": "https://github.com/geoffreylitt" }, "site": { "value": "https://www.geoffreylitt.com", "source": "https://github.com/geoffreylitt" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/934016?v=4", "source": "https://github.com/geoffreylitt" }, "known_for": { "value": "Malleable software, Notion", "source": "https://x.com/geoffreylitt" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary, long unsupervised runs", "surface": null, "source": "https://geoffreylitt.com/2025/10/24/code-like-a-surgeon.html", "seen": "2025-10-24", "quote": "Claude Code has been my go-to for long unsupervised sessions but Codex CLI is becoming a strong contender there too, possibly my new favorite.", "also": [ { "source": "https://geoffreylitt.com/2026/07/02/understanding-is-the-new-bottleneck.html", "seen": "2026-07-02", "quote": "One tiny example: you can now run Claude and Cursor agents in Notion. I do a lot of my coding that way now." } ], "history": [] }, { "name": "Cursor", "icon": "cursor", "role": "tab-complete for core design work", "surface": "ide", "source": "https://geoffreylitt.com/2025/10/24/code-like-a-surgeon.html", "seen": "2025-10-24", "quote": "(eg, I like Cursor tab-complete here)", "also": [ { "source": "https://geoffreylitt.com/2026/07/02/understanding-is-the-new-bottleneck.html", "seen": "2026-07-02", "quote": "One tiny example: you can now run Claude and Cursor agents in Notion. I do a lot of my coding that way now." } ], "history": [] }, { "name": "Codex CLI", "icon": "codex", "role": "contender for background runs (2025)", "surface": "cli", "source": "https://geoffreylitt.com/2025/10/24/code-like-a-surgeon.html", "seen": "2025-10-24", "quote": "Claude Code has been my go-to for long unsupervised sessions but Codex CLI is becoming a strong contender there too, possibly my new favorite.", "history": [] } ], "models": [ { "name": "Claude Opus 5", "icon": "claude", "source": "https://github.com/geoffreylitt/homepage/commit/ea712275a9ffd05f9b448652852726d37e2172d1", "seen": "2026-08-24", "quote": "Co-Authored-By: Claude Opus 5", "inferred": true, "history": [] }, { "name": "Claude Fable 5", "icon": "claude", "source": "https://github.com/geoffreylitt/homepage/commit/9025712970f04625457fa9366ed3544531cbff0b", "seen": "2026-07-02", "quote": "Co-Authored-By: Claude Fable 5", "inferred": true, "history": [] } ], "review_style": { "value": "Reads an explainer first, then the full diff; must pass a 5-question quiz before sending code on", "source": "https://geoffreylitt.com/2026/07/02/understanding-is-the-new-bottleneck.html", "seen": "2026-07-02", "quote": "I still read the code diff but I always read this first.", "kind": "reads-all", "also": [ { "source": "https://geoffreylitt.com/2026/07/02/understanding-is-the-new-bottleneck.html", "seen": "2026-07-02", "quote": "My rule: I won't send code to others until I can pass the quiz, and I do the same when reviewing others' code." } ] }, "harness": { "value": "Coding agents run from Notion; their plans land in shared pages the team comments on", "short": "Notion", "source": "https://geoffreylitt.com/2026/07/02/understanding-is-the-new-bottleneck.html", "seen": "2026-07-02", "quote": "And when those agents make a technical plan in Notion, it's in a collaborative page by default, so I can comment on it with my team and discuss immediately.", "history": [ { "value": "Secondary tasks run async in the background, even overnight", "source": "https://geoffreylitt.com/2025/10/24/code-like-a-surgeon.html", "seen": "2025-10-24", "quote": "I often find it useful to run these secondary tasks async in the background – while I’m eating lunch, or even literally overnight!" } ] }, "parallelism": null, "rules_files": { "value": "CLAUDE.md (imports AGENTS.md)", "short": "CLAUDE.md", "url": "https://github.com/geoffreylitt/super-skills/blob/2313512297e79fe020c4c18397b8598848890ae3/CLAUDE.md", "source": "https://github.com/geoffreylitt/super-skills/blob/2313512297e79fe020c4c18397b8598848890ae3/CLAUDE.md#L5-L6", "seen": "2026-08-13", "quote": "A demo client for the Notion Agent Plugins API. User-facing setup is in the" }, "skills": { "value": [ "explain-diff-html", "explain-diff-notion" ], "count": 2, "featured": [ "explain-diff-html", "explain-diff-notion" ], "source": "https://gist.github.com/geoffreylitt/a29df1b5f9865506e8952488eac3d524", "seen": "2026-06-28" }, "mcp_servers": null, "hooks": null, "signature": { "value": "/explain-diff", "kind": "skill", "caption": "with a quiz before he ships", "why": "Our pick: every agent change becomes a background-intuition-code explainer ending in five quiz questions, and he won't send code until he passes.", "editorial": true, "source": "https://geoffreylitt.com/2026/07/02/understanding-is-the-new-bottleneck.html", "seen": "2026-07-02", "quote": "Here's one answer. I made a skill called /explain-diff, which I use every day and many coworkers have found valuable." }, "archetype": { "value": "Purist", "editorial": true, "why": "He still reads every diff and quizzes himself until he understands what the agent wrote." }, "signature_quote": { "value": "Hot take: I think it's still important to understand the code that our agents write!", "source": "https://geoffreylitt.com/2026/07/02/understanding-is-the-new-bottleneck.html", "seen": "2026-07-02" } }, { "schema": 2, "handle": "geohot", "status": "ok", "checked": "2026-10-12", "name": { "value": "George Hotz", "source": "https://github.com/geohot" }, "x": { "value": "realGeorgeHotz", "source": "https://geohot.github.io/blog/" }, "github": { "value": "geohot", "source": "https://github.com/geohot" }, "site": { "value": "https://geohot.com", "source": "https://github.com/geohot" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/72895?v=4", "source": "https://github.com/geohot" }, "known_for": { "value": "tinygrad, comma.ai", "source": "https://en.wikipedia.org/wiki/George_Hotz" }, "agents": [ { "name": "OpenCode", "icon": "opencode", "role": "primary, on open models", "surface": null, "source": "https://geohot.github.io/blog/jekyll/update/2026/04/22/ai-has-no-moat.html", "seen": "2026-04-22", "quote": "In my opinion, opencode is legit the best coding agent. Not open source cope, like it’s actually number 1.", "also": [ { "source": "https://geohot.github.io/blog/jekyll/update/2026/07/12/i-love-llms.html", "seen": "2026-07-12", "quote": "I set up a Linux box with opencode on my local GLM-5.2 last week and wow like just saying install tmux with the geohot configuration works; the Year of the Linux Desktop is finally here!" }, { "source": "https://github.com/tinygrad/tinygrad/blob/5228f7bd069a605f62251d1025a2e63622ab9a40/opencode.json", "seen": "2025-12-20", "quote": "{\"$schema\": \"https://opencode.ai/config.json\", \"formatter\": false}" } ], "history": [ { "value": "Claude Code", "surface": "cli", "surface_source": { "source": "https://geohot.github.io/blog/jekyll/update/2025/12/18/computer-use-models.html", "seen": "2025-12-18", "quote": "Turns out the idea wasn’t a desktop emulator with a keyboard and mouse, it was just a command line." }, "source": "https://geohot.github.io/blog/jekyll/update/2025/12/18/computer-use-models.html", "seen": "2025-12-18", "quote": "I’m blown away with how good Claude Code is." } ] } ], "models": [ { "name": "Kimi", "icon": "kimi", "detail": "named in his PR titles", "source": "https://github.com/tinygrad/tinygrad/commit/9f6d3d882cfe0b8d5e98255b85d9c96dac2966b3", "seen": "2026-09-27", "quote": "variables are bound params (kimi) (#18444)", "history": [], "inferred": true }, { "name": "Claude Opus", "icon": "claude", "detail": "named in his PR titles", "source": "https://github.com/tinygrad/tinygrad/pull/18733", "seen": "2026-10-11", "quote": "Had opus and astra go back and forth on this", "inferred": true, "history": [ { "value": "Opus 4.5", "source": "https://geohot.github.io/blog/jekyll/update/2025/12/18/computer-use-models.html", "seen": "2025-12-18", "quote": "I tried GLM, Qwen3, and gpt-oss in Claude Code and they are all far worse than Opus 4.5." }, { "value": "GPT-5.4", "source": "https://geohot.github.io/blog/jekyll/update/2026/04/15/zappa-mitmproxy.html", "seen": "2026-04-15", "quote": "I used GPT-5.4 to code it." }, { "value": "GLM-5.2", "source": "https://geohot.github.io/blog/jekyll/update/2026/07/12/i-love-llms.html", "seen": "2026-07-12", "quote": "I set up a Linux box with opencode on my local GLM-5.2 last week" } ] } ], "review_style": { "value": "Reads and understands each line", "source": "https://geohot.github.io/blog/jekyll/update/2026/05/24/the-eternal-sloptember.html", "seen": "2026-05-24", "quote": "I haven’t seen anyone of them move to a model where they don’t carefully read and understand each line, except in some confined domains.", "kind": "spot-checks", "inferred": true, "also": [ { "source": "https://geohot.github.io/blog/jekyll/update/2026/05/24/the-eternal-sloptember.html", "seen": "2026-05-24", "quote": "I wrote some parts of tinygrad with agents. I reversed a USB <-> PCIe chip with agents. But each time I suspected I could have done it better and faster manually." }, { "source": "https://github.com/tinygrad/tinygrad/commit/dc04c7820e25987c31f2068875a515c881b79bb3", "seen": "2026-08-26", "quote": "* gpt sol review" } ] }, "harness": { "value": "opencode on a Linux box running local GLM-5.2", "short": "local models", "source": "https://geohot.github.io/blog/jekyll/update/2026/07/12/i-love-llms.html", "seen": "2026-07-12", "quote": "I set up a Linux box with opencode on my local GLM-5.2 last week", "history": [ { "value": "Claude Code with permissions bypassed", "source": "https://geohot.github.io/blog/jekyll/update/2025/12/18/computer-use-models.html", "seen": "2025-12-18", "quote": "One prompt, “bypass permissions on”" } ] }, "parallelism": null, "rules_files": { "value": "tinygrad AGENTS.md (8 lines)", "short": "AGENTS.md", "url": "https://github.com/tinygrad/tinygrad/blob/479e077ecfcf14d0cf6ea5b500c9d616db812b15/AGENTS.md", "source": "https://github.com/tinygrad/tinygrad/blob/479e077ecfcf14d0cf6ea5b500c9d616db812b15/AGENTS.md#L7-L8", "seen": "2026-09-05", "quote": "- Do not do amend commits. Always do a new commit if a force push to origin would be required.\n- tinygrad has user space PCI drivers for AMD and NVIDIA GPUs. Do not insert the unneeded kernel modules." }, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "(kimi) in the PR title", "kind": "habit", "caption": "names the model per PR", "why": "Our pick: his tinygrad PRs carry the name of the model that helped, a public log of which model wrote what.", "editorial": true, "source": "https://github.com/tinygrad/tinygrad/commit/9f6d3d882cfe0b8d5e98255b85d9c96dac2966b3", "seen": "2026-09-27", "quote": "variables are bound params (kimi) (#18444)" }, "archetype": { "value": "Skeptic", "editorial": true, "why": "Uses opencode on open models but wrote that agents cannot program and vibe-coded output is slop." }, "signature_quote": { "value": "I’m now pretty confident I’m getting better at using them and get some boost from the models.", "more": "It is a new skill, and it’s not like I haven’t constantly been trying them.", "source": "https://geohot.github.io/blog/jekyll/update/2026/07/12/i-love-llms.html", "seen": "2026-07-12" } }, { "schema": 2, "handle": "jarred-sumner", "status": "ok", "checked": "2026-10-06", "name": { "value": "Jarred Sumner", "source": "https://github.com/Jarred-Sumner" }, "x": { "value": "jarredsumner", "source": "https://github.com/Jarred-Sumner" }, "github": { "value": "Jarred-Sumner", "source": "https://github.com/Jarred-Sumner" }, "site": { "value": "https://jarredsumner.com", "source": "https://github.com/Jarred-Sumner" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/709451?v=4", "source": "https://github.com/Jarred-Sumner" }, "known_for": { "value": "Creator of Bun, now at Anthropic", "source": "https://claude.com/code-with-claude/session/sf-live-coding-session-with-boris-cherny-and-jarred-sumner" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary, dynamic workflows", "surface": "cli", "surface_source": { "source": "https://github.com/PiLastDigit/Code-With-Claude/blob/b8a0a74e5834e9cf04a300d46d5226649278b6bf/04%20-%20Live%20coding%20session%20with%20Boris%20Cherny%20and%20Jarred%20Sumner.txt", "seen": "2026-05-06", "quote": "you mostly use CLI? Yeah." }, "source": "https://bun.com/blog/bun-in-rust", "seen": "2026-07-08", "quote": "I rewrote Bun in Rust using about 50 dynamic workflows in Claude Code run continuously over the course of 11 days." }, { "name": "RoboBun", "icon": null, "role": "Bun's Claude bot, fixes issues", "surface": null, "source": "https://github.com/PiLastDigit/Code-With-Claude/blob/b8a0a74e5834e9cf04a300d46d5226649278b6bf/04%20-%20Live%20coding%20session%20with%20Boris%20Cherny%20and%20Jarred%20Sumner.txt", "seen": "2026-05-06", "quote": "So in Bun's repo, every time somebody submits an issue, we have a Claude bot automatically run and try to reproduce the issue." } ], "models": [ { "name": "Claude Fable 5", "icon": "claude", "detail": "pre-release", "source": "https://bun.com/blog/bun-in-rust", "seen": "2026-07-08", "quote": "I used a pre-release version of Claude Fable 5 for much of the Rust rewrite.", "history": [ { "value": "Opus 4.7", "source": "https://github.com/PiLastDigit/Code-With-Claude/blob/b8a0a74e5834e9cf04a300d46d5226649278b6bf/04%20-%20Live%20coding%20session%20with%20Boris%20Cherny%20and%20Jarred%20Sumner.txt", "seen": "2026-05-06", "quote": "it's so much better now, especially with Opus 4.7." } ] } ], "review_style": { "value": "Two adversarial Claude reviewers per implementer, then he reads the outputs", "source": "https://bun.com/blog/bun-in-rust", "seen": "2026-07-08", "quote": "1 implementer, 2 or more adversarial reviewers per implementer. The reviewer's only job: find bugs & reasons why the code does not work. The implementer doesn't review. The reviewer doesn't implement.", "kind": "agent-reviews", "also": [ { "source": "https://bun.com/blog/bun-in-rust", "seen": "2026-07-08", "quote": "I reviewed the original Rust rewrite PR by checking the adversarial code review agents were correctly catching discrepancies between the Zig code and the Rust code, that they were ensuring the porting guide and lifetime guide were being followed, and also manually reading a lot of the code myself side-by-side with the Zig vs Rust." }, { "source": "https://x.com/jarredsumner/status/2031099230702752065", "seen": "2026-03-09", "quote": "Been using this in Bun’s repo for weeks & giving lots of feedback\n\nThis, imo, is the best product in the code review category today." } ] }, "harness": { "value": "Claude Code dynamic workflows with /loop, one worktree per workflow shard, auto mode", "short": "workflows + /loop", "source": "https://x.com/jarredsumner/status/2060050583017373979", "seen": "2026-05-28", "quote": "Dynamic workflows, in my personal opinion, is the state of the art today for reliably using agents to complete medium - large projects. \n\nFor extra large projects, I especially like using workflows with /loop.", "history": [] }, "parallelism": { "value": 64, "short": "64 Claudes", "source": "https://bun.com/blog/bun-in-rust", "seen": "2026-07-08", "quote": "At peak, we were running 4 of these workflows at once each in a separate worktree, each with 16 Claudes per workflow. About 64 Claudes at a time.", "history": [] }, "rules_files": { "value": "CLAUDE.md in oven-sh/bun (240 lines, AGENTS.md symlinks to it)", "short": "CLAUDE.md", "url": "https://github.com/oven-sh/bun/blob/a2b69f7b0618cdcb1b5800247fb17452a31c0eb7/CLAUDE.md", "source": "https://github.com/oven-sh/bun/blob/a2b69f7b0618cdcb1b5800247fb17452a31c0eb7/CLAUDE.md#L205", "seen": "2026-09-21", "quote": "13. **If you need a paragraph-long comment to justify why the workaround is OK, the code is wrong — fix the code.**." }, "skills": { "value": [ "implementing-jsc-classes-cpp", "implementing-jsc-classes-rust", "javascriptcore-garbage-collector", "rust-system-calls", "slowest-tests", "sync-react-compiler", "verify", "writing-bundler-tests", "writing-dev-server-tests" ], "count": 9, "featured": [ "verify", "implementing-jsc-classes-rust", "slowest-tests" ], "source": "https://github.com/oven-sh/bun/tree/3f1765a6de030d00a98c33ff0c776c7a7e4b23e9/.claude/skills", "seen": "2026-08-21" }, "mcp_servers": null, "hooks": { "value": [ "PreToolUse Bash guard (pre-bash-guard.js)", "PostToolUse formatter on Write/Edit (post-edit-format.js)" ], "source": "https://github.com/oven-sh/bun/blob/3f1765a6de030d00a98c33ff0c776c7a7e4b23e9/.claude/hooks/pre-bash-guard.js#L126", "seen": "2026-06-25", "quote": "denyWithReason(\"error: Run `bun bd` without a timeout\");" }, "signature": { "value": "Adversarial review", "kind": "habit", "caption": "2 Claudes refute every change", "why": "Our pick: separate Claude reviewers who only see the diff and try to prove it wrong let him merge a million-line port.", "editorial": true, "source": "https://x.com/jarredsumner/status/2060050588574818763", "seen": "2026-05-28", "quote": "Step 2: Use 2 adversarial review agents to refute the bugfix. Uncover every flaw." }, "archetype": { "value": "Swarm Lord", "editorial": true, "why": "Ran about 64 Claudes at once in dynamic workflows to port Bun to Rust in 11 days." }, "signature_quote": { "value": "The Claude that wrote the code wants the code to get accepted. The Claude that reviews wants to find issues in the code.", "more": "1 implementer, 2 or more adversarial reviewers per implementer.", "source": "https://bun.com/blog/bun-in-rust", "seen": "2026-07-08" } }, { "schema": 2, "handle": "jessitron", "status": "ok", "checked": "2026-10-12", "name": { "value": "Jessica Kerr", "source": "https://github.com/jessitron" }, "x": null, "github": { "value": "jessitron", "source": "https://github.com/jessitron" }, "site": { "value": "https://jessitron.com", "source": "https://github.com/jessitron" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/1149737?v=4", "source": "https://github.com/jessitron" }, "known_for": { "value": "Symmathesy, Honeycomb", "source": "https://jessitron.com/2026/06/17/symmathesy-and-ai-with-kent-beck/" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary", "surface": null, "source": "https://jessitron.com/2026/04/27/communication-is-hard-but-sometimes-i-can-fix-it/", "seen": "2026-04-27", "quote": "The work: I’m getting Claude to build a program that turns Claude conversation logs into a vertical HTML comic.", "also": [ { "source": "https://jessitron.com/2026/04/27/communication-is-hard-but-sometimes-i-can-fix-it/", "seen": "2026-04-27", "quote": "This way, I learn something about the problem domain (how Claude Code records conversations)." }, { "source": "https://github.com/jessitron/mtg-deck-shuffler/blob/d463957d0b6fc318f18f4ae7fb570792dd310727/CLAUDE.md#L3", "seen": "2026-10-06", "quote": "This file provides guidance to Claude Code when working with code in this repository." } ], "history": [ { "value": "Augment Agent", "surface": "ide", "source": "https://bsky.app/profile/jessitron.bsky.social/post/3lmmzhxcwy22t", "seen": "2025-04-12", "quote": "Today I'm updating the app for my workshop at exploreddd.com next week, using @augmentcode.com", "surface_source": { "source": "https://bsky.app/profile/jessitron.bsky.social/post/3lmn2fxg7j22t", "seen": "2025-04-12", "quote": "biggest problem so far: the terminal in vscode grabs focus constantly." } } ] } ], "models": [ { "name": "Claude Opus 5.5", "icon": "claude", "inferred": true, "source": "https://github.com/jessitron/mtg-deck-shuffler/commit/d463957d0b6fc318f18f4ae7fb570792dd310727", "seen": "2026-10-06", "quote": "Co-Authored-By: Claude Opus 5.5 ", "history": [ { "value": "Claude Opus 4.7 (1M context)", "source": "https://github.com/jessitron/claude-log/commit/b94cab02db2ccfe9491598ae8d91856fcc6e5012", "seen": "2026-04-22", "quote": "Co-Authored-By: Claude Opus 4.7 (1M context) " }, { "value": "Claude Opus 4.8 (1M context)", "source": "https://github.com/jessitron/tronijess-skills/commit/ce53bb186a51202d87cf85e666f95f7fd0df6e12", "seen": "2026-06-02", "quote": "Co-Authored-By: Claude Opus 4.8 (1M context) " } ] }, { "name": "Claude Sonnet 5.5", "icon": "claude", "inferred": true, "source": "https://github.com/jessitron/mtg-deck-shuffler/commit/c46ce455543f0e947589341095a68a1d775e108b", "seen": "2026-10-06", "quote": "Co-Authored-By: Claude Sonnet 5.5 ", "history": [] } ], "review_style": { "value": "A reviewer agent checks each PR against a stated correctness condition; her own reviews fix the system", "source": "https://jessitron.com/2026/04/06/adding-correctness-conditions-to-code-changes/", "seen": "2026-04-06", "quote": "Verification – We can add a reviewer agent to check each PR for missed documentation updates.", "kind": "agent-reviews", "also": [ { "source": "https://jessitron.com/2026/04/06/adding-correctness-conditions-to-code-changes/", "seen": "2026-04-06", "quote": "Now my PR reviews are also system reviews: what about this PR should have been different? How can we change the agent’s context and feedback to make that different? Now test that system change on this PR before we fix it." }, { "source": "https://jessitron.com/2026/04/06/adding-correctness-conditions-to-code-changes/", "seen": "2026-04-06", "quote": "(This project does not deserve that level of scrutiny.)" } ], "inferred": true }, "harness": { "value": "Claude Code with her seamapping plugin; subagents spawned freely", "short": "subagents", "source": "https://github.com/jessitron/mtg-deck-shuffler/blob/d463957d0b6fc318f18f4ae7fb570792dd310727/CLAUDE.md#L10-L12", "seen": "2026-10-06", "quote": "- **Spawn subagents whenever it's a good idea** — don't wait to be asked.", "history": [] }, "parallelism": null, "rules_files": { "value": "CLAUDE.md (fleet-level, plus one per app)", "short": "CLAUDE.md", "url": "https://github.com/jessitron/mtg-deck-shuffler/blob/d463957d0b6fc318f18f4ae7fb570792dd310727/CLAUDE.md", "source": "https://github.com/jessitron/mtg-deck-shuffler/blob/d463957d0b6fc318f18f4ae7fb570792dd310727/CLAUDE.md#L13-L15", "seen": "2026-10-06", "quote": "- **Done work leaves no trace.** Don't keep `## Done` sections, tombstones, or \"decided" }, "skills": { "value": [ "begin-voyage", "establish-tracking", "orient", "drop-buoy", "request-skill-change", "create-owner", "give-yourself-a-script", "one-time-test" ], "count": 8, "featured": [ "orient", "create-owner", "give-yourself-a-script" ], "source": "https://github.com/jessitron/tronijess-skills/blob/f2e1d10e61376c756892d109f74fd50c076d3ff2/README.md#L15-L55", "seen": "2026-08-23" }, "mcp_servers": { "value": [ "honeycomb" ], "source": "https://github.com/jessitron/mtg-deck-shuffler/blob/d463957d0b6fc318f18f4ae7fb570792dd310727/.mcp.json#L3-L6", "seen": "2026-10-06", "quote": "\"url\": \"https://mcp.honeycomb.io/mcp\"" }, "hooks": null, "signature": { "value": "Capability Owners", "kind": "skill", "caption": "standing agents per feature", "why": "Our pick: her create-owner skill appoints a standing Owner for each capability, a knowledge base plus instructions a fresh subagent follows, so 'everything else still works' has someone to check it.", "editorial": true, "source": "https://github.com/jessitron/tronijess-skills/blob/6a0c93bac3617baf3a60c64e10449c760d5a1922/README.md#L33-L36", "seen": "2026-10-11", "quote": "instruction files that a fresh subagent follows. Owners are process steps, not skills." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Builds her own Claude Code plugins, skills and log viewers, and reshapes the agent's context after every miss." }, "signature_quote": { "value": "It’s all part of programming the agents to program our software.", "source": "https://jessitron.com/2026/04/06/adding-correctness-conditions-to-code-changes/", "seen": "2026-04-06" } }, { "schema": 2, "handle": "josevalim", "status": "ok", "checked": "2026-10-12", "name": { "value": "José Valim", "source": "https://github.com/josevalim" }, "x": { "value": "josevalim", "source": "https://x.com/josevalim" }, "github": { "value": "josevalim", "source": "https://github.com/josevalim" }, "site": { "value": "https://dashbit.co/", "source": "https://github.com/josevalim" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/9582?v=4", "source": "https://github.com/josevalim" }, "known_for": { "value": "Creator of Elixir, Tidewave", "source": "https://github.com/josevalim" }, "agents": [ { "name": "Codex", "icon": "codex", "role": "primary", "surface": null, "source": "https://x.com/josevalim/status/2083202612162077123", "seen": "2026-07-31", "quote": "It is the reason why I prefer Codex, which bundles fast mode in the subscription, compared to Claude Code, which bundles larger context windows.", "also": [ { "source": "https://x.com/josevalim/status/2064811215528906814", "seen": "2026-06-10", "quote": "One tiny feature I really like about Codex is that when I ask it a question, it answers the question." }, { "source": "https://x.com/josevalim/status/2066206036701184441", "seen": "2026-06-14", "quote": "I wanted Codex to find usages of [key: foo | bar] in Elixir's codebase." } ], "history": [ { "value": "Zed and Tidewave", "surface": "ide", "source": "https://changelog.com/friends/112", "seen": "2025-10-10", "quote": "I use Zed a lot, and Tidewave, and they all have notifications…", "inferred": true }, { "value": "Claude Code", "surface": null, "source": "https://x.com/josevalim/status/2064811215528906814", "seen": "2026-06-10", "quote": "Many times I've asked Claude Code, \"Do you think X would be better?\", and it would just go ahead and implement X instead of discussing the trade-offs first.", "inferred": true } ] }, { "name": "Tidewave", "icon": null, "role": "his own agent environment", "surface": "app", "surface_source": { "source": "https://tidewave.ai/blog/tidewave-connect", "seen": "2026-08-06", "quote": "Until now, using Tidewave meant working in the Tidewave app." }, "source": "https://x.com/josevalim/status/2093304967360548895", "seen": "2026-08-28", "quote": "Tidewave now ships with a design canvas. It has been so useful internally that we're kicking ourselves for not shipping it earlier!" } ], "models": [ { "name": "Sol 5.6", "icon": "openai", "detail": "in Codex", "source": "https://x.com/josevalim/status/2078177113560244355", "seen": "2026-07-17", "quote": "You probably heard me saying \"coding agents are useless for type system work 🥴\" but Sol is definitely an improvement.", "history": [ { "value": "ChatGPT-5.5 in Codex", "source": "https://github.com/elixir-lang/elixir/commit/23a431223bbbc156eb224c6aa55b68479393d3e5", "seen": "2026-05-26", "quote": "Assisted-by: ChatGPT-5.5/Codex for test case generation." } ] } ], "review_style": { "value": "Reviews agent PRs like any other code", "source": "https://x.com/josevalim/status/2096683774662611162", "seen": "2026-09-06", "quote": "Outside of that, I mainly ask the agent to break any large PR into small ones. Then I review them like any other code 😅", "kind": "spot-checks", "also": [ { "source": "https://x.com/josevalim/status/2108167477469327411", "seen": "2026-10-08", "quote": "I'm still reviewing code across Dashbit's projects and my open source projects. And, because of coding agents, I am reviewing significantly more code than before." }, { "source": "https://tidewave.ai/blog/code-review", "seen": "2026-04-15", "quote": "At Dashbit, we review all code written by coding agents before submitting a pull request." }, { "source": "https://x.com/josevalim/status/2078214085754171822", "seen": "2026-07-17", "quote": "Also, the AGENTS.md now asks for adversarial reviews when doing compiler, security, or performance work, and my gut feeling (no evals here) is that it has helped considerably. The subagent doing the review caught a few things so I didn't have to." }, { "source": "https://x.com/josevalim/status/2096192281086673043", "seen": "2026-09-05", "quote": "I run the experiment once already with one Elixir issue and all of Opus 5, Sol 5.6, and Astra 6 came up with the wrong solution. But the only conclusion I can draw from this is that it is still essential to review code." }, { "source": "https://github.com/josevalim/lynx/blob/35a4e2ef387099b9a610a08754afe4612c8aa916/README.md#L155-L158", "seen": "2026-10-05", "quote": "Proofs, which are validated by Lean's kernel, and the deserialization layer in Lean code were produced by coding agents with limited review." } ], "inferred": true }, "harness": { "value": "Tidewave's live code-review pane, marking hunks as reviewed", "short": "review pane", "source": "https://tidewave.ai/blog/code-review", "seen": "2026-04-15", "quote": "Tidewave updates the code review pane in real-time and allows you to mark which code sections (hunks in git terms) have been reviewed", "history": [] }, "parallelism": null, "rules_files": { "value": "AGENTS.md symlinked to CONTRIBUTING.md in Elixir; own AGENTS.md in Lynx", "short": "AGENTS.md", "url": "https://github.com/elixir-lang/elixir/blob/f6e177092eca20118eddf95d6a4e663f3d53d0c3/CONTRIBUTING.md#L198-L249", "source": "https://github.com/elixir-lang/elixir/commit/1d599978f7d36ea6896b294b3d26d110212be7ab", "seen": "2026-06-15", "quote": "Symlink AGENTS.md to CONTRIBUTING.md" }, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "Adversarial review agents", "kind": "rule", "caption": "on compiler & security work", "why": "Our pick: he makes a second set of agents try to disprove a change before any human sees it.", "editorial": true, "source": "https://github.com/elixir-lang/elixir/blob/f6e177092eca20118eddf95d6a4e663f3d53d0c3/CONTRIBUTING.md#L221-L225", "seen": "2026-08-16", "quote": "When automating AI usage on the Elixir codebase for performance improvements, security fixes, or correctness changes to the compiler or type system, pair it with a separate set of agents whose job is to argue against and try to invalidate any proposed change." }, "archetype": { "value": "Rules Lawyer", "editorial": true, "why": "Wrote Elixir's agent rules: adversarial reviewers, Assisted-by trailers and human review of all AI code." }, "signature_quote": { "value": "encode as many rules as possible into deterministic checks (first) and adversarial reviews (second), instead of relying exclusively on a bunch of AGENTS.md files to do the job.", "source": "https://x.com/josevalim/status/2075221914549686774", "seen": "2026-07-09" } }, { "schema": 2, "handle": "jph00", "status": "ok", "checked": "2026-10-07", "name": { "value": "Jeremy Howard", "source": "https://github.com/jph00" }, "x": { "value": "jeremyphoward", "source": "https://github.com/jph00" }, "github": { "value": "jph00", "source": "https://github.com/jph00" }, "site": { "value": "https://jeremy.fast.ai/", "source": "https://github.com/jph00" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/346999?v=4", "source": "https://github.com/jph00" }, "known_for": { "value": "fast.ai, Answer.AI, ULMFiT", "source": "https://github.com/jph00" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary, kernel-centric setup", "surface": "cli", "inferred": true, "surface_source": { "source": "https://github.com/AnswerDotAI/aai-coding/blob/bb010bf14189878ed1b2bb8db7158305c61696a3/README.md#L64", "seen": "2026-10-06", "quote": "claude --system-prompt-file ~/.claude/sysp --append-system-prompt-file /prompts/core.md" }, "source": "https://github.com/AnswerDotAI/aai-coding/blob/bb010bf14189878ed1b2bb8db7158305c61696a3/skills/persistent-python/SKILL.md#L3", "seen": "2026-10-06", "quote": "description: The persistent clikernel Python session — Jeremy's preferred single tool for nearly ALL Python work.", "also": [ { "source": "https://github.com/AnswerDotAI/aai-coding/blob/bb010bf14189878ed1b2bb8db7158305c61696a3/README.md#L29", "seen": "2026-10-06", "quote": "Claude Code uses a kernel-centric setup. Codex supports both kernel-centric and hybrid setups." }, { "source": "https://github.com/AnswerDotAI/fastcore/commit/42edbaaed64d27d62d357ed40ab573edc3f8e656", "seen": "2026-09-04", "quote": "Claude-Session: https://claude.ai/code/session_019kQ64967EEsdUQuzNU2mML" } ], "history": [ { "value": "Solveit for basically all his coding", "surface": "web", "source": "https://news.ycombinator.com/item?id=45455810", "seen": "2025-10-02", "quote": "I've enjoyed building and using our tool, \"solveit\", for the last year, and do basically all my coding, writing, reading, research, etc in it nowadays." } ] }, { "name": "Solveit", "icon": null, "role": "his company's notebook-style AI platform", "surface": "web", "source": "https://news.ycombinator.com/item?id=47302081", "seen": "2026-03-08", "quote": "Over the last couple of years we integrated LLMs with notebooks and nbdev to create Solveit, which everyone at our company uses for nearly all our work (even our lawyers, HR, etc).", "history": [] }, { "name": "Codex", "icon": "codex", "role": "second harness, hybrid mode", "surface": null, "source": "https://github.com/AnswerDotAI/aai-coding/blob/bb010bf14189878ed1b2bb8db7158305c61696a3/README.md#L3", "seen": "2026-10-06", "quote": "`aai-coding` contains Answer.AI's shared configuration, skills, and tools for Claude Code and Codex.", "history": [] } ], "models": [ { "name": "Claude Opus 4.5", "icon": "claude", "source": "https://github.com/AnswerDotAI/fastship/commit/89d1e4c93413ebaaf066d45697886feb06429dbd", "seen": "2026-01-27", "quote": "Co-Authored-By: Claude Opus 4.5 ", "history": [], "inferred": true } ], "review_style": { "value": "Small steps he follows one by one; the agent must propose before touching his published libraries", "source": "https://news.ycombinator.com/item?id=45455810", "seen": "2025-10-02", "quote": "I use small fast iterative steps and work to maximize my learning at each step.", "kind": "unknown", "inferred": true, "also": [ { "source": "https://github.com/AnswerDotAI/aai-coding/blob/bb010bf14189878ed1b2bb8db7158305c61696a3/prompts/sysp.md#L22", "seen": "2026-10-06", "quote": "Propose changes before making them when working on these libraries, and never add new API to a published library without confirming first; approval for a downstream change does not cover upstream additions." }, { "source": "https://www.fast.ai/posts/2025-10-30-build-to-last.html", "seen": "2025-10-30", "quote": "If you want to maintain craftsmanship while using AI, you need tight iteration loops so you can see what’s happening." } ] }, "harness": { "value": "Native file tools denied; all reading, editing and Python run through one persistent IPython kernel (clikernel)", "short": "persistent kernel", "source": "https://github.com/AnswerDotAI/aai-coding/blob/bb010bf14189878ed1b2bb8db7158305c61696a3/README.md#L31", "seen": "2026-10-06", "quote": "The kernel-centric setup denies native file tools. File reading, editing, searching, and Python execution use one persistent IPython kernel through clikernel.", "history": [] }, "parallelism": null, "rules_files": { "value": "Replacement Claude Code system prompt (sysp.md), core.md, coding_patterns pyskill", "short": "custom sysp", "url": "https://github.com/AnswerDotAI/aai-coding/blob/bb010bf14189878ed1b2bb8db7158305c61696a3/prompts/sysp.md", "source": "https://github.com/AnswerDotAI/aai-coding/blob/bb010bf14189878ed1b2bb8db7158305c61696a3/prompts/sysp.md#L32", "seen": "2026-10-06", "quote": "Their company does R&D and they are the boss: there is no deadline, and the goal is always the best design, never the shortest path from the code as it stands." }, "skills": { "value": [ "persistent-python", "pyskills", "clikernel", "notebook-dialog-editing", "coding_patterns", "nbreview", "write_prose", "write_docs", "harness_docs", "vision" ], "count": 10, "featured": [ "persistent-python", "pyskills", "write_prose" ], "source": "https://github.com/AnswerDotAI/aai-coding/blob/bb010bf14189878ed1b2bb8db7158305c61696a3/README.md", "seen": "2026-10-06" }, "mcp_servers": { "value": [ "clikernel" ], "source": "https://github.com/AnswerDotAI/aai-coding/blob/bb010bf14189878ed1b2bb8db7158305c61696a3/SETUP.md#L91", "seen": "2026-10-06", "quote": "Outcome: the clikernel MCP server is registered. Claude Code: a user-scope server named `clikernel` running `/bin/clikernel-mcp`." }, "hooks": { "value": [ "claude-air", "claude-slop", "claude-block-native-edit", "claude-bash-guard", "claude-prompt-submit", "claude-session-start", "claude-subagent-start", "safecmd" ], "source": "https://github.com/AnswerDotAI/aai-coding/blob/bb010bf14189878ed1b2bb8db7158305c61696a3/SETUP.md#L140", "seen": "2026-10-06", "quote": "PreToolUse matcher `Write|Edit|NotebookEdit` runs `aai-hook claude-block-native-edit`; PreToolUse matcher `Bash` runs `aai-hook claude-bash-guard`;" }, "signature": { "value": "claude-air", "kind": "hook", "caption": "nudges after 8 silent rounds", "why": "Our pick: a hook that interrupts long silent tool runs and makes the agent stop, say what it found and check its own last call, so he stays in the loop instead of the agent drifting.", "editorial": true, "source": "https://github.com/AnswerDotAI/aai-coding/blob/bb010bf14189878ed1b2bb8db7158305c61696a3/SETUP.md#L140", "seen": "2026-10-06", "quote": "(the come-up-for-air nudge: after 8 tool-call rounds with no text response of 100+ chars, it injects a reminder to surface and reassess, repeating every 5 further rounds)" }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Replaced Claude Code's system prompt, file tools and skill system with his own kernel, pyskills and hooks." }, "signature_quote": { "value": "If you want to maintain craftsmanship while using AI, you need tight iteration loops so you can see what’s happening.", "more": "You need a live workspace where you (and the AI) are manipulating actual state, not just writing text files.", "source": "https://www.fast.ai/posts/2025-10-30-build-to-last.html", "seen": "2025-10-30" } }, { "schema": 2, "handle": "jxnl", "status": "ok", "checked": "2026-10-07", "name": { "value": "Jason Liu", "source": "https://github.com/jxnl" }, "x": { "value": "jxnlco", "source": "https://github.com/jxnl" }, "github": { "value": "jxnl", "source": "https://github.com/jxnl" }, "site": { "value": "https://jxnl.co", "source": "https://github.com/jxnl" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/4852235?v=4", "source": "https://github.com/jxnl" }, "known_for": { "value": "Instructor creator, OpenAI DX", "source": "https://www.youtube.com/watch?v=il1c1a2FufU" }, "agents": [ { "name": "Codex", "icon": "codex", "role": "primary, desktop app", "surface": "app", "surface_source": { "source": "https://ai.engineer/talks/il1c1a2FufU-full-workshop-setting-yourself-up-for-success-jason-liu-openai-codex", "seen": "2026-07-24", "quote": "But generally, the desktop app has pretty good-- it has a pretty good experience. Primarily because everything I do is an AppShot." }, "source": "https://ai.engineer/talks/il1c1a2FufU-full-workshop-setting-yourself-up-for-success-jason-liu-openai-codex", "seen": "2026-07-24", "quote": "And all of this work is done effectively in the Codex app.", "also": [ { "source": "https://jxnl.co/writing/2026/05/10/codex-maxxing/", "seen": "2026-05-10", "quote": "The latest Codex app upgrades are the first thing I've used that make that broader mode feel native. Codex is still excellent for coding, but the more interesting shift is that it gives my work somewhere to live." }, { "source": "https://ai.engineer/talks/il1c1a2FufU-full-workshop-setting-yourself-up-for-success-jason-liu-openai-codex", "seen": "2026-07-24", "quote": "Sometimes I'll use the Codex CLI every once in a while if I wanna, like," } ] } ], "models": [ { "name": "GPT-5.5", "icon": "openai", "source": "https://ai.engineer/talks/il1c1a2FufU-full-workshop-setting-yourself-up-for-success-jason-liu-openai-codex", "seen": "2026-07-24", "quote": "Um, I found that, uh, the 5.5 models are pretty reluctant to take these like destructive actions.", "history": [] }, { "name": "GPT-5.3-Codex-Spark", "icon": "openai", "detail": "simple computer-use tasks", "source": "https://ai.engineer/talks/il1c1a2FufU-full-workshop-setting-yourself-up-for-success-jason-liu-openai-codex", "seen": "2026-07-24", "quote": "Like, uh, I think, I think I have a thing that just checks me into flights, and that is like a Spark agent.", "history": [] } ], "review_style": { "value": "Review skills modelled on colleagues' PR feedback; memory changes reviewed as git diffs", "kind": "agent-reviews", "inferred": true, "source": "https://ai.engineer/talks/il1c1a2FufU-full-workshop-setting-yourself-up-for-success-jason-liu-openai-codex", "seen": "2026-07-24", "quote": "And so I just have like a review my code like Charlie based off the past year of like feedback he's given on pull requests. Review my code like Dominic.", "also": [ { "source": "https://jxnl.co/writing/2026/05/10/codex-maxxing/", "seen": "2026-05-10", "quote": "When the agent updates the vault, I can read the diff and see what it thought was important enough to remember. That review step matters." }, { "source": "https://github.com/jxnl/dots/blob/34d197b7e2ede7c57d11b601250b42b8a2ae303a/agents/skills/audit-ai-code/SKILL.md#L3", "seen": "2026-06-15", "quote": "description: Audit, de-slop, parameterize, modularize, or safely clean up AI-generated or AI-shaped backend/general code." } ] }, "harness": { "value": "Codex app: pinned megathreads per workstream, an Obsidian vault in git as shared memory, Heartbeats", "short": "pinned threads", "source": "https://jxnl.co/writing/2026/05/10/codex-maxxing/", "seen": "2026-05-10", "quote": "These are not short chats. They are megathreads that I have been compacting for months.", "history": [] }, "parallelism": null, "rules_files": { "value": "Global agents/AGENTS.md (installed as ~/.claude/CLAUDE.md) plus a vault AGENTS.md", "short": "AGENTS.md", "url": "https://github.com/jxnl/dots/blob/34d197b7e2ede7c57d11b601250b42b8a2ae303a/agents/AGENTS.md", "source": "https://github.com/jxnl/dots/blob/34d197b7e2ede7c57d11b601250b42b8a2ae303a/agents/AGENTS.md#L20-L22", "seen": "2026-01-12", "quote": "## Autonomy\n- Default to action; only ask when a decision is truly blocking.\n- If details are missing, pick a reasonable default and state the assumption." }, "skills": { "value": [ "audit-ai-code", "audit-ai-frontend", "audit-ai-writing", "email-like-me", "file-organizer", "gh-address-comments", "gh-commit", "gh-fix-ci", "loop", "recording-to-video-essay", "self-improve", "simple-html-artifact", "slidev-presentation-kit", "transcript-to-blog", "tweet-like-me", "ultragoal", "yeet" ], "count": 17, "featured": [ "ultragoal", "audit-ai-code", "email-like-me" ], "source": "https://github.com/jxnl/dots/tree/34d197b7e2ede7c57d11b601250b42b8a2ae303a/agents/skills", "seen": "2026-07-14" }, "mcp_servers": { "value": [ "slack", "gmail", "calendar" ], "source": "https://jxnl.co/writing/2026/05/10/codex-maxxing/", "seen": "2026-05-10", "quote": "The ones I use most are `$slack`, `$gmail`, and `$calendar`, because Slack threads, inboxes, and calendars are where a lot of work shows up before it ever becomes code." }, "hooks": null, "signature": { "value": "$ultragoal", "kind": "skill", "caption": "goal kept in an editable file", "why": "Our pick: long Codex runs chase a goal kept in a file, so he can add scope while the loop is still running.", "editorial": true, "source": "https://ai.engineer/talks/il1c1a2FufU-full-workshop-setting-yourself-up-for-success-jason-liu-openai-codex", "seen": "2026-07-24", "quote": "In the, in the monorepo, I've also included a skill called Ultra Goal, and all it does is instead of setting the goal in the app, we set it in a file, right?" }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "Runs his work as pinned Codex threads that wake on schedules and message each other." }, "signature_quote": { "value": "Codex is still excellent for coding, but the more interesting shift is that it gives my work somewhere to live.", "source": "https://jxnl.co/writing/2026/05/10/codex-maxxing/", "seen": "2026-05-10" } }, { "schema": 2, "handle": "kentcdodds", "status": "ok", "checked": "2026-10-12", "name": { "value": "Kent C. Dodds", "source": "https://github.com/kentcdodds" }, "x": { "value": "kentcdodds", "source": "https://x.com/kentcdodds" }, "github": { "value": "kentcdodds", "source": "https://github.com/kentcdodds" }, "site": { "value": "https://kentcdodds.com", "source": "https://github.com/kentcdodds" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/1500684?v=4", "source": "https://github.com/kentcdodds" }, "known_for": { "value": "Testing Library, Epic Web", "source": "https://kentcdodds.com/resume" }, "agents": [ { "name": "Grok Bot", "icon": "grok", "role": "main driver", "surface": null, "source": "https://x.com/kentcdodds/status/2109131437006074355", "seen": "2026-10-11", "quote": "Grok bot (my main driver for everything)", "history": [] }, { "name": "Cursor", "icon": "cursor", "role": "primary, mostly Cloud Agents", "surface": "web", "surface_source": { "source": "https://gitnation.com/contents/how-i-build-web-applications-in-2026", "seen": "2026-02-26", "quote": "And then cursor.com agents. This is my addiction. I go to sleep with this." }, "source": "https://x.com/kentcdodds/status/2104285773574885610", "seen": "2026-09-27", "quote": "In the last couple weeks I've had Cursor make me two native Android apps for personal software I wanted because it's that easy", "also": [ { "source": "https://x.com/kentcdodds/status/2012540618937901110", "seen": "2026-01-17", "quote": "In my own usage of models inside Cursor, through both cloud agents and local agents working in repositories of all sizes, the biggest thing I’ve learned is that token cost is not really about how much text I generate." } ] } ], "models": [ { "name": "Grok 4.5", "icon": "grok", "detail": "coding sub-agents, fast xhigh", "source": "https://github.com/kentcdodds/kcd-skills/blob/ab6303e3b22fbbeec665f9a62ee654f379c8c600/skills/orchestrate/SKILL.md#L25-L28", "seen": "2026-08-27", "quote": "Prefer **Grok 4.5** (`grok-4.5-fast-xhigh`) for coding sub-agents; `composer-2.5-fast` for mechanical work.", "inferred": true, "history": [ { "value": "Composer 1.5", "source": "https://kentcdodds.com/blog/how-i-used-cursor-to-migrate-frameworks", "seen": "2026-02-20", "quote": "I gave Cursor (Composer 1.5) the following prompt:" } ] }, { "name": "Fable", "icon": "claude", "detail": "orchestrates", "source": "https://github.com/kentcdodds/kcd-skills/blob/ab6303e3b22fbbeec665f9a62ee654f379c8c600/skills/orchestrate/SKILL.md#L25-L28", "seen": "2026-08-27", "quote": "**Model split:** frontier model (Fable) orchestrates and integrates; cheap/fast models implement.", "history": [] } ], "review_style": { "value": "Trusts gates and AI reviewers, reads the outcome, not the diff", "kind": "agent-reviews", "source": "https://kentcdodds.com/blog/you-own-the-outcome", "seen": "2026-09-24", "quote": "I didn't read that diff line by line. I read the outcome: does it do what we decided it should do, and did the gates I trust pass?", "also": [ { "source": "https://kentcdodds.com/blog/you-own-the-outcome", "seen": "2026-09-24", "quote": "The agent wrote the implementation, tests, and PR. Automated gates and AI reviewers checked the work." }, { "source": "https://kentcdodds.com/blog/how-i-used-cursor-to-migrate-frameworks", "seen": "2026-02-20", "quote": "I have auto-fix enabled, so when BugBot finds issues, it automatically fires off an agent to fix the bug." }, { "source": "https://gitnation.com/contents/how-i-build-web-applications-in-2026", "seen": "2026-02-26", "quote": "And so I typically like to have the review agent and the actual agent do a couple iterations. So build it out, test it nicely, and then have the agent review it, iterate on each one of these steps. And only when all that's done, then I'm going to go and review it with my time." } ] }, "harness": { "value": "Cursor Cloud Agents open PRs; CI gates and Bugbot review; ship-pr skill merges and reports to Discord", "short": "cloud agents", "source": "https://github.com/kentcdodds/kentcdodds.com/blob/76fdb8987b95973066daf59d2807c2bb9bed722d/docs/agents/project-context.md#L184-L189", "seen": "2026-10-06", "quote": "Code-changing agents push their branch and create or update the pull request with Cursor Cloud `ManagePullRequest` (Kent C. Dodds account) before finishing.", "history": [ { "value": "Several cloud agents per project plus one local agent he guides", "source": "https://kentcdodds.com/blog/how-i-used-cursor-to-migrate-frameworks", "seen": "2026-02-20", "quote": "I'm often working on several projects at once with sometimes a couple cloud agents running on each while guiding an agent more directly locally." } ] }, "parallelism": { "value": 12, "short": "a dozen", "inferred": true, "source": "https://github.com/kentcdodds/kcd-skills/blob/ab6303e3b22fbbeec665f9a62ee654f379c8c600/skills/orchestrate/SKILL.md#L41-L42", "seen": "2026-08-27", "quote": "**Sweeps:** a dozen cheap agents testing every feature against the reference system (parity / perf / errors / readiness gaps).", "history": [ { "value": 25, "source": "https://kentcdodds.com/blog/how-i-used-cursor-to-migrate-frameworks", "seen": "2026-02-20", "quote": "I mean, just in this site, just this morning I have 25 agents that I triggered myself." } ] }, "rules_files": { "value": "AGENTS.md as a tiny map to docs/agents/", "short": "AGENTS.md", "url": "https://github.com/kentcdodds/kentcdodds.com/blob/b61c7ff4200c4aa0e35326d1a7baeb8a51d86673/AGENTS.md", "source": "https://github.com/kentcdodds/kentcdodds.com/blob/b61c7ff4200c4aa0e35326d1a7baeb8a51d86673/AGENTS.md#L3-L4", "seen": "2026-08-17", "quote": "This file is included in full context for every agent conversation. Keep it tiny and stable." }, "skills": { "value": [ "ask-kent", "review-and-recommend", "orchestrate", "visual-recap", "ship-pr", "prefer-local-kody-execute" ], "count": 6, "featured": [ "ship-pr", "orchestrate", "review-and-recommend" ], "source": "https://github.com/kentcdodds/kcd-skills/tree/ab6303e3b22fbbeec665f9a62ee654f379c8c600/skills", "seen": "2026-10-01" }, "mcp_servers": { "value": [ "kody" ], "source": "https://github.com/kentcdodds/kcd-skills/blob/ab6303e3b22fbbeec665f9a62ee654f379c8c600/skills/prefer-local-kody-execute/SKILL.md#L11-L12", "seen": "2026-10-01", "quote": "Default: run execute modules on the agent machine with `@kodycodes/cli execute --local`. Hosted MCP `execute` is the fallback." }, "hooks": { "value": [ "pre-push test gate (npm run test:push)", "composed with Cursor Cloud agent hooks" ], "source": "https://github.com/kentcdodds/kody/blob/65ac15100a5c1dde9b6a8387404f0bd775602a01/docs/contributing/cloud-agents.md#L75-L81", "seen": "2026-10-06", "quote": "`git push` then runs `npm run test:push` (`test:node` + `test:workers`)" }, "signature": { "value": "/ship-pr", "kind": "skill", "caption": "Bugbot loop, merge, Discord", "why": "Our pick: a cloud agent babysits its own PR through Bugbot and CI, merges low-risk changes, then posts a summary with the billed agent cost to his Discord.", "editorial": true, "source": "https://github.com/kentcdodds/kcd-skills/blob/ab6303e3b22fbbeec665f9a62ee654f379c8c600/skills/ship-pr/SKILL.md#L3-L5", "seen": "2026-10-01", "quote": "Babysit a PR. Iterate with AI reviewers and CI. Get it ready and maybe merge. Send summary message." }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "Cloud agents ship PRs end to end while his orchestrate skill splits work between a frontier model and cheap implementers." }, "signature_quote": { "value": "Agents produce output. You own the outcome.", "more": "That's not a consolation prize. Owning the outcome was always the valuable part.", "source": "https://kentcdodds.com/blog/you-own-the-outcome", "seen": "2026-09-24" } }, { "schema": 2, "handle": "kentonv", "status": "ok", "checked": "2026-10-06", "name": { "value": "Kenton Varda", "source": "https://github.com/kentonv" }, "x": { "value": "KentonVarda", "source": "https://x.com/KentonVarda" }, "github": { "value": "kentonv", "source": "https://github.com/kentonv" }, "site": null, "avatar": { "value": "https://avatars.githubusercontent.com/u/4001805?v=4", "source": "https://github.com/kentonv" }, "known_for": { "value": "Cloudflare Workers, Cap'n Proto", "source": "https://github.com/kentonv" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary", "surface": "cli", "surface_source": { "source": "https://news.ycombinator.com/item?id=44176170", "seen": "2025-06-04", "quote": "Dude. Claude Code has zero learning curve. You just open the terminal app in your code directory and you tell it what you want, in English." }, "source": "https://github.com/cloudflare/cloudflare-os/commit/895dd83e4ef3bfd640612f4ec8e4f59bebacc3f7", "seen": "2025-08-03", "quote": "Generated by Claude Code based on my instructions.", "also": [ { "source": "https://github.com/cloudflare/cloudflare-os/pull/639", "seen": "2026-10-02", "quote": "Claude suggested several small related cleanups as well which I told it to go ahead and do." } ], "inferred": true, "history": [ { "value": "Claude chat (claude.ai)", "surface": "web", "source": "https://github.com/cloudflare/workers-oauth-provider/commit/b1079e7beb5f4821a0ffc7ae07670c417b66d07d", "seen": "2025-03-04", "quote": "NOTE: At this point I transitioned from a chat session to using Claude Code." } ] }, { "name": "Cloudflare OS", "icon": null, "role": "his own agent platform", "surface": "web", "source": "https://github.com/cloudflare/cloudflare-os/pull/384", "seen": "2026-08-28", "quote": "This pull request was created entirely by my Cloudflare OS agent: https://github.com/kentonv/lanparty/pull/10" } ], "models": [ { "name": "Fable 5.1", "icon": "claude", "detail": "writes plans and code", "source": "https://github.com/cloudflare/cloudflare-os/pull/492", "seen": "2026-09-14", "quote": "Plan written by Fable 5.1, reviewed by me and Astra (Astra's review was mostly stupid, though). Each code commit written by Fable 5.1, reviewed by Sol.", "history": [ { "value": "Claude Sonnet 3.7", "source": "https://news.ycombinator.com/item?id=44159779", "seen": "2025-06-02", "quote": "It was almost entirely Claude Sonnet 3.7." } ] }, { "name": "Sol", "icon": "openai", "detail": "reviews each commit", "inferred": true, "source": "https://github.com/cloudflare/cloudflare-os/pull/492", "seen": "2026-09-14", "quote": "Each code commit written by Fable 5.1, reviewed by Sol.", "history": [] } ], "review_style": { "value": "Sol reviews every commit; he reviews the plan and the code", "kind": "agent-reviews", "source": "https://github.com/cloudflare/cloudflare-os/pull/275", "seen": "2026-08-19", "quote": "The code was almost entirely written by Fable with Sol acting as code reviewer at each step. I am pretty happy with the code I've reviewed but need to spend a bit more time looking before un-drafting this.", "also": [ { "source": "https://news.ycombinator.com/item?id=47939272", "seen": "2026-04-28", "quote": "So now we mandate that. And honestly, I love it, the AI reviewer spots all sorts of things that humans would probably miss." }, { "source": "https://news.ycombinator.com/item?id=44650155", "seen": "2025-07-22", "quote": "Whereas I carefully reviewed the AI's output for workers-oauth-provider." } ] }, "harness": { "value": "Plan first, review it, then implement; one PR split into reviewable commits", "short": "plan first", "source": "https://news.ycombinator.com/item?id=49191519", "seen": "2026-08-06", "quote": "My usual process for using AI is I first tell it what I want and have it write out a plan. I then review the plan, make comments, and go back and forth until I like it. Then I ask it to implement.", "history": [ { "value": "Prompted change by change, prompt pasted into each commit message", "source": "https://github.com/cloudflare/workers-oauth-provider/blob/78b674e14ab4ca48589ef900ff8c61d69bfedead/README.md#L201", "seen": "2025-03-15", "quote": "Check out the commit history to see how Claude was prompted and what code it produced." } ] }, "parallelism": null, "rules_files": { "value": "AGENTS.md (181 lines, now team-shared), began as his CLAUDE.md", "short": "AGENTS.md", "url": "https://github.com/cloudflare/cloudflare-os/blob/dcedc16aa3fb3f66e44c44fa3413ab0245815674/AGENTS.md", "source": "https://github.com/cloudflare/cloudflare-os/commit/cc0bb85a6a511e670684474e315ee97b470efbf1", "seen": "2025-08-10", "quote": "I've been adding stuff here whenever Claude does something wrong." }, "skills": { "value": [ "write-gatekeeper" ], "count": 1, "featured": [ "write-gatekeeper" ], "source": "https://github.com/cloudflare/cloudflare-os/tree/dd8b5c9e28593cf2595dec4a9862e1b645e5ca61/.agents/skills/write-gatekeeper", "seen": "2026-03-21" }, "mcp_servers": { "value": [ "playwright" ], "source": "https://github.com/cloudflare/cloudflare-os/commit/5c0e1a44882113853724a18347998f1076356499", "seen": "2025-08-10", "quote": "Used Playwright MCP! Claude can just see the warnings directly. Amazing." }, "hooks": null, "signature": { "value": "plans/ in the repo", "kind": "habit", "caption": "checked in for code review", "why": "Our pick: the agent-written plan is committed beside the code, so reviewers and later prompts work from the same document.", "editorial": true, "source": "https://news.ycombinator.com/item?id=49191519", "seen": "2026-08-06", "quote": "I've started checking in plans under `plans/` so that people can see the plan during code review, and also so I can refer to it when I prompt agents about things related to the plan." }, "archetype": { "value": "Purist", "editorial": true, "why": "Reviews the plan before any code, then reviews the code with Sol." }, "signature_quote": { "value": "Claude, buddy. You're gonna do the whole conversion in an hour, then we'll see if it works.", "source": "https://x.com/KentonVarda/status/2084293600142135365", "seen": "2026-08-03" } }, { "schema": 2, "handle": "kepano", "status": "ok", "checked": "2026-10-07", "name": { "value": "Steph Ango", "source": "https://github.com/kepano" }, "x": { "value": "kepano", "source": "https://x.com/kepano" }, "github": { "value": "kepano", "source": "https://github.com/kepano" }, "site": { "value": "https://stephango.com", "source": "https://github.com/kepano" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/10565871?v=4", "source": "https://github.com/kepano" }, "known_for": { "value": "CEO of Obsidian", "source": "https://stephango.com/now" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary (from his rules files)", "surface": "cli", "surface_source": { "source": "https://x.com/kepano/status/2010786788152070587", "seen": "2026-01-12", "quote": "1. install Claude Code\n2. open your terminal of choice in your Obsidian vault folder\n3. run > claude command", "inferred": true }, "source": "https://x.com/kepano/status/2008578873903206895", "seen": "2026-01-06", "quote": "I'm starting a set of Claude Skills for Obsidian... so far they're centered around helping Claude Code edit .md, .base, and .canvas files", "inferred": true } ], "models": [], "review_style": { "value": "Recorded test output must be read; fixture must fail before the fix", "source": "https://github.com/obsidianmd/obsidian-importer/blob/9abfc21ce41f41e2fd9f3db7a343026f571d19b7/CLAUDE.md#L101", "seen": "2026-08-26", "quote": "This writes the output and **fails intentionally**. Read the new files or `git diff`, then rerun without `UPDATE_EXPECTED`. A recording that nobody reviews is not a useful check.", "kind": "unknown", "inferred": true, "also": [ { "source": "https://github.com/obsidianmd/obsidian-help/blob/ec32d3964634686f60707b56827abf98f5d49423/.claude/skills/obsidian-publish-all/SKILL.md#L25", "seen": "2026-03-30", "quote": "Always run `--dry-run` first so the user can review the changeset before publishing." } ] }, "harness": { "value": "Agent verifies changes in the running Obsidian app through Obsidian CLI", "short": "Obsidian CLI", "source": "https://github.com/obsidianmd/obsidian-importer/blob/9abfc21ce41f41e2fd9f3db7a343026f571d19b7/CLAUDE.md#L128", "seen": "2026-08-26", "quote": "The check requires the Obsidian CLI, the plugin enabled in the target vault, and the current source built and deployed there. It uses `E2E_VAULT` when set; otherwise it derives the vault from `OBSIDIAN_PATH`. It writes `_e2e-check` and trashes that folder after the run.", "history": [] }, "parallelism": null, "rules_files": { "value": "AGENTS.md in Defuddle (CLAUDE.md until 2026-09-20), 188-line CLAUDE.md in Obsidian Importer", "short": "AGENTS.md", "url": "https://github.com/kepano/defuddle/blob/49d14ecc4089696565c2479b2f57db72be199adc/AGENTS.md", "source": "https://github.com/kepano/defuddle/blob/49d14ecc4089696565c2479b2f57db72be199adc/AGENTS.md#L19", "seen": "2026-09-20", "quote": "Always use `curl` to test the Worker or defuddle.md; never open those URLs in a browser." }, "skills": { "value": [ "obsidian-de-style", "obsidian-ko-style", "obsidian-localize-new", "obsidian-localize-update", "obsidian-publish-all" ], "count": 5, "featured": [ "obsidian-localize-update", "obsidian-localize-new", "obsidian-publish-all" ], "source": "https://github.com/obsidianmd/obsidian-help/tree/ec32d3964634686f60707b56827abf98f5d49423/.claude/skills", "seen": "2026-03-30" }, "mcp_servers": null, "hooks": null, "signature": { "value": "Fail-first fixtures", "kind": "rule", "caption": "before every extraction fix", "why": "Our pick: his rules make the agent prove a bug with a failing fixture before fixing it, and forbid editing expectations to fit.", "editorial": true, "source": "https://github.com/kepano/defuddle/blob/49d14ecc4089696565c2479b2f57db72be199adc/AGENTS.md#L17", "seen": "2026-09-20", "quote": "For extraction fixes, add a minimal HTML fixture in `tests/fixtures/` with correct expected output in `tests/expected/`. Verify the same fixture fails before the fix and passes afterward; do not change expectations to match buggy behavior." }, "archetype": { "value": "Rules Lawyer", "editorial": true, "why": "Writes long, precise rules files (security, fail-first fixtures, reviewed recordings) for the repos he codes in." }, "signature_quote": { "value": "As hyped as Claude Code + Obsidian is, I have no desire to upload my personal unencrypted vault data to Anthropic or anyone else.", "more": "I hope that the ideas of private inference and confidential computing that @moxie described take off. It's how all LLMs should work.", "source": "https://x.com/kepano/status/2011114444051394801", "seen": "2026-01-13" } }, { "schema": 2, "handle": "leerob", "status": "ok", "checked": "2026-10-06", "name": { "value": "Lee Robinson", "source": "https://github.com/leerob" }, "x": { "value": "leerob", "source": "https://x.com/leerob" }, "github": { "value": "leerob", "source": "https://github.com/leerob" }, "site": { "value": "https://leerob.com", "source": "https://github.com/leerob" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/9113740?v=4", "source": "https://github.com/leerob" }, "known_for": { "value": "ML at SpaceX, ex-Cursor, Vercel", "source": "https://leerob.com" }, "agents": [ { "name": "Grok Bot", "icon": "grok", "role": "always-on bots, most of his tokens", "surface": "app", "surface_source": { "source": "https://x.com/leerob/status/2089169319099777364", "seen": "2026-08-17", "quote": "This helps make the UI fast and responsive on desktop and mobile." }, "source": "https://x.com/leerob/status/2092786147156246906", "seen": "2026-08-27", "quote": "I started by using Bot for personal tasks instead of work things, but it quickly took over the majority of my tokens.", "history": [] }, { "name": "Cursor", "icon": "cursor", "role": "plan, then cloud agents", "surface": "web", "surface_source": { "source": "https://leerob.com/pixo", "seen": "2025-12", "quote": "I would go back and forth with Cursor to build a plan, and then start the agent running in the cloud, mostly with Codex." }, "source": "https://leerob.com/cursor", "seen": null, "undated": true, "quote": "I'm very opinionated about products for engineers, and a heavy user of Cursor myself, so I often gathered customer feedback and made sure we were building the right thing.", "history": [] } ], "models": [ { "name": "Claude Opus 4.5", "icon": "claude", "detail": "about 70% of spend", "source": "https://leerob.com/pixo", "seen": "2025-12", "quote": "The spend was about 70% Opus, 30% Codex due to their API pricing.", "history": [] }, { "name": "GPT-5.1-Codex-Max", "icon": "openai", "detail": "long-running cloud runs", "source": "https://leerob.com/pixo", "seen": "2025-12", "quote": "Throughout the project, I flipped between Opus 4.5 and GPT-5.1-Codex-Max. Both performed very well, although I used them differently. For example, Codex performed better for long-running tasks, while Opus was significantly better at writing.", "history": [] } ], "review_style": { "value": "Reviews most of the code and tests it; Bugbot reviews every PR", "kind": "spot-checks", "source": "https://leerob.com/pixo", "seen": "2025-12", "quote": "It’s worth noting that I did review most of the code and tested changes both manually and with automated tests. I also had Bugbot review all my PRs (example).", "also": [ { "source": "https://x.com/leerob/status/2091265397056016677", "seen": "2026-08-22", "quote": "I explicitly block any write/destructive actions and have it ask for my approval first." } ] }, "harness": { "value": "Always-running Grok Bots on their own cloud computer, connected to his X and Gmail", "short": "always-on bots", "source": "https://x.com/leerob/status/2091265397056016677", "seen": "2026-08-22", "quote": "For example, I have bots that are connected to my X, Gmail, etc.", "history": [ { "value": "Cursor plan mode, then Cursor cloud agents", "source": "https://leerob.com/pixo", "seen": "2025-12", "quote": "A large portion of this project was built while I was away from my keyboard." } ] }, "parallelism": null, "rules_files": { "value": "AGENTS.md (pixo), added only after the model made mistakes", "short": "AGENTS.md", "url": "https://github.com/leerob/pixo/blob/e6bf89c987311fcc6c4b2d02f35aad42268f1146/AGENTS.md", "source": "https://leerob.com/pixo", "seen": "2025-12", "quote": "I started with zero custom rules. I only added a rules file later after the model got a few things wrong (e.g. used the wrong linting command, couldn’t remember how to generate the binary)." }, "skills": { "value": [ "next-best-practices", "vercel-react-best-practices" ], "count": 2, "featured": [ "next-best-practices", "vercel-react-best-practices" ], "source": "https://github.com/leerob/march-arena/tree/95e10f33b695f8c535aedd6b27f88ba8fadbf6cb/.cursor/skills", "seen": "2026-03-16" }, "mcp_servers": null, "hooks": null, "signature": { "value": "Plan mode", "kind": "tool", "caption": "before every long agent run", "why": "Our pick: he plans each change with the model in Cursor, then hands it to a cloud agent and walks away. That is how he built a whole Rust library without writing code by hand.", "editorial": true, "source": "https://leerob.com/pixo", "seen": "2025-12", "quote": "Plan mode in Cursor is one of the most underrated features." }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "He plans, then sends agents to the cloud, and now runs always-on bots that use most of his tokens." }, "signature_quote": { "value": "The cost of abstractions with AI is very high.", "more": "Over-abstraction was always annoying and a code smell but now there’s an easy solution: spend tokens.", "source": "https://leerob.com/agents", "seen": "2025-12" } }, { "schema": 2, "handle": "lethain", "status": "ok", "checked": "2026-10-06", "name": { "value": "Will Larson", "source": "https://github.com/lethain" }, "x": { "value": "lethain", "source": "https://github.com/lethain" }, "github": { "value": "lethain", "source": "https://github.com/lethain" }, "site": { "value": "https://lethain.com", "source": "https://github.com/lethain" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/1923?v=4", "source": "https://github.com/lethain" }, "known_for": { "value": "Staff Engineer, CTO at Imprint", "source": "https://lethain.com/about/" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary, company-wide", "surface": null, "source": "https://lethain.com/software-factory-experiment/", "seen": "2026-09-20", "quote": "January: get every engineer onto Claude Code every single day", "also": [ { "source": "https://lethain.com/company-ai-adoption/", "seen": "2025-12-07", "quote": "building my rudimentary own agent platform using Claude code for implementation" }, { "source": "https://lethain.com/company-ai-adoption/", "seen": "2025-12-07", "quote": "That said, the vast majority of our Claude usage is done via AWS Bedrock, which we use to power Claude Code… and we use Claude Code quite a bit." } ] }, { "name": "Agent Fleet", "icon": null, "role": "Imprint's orchestrated harness, one-off tasks", "surface": null, "source": "https://lethain.com/software-factory-experiment/", "seen": "2026-09-20", "quote": "let’s roll out an orchestrated harness which internally we call “Agent Fleet”, along the lines of Stripe’s Minions" }, { "name": "Codex", "icon": "codex", "role": "rewrites agents into code", "surface": null, "source": "https://lethain.com/agents-as-scaffolding/", "seen": "2026-04-12", "quote": "generally I can take logs of some recent runs, the agent’s prompt, and some brief instructions, throw them into Codex/Claude, and get a working replacement in a few minutes." } ], "models": [ { "name": "GPT 5.4", "icon": "openai", "detail": "high reasoning, internal agents", "source": "https://lethain.com/agents-as-scaffolding/", "seen": "2026-04-12", "quote": "About a month ago, when I upgraded from GPT 4.1 to GPT 5.4 with high reasoning, I noticed that it got quite good at using the Github MCP to determine the appropriate owners for a given issue", "history": [ { "value": "GPT 4.1", "source": "https://lethain.com/agents-as-scaffolding/", "seen": "2026-04-12", "quote": "when I upgraded from GPT 4.1 to GPT 5.4 with high reasoning" } ] } ], "review_style": { "value": "Same harness reviews first, context cleared; humans after", "source": "https://lethain.com/revised-rules-of-engineering-leadership/", "seen": "2026-06-15", "quote": "The first pass of code review is done by the same harness that implements the changes, cleared of the context used to write the change, allowing humans to focus on higher value feedback.", "kind": "unknown", "also": [ { "source": "https://lethain.com/decisions-not-dates/", "seen": "2026-08-11", "quote": "Even with strong AI-enhanced workflows, I still find human review of load-bearing technical decisions (e.g. a decision that will meaningfully impact subsequent decisions) to be extremely valuable." }, { "source": "https://lethain.com/revised-rules-of-engineering-leadership/", "seen": "2026-06-15", "quote": "For example, the base case of code review from a human is slower and less effective than a good harness’ code review." } ], "inferred": true }, "harness": { "value": "~10 local workspaces, each with a checkout of every repo, for cross-repo PRs; Agent Fleet for remote work", "short": "~10 workspaces", "source": "https://lethain.com/software-factory-experiment/", "seen": "2026-09-20", "quote": "April: local development is bottlenecked on checkout and worktree model, instead create ~10 local workspaces which each have an independent checkout of every repository, and operate at the workspace level, not at the repository level, so it can generate cross-repository pull requests across frontend, backend, infrastructure and data monorepos", "history": [] }, "parallelism": { "value": 5, "short": "five or so projects", "source": "https://lethain.com/judgment-is-all-you-need/", "seen": "2026-03-11", "quote": "After I go beyond five or so concurrent projects, I tend to lose track of the necessary work to shepherd those projects to completion", "history": [], "inferred": true }, "rules_files": { "value": "AGENTS.md (156 lines), CLAUDE.md just imports it", "short": "AGENTS.md", "url": "https://github.com/lethain/workflows/blob/193edc36efdf0c59b2eb0a28d42b346256543ae2/AGENTS.md", "source": "https://github.com/lethain/workflows/blob/193edc36efdf0c59b2eb0a28d42b346256543ae2/AGENTS.md#L3-L5", "seen": "2025-12-25", "quote": "## IMPORTANT: Before Finishing Any Task\n\n**ALWAYS run linting, type checking, and tests before considering any task complete:**" }, "skills": { "value": [ "linear-project-loop" ], "count": 1, "featured": [ "linear-project-loop" ], "source": "https://lethain.com/software-factory-experiment/", "seen": "2026-09-20" }, "mcp_servers": { "value": [ "datadog" ], "source": "https://lethain.com/software-factory-experiment/", "seen": "2026-09-20", "quote": "this factory pattern depends on having Datadog MCP and Snowflake access available to manage goal-tracking" }, "hooks": null, "signature": { "value": "/linear-project-loop", "kind": "skill", "caption": "loops a project to its goal", "why": "Our pick: a software-factory skill that audits a Linear project's RFC and metrics, files missing issues, then works the unblocked ones.", "editorial": true, "source": "https://lethain.com/software-factory-experiment/", "seen": "2026-09-20", "quote": "An agent skill /linear-project-loop which reads in a Linear project and starts by auditing that project’s goal definition on these dimensions" }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "Runs agents at company scale: a fleet harness, ~10 workspaces and a skill that loops on project goals." }, "signature_quote": { "value": "coding agents have already generally solved the problem of time for our team.", "more": "We have, effectively, an unlimited amount of time, at a very affordable price, to complete our work.", "source": "https://lethain.com/judgment-is-all-you-need/", "seen": "2026-03-11" } }, { "schema": 2, "handle": "maggieappleton", "status": "ok", "checked": "2026-10-06", "name": { "value": "Maggie Appleton", "source": "https://github.com/MaggieAppleton" }, "x": { "value": "mappletons", "source": "https://github.com/MaggieAppleton" }, "github": { "value": "MaggieAppleton", "source": "https://github.com/MaggieAppleton" }, "site": { "value": "https://maggieappleton.com", "source": "https://github.com/MaggieAppleton" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/5599295?v=4", "source": "https://github.com/MaggieAppleton" }, "known_for": { "value": "Digital garden, GitHub Next", "source": "https://github.com/MaggieAppleton" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary", "surface": "web", "surface_source": { "source": "https://github.com/MaggieAppleton/maggieappleton.com-V3/pull/243", "seen": "2026-08-18", "quote": "_Generated by [Claude Code](https://claude.ai/code/session_017VSEopEqRz6NdHDhv1Rxn7)_", "inferred": true }, "source": "https://maggieappleton.com/gastown", "seen": "2026-01-23", "quote": "I currently juggle a handful of consecutive Claude Code and OpenCode agents, but pay close attention to the diffs and regularly check code in an IDE.", "also": [ { "source": "https://github.com/MaggieAppleton/elves/commit/1c7a3906b1c8cd1b71827441628adff3129e2c12", "seen": "2026-09-02", "quote": "Co-Authored-By: Claude Opus 5 (1M context) " } ], "history": [ { "value": "Cursor", "surface": "ide", "source": "https://maggieappleton.com/2025-08-vibe-legacy-code", "seen": "2025-08-02", "quote": "Like many developers, I’m constantly grappling with how much understanding I’m willing to hand over to Cursor or Claude Code." } ] }, { "name": "Codex", "icon": "codex", "role": "recent site PRs", "surface": null, "source": "https://github.com/MaggieAppleton/maggieappleton.com-V3/pull/304", "seen": "2026-09-27", "quote": "Merge remote-tracking branch 'origin/maggie/writing-assist' into codex/writing-assist-reverse-outline", "inferred": true }, { "name": "OpenCode", "icon": "opencode", "role": "named Jan 2026", "surface": null, "source": "https://maggieappleton.com/gastown", "seen": "2026-01-23", "quote": "I currently juggle a handful of consecutive Claude Code and OpenCode agents, but pay close attention to the diffs and regularly check code in an IDE." } ], "models": [ { "name": "Claude Opus 5", "icon": "claude", "detail": "1M context", "source": "https://github.com/MaggieAppleton/elves/commit/1c7a3906b1c8cd1b71827441628adff3129e2c12", "seen": "2026-09-02", "quote": "Co-Authored-By: Claude Opus 5 (1M context) ", "inferred": true, "history": [ { "value": "Opus 4.6", "source": "https://maggieappleton.com/zero-alignment", "seen": "2026-04-13", "quote": "I pick which model I want to use. Opus 4.6." }, { "value": "Claude Opus 4.8 (1M context)", "source": "https://github.com/MaggieAppleton/elves/commit/bc21e4909c38cb8154a766b9495a40c235ee72c0", "seen": "2026-07-08", "quote": "Co-authored-by: Claude Opus 4.8 (1M context) ", "inferred": true } ] }, { "name": "Claude Sonnet 4.6", "icon": "claude", "detail": "gh-aw planner + implementer", "source": "https://github.com/MaggieAppleton/maggieappleton.com-V3/blob/ca41b010e1910fb753529f0ec4f9be2e4089b576/.github/workflows/planner.md#L13-L15", "seen": "2026-04-25", "quote": "model: claude-sonnet-4-6", "history": [] } ], "review_style": { "value": "Watches the diffs, checks code in an IDE", "source": "https://maggieappleton.com/gastown", "seen": "2026-01-23", "quote": "I currently juggle a handful of consecutive Claude Code and OpenCode agents, but pay close attention to the diffs and regularly check code in an IDE.", "kind": "spot-checks", "also": [ { "source": "https://maggieappleton.com/gastown", "seen": "2026-01-23", "quote": "My development velocity is far slower than Yegge since I only wrangle a few agents at a time and keep my eyes and hands on the code." }, { "source": "https://github.com/MaggieAppleton/maggieappleton.com-V3/blob/ca41b010e1910fb753529f0ec4f9be2e4089b576/.agents/skills/pr-hygiene/SKILL.md#L12", "seen": "2026-09-26", "quote": "Ask for an independent review of every changed test." }, { "source": "https://maggieappleton.com/gastown", "seen": "2026-01-23", "quote": "Language is a poor medium for designing easing curves and describing aesthetic feelings – I always need to touch the CSS, and it’s often faster to just tweak directly than try to explain what I want." } ], "inferred": true }, "harness": { "value": "Superpowers: spec, plan, subagent-driven development", "short": "Superpowers", "source": "https://github.com/MaggieAppleton/elves/blob/8573ed05224e298270d35c08c20b09f25853c67a/docs/superpowers/plans/2026-08-30-annotation-thread-reply-composer.md#L3", "seen": "2026-08-30", "quote": "> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.", "inferred": true, "history": [] }, "parallelism": null, "rules_files": { "value": "AGENTS.md (60 lines) in her garden repo", "short": "AGENTS.md", "url": "https://github.com/MaggieAppleton/maggieappleton.com-V3/blob/d20e3ce31227f700ed52bc0810b1d072f4c80d1e/AGENTS.md", "source": "https://github.com/MaggieAppleton/maggieappleton.com-V3/blob/d20e3ce31227f700ed52bc0810b1d072f4c80d1e/AGENTS.md#L25", "seen": "2026-09-26", "quote": "Before creating a PR, use the [PR hygiene skill](.agents/skills/pr-hygiene/SKILL.md) to clean up process notes and review changed tests." }, "skills": { "value": [ "12-principles-of-animation", "baseline-ui", "brainstorming", "design-lab", "design-with-taste", "feature-kickoff", "find-skills", "fixing-accessibility", "fixing-motion-performance", "frontend-design", "grill-me", "interaction-design", "interface-craft", "maggie-design", "pr-hygiene", "prd-to-issues", "prd-to-plan", "systematic-debugging", "verification-before-completion", "web-animation-design", "write-a-prd", "write-a-skill", "writing-plans" ], "count": 23, "featured": [ "maggie-design", "pr-hygiene", "grill-me" ], "source": "https://github.com/MaggieAppleton/maggieappleton.com-V3/tree/ca41b010e1910fb753529f0ec4f9be2e4089b576/.claude/skills", "seen": "2026-09-26" }, "mcp_servers": { "value": [ "elves" ], "source": "https://github.com/MaggieAppleton/elves/blob/49d153817304e6b22a6e72167f809440133444f4/.mcp.json#L3", "seen": "2026-07-01", "quote": "\"elves\": {" }, "hooks": null, "signature": { "value": "maggie-design", "kind": "skill", "caption": "her design taste, as a skill", "why": "Our pick: she says design, not build time, is always her bottleneck, so she wrote her own design philosophy down for the agent.", "editorial": true, "source": "https://github.com/MaggieAppleton/maggieappleton.com-V3/blob/ca41b010e1910fb753529f0ec4f9be2e4089b576/.claude/skills/design-and-ui/maggie-design/SKILL.md#L8", "seen": "2026-04-25", "quote": "My personal design philosophy. The core is: **minimalist, progressive, and dynamic.** Interfaces should feel spare, teach through interaction, and behave like live, responsive widgets rather than static documents." }, "archetype": { "value": "Purist", "editorial": true, "why": "Calls herself agentically conservative: a few agents, eyes on every diff, hands on the CSS." }, "signature_quote": { "value": "My development velocity is far slower than Yegge since I only wrangle a few agents at a time and keep my eyes and hands on the code.", "more": "But the build time is rarely what holds me up.", "source": "https://maggieappleton.com/gastown", "seen": "2026-01-23" } }, { "schema": 2, "handle": "matklad", "status": "ok", "checked": "2026-10-06", "name": { "value": "Alex Kladov", "source": "https://github.com/matklad" }, "x": null, "github": { "value": "matklad", "source": "https://github.com/matklad" }, "site": { "value": "https://matklad.github.io", "source": "https://github.com/matklad" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/1711539?v=4", "source": "https://github.com/matklad" }, "known_for": { "value": "rust-analyzer, TigerBeetle", "source": "https://matklad.github.io/2025/12/06/mechanical-habits.html" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary, implements from his spec", "surface": "cli", "surface_source": { "source": "https://matklad.github.io/2025/08/31/vibe-coding-terminal-editor.html", "seen": "2025-08-31", "quote": "I originally tried to use claude code normally, by iteratively prompting in the terminal until I get the output I want." }, "source": "https://matklad.github.io/2026/01/20/vibecoding-2.html", "seen": "2026-01-20", "quote": "With the spec document in hand, I turned over to Claude code for actual implementation work.", "history": [] } ], "models": [ { "name": "Claude 3.7 Sonnet", "icon": "claude", "detail": "q alias, llm CLI", "source": "https://github.com/matklad/config/blob/2baa99f355e1e6a39febc58a0a93b8e3cd345f5c/home/.config/fish/config.fish#L29", "seen": "2026-05-11", "quote": "alias q \"llm -s \\\"Answer in as few words as possible. Use a brief style with short replies.\\\" -m claude-3.7-sonnet\";", "history": [] } ], "review_style": { "value": "Owns the structure, runs the code, does not vet the guts", "kind": "spot-checks", "source": "https://matklad.github.io/2026/01/20/vibecoding-2.html", "seen": "2026-01-20", "quote": "I want to be careful — I can’t vouch for correctness and especially completeness of the above snippet. However, given that the nature of the problem is such that I can just run the code and see the result, I am fine with it.", "also": [ { "source": "https://matklad.github.io/2026/01/20/vibecoding-2.html", "seen": "2026-01-20", "quote": "This is how this overall vibe-coding session transpired — I was providing structure, Claude was painting by the numbers." }, { "source": "https://matklad.github.io/2025/08/31/vibe-coding-terminal-editor.html", "seen": "2025-08-31", "quote": "After claude finishes iterating on a step you look at the diff and interactively prompt for necessary corrections." } ] }, "harness": { "value": "ChatGPT expands a rough idea into SPEC.md, Claude Code implements against hand-typed skeletons", "short": "spec + skeleton", "source": "https://matklad.github.io/2026/01/20/vibecoding-2.html", "seen": "2026-01-20", "quote": "Then I asked ChatGPT to expand those two paragraphs into a more or less complete spec to hand down to an agent for implementation.", "history": [ { "value": "plan.md task list, one step per conversation, then a mutable spec.md", "source": "https://matklad.github.io/2025/08/31/vibe-coding-terminal-editor.html", "seen": "2025-08-31", "quote": "When I want to apply a change to terminal-editor now, I prompt claude to update the spec first (unchecking any items that need re-doing), manually review/touch-up the spec, and use a canned prompt to align the code and tests with the spec." } ] }, "parallelism": null, "rules_files": { "value": "plan.md task list the prompt points Claude at (no CLAUDE.md)", "short": "plan.md", "url": "https://github.com/matklad/terminal-editor/blob/7c224ba85caceea40edeee967127de3fb4b1a5ad/plan.md", "source": "https://matklad.github.io/2025/08/31/vibe-coding-terminal-editor.html", "seen": "2025-08-31", "quote": "Read @plan.md, complete the next task, and mark it with X." }, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "Hand-typed skeletons", "kind": "habit", "caption": "Claude fills in the bodies", "why": "Our pick: he types the types and function signatures himself and has Claude complete only the bodies, so the code keeps his structure.", "editorial": true, "source": "https://matklad.github.io/2026/01/20/vibecoding-2.html", "seen": "2026-01-20", "quote": "The hand-written skeleton/vibe-coded guts worked not only for the CLI." }, "archetype": { "value": "Purist", "editorial": true, "why": "Writes the structure by hand and lets Claude fill in only the parts he calls trivia." }, "signature_quote": { "value": "I was providing structure, Claude was painting by the numbers.", "source": "https://matklad.github.io/2026/01/20/vibecoding-2.html", "seen": "2026-01-20" } }, { "schema": 2, "handle": "mckaywrigley", "status": "ok", "checked": "2026-10-06", "name": { "value": "McKay Wrigley", "source": "https://github.com/mckaywrigley" }, "x": { "value": "mckaywrigley", "source": "https://github.com/mckaywrigley" }, "github": { "value": "mckaywrigley", "source": "https://github.com/mckaywrigley" }, "site": { "value": "https://mckaywrigley.com", "source": "https://github.com/mckaywrigley" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/29221284?v=4", "source": "https://github.com/mckaywrigley" }, "known_for": { "value": "Founder, Takeoff AI", "source": "https://github.com/mckaywrigley" }, "agents": [ { "name": "Codex", "icon": "codex", "role": "primary for code", "surface": "app", "inferred": true, "surface_source": { "source": "https://x.com/mckaywrigley/status/2018389927936090373", "seen": "2026-02-02", "quote": "i got to use the codex app this past week and it’s really good!\n\neasily 2x’d my codex usage." }, "source": "https://x.com/mckaywrigley/status/2047806060061507971", "seen": "2026-04-24", "quote": "codex feels like an engineer (which is great for coding!), whereas claude still feels like a general purpose coworker." }, { "name": "Claude Code", "icon": "claudecode", "role": "terminal, agent swarms", "surface": "cli", "surface_source": { "source": "https://mckaywrigley.substack.com/p/my-thoughts-on-claude-opus-45", "seen": "2025-12-06", "quote": "I use it in the terminal, but you can also use Claude Code Desktop if you prefer a GUI." }, "source": "https://x.com/mckaywrigley/status/2008637181355847736", "seen": "2026-01-06", "quote": "claude code + opus 4.5 injected the immaculate hacker vibes back into ai that we haven’t had since gpt-4.", "history": [] } ], "models": [ { "name": "Claude Opus 5.5", "icon": "claude", "source": "https://x.com/mckaywrigley/status/2102447442147086563", "seen": "2026-09-22", "quote": "i honestly don’t really see a reason to use another model right now?", "history": [ { "value": "Claude Opus 4.5", "source": "https://mckaywrigley.substack.com/p/my-thoughts-on-claude-opus-45", "seen": "2025-12-06", "quote": "Claude Code + Opus 4.5 is the best AI coding tool in the world." }, { "value": "GPT-5.4 xhigh", "source": "https://x.com/mckaywrigley/status/2034399600325587452", "seen": "2026-03-18", "quote": "gpt-5.4 xhigh fundamentally changed how ambitious i am which is my new favorite benchmark" }, { "value": "Claude Fable 5", "source": "https://x.com/mckaywrigley/status/2084682154240557205", "seen": "2026-08-04", "quote": "there are like a half dozen things it is perfectly capable of that i can offload from my fable 5 workflow." } ] }, { "name": "GPT-6 Astra", "icon": "openai", "source": "https://x.com/mckaywrigley/status/2096264183482552753", "seen": "2026-09-05", "quote": "with gpt-6 astra i’m up another 3-4x on token usage.", "history": [] } ], "review_style": { "value": "Approves without reading the code", "source": "https://x.com/mckaywrigley/status/2014743894840451263", "seen": "2026-01-23", "quote": "me after hitting \"approve\" for the 9,646th time in claude code without looking at a single line of code it wrote", "kind": "no-review", "inferred": true, "also": [ { "source": "https://mckaywrigley.substack.com/p/my-thoughts-on-claude-opus-45", "seen": "2025-12-06", "quote": "Trust the model. Give it more complex tasks. Let it work for longer. Look over its shoulder less." } ] }, "harness": { "value": "Claude Code agent swarms with the multi-agent tmux view", "short": "agent swarms", "source": "https://x.com/mckaywrigley/status/2019557279222439962", "seen": "2026-02-05", "quote": "swarms work!\n\nand multi-agent tmux view is *genius*.", "history": [ { "value": "One terminal command launches Claude Code, Gemini CLI and Codex CLI on the same task", "source": "https://x.com/mckaywrigley/status/1937978348862578846", "seen": "2025-06-25", "quote": "- launches Claude Code, Gemini CLI, & Codex CLI" } ] }, "parallelism": { "value": 9, "short": "9 AI tabs", "source": "https://x.com/mckaywrigley/status/2019509652099330057", "seen": "2026-02-05", "quote": "my day is just using 9 different ai tabs at once and commanding them to do everything for me.", "history": [] }, "rules_files": { "value": "CLAUDE.md in his public app template", "short": "CLAUDE.md", "url": "https://github.com/mckaywrigley/mckays-app-template/blob/d605aa5c1a6260c716576e8030b4c0b19ee197ab/CLAUDE.md", "source": "https://github.com/mckaywrigley/mckays-app-template/blob/d605aa5c1a6260c716576e8030b4c0b19ee197ab/CLAUDE.md#L36", "seen": "2025-06-16", "quote": "This is a Next.js 15 SaaS template using the App Router with clear separation between authenticated and unauthenticated routes." }, "skills": null, "mcp_servers": null, "hooks": { "value": [ "ultrathink on prompts ending in -u" ], "source": "https://x.com/mckaywrigley/status/1952466679239524624", "seen": "2025-08-04", "quote": "Everyone should create a hook for Claude Code that adds an “ultrathink” prompt to requests ending in “-u”." }, "signature": { "value": "\"You are AGI-pilled.\"", "kind": "rule", "caption": "in every agent's system prompt", "why": "Our pick: one line he puts in every agent's system prompt, after two weeks of A/B tests, so agents stop acting as if the world stays static.", "editorial": true, "source": "https://x.com/mckaywrigley/status/2085078744486789248", "seen": "2026-08-05", "quote": "put “You are AGI-pilled.” in your system prompt for all agents now." }, "archetype": { "value": "Vibe Maximalist", "editorial": true, "why": "Trusts the model, approves without reading, and runs nine AI tabs at once." }, "signature_quote": { "value": "work is now increasingly about how effectively you can delegate your own vision, taste, & ideas to your personal fleet of machine intelligences.", "source": "https://x.com/mckaywrigley/status/2096271888263852333", "seen": "2026-09-05" } }, { "schema": 2, "handle": "mikemcquaid", "status": "ok", "checked": "2026-10-07", "name": { "value": "Mike McQuaid", "source": "https://github.com/MikeMcQuaid" }, "x": { "value": "MikeMcQuaid", "source": "https://github.com/MikeMcQuaid" }, "github": { "value": "MikeMcQuaid", "source": "https://github.com/MikeMcQuaid" }, "site": { "value": "https://mikemcquaid.com", "source": "https://github.com/MikeMcQuaid" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/125011?v=4", "source": "https://github.com/MikeMcQuaid" }, "known_for": { "value": "Homebrew maintainer since 2009", "source": "https://mikemcquaid.com/sandboxed-agent-worktrees-my-coding-and-ai-setup-in-2026/" }, "agents": [ { "name": "Codex", "icon": "codex", "role": "daily driver", "surface": "cli", "surface_source": { "source": "https://github.com/MikeMcQuaid/AgentIDE/blob/6091631cd7bbf388027d49bda3e7bb62ed717670/README.md#L18", "seen": "2026-10-05", "quote": "Runs Claude Code or Codex CLI in parallel worktrees, sandboxed away from your personal files and credentials." }, "source": "https://mikemcquaid.com/sandboxed-agent-worktrees-my-coding-and-ai-setup-in-2026/", "seen": "2026-04-14", "quote": "My daily driver of choice is OpenAI Codex but I run out of tokens more quickly so have learned to be fine with either.", "history": [ { "value": "GitHub Copilot (code completion, internal alpha)", "surface": null, "source": "https://mikemcquaid.com/sandboxed-agent-worktrees-my-coding-and-ai-setup-in-2026/", "seen": "2026-04-14", "quote": "I've been using AI tools since early 2021 when I was invited to test out the Copilot internal alpha at GitHub (where I spent 10 years)." } ] }, { "name": "Claude Code", "icon": "claudecode", "role": "built AgentIDE with it", "surface": "cli", "surface_source": { "source": "https://mikemcquaid.com/sandboxed-agent-worktrees-my-coding-and-ai-setup-in-2026/", "seen": "2026-04-14", "quote": "Once installed (`brew install sandvault`), you can run `sv codex` or `sv claude` to start your agent of choice or `sv shell` to start a shell." }, "source": "https://mikemcquaid.com/vibe-coding-developer-productivity-tools/", "seen": "2026-09-24", "quote": "By the end of a weekend with Claude Max and Anthropic's Fable 5 model and ~4 hours interacting with my computer I had something that sort of worked." }, { "name": "OpenCode", "icon": "opencode", "role": "local Ollama models", "surface": "cli", "source": "https://github.com/MikeMcQuaid/dotfiles/blob/3e74ea9e29e03b55ee38d81a857351a8155d9f7a/shrc.sh#L170-L171", "seen": "2026-09-17", "quote": "# Slower local coding model; optimal for OpenCode plan/review and harder\n# debugging." } ], "models": [ { "name": "Fable 5", "icon": "claude", "detail": "via Claude Max", "inferred": true, "source": "https://mikemcquaid.com/vibe-coding-developer-productivity-tools/", "seen": "2026-09-24", "quote": "By the end of a weekend with Claude Max and Anthropic's Fable 5 model and ~4 hours interacting with my computer I had something that sort of worked.", "history": [ { "value": "Opus 4.6 max", "source": "https://mikemcquaid.com/sandboxed-agent-worktrees-my-coding-and-ai-setup-in-2026/", "seen": "2026-04-14", "quote": "Claude Code (Opus 4.6 `max`): fairly stupid by default, can be nudged with aggressive hooks/tools/prompts to being less so" } ] }, { "name": "OpenAI Codex 5.4", "icon": "openai", "detail": "xhigh", "source": "https://mikemcquaid.com/sandboxed-agent-worktrees-my-coding-and-ai-setup-in-2026/", "seen": "2026-04-14", "quote": "OpenAI Codex (5.4 `xhigh`): takes a while, generally does things mostly right first time, doesn't need much prompting/steering", "history": [] } ], "review_style": { "value": "Reviews agent work before pushing it himself; depth depends on the code", "kind": "spot-checks", "inferred": true, "source": "https://mikemcquaid.com/sandboxed-agent-worktrees-my-coding-and-ai-setup-in-2026/", "seen": "2026-04-14", "quote": "I always review any agent-produced work as my normal user in the same shared worktree before I share it with others.", "also": [ { "source": "https://mikemcquaid.com/sandboxed-agent-worktrees-my-coding-and-ai-setup-in-2026/", "seen": "2026-04-14", "quote": "How many of these I do depends on my familiarity with the code, its criticality and how confident I am in other guardrails e.g. CI, human review." }, { "source": "https://mikemcquaid.com/vibe-coding-developer-productivity-tools/", "seen": "2026-09-24", "quote": "The wildest part: I have not (deliberately) read a single line of Swift code." }, { "source": "https://mikemcquaid.com/vibe-coding-developer-productivity-tools/", "seen": "2026-09-24", "quote": "open a PR, wait for CI results and code review from humans or Copilot" } ] }, "harness": { "value": "AgentIDE (his own Mac app) on sandvault, herdr and git worktrees", "short": "AgentIDE", "source": "https://mikemcquaid.com/vibe-coding-developer-productivity-tools/", "seen": "2026-09-24", "quote": "I'm now completely off Superset and almost entirely off Fork and Zed by (vibe) building my own tool: [AgentIDE](https://github.com/MikeMcQuaid/AgentIDE/).", "history": [ { "value": "Superset worktrees running agents under sandvault, Fork and Zed for review", "source": "https://mikemcquaid.com/sandboxed-agent-worktrees-my-coding-and-ai-setup-in-2026/", "seen": "2026-04-14", "quote": "Superset creates the worktree under `/Users/Shared/sv-mike/worktrees` and launches the agent as the Sandvault user inside it." } ] }, "parallelism": null, "rules_files": { "value": "AGENTS-GLOBAL.md (19 lines), shared by Claude and Codex, kept minimal", "short": "AGENTS.md", "url": "https://github.com/MikeMcQuaid/dotfiles/blob/c6f8883e71a4e5fdcc4074862acdfdec62243928/AGENTS-GLOBAL.md", "source": "https://github.com/MikeMcQuaid/dotfiles/blob/c6f8883e71a4e5fdcc4074862acdfdec62243928/AGENTS-GLOBAL.md#L8", "seen": "2026-09-10", "quote": "Before finishing, check your work, identify what the user may not have considered in relation to their goals and ask useful follow-up questions, including what they should have asked." }, "skills": null, "mcp_servers": null, "hooks": { "value": [ "git pre-push hook rejecting AI attribution trailers", "refuse-coding-agent guard for scripts" ], "source": "https://github.com/MikeMcQuaid/dotfiles/blob/a340d0ef58084694861946d587a0832a80782898/git-hooks/pre-push#L51", "seen": "2026-09-10", "quote": "reject \"has AI attribution; remove it and disclose AI use in the PR description.\"" }, "signature": { "value": "AgentIDE", "kind": "tool", "caption": "his own vibe-built agent IDE", "why": "Our pick: he replaced Superset, Fork and Zed with a native app he built for sandboxed agents in parallel worktrees, now in Homebrew.", "editorial": true, "source": "https://mikemcquaid.com/vibe-coding-developer-productivity-tools/", "seen": "2026-09-24", "quote": "It's an \"agentic IDE\" built by me just for my workflows and it has made me more productive and happy working with agents." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Built his own agent IDE on sandvault and worktrees instead of adapting someone else's." }, "signature_quote": { "value": "The combination of sandboxing and worktrees means I can just throw more tokens at more problems and get more done.", "source": "https://mikemcquaid.com/sandboxed-agent-worktrees-my-coding-and-ai-setup-in-2026/", "seen": "2026-04-14" } }, { "schema": 2, "handle": "mikeyk", "status": "ok", "checked": "2026-10-06", "name": { "value": "Mike Krieger", "source": "https://github.com/mikeyk" }, "x": { "value": "mikeyk", "source": "https://x.com/mikeyk" }, "github": { "value": "mikeyk", "source": "https://github.com/mikeyk" }, "site": null, "avatar": { "value": "https://avatars.githubusercontent.com/u/21233?v=4", "source": "https://github.com/mikeyk" }, "known_for": { "value": "Instagram co-founder, Anthropic Labs", "source": "https://podscripts.co/podcasts/big-technology-podcast/anthropics-labs-lead-on-fables-capabilities-building-ai-native-products-with-mike-krieger" }, "agents": [ { "name": "Claude Desktop", "icon": "claude", "role": "primary, Projects with a coordinator Claude", "surface": "app", "source": "https://x.com/mikeyk/status/2100653182708035658", "seen": "2026-09-17", "quote": "Most of my Claude Desktop work these days is in these new-style projects and I've really enjoyed how easy it is to kick off a bunch of work while still having a coordinator Claude responding quickly", "history": [] } ], "models": [ { "name": "Claude Opus 5.5", "icon": "claude", "detail": "most work", "source": "https://x.com/mikeyk/status/2104643948442743060", "seen": "2026-09-28", "quote": "I still reach for Opus 5.5 for most work, but Sonnet 5.5 is a great complement.", "history": [ { "value": "Claude Fable", "source": "https://podscripts.co/podcasts/big-technology-podcast/anthropics-labs-lead-on-fables-capabilities-building-ai-native-products-with-mike-krieger", "seen": "2026-06-24", "quote": "Fable, like enough chunky work to last." }, { "value": "Claude Opus 5", "source": "https://x.com/mikeyk/status/2080702940445397167", "seen": "2026-07-24", "quote": "Opus 5 quickly became my daily driver both at work and on the weekends." } ] }, { "name": "Claude Sonnet 5.5", "icon": "claude", "detail": "design, features, Artifacts", "source": "https://x.com/mikeyk/status/2104643948442743060", "seen": "2026-09-28", "quote": "I love its design skills and it's great for building features — I use it to make Artifacts often now.", "history": [] } ], "review_style": { "value": "Hands over whole goals and has Claude verify, then double-verify", "kind": "agent-reviews", "source": "https://podscripts.co/podcasts/big-technology-podcast/anthropics-labs-lead-on-fables-capabilities-building-ai-native-products-with-mike-krieger", "seen": "2026-06-24", "quote": "Go off and do it. Go plan it. Go execute it. Go verify the work. Double verify the work.", "inferred": true }, "harness": { "value": "Claude Desktop Projects: a coordinator Claude plus background work; dynamic /workflows", "short": "coordinator Claude", "source": "https://x.com/mikeyk/status/2100653182708035658", "seen": "2026-09-17", "quote": "Most of my Claude Desktop work these days is in these new-style projects and I've really enjoyed how easy it is to kick off a bunch of work while still having a coordinator Claude responding quickly", "history": [ { "value": "Bespoke harness and verification system", "source": "https://x.com/mikeyk/status/2080721530699587798", "seen": "2026-07-24", "quote": "Compared to earlier in the year where I had a bespoke harness/verification system, the models can do so much more with so much less instruction." } ] }, "parallelism": null, "rules_files": null, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "iOS Claude pings Android Claude", "kind": "habit", "caption": "to port each feature", "why": "Our pick: he keeps an iOS and an Android version of a Labs app in step by having the agent on the iOS side message the agent on the Android side after each change.", "editorial": true, "source": "https://podscripts.co/podcasts/big-technology-podcast/anthropics-labs-lead-on-fables-capabilities-building-ai-native-products-with-mike-krieger", "seen": "2026-06-24", "quote": "And that project, I am maintaining an iOS and an Android version of, and I basically have the cloud that works on the iOS one, basically like ping the Android one and be like," }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "He hands whole goals to a coordinator Claude and queues a night's work before bed." }, "signature_quote": { "value": "Claude does the deciding about how to get there, while I get to keep my thinking on the work itself.", "source": "https://x.com/mikeyk/status/2100259777528177030", "seen": "2026-09-16" } }, { "schema": 2, "handle": "minimaxir", "status": "ok", "checked": "2026-10-06", "name": { "value": "Max Woolf", "source": "https://github.com/minimaxir" }, "x": { "value": "minimaxir", "source": "https://x.com/minimaxir" }, "github": { "value": "minimaxir", "source": "https://github.com/minimaxir" }, "site": { "value": "https://minimaxir.com", "source": "https://github.com/minimaxir" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/2179708?v=4", "source": "https://github.com/minimaxir" }, "known_for": { "value": "Ex-BuzzFeed data scientist", "source": "https://minimaxir.com/2026/07/agent-quota-reset/" }, "agents": [ { "name": "Zed Agent", "icon": null, "role": "primary, on his Codex plan", "surface": "ide", "surface_source": { "source": "https://bsky.app/profile/minimaxir.bsky.social/post/3moazy7myrc2q", "seen": "2026-06-14", "quote": "yes (I am using Zed agent which does not inherently support compaction)" }, "source": "https://minimaxir.com/2026/09/agentic-iteration/", "seen": "2026-09-21", "quote": "Running a prompt from a Markdown document (right) by tagging the file (left), using Zed Agent.", "history": [ { "value": "GitHub Copilot in VS Code", "surface": "ide", "source": "https://minimaxir.com/2026/02/ai-agent-coding/", "seen": "2026-02-27", "quote": "For my data science work, Sonnet 4.5 in Copilot was not helpful and tended to create overly verbose Jupyter Notebooks so I was not impressed." }, { "value": "Claude Code in the terminal", "surface": "cli", "source": "https://minimaxir.com/2026/02/ai-agent-coding/", "seen": "2026-02-27", "quote": "From this point going forward, I will also switch to the terminal Claude Code, since my pipeline is simple enough and doesn’t warrant a UI or other shenanigans." }, { "value": "Cancelled Claude Code, kept Codex", "surface": null, "source": "https://bsky.app/profile/minimaxir.bsky.social/post/3ml75bz4swc2r", "seen": "2026-05-06", "quote": "The 5-hour rate limit shrinking was indeed an efficiency blocker that finally convinced me to cancel, but being only able to get 4 full sessions a week doesn't compell me to resubscribe." } ] }, { "name": "Codex", "icon": "codex", "role": "$100/mo plan; codex exec subagents", "surface": "cli", "surface_source": { "source": "https://minimaxir.com/2026/09/agentic-iteration/", "seen": "2026-09-21", "quote": "I wanted to use a cheaper, small model like GPT-5.6 Luna for the subagents since they don’t need to write code, so I came up with a galaxy brain solution that works regardless of which parent harness is used: tell the model to run independent CLI commands that themselves invoke the agent:" }, "source": "https://minimaxir.com/2026/07/agent-quota-reset/", "seen": "2026-07-18", "quote": "I was on the $20/mo Codex plan but with the promise of GPT-5.6 on the horizon and with the frequency I kept hitting the 5-hour limits, I upgraded to the $100/mo plan.", "history": [ { "value": "Codex app", "surface": "app", "source": "https://minimaxir.com/2026/02/ai-agent-coding/", "seen": "2026-02-27", "quote": "This month, OpenAI announced their Codex app and my coworkers were asking questions. So I downloaded it, and as a test case for the GPT-5.2-Codex (high) model, I asked it to reimplement the UMAP algorithm in Rust." } ] }, { "name": "Claude Code", "icon": "claudecode", "role": "testing Opus 5.5", "surface": "cli", "surface_source": { "source": "https://minimaxir.com/2026/02/ai-agent-coding/", "seen": "2026-02-27", "quote": "From this point going forward, I will also switch to the terminal Claude Code, since my pipeline is simple enough and doesn’t warrant a UI or other shenanigans." }, "source": "https://bsky.app/profile/minimaxir.bsky.social/post/3mwueh5ltv22e", "seen": "2026-10-02", "quote": "Testing out Opus 5.5 and does Claude Code really edit files by running Python's replace function?", "also": [ { "source": "https://bsky.app/profile/minimaxir.bsky.social/post/3mwrj77t3mc2a", "seen": "2026-09-30", "quote": "testing Opus 5.5 for my agentic benchmaxxing pipelines" } ] } ], "models": [ { "name": "GPT-6 Astra", "icon": "openai", "detail": "parent agent", "source": "https://bsky.app/profile/minimaxir.bsky.social/post/3muwwfuyoms2t", "seen": "2026-09-07", "quote": "GPT-6 Astra verdict: for what I use it for I like it and will likely keep using it. However, it did need a few more AGENTS[.]md tricks to keep it on track.", "history": [ { "value": "Claude Opus 4.5", "source": "https://minimaxir.com/2026/02/ai-agent-coding/", "seen": "2026-02-27", "quote": "Opus 4.5 is good enough to let me finally do side projects where I know precisely what I want but not necessarily how to implement it." }, { "value": "GPT 5.5 over Opus 4.7", "source": "https://bsky.app/profile/minimaxir.bsky.social/post/3mkyosz3g7s2r", "seen": "2026-05-04", "quote": "I was paying for both Claude Code and Codex, but Opus 4.7 is a pain to work with due to shrinking session limits and GPT 5.5 is surprisingly more reliable for what I am doing with better session limits." }, { "value": "GPT-5.6 Sol", "source": "https://minimaxir.com/2026/07/agent-quota-reset/", "seen": "2026-07-18", "quote": "GPT-5.6 Sol is indeed a great model that does live up to the hype, and I’ve had to create new projects just to have an excuse to test its limits." } ] }, { "name": "GPT-5.6 Luna", "icon": "openai", "detail": "cheap subagents, high effort", "source": "https://gist.github.com/minimaxir/933cd6354d96e1fbb45bea13e0940952/a09cf9b178bd870050c1d2873405815cca0ef73d#file-ur-prompt-md-L9", "seen": "2026-09-19", "quote": "These subagents MUST use `gpt-5.6-luna` in Codex, e.g.", "history": [] } ], "review_style": { "value": "Luna subagents re-check until all satisfied; he watches benchmark diffs", "source": "https://gist.github.com/minimaxir/933cd6354d96e1fbb45bea13e0940952/a09cf9b178bd870050c1d2873405815cca0ef73d#file-ur-prompt-md-L17", "seen": "2026-09-19", "quote": "After you are done making changes, before handing off to the user, spin up the subagents again to confirm your implementation matches their hypotheses and ask for potential further areas of improvement. Keep iterating until ALL subagents are satisfied with your implementation.", "kind": "agent-reviews", "also": [ { "source": "https://minimaxir.com/2026/09/agentic-iteration/", "seen": "2026-09-21", "quote": "It’s easy to see if an algo gamed the benchmarks if you see the benchmark file in the git diff—agents can’t cheat that (easily, anyways)." }, { "source": "https://minimaxir.com/2026/09/agentic-iteration/", "seen": "2026-09-21", "quote": "That said, it’s more hands-off than I’d like and there’s a lot of explicit faith that the newer agents aren’t just really good at cheating the benchmarks in ways I cannot detect." }, { "source": "https://minimaxir.com/2026/02/ai-agent-coding/", "seen": "2026-02-27", "quote": "Once the work is completed and manually reviewed, I manually commit the work to git, with the message referencing the specific prompt file so I have good internal tracking." } ] }, "harness": { "value": "Zed Agent, long Markdown prompt files, prompts queued overnight", "short": "prompt files", "source": "https://minimaxir.com/2026/09/agentic-iteration/", "seen": "2026-09-21", "quote": "The Ur-Prompt, combined with the rules and constraints in my AGENTS.md + the two breakthroughmaxxing prompts + a competition prompt pass + a refactor prompt pass with GPT-6 Astra allows me to just queue the prompts overnight, go to bed, and wake up to superfast software.", "history": [ { "value": "Terminal Claude Code, prompt files tagged in, manual commits", "source": "https://minimaxir.com/2026/02/ai-agent-coding/", "seen": "2026-02-27", "quote": "My approach to prompting is to write the potentially-very-large individual prompt in its own Markdown file (which can be tracked in git), then tag the agent with that prompt and tell it to implement that Markdown file." } ] }, "parallelism": { "value": 12, "short": "7–12 subagents", "source": "https://gist.github.com/minimaxir/933cd6354d96e1fbb45bea13e0940952/a09cf9b178bd870050c1d2873405815cca0ef73d#file-ur-prompt-md-L9", "seen": "2026-09-19", "quote": "To best accomplish innovative implementations, you **MUST** spin up 7-12 independent distinct \"subagents\" by running a long-duration CLI command (do not use the subagent tool) which can explore and evaluate different feasible hypotheses for improving the performance, usability, and security of this crate.", "history": [], "inferred": true }, "rules_files": { "value": "AGENTS.md for Rust (245 lines) and for Python, shared as gists, ALL-CAPS rules", "short": "AGENTS.md", "url": "https://gist.github.com/minimaxir/86de3cc8f628079d8337e70924b3411d/b640fcd77e43f00658b8819c4bd8b2d67f6fb867", "source": "https://gist.github.com/minimaxir/86de3cc8f628079d8337e70924b3411d/b640fcd77e43f00658b8819c4bd8b2d67f6fb867#file-agents-md-L17", "seen": "2026-09-19", "quote": "If the code is not fully optimized before handing off to the user, you will be fined $100. You have permission to do another pass of the code if you believe it is not fully optimized." }, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "codex exec Luna subagents", "kind": "habit", "caption": "7–12 cheap critics per pass", "why": "Our pick: instead of the harness's full-price subagent tool, his prompt makes the agent shell out to cheap read-only Codex runs that pitch ideas and then sign off on the result.", "editorial": true, "source": "https://minimaxir.com/2026/09/agentic-iteration/", "seen": "2026-09-21", "quote": "I wanted to use a cheaper, small model like GPT-5.6 Luna for the subagents since they don’t need to write code, so I came up with a galaxy brain solution that works regardless of which parent harness is used: tell the model to run independent CLI commands that themselves invoke the agent:" }, "archetype": { "value": "Rules Lawyer", "editorial": true, "why": "Long ALL-CAPS AGENTS.md rules and prompt files, down to a $100 fine for unoptimized code." }, "signature_quote": { "value": "Prompt engineering is less important nowadays since agents can handle some ambiguity, but the less ambiguity, the more **productive** the agent is.", "more": "It also adds checks against being too smart and cheating.", "source": "https://bsky.app/profile/minimaxir.bsky.social/post/3mr3n2hvc5k2e", "seen": "2026-07-20" } }, { "schema": 2, "handle": "natolambert", "status": "ok", "checked": "2026-10-07", "name": { "value": "Nathan Lambert", "source": "https://github.com/natolambert" }, "x": { "value": "natolambert", "source": "https://github.com/natolambert" }, "github": { "value": "natolambert", "source": "https://github.com/natolambert" }, "site": { "value": "https://natolambert.com", "source": "https://github.com/natolambert" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/10695622?v=4", "source": "https://github.com/natolambert" }, "known_for": { "value": "RLHF Book, Interconnects", "source": "https://github.com/natolambert/rlhf-book" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary", "surface": "cli", "surface_source": { "source": "https://www.interconnects.ai/p/opus-46-vs-codex-53", "seen": "2026-02-09", "inferred": true, "quote": "Claude Code’s harness has a terrible bug that makes subagents brick the terminal, where new messages say you must compact or clear, but compaction fails." }, "source": "https://www.interconnects.ai/p/i-wrote-an-ai-textbook-how-long-until", "seen": "2026-08-12", "quote": "I would have Claude Code navigate to each comment, print the context before and after, and let me know if it was an easy typo fix or something more nuanced.", "also": [ { "source": "https://www.interconnects.ai/p/glm-52-is-the-step-change-for-open", "seen": "2026-06-22", "quote": "I put it to work helping make content for my post-training course with Fireworks’ API in Claude Code (setting this up was very easy)." } ] }, { "name": "Codex", "icon": "codex", "role": "second agent, used more evenly since 5.3", "surface": null, "source": "https://www.interconnects.ai/p/opus-46-vs-codex-53", "seen": "2026-02-09", "quote": "For the last few days, I’ve been using both of the models much more evenly.", "history": [ { "value": "Codex with GPT 5.2 xhigh, not yet working for him", "surface": null, "source": "https://www.interconnects.ai/p/get-good-at-agents", "seen": "2026-01-21", "quote": "Codex with GPT 5.2 on xhigh thinking effort alone feels very capable, more meticulous than Claude even, but I haven’t yet figured out how to get the best out of it." } ] } ], "models": [ { "name": "Claude Fable 5", "icon": "claude", "detail": "in Claude Code", "inferred": true, "source": "https://github.com/natolambert/rlhf-book/commit/1042143366d73ff654727e4d18a1d3a21209c3a3", "seen": "2026-08-12", "quote": "Co-authored-by: Claude Fable 5 ", "history": [ { "value": "Claude Opus 4.5 in Claude Code", "source": "https://www.interconnects.ai/p/get-good-at-agents", "seen": "2026-01-21", "quote": "My default workflow now is GPT 5 Pro for planning, Claude Code with Opus 4.5 for implementation." } ] }, { "name": "GPT 5 Pro", "icon": "openai", "detail": "planning and deep search", "source": "https://www.interconnects.ai/p/get-good-at-agents", "seen": "2026-01-21", "quote": "I often have Claude Code pass information back to GPT 5 Pro for a deep search when stuck with a very detailed prompt.", "history": [] } ], "review_style": { "value": "Lets agents catch their own mistakes, then checks the results himself", "kind": "agent-reviews", "source": "https://www.interconnects.ai/p/get-good-at-agents", "seen": "2026-01-21", "quote": "They make mistakes, they catch most of their own mistakes, and they’re fairly slow too, but they’re capable. I can’t wait to go home and check on what my Claudes were up to.", "also": [ { "source": "https://lexfridman.com/ai-sota-2026-transcript", "seen": "2026-01-31", "quote": "And then I have enough situational awareness to be like, “Okay, these trends obviously make sense,” and you can check things." } ] }, "harness": { "value": "Claude Code instances left running on a DGX Spark while he is out", "short": "DGX Spark", "source": "https://www.interconnects.ai/p/get-good-at-agents", "seen": "2026-01-21", "quote": "I’ve taken to leaving Claude Code instances running on my DGX Spark trying to implement new features in our RL codebase when I’m at dinner or work.", "history": [] }, "parallelism": null, "rules_files": { "value": "CLAUDE.md per repo, AGENTS.md symlinked to it for Codex", "short": "CLAUDE.md", "url": "https://github.com/natolambert/rlhf-book/blob/6063b3de36fff7cb57892e99519dbdd7d72f0225/CLAUDE.md", "source": "https://github.com/natolambert/rlhf-book/blob/6063b3de36fff7cb57892e99519dbdd7d72f0225/CLAUDE.md#L31", "seen": "2026-07-30", "quote": "`AGENTS.md` is intentionally a symlink to `CLAUDE.md` for Codex compatibility. If you are Codex and the symlink is present, edit `CLAUDE.md` directly; do not replace, unlink, or edit through `AGENTS.md`." }, "skills": { "value": [ "gemini-feedback", "pre-submit-pr", "push-to-pr", "run-rlhf-code-experiment", "serve-course-lecture", "update-pr-body" ], "count": 6, "featured": [ "gemini-feedback", "serve-course-lecture", "push-to-pr" ], "source": "https://github.com/natolambert/rlhf-book/tree/eecc49e1e1daf1be670d7242eb090240b5bb0f04/.claude/skills", "seen": "2026-09-02" }, "mcp_servers": null, "hooks": { "value": [ "PreToolUse changelog reminder on git commit, git push, gh pr create" ], "source": "https://github.com/natolambert/colloquium/blob/2e871435b33bb84b3bd767010f637d2c88511184/.claude/hooks/changelog-reminder.sh#L7-L8", "seen": "2026-03-25", "quote": "if echo \"$COMMAND\" | grep -qE 'git commit|git push|gh pr create'; then\n echo \"Reminder: Update CHANGELOG.md with ONE line per PR (not per commit) and include the PR link.\" >&2" }, "signature": { "value": "/gemini-feedback", "kind": "skill", "caption": "Gemini critiques each diagram", "why": "Our pick: Claude Code draws the textbook figures, then a skill sends each one to Gemini for a second model's review.", "editorial": true, "source": "https://github.com/natolambert/rlhf-book/blob/6063b3de36fff7cb57892e99519dbdd7d72f0225/CLAUDE.md#L36", "seen": "2026-07-30", "quote": "For diagram review, use `.claude/skills/gemini-feedback/SKILL.md`; it sends a generated diagram and relevant mathematical context to Gemini for textbook-quality feedback." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Repo skills, a changelog hook and his own agent-native slide tool around Claude Code." }, "signature_quote": { "value": "Every researcher needs to learn how to run a lab.", "more": "Agents push the humans up the org chart.", "source": "https://www.interconnects.ai/p/get-good-at-agents", "seen": "2026-01-21" } }, { "schema": 2, "handle": "nedbat", "status": "ok", "checked": "2026-10-06", "name": { "value": "Ned Batchelder", "source": "https://github.com/nedbat" }, "x": null, "github": { "value": "nedbat", "source": "https://github.com/nedbat" }, "site": { "value": "https://nedbatchelder.com", "source": "https://github.com/nedbat" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/23789?v=4", "source": "https://github.com/nedbat" }, "known_for": { "value": "coverage.py", "source": "https://nedbatchelder.com/" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary, with a plan-first CLAUDE.md", "surface": "cli", "source": "https://github.com/nedbat/dotfiles/blob/003ae2618eebc9877f69e7a01f610d43070bd92c/dot_config/shellrc/claude.sh#L1-L2", "seen": "2026-02-24", "quote": "# Don't change the title of my terminal tab, because it doesn't change back when I exit claude.\nexport CLAUDE_CODE_DISABLE_TERMINAL_TITLE=1", "also": [ { "source": "https://github.com/coveragepy/coveragepy/commit/cb5f925f5198118c6c19bc9e03ee883926b6ed21", "seen": "2025-12-25", "quote": "Claude did this with my direction." }, { "source": "https://hachyderm.io/@nedbat/114585281313085432", "seen": "2025-05-28", "quote": "Also, Claude made the code changes, and did a good job!" } ] }, { "name": "GitHub Copilot", "icon": "githubcopilot", "role": "inline completions in Vim", "surface": "ide", "source": "https://github.com/nedbat/dotfiles/blob/003ae2618eebc9877f69e7a01f610d43070bd92c/dot_config/vim/vimrc#L400-L413", "seen": "2026-02-09", "quote": "Plug 'github/copilot.vim'" } ], "models": [], "review_style": { "value": "A person owns and reviews every commit, Claude or not", "kind": "unknown", "inferred": true, "source": "https://hachyderm.io/@nedbat/116134732094799152", "seen": "2026-02-26", "quote": "Every commit is owned by a person who has vouched for the quality of the commit, and every commit is reviewed by a person who has also vouched for it. Making a commit with Claude's help doesn't change that.", "also": [ { "source": "https://hachyderm.io/@nedbat/116133445557306539", "seen": "2026-02-25", "quote": "Claude can write wrong code. New contributors can write wrong code. What matters is what you do with that code after it's been written." }, { "source": "https://github.com/nedbat/dotfiles/blob/003ae2618eebc9877f69e7a01f610d43070bd92c/dot_config/claude/nedbat-main.md#L8", "seen": "2025-06-30", "quote": "You should make a detailed plan and confirm it with me." }, { "source": "https://github.com/nedbat/dotfiles/blob/003ae2618eebc9877f69e7a01f610d43070bd92c/dot_config/claude/nedbat-main.md#L15-L16", "seen": "2026-08-07", "quote": "When showing me code to approve running, ALWAYS provide a one-sentence sumamry\nof what the code does and why you are running it. Tell me the goal of the code." } ] }, "harness": { "value": "Claude Code in a terminal tab, LSP tool on, prompt suggestions off", "short": "LSP tool", "source": "https://github.com/nedbat/dotfiles/blob/003ae2618eebc9877f69e7a01f610d43070bd92c/dot_config/shellrc/claude.sh#L4-L8", "seen": "2026-04-16", "quote": "# Don't suggest the next prompt.\nexport CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION=false\n\n# https://karanbansal.in/blog/claude-code-lsp/\nexport ENABLE_LSP_TOOL=1", "history": [] }, "parallelism": null, "rules_files": { "value": "Global CLAUDE.md importing nedbat-main.md (56 lines) from his dotfiles", "short": "CLAUDE.md", "url": "https://github.com/nedbat/dotfiles/blob/003ae2618eebc9877f69e7a01f610d43070bd92c/dot_config/claude/nedbat-main.md", "source": "https://github.com/nedbat/dotfiles/blob/003ae2618eebc9877f69e7a01f610d43070bd92c/dot_config/claude/README.md#L1-L4", "seen": "2025-09-19", "quote": "Put this line into ~/.claude/CLAUDE.md:\n```\n@~/.config/claude/nedbat-main.md\n```" }, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "Be introspective", "kind": "rule", "caption": "Claude fixes its own rules", "why": "Our pick: when a workflow goes wrong, his CLAUDE.md has Claude work out why and propose changes to its own instructions.", "editorial": true, "source": "https://github.com/nedbat/dotfiles/blob/003ae2618eebc9877f69e7a01f610d43070bd92c/dot_config/claude/nedbat-main.md#L11-L13", "seen": "2026-03-13", "quote": "Be introspective. When something goes wrong with a workflow or process, think\ncarefully about why it happened. Offer to discuss the problem, and suggest\nchanges to instructions that could prevent it in the future." }, "archetype": { "value": "Purist", "editorial": true, "why": "Makes Claude plan and get his sign-off first, and keeps his fun projects AI-free." }, "signature_quote": { "value": "Another choice I made: no AI.", "more": "I like writing programs. I get a deeper sense of the thing I am building when I have my fingers in the clay.", "source": "https://nedbatchelder.com/blog/202607/acidica", "seen": "2026-07-26" } }, { "schema": 2, "handle": "nolanlawson", "status": "ok", "checked": "2026-10-07", "name": { "value": "Nolan Lawson", "source": "https://github.com/nolanlawson" }, "x": null, "github": { "value": "nolanlawson", "source": "https://github.com/nolanlawson" }, "site": { "value": "https://nolanlawson.com", "source": "https://github.com/nolanlawson" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/283842?v=4", "source": "https://github.com/nolanlawson" }, "known_for": { "value": "PouchDB, emoji-picker-element", "source": "https://github.com/nolanlawson" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary", "surface": null, "source": "https://nolanlawson.com/2026/08/23/the-asteroid-currently-hitting-frontend-web-development/", "seen": "2026-08-23", "quote": "These days, given a slow-loading website, even as someone who’s written about browser performance for years and who literally worked on a browser performance team, I would probably just chuck a Chrome trace at Claude Code and have it suggest improvements. In fact I’ve done this very thing in my day job and gotten some good results.", "history": [] }, { "name": "Codex", "icon": "codex", "role": "second reviewer", "surface": "cli", "surface_source": { "source": "https://gist.github.com/nolanlawson/4150b0ca9640654c256b324fac0d5253", "seen": "2026-05-27", "quote": "allowed-tools: Bash(gh:*) Bash(codex:*) Bash(cat:*) Bash(tee:*) Bash(sleep:*) Agent Read Grep Glob Write(/tmp/*)", "inferred": true }, "source": "https://nolanlawson.com/2026/05/25/using-ai-to-write-better-code-more-slowly/", "seen": "2026-05-26", "quote": "Note that you’ll need gh installed, and also that I use Claude with Opus 4.7 on xhigh thinking and Codex with GPT 5.5 on high thinking.", "history": [] }, { "name": "Cursor Bugbot", "icon": "cursor", "role": "third reviewer on PRs", "surface": "web", "inferred": true, "source": "https://nolanlawson.com/2026/05/25/using-ai-to-write-better-code-more-slowly/", "seen": "2026-05-25", "quote": "Run a Claude sub-agent, Codex, and Cursor Bugbot to find bugs in this PR ranked by critical/high/medium/low. Once they’re all done, review their findings, do your own research to rule out false positives, and write a final report.", "history": [] } ], "models": [ { "name": "Claude Opus 4.7", "icon": "claude", "detail": "xhigh thinking", "source": "https://nolanlawson.com/2026/05/25/using-ai-to-write-better-code-more-slowly/", "seen": "2026-05-26", "quote": "Note that you’ll need gh installed, and also that I use Claude with Opus 4.7 on xhigh thinking and Codex with GPT 5.5 on high thinking.", "history": [] }, { "name": "GPT 5.5", "icon": "openai", "detail": "high thinking, in Codex", "source": "https://nolanlawson.com/2026/05/25/using-ai-to-write-better-code-more-slowly/", "seen": "2026-05-26", "quote": "Note that you’ll need gh installed, and also that I use Claude with Opus 4.7 on xhigh thinking and Codex with GPT 5.5 on high thinking.", "history": [] } ], "review_style": { "value": "Three agents hunt bugs in every PR; he triages the ranked report and guides the fixes", "source": "https://nolanlawson.com/2026/05/25/using-ai-to-write-better-code-more-slowly/", "seen": "2026-05-25", "quote": "Run a Claude sub-agent, Codex, and Cursor Bugbot to find bugs in this PR ranked by critical/high/medium/low. Once they’re all done, review their findings, do your own research to rule out false positives, and write a final report.", "kind": "agent-reviews", "also": [ { "source": "https://nolanlawson.com/2026/05/25/using-ai-to-write-better-code-more-slowly/", "seen": "2026-05-25", "quote": "Have an agent fix all the criticals and highs (with my guidance on the proper solution), then repeat until no criticals/highs" }, { "source": "https://nolanlawson.com/2026/08/16/you-can-just-choose-how-many-bugs-you-want-now/", "seen": "2026-08-16", "quote": "For example, I find myself spending a lot of time these days in code review, using tools like my triple-agent code review skill as well as Geoffrey Litt’s explain-diff skill." } ] }, "harness": null, "parallelism": null, "rules_files": null, "skills": { "value": [ "code-review-turbo", "explain-diff" ], "count": 2, "featured": [ "code-review-turbo", "explain-diff" ], "source": "https://nolanlawson.com/2026/08/16/you-can-just-choose-how-many-bugs-you-want-now/", "seen": "2026-08-16" }, "mcp_servers": null, "hooks": null, "signature": { "value": "code-review-turbo", "kind": "skill", "caption": "three-agent PR bug hunt", "why": "Our pick: Claude, Codex and Cursor Bugbot each review the PR, then the main agent cross-checks their findings to drop hallucinated bugs.", "editorial": true, "source": "https://nolanlawson.com/2026/05/25/using-ai-to-write-better-code-more-slowly/", "seen": "2026-05-25", "quote": "In my experience, this skill always finds tons of bugs in a PR, and the false positive rate is near zero." }, "archetype": { "value": "Purist", "editorial": true, "why": "He uses agents to make code better, not faster: multi-agent review on every PR, and no AI in his writing." }, "signature_quote": { "value": "And you can use them just as effectively to write high-quality code more slowly.", "source": "https://nolanlawson.com/2026/05/25/using-ai-to-write-better-code-more-slowly/", "seen": "2026-05-25" } }, { "schema": 2, "handle": "orta", "status": "ok", "checked": "2026-10-06", "name": { "value": "Orta Therox", "source": "https://github.com/orta" }, "x": null, "github": { "value": "orta", "source": "https://github.com/orta" }, "site": { "value": "https://orta.io", "source": "https://github.com/orta" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/49038?v=4", "source": "https://github.com/orta" }, "known_for": { "value": "Puzzmo, CocoaPods, Danger", "source": "https://github.com/orta" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary, only harness", "surface": "ide", "source": "https://blog.puzzmo.com/posts/2026/07/03/twelve-months-of-claude/", "seen": "2026-07-03", "quote": "I use VS Code for all my coding with Claude Code as an extension.", "also": [ { "source": "https://blog.puzzmo.com/posts/2026/07/03/twelve-months-of-claude/", "seen": "2026-07-03", "quote": "I don’t plan on trying anything other than Claude Code, there is still so much to master." } ], "history": [ { "value": "Cursor", "surface": "ide", "source": "https://blog.puzzmo.com/posts/2025/06/07/orta-on-claude/", "seen": "2025-06-07", "quote": "I’ve gone through two epoch moments with these tools, the first was adopting Cursor as my text editor." }, { "value": "Claude Code in the terminal", "surface": "cli", "source": "https://blog.puzzmo.com/posts/2025/06/07/orta-on-claude/", "seen": "2025-06-07", "quote": "Claude Code is a background app which runs in the terminal which is sortof a long-running connection to an LLM which can do very long running tasks and has a sense of how to verify things in the way that you would when writing code." } ] }, { "name": "Claude Code on the web", "icon": "claudecode", "role": "scheduled doc checks", "surface": "web", "source": "https://bsky.app/profile/orta.io/post/3miioilpuds2e", "seen": "2026-04-02", "quote": "I've been trying to use the new scheduled claude code (web) features to do this, so far it's caught a few typos and links" } ], "models": [ { "name": "Fable", "icon": "claude", "source": "https://blog.puzzmo.com/posts/2026/07/03/twelve-months-of-claude/", "seen": "2026-07-03", "quote": "For most of the year, the models themselves have come and gone. I didn’t really think much about them. This changed with Fable, which I think was a significant jump in capabilities.", "history": [ { "value": "Sonnet 4.5", "source": "https://blog.puzzmo.com/posts/2025/12/9/tech-2025/", "seen": "2025-12-09", "quote": "I’m finding myself using Claude Code almost every time I open up my editor, and it’s capabilities feels to have had quite a boost with the introduction of Sonnet 4.5." } ] } ], "review_style": { "value": "Reads all Puzzmo code; Claude reviews every PR too", "source": "https://blog.puzzmo.com/posts/2026/09/28/re-evaluating-principles/", "seen": "2026-09-28", "quote": "I give pretty specific instructions, read all the code for Puzzmo systems and am very present in reviewing the output of an LLM but every line of code used to be a journey and now it’s a transaction.", "kind": "reads-all", "also": [ { "source": "https://bsky.app/profile/orta.io/post/3mwls4ka2xs2y", "seen": "2026-09-28", "quote": "Yeah, I am definitely the one making that pile bigger and so I have to triple review my own PRs nowadays!" }, { "source": "https://blog.puzzmo.com/posts/2026/07/03/twelve-months-of-claude/", "seen": "2026-07-03", "quote": "I have Claude automatically review all Pull Requests in Puzzmo. For me, this is my main review partner on a pull request (outside of future me) - I’ve found it to be a good ‘you missed this’ sorta review, but it is sorta marking itself and so reviews have a tendency to be ‘This is great.’" } ] }, "harness": { "value": "VS Code with the Claude Code extension, multiple clones of the monorepo instead of worktrees", "short": "clones", "source": "https://blog.puzzmo.com/posts/2026/07/03/twelve-months-of-claude/", "seen": "2026-07-03", "quote": "I still operate with multiple clones, though I’ve raised my limit to 3 clones of the Puzzmo monorepo.", "history": [ { "value": "Two clones, different VS Code profiles", "source": "https://blog.puzzmo.com/posts/2025/07/30/six-weeks-of-claude-code/", "seen": "2025-07-30", "quote": "Anthropic have information about how to use worktrees - I would like to argue for a simpler approach. Two clones, different VS Code profiles." } ] }, "parallelism": { "value": 3, "short": "3 clones", "source": "https://blog.puzzmo.com/posts/2026/07/03/twelve-months-of-claude/", "seen": "2026-07-03", "quote": "I still operate with multiple clones, though I’ve raised my limit to 3 clones of the Puzzmo monorepo.", "history": [ { "value": 2, "source": "https://blog.puzzmo.com/posts/2025/07/30/six-weeks-of-claude-code/", "seen": "2025-07-30", "quote": "Two clones, different VS Code profiles." } ] }, "rules_files": { "value": "CLAUDE.md (132 lines, keytrace)", "short": "CLAUDE.md", "url": "https://github.com/orta/keytrace/blob/e2bca7b461854eae26c5e408a7cdec7ba6355540/CLAUDE.md", "source": "https://github.com/orta/keytrace/blob/e2bca7b461854eae26c5e408a7cdec7ba6355540/CLAUDE.md#L11", "seen": "2026-07-30", "quote": "- Do not push to remote or deploy unless explicitly asked." }, "skills": { "value": [ "commit" ], "count": 1, "featured": [ "commit" ], "source": "https://blog.puzzmo.com/posts/2026/07/03/twelve-months-of-claude/", "seen": "2026-07-03" }, "mcp_servers": null, "hooks": null, "signature": { "value": "/commit", "kind": "skill", "caption": "human writes the headline", "why": "Our pick: he types the first line of every commit himself and Claude only adds the body from the diff, so humans still write for humans.", "editorial": true, "source": "https://blog.puzzmo.com/posts/2026/07/03/twelve-months-of-claude/", "seen": "2026-07-03", "quote": "The argument to `/commit` is the first line of the commit message. Use it verbatim — no reformatting, no conventional-commit prefix, no capitalization fixes. If there is no argument, stop and ask the user for one; the point of this skill is that the human writes the headline." }, "archetype": { "value": "Purist", "editorial": true, "why": "One harness in his editor, no MCP, and he reads all the Puzzmo code himself." }, "signature_quote": { "value": "every line of code used to be a journey and now it’s a transaction.", "source": "https://blog.puzzmo.com/posts/2026/09/28/re-evaluating-principles/", "seen": "2026-09-28" } }, { "schema": 2, "handle": "paulirish", "status": "ok", "checked": "2026-10-07", "name": { "value": "Paul Irish", "source": "https://github.com/paulirish" }, "x": { "value": "paul_irish", "source": "https://x.com/paul_irish" }, "github": { "value": "paulirish", "source": "https://github.com/paulirish" }, "site": { "value": "https://paul.irish", "source": "https://github.com/paulirish" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/39191?v=4", "source": "https://github.com/paulirish" }, "known_for": { "value": "Chrome DevTools, Lighthouse", "source": "https://toot.cafe/@paul_irish" }, "agents": [ { "name": "Codex", "icon": "codex", "role": "opens PRs and reviews them (cloud)", "surface": "web", "source": "https://github.com/paulirish/carlo/pull/3", "seen": "2026-09-25", "quote": "[Codex Task](https://chatgpt.com/codex/cloud/tasks/task_e_6aaa2a657c0083318d7bebc85c0204b8)", "inferred": true, "history": [ { "value": "Gemini CLI and Claude Code (skills linked into both)", "surface": "cli", "source": "https://github.com/paulirish/dotfiles/blob/39644bfba35e30922495fc9a900dc5f50c9ef68d/agents/paulirish-skills/install-skills.sh#L6-L7", "seen": "2026-02-23", "quote": "GEMINI_TARGET_DIR=\"$HOME/.gemini/skills\"\nCLAUDE_TARGET_DIR=\"$HOME/.claude/skills\"" } ] }, { "name": "GitHub Copilot coding agent", "icon": "githubcopilot", "role": "fixes review findings on PRs", "surface": "web", "source": "https://github.com/paulirish/carlo/pull/3#issuecomment-5827505180", "seen": "2026-09-25", "quote": "@copilot address the review feedback and fix the falling CI" }, { "name": "Jules", "icon": null, "role": "cloud PRs", "surface": "web", "source": "https://github.com/paulirish/carlo/pull/2", "seen": "2026-09-07", "quote": "*PR created automatically by Jules for task [13436147436147156985](https://jules.google.com/task/13436147436147156985) started by @paulirish*" } ], "models": [ { "name": "Gemini 3 Flash (preview)", "icon": "gemini", "detail": "commit messages via llm", "source": "https://github.com/paulirish/dotfiles/blob/d27a39a78a36bc5548e320f9bcc064f13b3c0323/fish/aliases.fish#L41", "seen": "2026-04-02", "quote": "abbr write-commit-message \"git diffbranch | llm --system-fragment ~/code/dotfiles/private/prompts/devtools-commit-message.md --model=gemini-3-flash-preview\"", "also": [ { "source": "https://github.com/paulirish/dotfiles/blob/d27a39a78a36bc5548e320f9bcc064f13b3c0323/fish/aliases.fish#L155", "seen": "2026-04-02", "quote": "abbr geminif 'gemini --model gemini-3-flash-preview'" } ], "history": [] } ], "review_style": { "value": "Codex bot reviews every PR; he has Copilot fix the findings", "kind": "agent-reviews", "source": "https://github.com/paulirish/carlo/pull/3#issuecomment-5827505180", "seen": "2026-09-25", "quote": "@copilot address the review feedback and fix the falling CI", "also": [ { "source": "https://github.com/paulirish/carlo/pull/3#issuecomment-5842010944", "seen": "2026-09-26", "quote": "@copilot 12 progress checks??? I'm not sure that's a fast lane...." }, { "source": "https://github.com/paulirish-agent", "seen": null, "undated": true, "quote": "Experimenting w/ putting agent commits under a dedicated GH account. Especially when there's less code review. We'll see. ~@paulirish" } ], "inferred": true }, "harness": { "value": "Cloud agents on GitHub PRs, steered from PR comments", "short": "cloud PRs", "source": "https://github.com/paulirish/carlo/pull/3#issuecomment-5827627920", "seen": "2026-09-25", "quote": "@copilot do our ci tests need a rethink given that we're just starting the modernization plan?", "history": [] }, "parallelism": null, "rules_files": { "value": "AGENTS.md per repo (CLAUDE.md just imports it), linked to a LEXICON.md", "short": "AGENTS.md", "url": "https://github.com/paulirish/chromestatus/blob/f999d25bc6964083f7e51c8b2303cc3ef0d6e2f7/AGENTS.md", "source": "https://github.com/paulirish/chromestatus/blob/f999d25bc6964083f7e51c8b2303cc3ef0d6e2f7/AGENTS.md#L9", "seen": "2026-10-03", "quote": "Future code will be reasoned about, rewritten, and maintained by LLMs and high-level humans who operate on semantic representations." }, "skills": { "value": [ "buildless-types", "distill-page", "fetch-as-markdown", "github-pr-comments", "google-ai-search", "heap-snapshot-inspection", "hot-reload-chrome-ext", "lexicon", "modern-css", "npm-trusted-publishing", "pauls-project-setup", "qmd-expert" ], "count": 12, "featured": [ "lexicon", "pauls-project-setup", "github-pr-comments" ], "source": "https://github.com/paulirish/dotfiles/tree/d7607a4c2383bf7300032b2c7b89da87d6f9dcc5/agents/skills", "seen": "2026-10-03" }, "mcp_servers": null, "hooks": null, "signature": { "value": "paulirish-agent", "kind": "habit", "caption": "a GitHub account for agents", "why": "Our pick: agent commits land under their own GitHub identity, so his history shows what he wrote and what the agent wrote.", "editorial": true, "source": "https://github.com/paulirish-agent", "seen": null, "undated": true, "quote": "Experimenting w/ putting agent commits under a dedicated GH account. Especially when there's less code review. We'll see. ~@paulirish" }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "Hands PRs to Codex, Jules and Copilot in the cloud and steers them from PR comments." }, "signature_quote": { "value": "This allows coding agents to iterate and test logic instantly via CLI/Node unit tests without browser automation overhead (which is ~20x slower).", "source": "https://github.com/paulirish/dotfiles/blob/384da0d2aaca4cb0287f046e1a9791a405f00b17/agents/skills/pauls-project-setup/SKILL.md#L19", "seen": "2026-08-10" } }, { "schema": 2, "handle": "rasbt", "status": "ok", "checked": "2026-10-07", "name": { "value": "Sebastian Raschka", "source": "https://github.com/rasbt" }, "x": { "value": "rasbt", "source": "https://github.com/rasbt" }, "github": { "value": "rasbt", "source": "https://github.com/rasbt" }, "site": { "value": "https://sebastianraschka.com", "source": "https://github.com/rasbt" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/5618407?v=4", "source": "https://github.com/rasbt" }, "known_for": { "value": "Build an LLM From Scratch author", "source": "https://github.com/rasbt/LLMs-from-scratch" }, "agents": [ { "name": "Codex", "icon": "codex", "role": "daily driver, alternates with Claude Code", "surface": null, "source": "https://magazine.sebastianraschka.com/p/using-local-coding-agents", "seen": "2026-06-27", "quote": "I have to admit that I still primarily alternate between Codex and Claude Code as my daily drivers, for now (and just to keep up with the new tooling and functions that are constantly being added).", "history": [ { "value": "Codex plugin for VS Code", "surface": "ide", "source": "https://podscripts.co/podcasts/lex-fridman-podcast/490-state-of-ai-in-2026-llms-coding-scaling-laws-china-agents-gpus-agi", "seen": "2026-01-31", "quote": "So I use the Codex plugin for VS code." } ] }, { "name": "Claude Code", "icon": "claudecode", "role": "daily driver, alternates with Codex", "surface": null, "source": "https://magazine.sebastianraschka.com/p/using-local-coding-agents", "seen": "2026-06-27", "quote": "I have to admit that I still primarily alternate between Codex and Claude Code as my daily drivers, for now (and just to keep up with the new tooling and functions that are constantly being added)." }, { "name": "Qwen Code", "icon": "qwen", "role": "fully local stack, with Qwen3.6", "surface": "cli", "surface_source": { "source": "https://magazine.sebastianraschka.com/p/using-local-coding-agents", "seen": "2026-06-27", "quote": "After completing the installation, we can now launch the Qwen-Code client via the qwen command from the terminal to complete the setup and connect to the locally served LLM." }, "source": "https://magazine.sebastianraschka.com/p/using-local-coding-agents", "seen": "2026-06-27", "quote": "However, I’ve been using local solutions for a while, too, to test things and because it somehow gives me joy to have and use a fully local setup (versus proprietary services).", "also": [ { "source": "https://magazine.sebastianraschka.com/p/using-local-coding-agents", "seen": "2026-06-27", "quote": "I can run both Codex (with the latest GPT model) and Qwen-Code with a local Qwen model side by side on the same machine without having to switch manually back and forth between models." } ] } ], "models": [ { "name": "GPT-5.5", "icon": "openai", "detail": "in Codex", "inferred": true, "source": "https://magazine.sebastianraschka.com/p/using-local-coding-agents", "seen": "2026-06-27", "quote": "Ask a trusted agent you used before (like GPT 5.5 in Codex or Opus 4.8 in Claude Code) to review it with a focused prompt.", "history": [] }, { "name": "Qwen3.6-35B-A3B", "icon": "qwen", "detail": "local, via Ollama", "source": "https://magazine.sebastianraschka.com/p/using-local-coding-agents", "seen": "2026-06-27", "quote": "If we accept the reported findings and risks (personally, I didn’t see any red flags), we can now proceed with the installation and hook up our local Qwen3.6-35B-A3B model to Qwen Code (and Codex and Claude Code in the next sections).", "history": [] } ], "review_style": { "value": "Wants to see what the agent is doing; writes the code he cares about himself", "kind": "unknown", "inferred": true, "source": "https://lexfridman.com/ai-sota-2026-transcript", "seen": "2026-01-31", "quote": "I’m not quite there yet where I’m comfortable with that because maybe I’m a control freak, but I still like to see what’s going on.", "also": [ { "source": "https://magazine.sebastianraschka.com/p/state-of-llms-2025", "seen": "2025-12-30", "quote": "Today, I still write most of the code I care about myself. With “care about,” I mean in contexts where it matters that I understand the code and that the code is correct." } ] }, "harness": { "value": "Agents run on a separate DGX Spark; local models served by Ollama, tunnelled to the Mac over SSH", "short": "DGX Spark", "source": "https://magazine.sebastianraschka.com/p/using-local-coding-agents", "seen": "2026-06-27", "quote": "By the way, personally, I run my agents almost exclusively on my DGX Spark because I don’t want my Mac Mini to get too hot and I want to have the RAM available for other tasks.", "history": [] }, "parallelism": null, "rules_files": null, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "Audit the agent first", "kind": "habit", "caption": "before installing any harness", "why": "Our pick: before trying a new coding agent he clones its repo and has a trusted agent audit it for egress and blast radius.", "editorial": true, "source": "https://magazine.sebastianraschka.com/p/using-local-coding-agents", "seen": "2026-06-27", "quote": "For instance, when trying new (coding) agents, I like to\nDo an audit of the (open-source) agent code base first." }, "archetype": { "value": "Purist", "editorial": true, "why": "Keeps the code he cares about in his own hands and keeps a fully local agent stack." }, "signature_quote": { "value": "Today, I still write most of the code I care about myself.", "more": "With “care about,” I mean in contexts where it matters that I understand the code and that the code is correct.", "source": "https://magazine.sebastianraschka.com/p/state-of-llms-2025", "seen": "2025-12-30" } }, { "schema": 2, "handle": "rauchg", "status": "ok", "checked": "2026-10-07", "name": { "value": "Guillermo Rauch", "source": "https://github.com/rauchg" }, "x": { "value": "rauchg", "source": "https://x.com/rauchg" }, "github": { "value": "rauchg", "source": "https://github.com/rauchg" }, "site": { "value": "https://rauchg.com", "source": "https://x.com/rauchg" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/13041?v=4", "source": "https://github.com/rauchg" }, "known_for": { "value": "CEO of Vercel", "source": "https://x.com/rauchg" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary (from commits)", "surface": null, "inferred": true, "source": "https://github.com/rauchg/typing-stats/commit/2b7bd1c99f596d8ba111564007ca122faabe63fc", "seen": "2026-05-10", "quote": "Co-Authored-By: Claude Opus 4.6 (1M context) ", "also": [ { "source": "https://github.com/rauchg/typing-stats/commit/0d6903f98e739fdb99174ed964359baec605b208", "seen": "2026-01-06", "quote": "🤖 Generated with [Claude Code](https://claude.com/claude-code)" } ], "history": [] } ], "models": [ { "name": "Claude Opus 4.6", "icon": "claude", "detail": "1M context", "inferred": true, "source": "https://github.com/rauchg/typing-stats/commit/2b7bd1c99f596d8ba111564007ca122faabe63fc", "seen": "2026-05-10", "quote": "Co-Authored-By: Claude Opus 4.6 (1M context) ", "history": [ { "value": "Claude Opus 4.5", "inferred": true, "source": "https://github.com/rauchg/typing-stats/commit/3729fdbb0db7ec6400aa26737912f3d4bf65497f", "seen": "2026-01-19", "quote": "Co-Authored-By: Claude Opus 4.5 " } ] } ], "review_style": null, "harness": null, "parallelism": null, "rules_files": { "value": "AGENTS.md per repo, written as a first-person product spec", "short": "AGENTS.md", "url": "https://github.com/rauchg/typing-stats/blob/0d6903f98e739fdb99174ed964359baec605b208/AGENTS.md", "source": "https://github.com/rauchg/typing-stats/blob/0d6903f98e739fdb99174ed964359baec605b208/AGENTS.md#L5", "seen": "2026-01-06", "quote": "- No xcode garbage. I want to build from CLI" }, "skills": { "value": [ "gdp-ts", "ui-recording-timeline" ], "count": 2, "featured": [ "gdp-ts", "ui-recording-timeline" ], "source": "https://www.skills.sh/rauchg", "seen": "2026-10-04" }, "mcp_servers": null, "hooks": null, "signature": { "value": "gdp-ts", "kind": "skill", "caption": "auth checks as type proofs", "why": "Our pick: instead of reviewing more agent code, he wrote a library, linter and skill that make the type checker reject missing authorization checks.", "editorial": true, "source": "https://x.com/rauchg/status/2107119811444748555", "seen": "2026-10-05", "quote": "The typechecker verifies these proofs at compile time, preventing your team and agents from shipping catastrophic security (and other kinds of) bugs." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Publishes his own agent skills, including a type-level guardrail library built for agents." }, "signature_quote": { "value": "Agents are writing more code than we can review, and they *thrive* in tight loops with hard constraints that would frustrate us.", "source": "https://x.com/rauchg/status/2107119811444748555", "seen": "2026-10-05" } }, { "schema": 2, "handle": "ry", "status": "ok", "checked": "2026-10-06", "name": { "value": "Ryan Dahl", "source": "https://github.com/ry" }, "x": { "value": "rough__sea", "source": "https://github.com/ry" }, "github": { "value": "ry", "source": "https://github.com/ry" }, "site": { "value": "https://tinyclouds.org", "source": "https://github.com/ry" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/80?v=4", "source": "https://github.com/ry" }, "known_for": { "value": "Creator of Node.js and Deno", "source": "https://bsky.app/profile/tinyclouds.org" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary", "surface": null, "source": "https://deno.com/blog/clawpatrol", "seen": "2026-05-21", "quote": "Most of what we run (Claude Code, Codex) is code we install, not code we wrote.", "inferred": true, "also": [ { "source": "https://github.com/denoland/clawpatrol/commit/a1f7d672254f786caa232cd0de363c8c5cca1c8f", "seen": "2026-05-12", "quote": "AGENTS.md is a symlink to\n.claude/CLAUDE.md so the agents.md convention (Codex) and the\n.claude/ convention (Claude Code) point at the same source of truth." }, { "source": "https://github.com/ry/tinyclouds/commit/003bd7ea0c312fdb39be9d31d944b460f946922c", "seen": "2025-07-30", "quote": "🤖 Generated with [Claude Code](https://claude.ai/code)" } ] }, { "name": "Codex", "icon": "codex", "role": "second agent", "surface": null, "source": "https://github.com/denoland/clawpatrol/commit/a1f7d672254f786caa232cd0de363c8c5cca1c8f", "seen": "2026-05-12", "quote": "AGENTS.md is a symlink to\n.claude/CLAUDE.md so the agents.md convention (Codex) and the\n.claude/ convention (Claude Code) point at the same source of truth.", "also": [ { "source": "https://deno.com/blog/clawpatrol", "seen": "2026-05-21", "quote": "Most of what we run (Claude Code, Codex) is code we install, not code we wrote." } ] } ], "models": [ { "name": "Claude Opus 4.6", "icon": "claude", "detail": "1M context", "source": "https://github.com/denoland/ronly/commit/4e89830a331e56fb195ad16a85d2f96895fb5507", "seen": "2026-03-24", "quote": "Co-Authored-By: Claude Opus 4.6 (1M context) ", "history": [], "inferred": true } ], "review_style": null, "harness": { "value": "Agents on standalone VMs, 24/7, traffic gated by Claw Patrol", "short": "standalone VMs", "source": "https://deno.com/blog/clawpatrol", "seen": "2026-05-21", "quote": "We already run our agents on standalone VMs; for us these are only marginally useful.", "history": [] }, "parallelism": null, "rules_files": { "value": "Two-line CLAUDE.md that points the agent at spec.md", "short": "CLAUDE.md", "url": "https://github.com/denoland/ronly/blob/fc49a71bc269e7f3da08b223c3d16115fa1057f8/CLAUDE.md", "source": "https://github.com/denoland/ronly/blob/fc49a71bc269e7f3da08b223c3d16115fa1057f8/CLAUDE.md#L1-L3", "seen": "2026-03-24", "quote": "read spec.md\n\nremember to keep the README updated - it's the main documentation" }, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "Claw Patrol", "kind": "tool", "caption": "firewall between agents and prod", "why": "Our pick: rather than trust the agent, Deno routes its agents' production traffic through a gateway that parses SQL, Kubernetes and HTTP and enforces rules.", "editorial": true, "source": "https://deno.com/blog/clawpatrol", "seen": "2026-05-21", "quote": "Most of what we run (Claude Code, Codex) is code we install, not code we wrote. Any solution has to sit outside the agent." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Builds infrastructure for agents: Claw Patrol, ronly and t4a." }, "signature_quote": { "value": "the era of humans writing code is over.", "more": "That's not to say SWEs don't have work to do, but writing syntax directly is not it.", "source": "https://x.com/rough__sea/status/2013280952370573666", "seen": "2026-01-19" } }, { "schema": 2, "handle": "sgoedecke", "status": "ok", "checked": "2026-10-12", "name": { "value": "Sean Goedecke", "source": "https://github.com/sgoedecke" }, "x": null, "github": { "value": "sgoedecke", "source": "https://github.com/sgoedecke" }, "site": { "value": "https://www.seangoedecke.com", "source": "https://github.com/sgoedecke" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/19204567?v=4", "source": "https://github.com/sgoedecke" }, "known_for": { "value": "Writing on AI and big tech", "source": "https://www.seangoedecke.com/about/" }, "agents": [ { "name": "GitHub Copilot app", "icon": "githubcopilot", "role": "primary, tens of sessions a day", "surface": "app", "source": "https://www.seangoedecke.com/how-i-use-llms-in-2026/", "seen": "2026-05-17", "quote": "Now I use the GitHub Copilot app a lot (tens of sessions per day).", "history": [ { "value": "VS Code windows", "surface": "ide", "source": "https://www.seangoedecke.com/how-i-use-llms-in-2026/", "seen": "2026-05-17", "quote": "In late 2025 I used a lot of open VSCode windows." }, { "value": "Copilot CLI in terminal tabs", "surface": "cli", "source": "https://www.seangoedecke.com/how-i-use-llms-in-2026/", "seen": "2026-05-17", "quote": "In early 2026, that changed to terminal tabs with the Copilot CLI, particularly when I needed to make changes across multiple repos at the same time." } ] }, { "name": "Copilot CLI", "icon": "githubcopilot", "role": "local setup and config chores", "surface": "cli", "source": "https://www.seangoedecke.com/how-i-use-llms-in-2026/", "seen": "2026-05-17", "quote": "For instance, if my nvm installation is not switching my Node version correctly, I will often open a Copilot CLI agent and ask it to figure it out.", "history": [] } ], "models": [ { "name": "GPT-6-Astra", "icon": "openai", "source": "https://www.seangoedecke.com/you-should-all-be-asking-way-more-questions/", "seen": "2026-09-25", "quote": "I mostly work with GPT-6-Astra and Claude Opus 5.5.", "history": [] }, { "name": "Claude Opus 5.5", "icon": "claude", "source": "https://www.seangoedecke.com/you-should-all-be-asking-way-more-questions/", "seen": "2026-09-25", "quote": "I mostly work with GPT-6-Astra and Claude Opus 5.5.", "history": [] } ], "review_style": { "value": "Skims each attempt for ~30 seconds, rejects most, then fully reviews the one he keeps", "source": "https://www.seangoedecke.com/how-i-use-llms-in-2026/", "seen": "2026-05-17", "quote": "I do a lot of skimming through and evaluating agent changes. Most of the time I reject them entirely, just based on “eh, that’s not what I was thinking”. On average it takes me about thirty seconds to make this initial assessment.", "kind": "reads-all", "also": [ { "source": "https://www.seangoedecke.com/how-i-use-llms-in-2026/", "seen": "2026-05-17", "quote": "If the change looks alright after that, I’ll dig in and do a proper review to make sure I understand it and it’s doing the right thing." }, { "source": "https://www.seangoedecke.com/you-should-all-be-asking-way-more-questions/", "seen": "2026-09-25", "quote": "About half the questions I ask get answers that convince me the model has made a mistake: it should have reused the existing subsystem for X, or authed to Y differently, or kept the Z implementation simple, and so on." }, { "source": "https://www.seangoedecke.com/how-to-read-code/", "seen": "2026-10-07", "quote": "Once I’m confident I understand the diff, only then will I sit down and carefully read it end-to-end." } ] }, "harness": { "value": "GitHub Copilot app; separate sessions per task for exploring, implementing, reviewing and testing", "short": "Copilot app", "source": "https://www.seangoedecke.com/how-to-keep-thinking/", "seen": "2026-08-07", "quote": "I routinely use six or seven different agent sessions on the same task: one for exploration, two or three for trying out different implementations, two or three for review, one for manual testing, and so on.", "history": [ { "value": "Terminal tabs with Copilot CLI across repos", "source": "https://www.seangoedecke.com/how-i-use-llms-in-2026/", "seen": "2026-05-17", "quote": "In early 2026, that changed to terminal tabs with the Copilot CLI, particularly when I needed to make changes across multiple repos at the same time." } ] }, "parallelism": { "value": 7, "short": "6–7 sessions per task", "source": "https://www.seangoedecke.com/how-to-keep-thinking/", "seen": "2026-08-07", "quote": "I routinely use six or seven different agent sessions on the same task: one for exploration, two or three for trying out different implementations, two or three for review, one for manual testing, and so on.", "inferred": true, "history": [] }, "rules_files": null, "skills": { "value": [ "extract-features-clamp-inference" ], "count": 1, "featured": [ "extract-features-clamp-inference" ], "source": "https://github.com/sgoedecke/skills/tree/0f38b38c2ef1691a28340d8faa9abfe39467c99d/skills", "seen": "2026-02-17" }, "mcp_servers": { "value": [ "runpod" ], "source": "https://www.seangoedecke.com/tell-agents-the-why/", "seen": "2026-09-15", "quote": "Hello. You should have Runpod access via MCP (if not, tell me and I’ll fix it)." }, "hooks": null, "signature": { "value": "30-second triage", "kind": "habit", "caption": "rejects most agent attempts", "why": "Our pick: he lets agents attempt every change, throws most away after a half-minute skim, and only reviews in depth the attempt worth keeping.", "editorial": true, "source": "https://www.seangoedecke.com/how-i-use-llms-in-2026/", "seen": "2026-05-17", "quote": "For difficult tasks, I’ll often reject five or six (or more!) agent attempts before accepting one as good enough to work with, or giving up and making the change by hand." }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "Six or seven sessions per task, each with a role: explore, implement, review, test." }, "signature_quote": { "value": "But now I start every single change by asking an agent to solve the problem, and usually push the PR after a single editing pass.", "source": "https://www.seangoedecke.com/how-i-use-llms-in-2026/", "seen": "2026-05-17" } }, { "schema": 2, "handle": "slavingia", "status": "ok", "checked": "2026-10-12", "name": { "value": "Sahil Lavingia", "source": "https://github.com/slavingia" }, "x": { "value": "shl", "source": "https://github.com/slavingia" }, "github": { "value": "slavingia", "source": "https://github.com/slavingia" }, "site": { "value": "https://sahillavingia.com", "source": "https://www.lennysnewsletter.com/p/gumroad-ceos-playbook-to-40x-his" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/74396?v=4", "source": "https://github.com/slavingia" }, "known_for": { "value": "Gumroad founder", "source": "https://www.lennysnewsletter.com/p/gumroad-ceos-playbook-to-40x-his" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary (from his repos)", "surface": null, "inferred": true, "source": "https://github.com/slavingia/prototypes/blob/9c921329300d64ecb328e8fdb0fd633fc087d810/README.md#L17-L18", "seen": "2026-07-02", "quote": "`serve.sh` wraps `python3 -m http.server` and opens your browser. (In Claude\nCode you can also just run the `/serve` skill.)", "history": [ { "value": "v0, then Devin, then Cursor", "surface": null, "source": "https://www.lennysnewsletter.com/p/gumroad-ceos-playbook-to-40x-his", "seen": "2025-04-22", "quote": "So generally my flow is V zero Devin cursor is probably how I would say it." } ] }, { "name": "Codex", "icon": "codex", "role": "on Gumroad PRs, since Sep", "surface": null, "source": "https://github.com/antiwork/gumroad/pull/8193", "seen": "2026-10-06", "quote": "Written with OpenAI Codex (GPT-6).", "history": [] } ], "models": [ { "name": "Claude Fable 5", "icon": "claude", "inferred": true, "source": "https://github.com/slavingia/prototypes/commit/8985f8232b18e2068274f3a98c4d1fa31111f658", "seen": "2026-08-28", "quote": "Co-authored-by: Claude Fable 5 ", "history": [ { "value": "Claude Opus 4.5", "inferred": true, "source": "https://github.com/slavingia/bug-catcher-game/commit/7b75348543a7d6220194dcbf1590387f7d46c24a", "seen": "2026-01-10", "quote": "Co-Authored-By: Claude Opus 4.5 " } ] }, { "name": "GPT-6", "icon": "openai", "detail": "via Codex", "source": "https://github.com/antiwork/gumroad/pull/8193", "seen": "2026-10-06", "quote": "Written with OpenAI Codex (GPT-6).", "history": [] } ], "review_style": { "value": "Does not just hit merge: QA, then asks a colleague to check", "kind": "unknown", "inferred": true, "source": "https://www.lennysnewsletter.com/p/gumroad-ceos-playbook-to-40x-his", "seen": "2025-04-22", "quote": "I don't normally just hit merge on these things.", "also": [ { "source": "https://github.com/slavingia/prototypes/blob/9c921329300d64ecb328e8fdb0fd633fc087d810/CONTRIBUTING.md#L37-L42", "seen": "2026-08-25", "quote": "- **Screenshots are generated, and never stale.** Committed screenshots are\n produced by a script (e.g. [`archive/omnibar-in-app/tools/shots.mjs`](archive/omnibar-in-app/tools/shots.mjs)),\n not captured by hand. Whenever you change a prototype's look or behavior, **you\n must re-run its screenshot generator in the same change** so the images (and\n any gallery/README/PR that shows them) stay accurate. Never commit a change\n that leaves a screenshot showing an out-of-date design." } ] }, "harness": { "value": "GitHub issue per feedback round, agent opens a PR with preview link and generated screenshots", "short": "issue → PR", "inferred": true, "source": "https://github.com/slavingia/prototypes/pull/2", "seen": "2026-06-29", "quote": "Auto-generated prototype in response to the issue.", "history": [ { "value": "v0 prototype, Devin implements, Cursor finishes", "source": "https://www.lennysnewsletter.com/p/gumroad-ceos-playbook-to-40x-his", "seen": "2025-04-22", "quote": "Then I go to Devon and if Devon sort of fails to completely finish," } ] }, "parallelism": null, "rules_files": { "value": "CONTRIBUTING.md hard rules, for humans and AI assistants", "short": "CONTRIBUTING.md", "url": "https://github.com/slavingia/prototypes/blob/9c921329300d64ecb328e8fdb0fd633fc087d810/CONTRIBUTING.md", "source": "https://github.com/slavingia/prototypes/blob/9c921329300d64ecb328e8fdb0fd633fc087d810/CONTRIBUTING.md#L30-L36", "seen": "2026-08-25", "quote": "- **Follow the design system.** Every prototype — new or updated — must match\n [`DESIGN.md`](DESIGN.md), the IRS ODG (USWDS-based) styling standard. This is\n not optional: use its color tokens, Source Sans 3 type ramp, the expandable\n USWDS government banner, square containers/inputs (`4px` buttons), the\n blue/navy step indicators, and the disclaimer footer. Run the checklist at the\n bottom of `DESIGN.md` before committing. This applies equally to humans and to\n AI assistants working in this repo." }, "skills": { "value": [ "serve" ], "count": 1, "featured": [ "serve" ], "source": "https://github.com/slavingia/prototypes/tree/9c921329300d64ecb328e8fdb0fd633fc087d810/.claude/skills", "seen": "2026-07-01" }, "mcp_servers": null, "hooks": null, "signature": { "value": "Fourth-wall check", "kind": "rule", "caption": "enforced in CI on every PR", "why": "Our pick: a hard rule, checked by a linter in CI, that prototype copy never calls itself a prototype or explains its changes.", "editorial": true, "source": "https://github.com/slavingia/prototypes/blob/9c921329300d64ecb328e8fdb0fd633fc087d810/CONTRIBUTING.md#L17-L18", "seen": "2026-08-25", "quote": "- **Never break the fourth wall.** Every prototype must read as a *real,\n shipping IRS product*." }, "archetype": { "value": "Rules Lawyer", "editorial": true, "why": "Hard repo rules for agents, with a CI linter that enforces one of them." }, "signature_quote": { "value": "I think of it like flying a plane, like humans will take off, decide where to go and land typically", "source": "https://www.lennysnewsletter.com/p/gumroad-ceos-playbook-to-40x-his", "seen": "2025-04-22" } }, { "schema": 2, "handle": "snarktank", "status": "ok", "checked": "2026-10-06", "name": { "value": "Ryan Carson", "source": "https://github.com/snarktank" }, "x": { "value": "ryancarson", "source": "https://github.com/snarktank" }, "github": { "value": "snarktank", "source": "https://github.com/snarktank" }, "site": { "value": "https://ryancarson.com", "source": "https://github.com/snarktank" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/152063952?v=4", "source": "https://github.com/snarktank" }, "known_for": { "value": "Founder of Untangle", "source": "https://www.chatprd.ai/how-i-ai/how-ryan-carson-manages-40-prs-a-day-with-devin-and-codex" }, "agents": [ { "name": "Devin", "icon": null, "role": "primary, cloud agents", "surface": "web", "source": "https://x.com/ryancarson/status/2087538698543907048", "seen": "2026-08-12", "quote": "I would say 95% of my engineering is in the cloud now with @DevinAI, and I assume it will eventually be 100%.", "also": [ { "source": "https://x.com/ryancarson/status/2083538848131190806", "seen": "2026-08-01", "quote": "The majority of work I do now is on my phone with @DevinAI cloud agents." } ], "history": [ { "value": "Amp", "surface": null, "source": "https://x.com/ryancarson/status/2021974832657956927", "seen": "2026-02-12", "quote": "I still have ~$5k of credits from Amp but I'm now actively trying out Claude Code and Codex, which I'm paying for myself." }, { "value": "Codex as main driver", "surface": null, "source": "https://x.com/ryancarson/status/2022250412376531424", "seen": "2026-02-13", "quote": "Trying out Codex as my main driver now." }, { "value": "Symphony + Codex Mac app", "surface": "app", "source": "https://x.com/ryancarson/status/2031755002205655511", "seen": "2026-03-11", "quote": "It uses Symphony from @alex_frantic (oss) + Codex Mac app + @linear" } ] }, { "name": "Codex", "icon": "codex", "role": "local frontend work", "surface": null, "source": "https://x.com/ryancarson/status/2057798991086014655", "seen": "2026-05-22", "quote": "3. frontend eng local = codex" } ], "models": [ { "name": "Claude Opus 5.5", "icon": "claude", "detail": "engineering manager", "source": "https://x.com/ryancarson/status/2103926098081763798", "seen": "2026-09-26", "quote": "Engineering manager Opus 5.5 + @DevinAI + @linear", "history": [ { "value": "gpt-5.4 xhigh in Codex", "source": "https://x.com/ryancarson/status/2029919931328802863", "seen": "2026-03-06", "quote": "omg love gpt-5.4 xhigh in Codex" }, { "value": "Opus 4.8 in Devin", "source": "https://x.com/ryancarson/status/2072331594363543988", "seen": "2026-07-01", "quote": "Need to do then cost analysis now and compare it to Opus 4.8, which was my main choice in Devin before." } ] }, { "name": "SWE 1.7", "icon": null, "detail": "bulk coding in Devin", "source": "https://x.com/ryancarson/status/2076975558756339847", "seen": "2026-07-14", "quote": "Did almost all my engineering on SWE 1.7 yesterday.", "history": [] } ], "review_style": { "value": "Agents review and land; he watches a test video first", "source": "https://x.com/ryancarson/status/2061393372737085816", "seen": "2026-06-01", "quote": "Agents write/review/land 100% of our code now.", "kind": "agent-reviews", "also": [ { "source": "https://www.chatprd.ai/how-i-ai/how-ryan-carson-manages-40-prs-a-day-with-devin-and-codex", "seen": "2026-08-20", "quote": "Ryan watches the short video to visually confirm the feature works as expected. He then gives the final command: Video approved. Land it." } ] }, "harness": { "value": "Devin cloud sessions sorted into P0/P1/Bugs folders, Linear, Opus 5.5 as manager, zero local dev", "short": "Devin + Linear", "source": "https://x.com/ryancarson/status/2103926098081763798", "seen": "2026-09-26", "quote": "We do zero local dev at @HelloUntangle", "history": [ { "value": "Three local checkouts in split Ghostty panes", "source": "https://x.com/ryancarson/status/2024187188032778350", "seen": "2026-02-18", "quote": "I have three local checkouts so I can work on up to three branches in parallel without having to use git worktrees (which mess with my mind)." }, { "value": "Symphony code factory, 5 agents", "source": "https://x.com/ryancarson/status/2031755002205655511", "seen": "2026-03-11", "quote": "This is how I run 5 agents concurrently in a Code Factory to write/ship 100% of our code." } ] }, "parallelism": null, "rules_files": null, "skills": null, "mcp_servers": { "value": [ "next-devtools" ], "source": "https://x.com/ryancarson/status/2023886869231239485", "seen": "2026-02-17", "quote": "The next-devtools mcp is killer." }, "hooks": { "value": [ "pre-commit design-system linter" ], "source": "https://x.com/ryancarson/status/2028901454996447296", "seen": "2026-03-03", "quote": "design system with tokens (colors, spacing, sizes) enforced with a design-system custom linter on your pre-commit hook" }, "signature": { "value": "!land playbook", "kind": "skill", "caption": "refreshes skills on every PR", "why": "Our pick: his Devin playbook for every merge starts a child session that updates skills and docs, so they never go stale.", "editorial": true, "source": "https://x.com/ryancarson/status/2089719890097664413", "seen": "2026-08-18", "quote": "2. Make one of the playbook steps to start a child session to sweep all skills/docs and delete/upsert based on what changed in the PR." }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "Runs throngs of Devin sessions in priority folders, with Opus 5.5 as their manager." }, "signature_quote": { "value": "What the agents are doing is none of my business.", "source": "https://www.chatprd.ai/how-i-ai/how-ryan-carson-manages-40-prs-a-day-with-devin-and-codex", "seen": "2026-08-20" } }, { "schema": 2, "handle": "steveklabnik", "status": "ok", "checked": "2026-10-06", "name": { "value": "Steve Klabnik", "source": "https://github.com/steveklabnik" }, "x": null, "github": { "value": "steveklabnik", "source": "https://github.com/steveklabnik" }, "site": { "value": "https://steveklabnik.com", "source": "https://github.com/steveklabnik" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/27786?v=4", "source": "https://github.com/steveklabnik" }, "known_for": { "value": "Co-author of the Rust book, jj", "source": "https://steveklabnik.com/" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary, writes with Fable", "surface": "cli", "inferred": true, "surface_source": { "source": "https://steveklabnik.com/writing/getting-started-with-claude-for-software-development/", "seen": "2026-01-07", "quote": "Once you've installed it and logged in, you'll be at a terminal prompt." }, "source": "https://bsky.app/profile/steveklabnik.com/post/3mrvyzjfdac2c", "seen": "2026-07-31", "quote": "claude code and the codex app. i copy/paste things between them when i want them to talk to each other, nothing fancy yet", "also": [ { "source": "https://bsky.app/profile/steveklabnik.com/post/3mqmrr2qimc2w", "seen": "2026-07-14", "quote": "\"claude did that for me while i did other actual work and validated that there were no regressions\"" } ] }, { "name": "Codex", "icon": "codex", "role": "PR review, long goals", "surface": "app", "source": "https://bsky.app/profile/steveklabnik.com/post/3mrvyzjfdac2c", "seen": "2026-07-31", "quote": "claude code and the codex app. i copy/paste things between them when i want them to talk to each other, nothing fancy yet", "also": [ { "source": "https://bsky.app/profile/steveklabnik.com/post/3mqnbd3nddc24", "seen": "2026-07-14", "quote": "you just tell codex to do things and it goes and does them." } ] } ], "models": [ { "name": "Claude Fable", "icon": "claude", "inferred": true, "source": "https://bsky.app/profile/steveklabnik.com/post/3mpted2uoec2n", "seen": "2026-07-04", "quote": "During a loop overnight, Fable found a heap overflow in @rue-lang.dev’s standard library", "history": [ { "value": "Claude Opus 4.5", "source": "https://steveklabnik.com/writing/getting-started-with-claude-for-software-development/", "seen": "2026-01-07", "quote": "Opus 4.5 is wonderful." } ] }, { "name": "GPT Sol", "icon": "openai", "detail": "also his cross-reviewer", "source": "https://bsky.app/profile/steveklabnik.com/post/3mqexwgdemk2q", "seen": "2026-07-11", "quote": "Okay I am digging Sol. Like a lot\n\nSorry Claude", "history": [] } ], "review_style": { "value": "Codex reviews Fable's PRs; humans stay responsible for what they submit", "source": "https://bsky.app/profile/steveklabnik.com/post/3mstwd7yyyk2g", "seen": "2026-08-12", "quote": "Codex doing PR review for Fable", "kind": "unknown", "also": [ { "source": "https://bsky.app/profile/steveklabnik.com/post/3mth3h4po2k2c", "seen": "2026-08-19", "quote": "We use agents to develop software but have a strong culture of “you are responsible for the code you submit,” not “agents write and merge code with no review”" }, { "source": "https://github.com/rue-language/rue/blob/5200b119000163884fb28310fd49107c662761af/docs/process/codex.md#L9-L12", "seen": "2026-09-04", "quote": "The coordinating agent investigates, writes a precise brief, reviews the returned change, integrates it, and owns publication and cleanup." }, { "source": "https://bsky.app/profile/steveklabnik.com/post/3mrvt5237u22c", "seen": "2026-07-31", "quote": "i tend to use sol, but luna sometimes yeah" } ], "inferred": true }, "harness": { "value": "Claude Code and the Codex app side by side, bridged by copy/paste; a jj workspace per agent", "short": "jj workspaces", "source": "https://bsky.app/profile/steveklabnik.com/post/3mrvyzjfdac2c", "seen": "2026-07-31", "quote": "claude code and the codex app. i copy/paste things between them when i want them to talk to each other, nothing fancy yet", "also": [ { "source": "https://steveklabnik.com/writing/agentic-development-basics/", "seen": "2026-01-22", "quote": "I never bothered with `git worktree`s in the past, but I use `jj workspace`s with agents all the time now." } ], "history": [] }, "parallelism": { "value": 90, "short": "90 agents", "inferred": true, "source": "https://bsky.app/profile/steveklabnik.com/post/3mpz2kteeik25", "seen": "2026-07-06", "quote": "> 8/90 agents done\n\nlisten", "history": [] }, "rules_files": { "value": "AGENTS.md (257 lines, tool-neutral), CLAUDE.md points to it", "short": "AGENTS.md", "url": "https://github.com/rue-language/rue/blob/5200b119000163884fb28310fd49107c662761af/AGENTS.md", "source": "https://github.com/rue-language/rue/blob/5200b119000163884fb28310fd49107c662761af/AGENTS.md#L3-L5", "seen": "2026-09-05", "quote": "This is the canonical guidance for coding agents working in Rue. Tool-specific\ninstruction files (`CLAUDE.md` and the like) point here rather than copying\nthese rules, and nothing here assumes a particular agent product." }, "skills": { "value": [ "integrate-worker" ], "count": 1, "featured": [ "integrate-worker" ], "source": "https://github.com/rue-language/rue/tree/5200b119000163884fb28310fd49107c662761af/.claude/skills", "seen": "2026-09-03" }, "mcp_servers": null, "hooks": { "value": [ "bookmark-guard (PreToolUse on Edit/Write: blocks edits on a pushed PR branch)", "SessionStart: source ~/.profile" ], "source": "https://github.com/rue-language/rue/blob/5200b119000163884fb28310fd49107c662761af/.claude/hooks/bookmark-guard.sh#L2-L4", "seen": "2026-07-28", "quote": "# PreToolUse guard on Edit/Write: block edits while the working copy (@) sits\n# on a pushed PR branch. Editing @ then pushing mutates a queued PR — this\n# exact mistake put a blog post into PR #981's branch (2026-06-11) and cost" }, "signature": { "value": "rue-overnight-hunt", "kind": "tool", "caption": "bugs filed only with repros", "why": "Our pick: an overnight workflow that fans out bug finders across one compiler subsystem and files only findings whose repro was actually run.", "editorial": true, "source": "https://github.com/rue-language/rue/blob/5200b119000163884fb28310fd49107c662761af/.claude/workflows/rue-overnight-hunt.js#L3", "seen": "2026-07-02", "quote": "Autonomous Rue compiler bug hunt for one subsystem area: fan-out finders, verify with repros, return findings to file in Linear" }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "Runs Claude and OpenAI models as subagents side by side, with overnight fan-out hunts and one model reviewing another's PRs." }, "signature_quote": { "value": "The trick for software work is you don’t really talk to it.", "more": "You tell it to do stuff and then review the stuff it does", "source": "https://bsky.app/profile/steveklabnik.com/post/3mtan7eknis2y", "seen": "2026-08-17" } }, { "schema": 2, "handle": "swyxio", "status": "ok", "checked": "2026-10-06", "name": { "value": "swyx", "source": "https://x.com/swyx" }, "x": { "value": "swyx", "source": "https://x.com/swyx" }, "github": { "value": "swyxio", "source": "https://github.com/swyxio" }, "site": { "value": "https://www.swyx.io", "source": "https://github.com/swyxio" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/6764957?v=4", "source": "https://github.com/swyxio" }, "known_for": { "value": "Latent Space, AI Engineer", "source": "https://www.youtube.com/watch?v=zepu8Kk6FBQ" }, "agents": [ { "name": "Codex", "icon": "codex", "role": "opens most of his PRs", "surface": null, "inferred": true, "source": "https://github.com/swyxio/swyxdotio/pull/612", "seen": "2026-09-30", "quote": "codex/search-cmd-k-only", "also": [ { "source": "https://github.com/swyxio/skills/blob/038ef3441a4c7bfff16ae3241dc1730746a69a5b/README.md#L46-L47", "seen": "2026-10-01", "quote": "The author also maintains a legacy `~/.codex/skills/` installation;\nthe table follows current documented directories for new installs." }, { "source": "https://gist.github.com/swyxio/7fa1009e460ecb818d5e6d9ca4616a05/e454b0314bd9f2459f37f040b0f3b034d9b84f2e#file-zshrc-L92", "seen": "2026-06-19", "quote": "alias co=\"codex --dangerously-bypass-approvals-and-sandbox\"" } ] }, { "name": "Claude Code", "icon": "claudecode", "role": "with Chrome", "surface": "cli", "source": "https://gist.github.com/swyxio/7fa1009e460ecb818d5e6d9ca4616a05/e454b0314bd9f2459f37f040b0f3b034d9b84f2e#file-zshrc-L91", "seen": "2026-06-19", "quote": "alias cc=\"claude --dangerously-skip-permissions --chrome\"", "history": [ { "value": "Claude Cowork", "surface": "app", "source": "https://x.com/swyx/status/2017482739285684708", "seen": "2026-01-31", "quote": "CLAUDE COWORK IS A HARNESS ON TOP OF CLAUDE CODE IS A HARNESS ON TOP OF CLAUDE OPUS" } ] }, { "name": "Devin", "icon": null, "role": "for his non-technical team", "surface": "web", "surface_source": { "source": "https://www.youtube.com/watch?v=zepu8Kk6FBQ&t=289s", "seen": "2026-05-01", "quote": "hooked up Devin to Figma and we uh in very short order, we have a perfectly functioning website", "inferred": true }, "source": "https://www.youtube.com/watch?v=zepu8Kk6FBQ&t=239s", "seen": "2026-05-01", "quote": "Um I just added Devin to it." } ], "models": [ { "name": "Claude Opus 4.6", "icon": "claude", "detail": "1M context", "inferred": true, "source": "https://github.com/swyxio/skills/commit/badecfdc732c0bb8e73174c8a07ae1f9b3eb0a54", "seen": "2026-04-10", "quote": "Co-Authored-By: Claude Opus 4.6 (1M context) ", "history": [] } ], "review_style": { "value": "Devin Review bot comments on each agent PR, often after merge", "inferred": true, "kind": "no-review", "source": "https://github.com/swyxio/swyxdotio/pull/612", "seen": "2026-09-30", "quote": "**Devin Review** found 2 potential issues.", "also": [ { "source": "https://github.com/swyxio/swyxdotio/pull/612", "seen": "2026-09-30", "quote": "Validation: Browser checks confirm Command alone, Cmd+C, Cmd+Shift+K and slash stay closed; Cmd+K and Ctrl+K open the menu. Svelte check reports zero errors and warnings, and the production build passes." } ] }, "harness": { "value": "Ghostty + cmux, Cursor as editor", "short": "Ghostty + cmux", "source": "https://github.com/swyxio/skills/blob/038ef3441a4c7bfff16ae3241dc1730746a69a5b/new-mac-setup/SKILL.md#L26", "seen": "2026-06-24", "quote": "- **Terminal**: Ghostty (fast, GPU-accelerated, Zig) + cmux (vertical tabs, split panes, socket API for AI agents). cmux reads Ghostty's config, so one config file serves both.", "history": [] }, "parallelism": null, "rules_files": { "value": "AGENTS.md in his skills repo (one rule)", "short": "AGENTS.md", "url": "https://github.com/swyxio/skills/blob/038ef3441a4c7bfff16ae3241dc1730746a69a5b/AGENTS.md", "source": "https://github.com/swyxio/skills/blob/038ef3441a4c7bfff16ae3241dc1730746a69a5b/AGENTS.md#L1", "seen": "2026-03-28", "quote": "For long-running scripts or commands, always report elapsed times for major steps and batches when giving progress updates." }, "skills": { "value": [ "accelevents-api", "accelevents-speaker-sync", "agents-sdk", "ai-devblog", "ai-engineering", "ai-readme", "align-me", "antislop-codebase", "app-ux-paradigms", "autoreview", "babysit-runs", "blog-system-design", "cassandra-creator", "ceo-creator", "claude-session-introspect", "cli-ux", "cloudflare-do-turn-based-multiplayer", "cloudflare-email-service", "cloudflare-one-migrations", "cloudflare-one", "cloudflare-production-builder", "cloudflare", "conference-developer-endpoints", "conference-transcribe", "coordinate", "data-chatbots", "data-visualization-quality", "deep-trajectory-analysis", "design-apps-with-imagegen", "design-preferences", "diy-netlify", "download-video", "download-x-video", "durable-objects", "europe-developer-api", "face-matching", "forge", "future-only", "generalize", "gsuite-setup", "keep-it-simple", "latent-space-thumbnail-director", "live-ai-pipelines", "location-input", "long-running-operation-ux", "media-heavy-workflows", "media-to-wiki-pipeline", "media-transform", "mobile-native", "mobile-webapp-ux", "multimodal-extraction", "new-mac-setup", "next-steps", "observability-hardening", "person-profile-writing", "podcast-publishing-assistant", "productionize-app-with-services", "programmatic-agents", "provision-model-keys", "public-qa-chatbot", "pulp-fiction-writing", "release-readiness-hardening", "report-progress-eta-analyze", "research-grounded-writing", "reserved-handle-policy", "resilient-computer-use", "review-thread", "sandbox-sdk", "schedule-design", "security-hardening", "sessionize-automation", "skill-cutter", "slackbot-builder", "smart-entity-resolution", "summarize-anything", "swyx-writing", "sync-accelevents", "sync-url-navigation", "test-strategy-hardening", "testing-schedule-preview", "thumbnail-extraction", "transcribe-anything", "turnstile-spin", "twitter-x-scraping", "vercel-production-cost-review", "video-talk-to-essay", "visual-playtest", "web-animation-perf", "web-perf", "website-searchbar", "workers-best-practices", "wrangler", "youtube-api", "youtube-channel-operator", "youtube-publish", "youtube-studio-batch-upload", "youtube-studio-computer-use", "youtube-thumbnails", "zoom-download" ], "count": 99, "featured": [ "cassandra-creator", "ceo-creator", "review-thread" ], "source": "https://github.com/swyxio/skills/tree/038ef3441a4c7bfff16ae3241dc1730746a69a5b", "seen": "2026-10-05" }, "mcp_servers": null, "hooks": { "value": [ "peon-ping notifications shared across Claude Code, Codex and Cursor" ], "source": "https://github.com/swyxio/skills/blob/038ef3441a4c7bfff16ae3241dc1730746a69a5b/new-mac-setup/SKILL.md#L34", "seen": "2026-06-24", "quote": "- **Agent notifications**: peon-ping shared across Claude Code, Codex, and Cursor, tuned to notify only when useful" }, "signature": { "value": "$cassandra-creator", "kind": "skill", "caption": "a standing dissent agent", "why": "Our pick: next to a CEO agent that runs a project, he sets up an independent agent whose job is to challenge it.", "editorial": true, "source": "https://github.com/swyxio/skills/blob/038ef3441a4c7bfff16ae3241dc1730746a69a5b/cassandra-creator/SKILL.md#L8", "seen": "2026-08-10", "quote": "Create a durable independent dissent function, not a generic contrarian persona or one-off review." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Keeps 99 skills, a multi-CLI agent runner and his own session inspectors." }, "signature_quote": { "value": "benefit of agents is that they save you the yak shaves.", "more": "Like all the dependency tree crawling of like, \"Oh, no, I have to do that first. Oh, no, I have to do that first.\"", "source": "https://www.youtube.com/watch?v=zepu8Kk6FBQ&t=256s", "seen": "2026-05-01" } }, { "schema": 2, "handle": "teorth", "status": "ok", "checked": "2026-10-06", "name": { "value": "Terence Tao", "source": "https://github.com/teorth/tao-web" }, "x": null, "github": { "value": "teorth", "source": "https://github.com/teorth" }, "site": { "value": "https://terrytao.wordpress.com", "source": "https://terrytao.wordpress.com/author/teorth/" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/199308?v=4", "source": "https://github.com/teorth" }, "known_for": { "value": "Fields Medal mathematician, UCLA", "source": "https://en.wikipedia.org/wiki/Terence_Tao" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary", "surface": null, "source": "https://mathstodon.xyz/@tao/116517314420904580", "seen": "2026-05-04", "quote": "This time around, I uploaded the report to the Claude Code agent, which was able to inspect the report, the LaTeX source, and the PDF version of the paper, identify the corrections, and propose unambiguous fixes for eleven of the twelve corrections", "also": [ { "source": "https://www.youtube.com/watch?v=JHEO7cplfk8", "seen": "2026-03-07", "quote": "this time use a recent version of Claude Code to agentically do most of the formalization, while still leaving enough interactivity to stay manually engaged with the formalization task." }, { "source": "https://teorth.github.io/tao-web/apps/porting-legacy-applets.html", "seen": "2026-07-14", "quote": "A curated transcript of the conversation (with Claude Code) that ported Terence Tao's old Java-1.0 teaching applets" } ], "history": [ { "value": "GitHub Copilot", "surface": null, "source": "https://www.youtube.com/watch?v=c1ixXMtmfS8", "seen": "2025-05-17", "quote": "In this video, I test how Github Copilot performs in formalizing some basic \"epsilon delta\" type proofs in analysis." }, { "value": "ChatGPT Pro", "surface": "web", "source": "https://mathstodon.xyz/@tao/115493667607261044", "seen": "2025-11-04", "quote": "I therefore uploaded the paper to ChatGPT Pro and asked it first to summarize the construction in informal language, and then formalize step by step." } ] } ], "models": [ { "name": "Claude Opus 5", "icon": "claude", "inferred": true, "source": "https://github.com/teorth/tao-web/commit/90fdcfc719e65aba141c75937e7168fe610f4f7a", "seen": "2026-10-03", "quote": "Co-Authored-By: Claude Opus 5 ", "history": [ { "value": "Claude Opus 4.8", "source": "https://teorth.github.io/tao-web/apps/zeta-process-making-of.html", "seen": "2026-07-16", "quote": "Built with Claude Code (Opus 4.8)." } ] } ], "review_style": { "value": "Checks results, doesn't read machine code", "inferred": true, "source": "https://github.com/teorth/IEANTN/blob/0f4454a3bdc92e85cd82722724b68c80fd14cb55/CLAUDE.md#L89-L91", "seen": "2026-09-25", "quote": "Solutions are exempt: they are machine-generated, verified once, and never read.", "kind": "tests-only", "also": [ { "source": "https://terrytao.wordpress.com/2026/07/11/old-and-new-apps-via-modern-coding-agents/", "seen": "2026-07-11", "quote": "While I have playtested the app somewhat, I would be interested in receiving further feedback on this “alpha” version of the applet, as I am sure (especially given the LLM-generated nature of the code) that there are still some bugs and rough edges to be ironed out." }, { "source": "https://mathstodon.xyz/@tao/116517314420904580", "seen": "2026-05-04", "quote": "I then reviewed all proposed changes, made the selection for the twelfth fix, and asked Claude to implement them all; the entire process took less than fifteen minutes." } ] }, "harness": { "value": "Spec-first recipe per app, background sub-agents", "short": "sub-agents", "source": "https://teorth.github.io/tao-web/apps/porting-legacy-applets.html", "seen": "2026-07-14", "quote": "Unlike the original apps, most ports were driven by a spec-first, per-app recipe run partly through background sub-agents; the shared method matters more than any single exchange.", "history": [] }, "parallelism": null, "rules_files": { "value": "CLAUDE.md for the IEANTN project (94 lines); AGENTS.md points to it", "short": "CLAUDE.md", "url": "https://github.com/teorth/IEANTN/blob/0f4454a3bdc92e85cd82722724b68c80fd14cb55/CLAUDE.md", "source": "https://github.com/teorth/IEANTN/blob/0f4454a3bdc92e85cd82722724b68c80fd14cb55/CLAUDE.md#L41-L42", "seen": "2026-09-25", "quote": "**Never weaken a statement to make a proof work.** The source paper is ground truth. If the\n stated bound is wrong, say so — do not quietly adjust it." }, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "Five favorable conditions", "kind": "habit", "caption": "before handing code to an agent", "why": "Our pick: he delegates all the coding only when the work is non-critical, stand-alone, sandboxed, off his core skills and not competing with humans.", "editorial": true, "source": "https://terrytao.wordpress.com/2026/07/16/two-more-apps-visualizing-the-zeta-process-and-the-motions-of-the-heavens/", "seen": "2026-07-16", "quote": "I would however caution against unrestricted LLM use when one or more of the above five favorable situations is not in effect." }, "archetype": { "value": "Skeptic", "editorial": true, "why": "Hands code to Claude Code only when his five conditions limit the risk, and checks the results instead of trusting them." }, "signature_quote": { "value": "I am perfectly willing to forego the opportunity to keep my Javascript skills at a high level, as this is a tertiary skill for me at best in my chosen profession.", "more": "(I continue to manually program in Lean and in Python to keep in practice with programming in general.)", "source": "https://terrytao.wordpress.com/2026/07/16/two-more-apps-visualizing-the-zeta-process-and-the-motions-of-the-heavens/", "seen": "2026-07-16" } }, { "schema": 2, "handle": "thariqs", "status": "ok", "checked": "2026-10-07", "name": { "value": "Thariq Shihipar", "source": "https://github.com/ThariqS" }, "x": { "value": "trq212", "source": "https://x.com/trq212" }, "github": { "value": "ThariqS", "source": "https://github.com/ThariqS" }, "site": { "value": "https://thariq.io", "source": "https://github.com/ThariqS" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/140827?v=4", "source": "https://github.com/ThariqS" }, "known_for": { "value": "Claude Code team, Anthropic", "source": "https://github.com/ThariqS" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary: one focus session", "surface": "cli", "surface_source": { "source": "https://www.youtube.com/watch?v=aVO6E181cNU&t=1859s", "seen": "2026-07-19", "quote": "I'm a mix of terminal and desktop." }, "source": "https://x.com/trq212/status/2073100352921215386", "seen": "2026-07-03", "quote": "I ask Claude Code to keep a temporary ‘implementation-notes.md’ (or .html) file where it keeps track of decisions it makes so we can learn from our next attempt.", "history": [ { "value": "Claude Code as a general agent in ~", "surface": "cli", "source": "https://x.com/trq212/status/1944877531028725840", "seen": "2025-07-14", "quote": "I use Mac OSX and run Claude Code in my home directory (~)." } ] }, { "name": "Claude Tag", "icon": "claude", "role": "parallel work in Slack", "surface": null, "source": "https://x.com/trq212/status/2069474338450108705", "seen": "2026-06-23", "quote": "Have a personal channel e.g. I have \"#thariq-claude\" where you can tag Claude for your own work and give it instructions specific to you." } ], "models": [ { "name": "Claude Fable 5", "icon": "claude", "source": "https://x.com/trq212/status/2073100352921215386", "seen": "2026-07-03", "quote": "Working with Claude Fable 5 keeps re-teaching me an old lesson: the map is not the territory.", "history": [] }, { "name": "Claude Opus 5.5", "icon": "claude", "detail": "effort per task", "source": "https://x.com/trq212/status/2103576349499855160", "seen": "2026-09-25", "quote": "I tried doing the same tasks at several different effort levels on Opus 5.5 understand the work it would do.", "inferred": true, "history": [ { "value": "Opus 4.7, 1M context", "source": "https://x.com/trq212/status/2052809885763747935", "seen": "2026-05-08", "quote": "With the 1MM context window in Opus 4.7, the increased token usage is not really noticeable in the context window." } ] } ], "review_style": { "value": "Claude writes an HTML report and a quiz on the change; he merges only after passing it", "source": "https://x.com/trq212/status/2073100352921215386", "seen": "2026-07-03", "quote": "Asking Claude to quiz me about the change after giving me a bunch of context helps me understand what happens. I only merge after I pass the quiz perfectly.", "kind": "unknown", "also": [ { "source": "https://x.com/trq212/status/2103576349499855160", "seen": "2026-09-25", "quote": "For normal software engineering, I’m now running a loop of getting the model to interview me then implementing on low/medium effort, reviewing what it built, and then running verification on high effort." }, { "source": "https://x.com/trq212/status/2052809885763747935", "seen": "2026-05-08", "quote": "I find this often works better than the default Github diff view, and I attach a HTML code explainer to every PR I make now." } ], "inferred": true }, "harness": { "value": "One active Claude Code session (terminal or desktop); background work in Claude Tag threads in Slack", "short": "Claude Tag", "source": "https://www.youtube.com/watch?v=aVO6E181cNU&t=1200s", "seen": "2026-07-19", "quote": "now like it's mostly one active Claude Code session and then a bunch of Claude Tag sessions.", "history": [] }, "parallelism": { "value": 1, "short": "1 + a bunch in Tag", "source": "https://www.youtube.com/watch?v=aVO6E181cNU&t=1200s", "seen": "2026-07-19", "quote": "now like it's mostly one active Claude Code session and then a bunch of Claude Tag sessions.", "history": [ { "value": 5, "source": "https://www.youtube.com/watch?v=aVO6E181cNU&t=1200s", "seen": "2026-07-19", "quote": "before I was having you know like five different Claudes uh in my Claude Code" } ] }, "rules_files": { "value": "Claude.MD in ~ for a general-agent setup (2025)", "short": "Claude.MD", "url": "https://gist.github.com/ThariqS/2c35650d1cb81e87392e9f0fee5d5a10/9f1c174813ec3797b03c492fc0b2837c466cdf54", "source": "https://gist.github.com/ThariqS/2c35650d1cb81e87392e9f0fee5d5a10/9f1c174813ec3797b03c492fc0b2837c466cdf54", "seen": "2025-07-14", "quote": "You should search memories to see if there's an relevant information for my query, especially if you feel like you're missing context." }, "skills": { "value": [ "journal", "todo" ], "count": 2, "featured": [ "journal", "todo" ], "source": "https://x.com/trq212/status/1944877541254422660", "seen": "2025-07-14" }, "mcp_servers": null, "hooks": null, "signature": { "value": "Change quiz", "kind": "habit", "caption": "must pass before merging", "why": "Our pick: instead of trusting a diff, he has Claude explain the change in HTML and quiz him on it, and he merges only after a perfect score.", "editorial": true, "source": "https://x.com/trq212/status/2073100352921215386", "seen": "2026-07-03", "quote": "I only merge after I pass the quiz perfectly." }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "He keeps one focus session locally and sends parallel work to Claude Tag threads he tracks with a status pin." }, "signature_quote": { "value": "Fable is the first model where I find the quality of the work is bottlenecked by my ability to clarify its unknowns.", "more": "I’ve found that working with Fable is an iterative process of discovering my unknowns before, during, and after implementation.", "source": "https://x.com/trq212/status/2073100352921215386", "seen": "2026-07-03" } }, { "schema": 2, "handle": "tiangolo", "status": "ok", "checked": "2026-10-07", "name": { "value": "Sebastián Ramírez", "source": "https://github.com/tiangolo" }, "x": { "value": "tiangolo", "source": "https://x.com/tiangolo" }, "github": { "value": "tiangolo", "source": "https://github.com/tiangolo" }, "site": { "value": "https://tiangolo.com", "source": "https://github.com/tiangolo" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/1326112?v=4", "source": "https://github.com/tiangolo" }, "known_for": { "value": "FastAPI, Typer, SQLModel", "source": "https://github.com/tiangolo" }, "agents": [ { "name": "Codex", "icon": "codex", "role": "primary", "surface": null, "source": "https://github.com/fastapi/sqlmodel/pull/2112", "seen": "2026-10-06", "quote": "Made with the help of AI, using Codex with `gpt-6-astra`, manually reviewed.", "also": [ { "source": "https://bsky.app/profile/tiangolo.com/post/3mma4kbywws2h", "seen": "2026-05-19", "quote": "Codex with /goal is quite good for the things it claims to be good." } ], "history": [ { "value": "Claude (Opus 4.5 and 4.6)", "surface": null, "source": "https://github.com/Kludex/starlette/pull/3135", "seen": "2026-02-10", "quote": "Using Claude Opus 4.5 (and 4.6)." } ] } ], "models": [ { "name": "GPT-6 Astra", "icon": "openai", "source": "https://github.com/fastapi/sqlmodel/pull/2112", "seen": "2026-10-06", "quote": "Made with the help of AI, using Codex with `gpt-6-astra`, manually reviewed.", "history": [ { "value": "Claude Opus 4.6", "source": "https://github.com/fastapi/fastapi/pull/14906", "seen": "2026-02-12", "quote": "This was iteratively written with Claude Opus 4.6." }, { "value": "GPT 5.5", "source": "https://github.com/fastapi/fastapi/pull/15563", "seen": "2026-05-19", "quote": "Most of this was done with Codex on GPT 5.5 and a `/goal` with a somewhat detailed description." }, { "value": "gpt-5.6-sol", "source": "https://github.com/pydantic/pydantic-settings/pull/952", "seen": "2026-08-28", "quote": "Made with the help of AI, using Codex with `gpt-5.6-sol`, manually reviewed." } ] } ], "review_style": { "value": "Reviews and adjusts all AI code by hand", "kind": "reads-all", "source": "https://x.com/tiangolo/status/2103083765308248228", "seen": "2026-09-24", "quote": "I will keep reviewing and adjusting any AI generated code until the models can consistently one-shot produce something I would consider acceptable.", "also": [ { "source": "https://github.com/fastapi/sqlmodel/pull/2112", "seen": "2026-10-06", "quote": "Made with the help of AI, using Codex with `gpt-6-astra`, manually reviewed." }, { "source": "https://github.com/Kludex/starlette/pull/3135", "seen": "2026-02-10", "quote": "I used AI to help write the code and tests, then I manually reviewed carefully all the code and adjusted, reverted (and unreverted :sweat_smile: ) some changes." } ] }, "harness": { "value": "Codex with /goal and a detailed task description", "short": "/goal", "source": "https://github.com/fastapi/fastapi/pull/15563", "seen": "2026-05-19", "quote": "Most of this was done with Codex on GPT 5.5 and a `/goal` with a somewhat detailed description.", "history": [] }, "parallelism": null, "rules_files": null, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "AI Disclaimer", "kind": "habit", "caption": "names agent and model in PRs", "why": "Our pick: each agent-made PR states the agent, the exact model and that he reviewed it by hand.", "editorial": true, "source": "https://github.com/fastapi/sqlmodel/pull/2112", "seen": "2026-10-06", "quote": "Made with the help of AI, using Codex with `gpt-6-astra`, manually reviewed." }, "archetype": { "value": "Purist", "editorial": true, "why": "Codes with Codex, but reviews every change by hand and does not trust one-shots yet." }, "signature_quote": { "value": "I haven't been able to just \"trust the model\".", "more": "But anyway, trust is earned, applies to humans, companies, products, AI.", "source": "https://x.com/tiangolo/status/2103083765308248228", "seen": "2026-09-24" } }, { "schema": 2, "handle": "tibo-openai", "status": "ok", "checked": "2026-10-07", "name": { "value": "Thibault Sottiaux", "source": "https://github.com/tibo-openai" }, "x": { "value": "thsottiaux", "source": "https://x.com/thsottiaux" }, "github": { "value": "tibo-openai", "source": "https://github.com/tibo-openai" }, "site": null, "avatar": { "value": "https://avatars.githubusercontent.com/u/177053821?v=4", "source": "https://github.com/tibo-openai" }, "known_for": { "value": "Codex & ChatGPT at OpenAI", "source": "https://x.com/thsottiaux" }, "agents": [ { "name": "Codex", "icon": "codex", "role": "primary, Codex app on Mac", "surface": "app", "source": "https://x.com/thsottiaux/status/2068144722460475527", "seen": "2026-06-20", "quote": "These days I spend more time in the codex app than all other apps combined on my Mac.", "history": [ { "value": "Codex CLI", "surface": "cli", "source": "https://x.com/thsottiaux/status/2014679156215448040", "seen": "2026-01-23", "quote": "You can now /fork conversations in Codex CLI to try something else without affecting the main session. I often use it to try different alternatives or to chat about a session without impacting it." } ] } ], "models": [ { "name": "GPT-5.6 Sol", "icon": "openai", "source": "https://x.com/thsottiaux/status/2085597685948813610", "seen": "2026-08-07", "quote": "I talk to it for 5 minutes straight with things that just seem to require weeks of work, get up to get something in the fridge, pet the dog, come back and ... it's done.", "history": [ { "value": "GPT-5.2-Codex, medium default, xhigh in background", "source": "https://x.com/thsottiaux/status/2005564827360809232", "seen": "2025-12-29", "quote": "This is why I recommend using medium as your default and using xhigh for background multitasking." }, { "value": "GPT-5.5", "source": "https://x.com/thsottiaux/status/2050395190847066270", "seen": "2026-05-02", "quote": "I've had the best experience with GPT-5.5 so far, it's night and day compared to two weeks ago." } ] }, { "name": "GPT-6 Astra", "icon": "openai", "source": "https://x.com/thsottiaux/status/2097193293532848288", "seen": "2026-09-08", "inferred": true, "quote": "Looking at this 28 page deck right now where Codex is keeping track of all the individual launches. Each one of them I can't wait to see what people do with it.\n\nBrought to you by an incredible team (and Astra)", "history": [] } ], "review_style": { "value": "Reviews what the agent presents; uses a fast model to understand code, never delegates understanding", "kind": "unknown", "inferred": true, "source": "https://every.to/podcast/transcript-how-openai-s-codex-team-uses-their-coding-agent", "seen": "2026-02-18", "quote": "You cannot delegate understanding—you’re trying to get to understanding something, and speed there is a real advantage.", "also": [ { "source": "https://every.to/podcast/transcript-how-openai-s-codex-team-uses-their-coding-agent", "seen": "2026-02-18", "quote": "You’re already reviewing the code the first time because the agent presents it to you, and then you have to review code produced by your peers. There are two rounds of review." } ] }, "harness": { "value": "Codex app with scheduled automations driven by skills; one fast in-flow thread", "short": "automations", "source": "https://every.to/podcast/transcript-how-openai-s-codex-team-uses-their-coding-agent", "seen": "2026-02-18", "quote": "Now I can do less multitasking and stay more in the flow. That felt really good.\nIt also feels satisfying to kick off automations using skills.", "history": [ { "value": "Long-running parallel tasks", "source": "https://every.to/podcast/transcript-how-openai-s-codex-team-uses-their-coding-agent", "seen": "2026-02-18", "quote": "I’d been optimizing for long-running, multitasking workflows—this task will take 10 or 15 minutes, I’ll kick off four different things and come back." } ] }, "parallelism": { "value": 4, "short": "4 tasks", "source": "https://every.to/podcast/transcript-how-openai-s-codex-team-uses-their-coding-agent", "seen": "2026-02-18", "quote": "I’d been optimizing for long-running, multitasking workflows—this task will take 10 or 15 minutes, I’ll kick off four different things and come back.", "history": [] }, "rules_files": null, "skills": { "value": [ "ask-questions-if-underspecified", "daily marketing-research skill" ], "count": 2, "featured": [ "ask-questions-if-underspecified" ], "source": "https://x.com/thsottiaux/status/2006624682515247604", "seen": "2026-01-01", "quote": "also codified it a bit more in a `$ask-questions-if-underspecified` skill that I use regularly.", "also": [ { "source": "https://every.to/podcast/transcript-how-openai-s-codex-team-uses-their-coding-agent", "seen": "2026-02-18", "quote": "It runs daily with a specific skill I’ve tuned over time to do deep marketing research." } ] }, "mcp_servers": { "value": [ "linear" ], "source": "https://every.to/podcast/transcript-how-openai-s-codex-team-uses-their-coding-agent", "seen": "2026-02-18", "inferred": true, "quote": "I find it much more reliable for things like going through Twitter replies and summarizing important themes, or filing bugs in Linear and using automation so things get implemented daily." }, "hooks": null, "signature": { "value": "Random-file bug hunt", "kind": "tool", "caption": "automation, several times a day", "why": "Our pick: a scheduled Codex automation literally runs Python random to pick a file, then finds and fixes a latent bug in it.", "editorial": true, "source": "https://every.to/podcast/transcript-how-openai-s-codex-team-uses-their-coding-agent", "seen": "2026-02-18", "quote": "One automation I have runs multiple times a day. It picks a random file and finds and fixes a subtle bug. It actually does pick a random file—it runs Python random, finds a file, and starts from there." }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "Scheduled automations hunt bugs and research Codex chatter while he steers one fast thread." }, "signature_quote": { "value": "You cannot delegate understanding", "more": "you’re trying to get to understanding something, and speed there is a real advantage.", "source": "https://every.to/podcast/transcript-how-openai-s-codex-team-uses-their-coding-agent", "seen": "2026-02-18" } }, { "schema": 2, "handle": "tmcw", "status": "ok", "checked": "2026-10-06", "name": { "value": "Tom MacWright", "source": "https://github.com/tmcw" }, "x": null, "github": { "value": "tmcw", "source": "https://github.com/tmcw" }, "site": { "value": "https://macwright.com", "source": "https://github.com/tmcw" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/32314?v=4", "source": "https://github.com/tmcw" }, "known_for": { "value": "Val Town, Placemark, ex-Mapbox", "source": "https://macwright.com/about/" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary, from Neovim", "surface": "cli", "surface_source": { "source": "https://macwright.com/2026/01/13/mind-changing", "seen": "2026-01-13", "quote": "Partly the watershed moment was the Claude Code form factor: in-editor AI suggestions are annoying and disruptive, whereas the CLI editor-lite direction is a lot better." }, "source": "https://macwright.com/2026/03/18/effect-devlog", "seen": "2026-03-18", "quote": "Obligatory LLM discourse: my usual AI tool, Claude Code with Opus 4.6, is decent at using Effect but stumbles in the same places that the documentation is lacking", "also": [ { "source": "https://github.com/val-town/codemirror-json-schema/pull/2", "seen": "2026-09-29", "quote": "🤖 Generated with [Claude Code](https://claude.com/claude-code)" } ], "history": [ { "value": "Windsurf, a little", "surface": "ide", "source": "https://bsky.app/profile/macwright.com/post/3lqnrq2c7i22s", "seen": "2025-06-02", "quote": "i use AI \"a little bit\" (windsurf to write aws junk, chatgpt for silly questions) and it's not terrible, it's fairly good." } ] }, { "name": "Pi", "icon": "pi", "role": "security research", "surface": null, "source": "https://blog.val.town/security", "seen": "2026-07-31", "quote": "Swapped out Claude Code with Pi for these tasks. Added lazy-pi so it has subagents and todo lists and all kinds of gizmos." } ], "models": [ { "name": "Claude Sonnet 5", "icon": "claude", "detail": "in Claude Code", "inferred": true, "source": "https://github.com/val-town/codemirror-json-schema/commit/e3a1518c621d092502a0d86676f5c26ae0d75b1b", "seen": "2026-09-29", "quote": "Co-Authored-By: Claude Sonnet 5 ", "history": [ { "value": "Claude Opus 4.6", "source": "https://macwright.com/2026/03/18/effect-devlog", "seen": "2026-03-18", "quote": "my usual AI tool, Claude Code with Opus 4.6, is decent at using Effect" } ] }, { "name": "GLM 5.2", "icon": "zai", "detail": "security research, via z.ai", "source": "https://blog.val.town/security", "seen": "2026-07-31", "quote": "First, I use GLM 5.2 for security research.", "history": [] } ], "review_style": { "value": "Careful human review of LLM fixes", "source": "https://blog.val.town/security", "seen": "2026-07-31", "quote": "I've been using LLMs much more for remediation, with lots of careful human review involved.", "kind": "unknown", "also": [ { "source": "https://macwright.com/2026/07/16/effect-update", "seen": "2026-07-16", "quote": "they still leave some refactoring debt where they call Effect.runPromise repeatedly instead of combining calls with Effect.gen." }, { "source": "https://macwright.com/2025/11/07/hallucination-city", "seen": "2025-11-07", "quote": "they'll leave their calling cards, like sloppy types, redundant methods, copypasta code, and comments." } ], "inferred": true }, "harness": { "value": "Neovim (LazyVim) with coder/claudecode.nvim", "short": "Neovim", "source": "https://github.com/tmcw/nvim/blob/5c1ef283babf541b31ded623c77225fc614dc928/lua/plugins/claude.lua#L1-L2", "seen": "2026-07-13", "quote": "-- coder/claudecode.nvim: IDE-style Claude Code integration\n-- (diff review/approve, @-file mentions, send selection, etc.).", "history": [ { "value": "Neovim with greggh/claude-code.nvim", "source": "https://github.com/tmcw/nvim/blob/d7b6b02d2892ecfc19eb009e8e3377c66042b65f/lua/plugins/claude.lua#L2", "seen": "2025-10-30", "quote": "\"greggh/claude-code.nvim\"," } ] }, "parallelism": null, "rules_files": null, "skills": { "value": [ "Cloudflare security audit skill" ], "count": 1, "featured": [ "Cloudflare security audit skill" ], "source": "https://blog.val.town/security", "seen": "2026-07-31" }, "mcp_servers": null, "hooks": null, "signature": { "value": "Pi on GLM 5.2", "kind": "tool", "caption": "as the security bug hunter", "why": "Our pick: a cheap model that does not refuse security work, in a separate harness with Cloudflare's audit skill and a local OWASP cheatsheet checkout.", "editorial": true, "source": "https://blog.val.town/security", "seen": "2026-07-31", "quote": "Swapped out Claude Code with Pi for these tasks. Added lazy-pi so it has subagents and todo lists and all kinds of gizmos." }, "archetype": { "value": "Skeptic", "editorial": true, "why": "Uses Claude Code as his usual tool while writing that it isn't satisfying and has ethical problems." }, "signature_quote": { "value": "I feel deeply conflicted about it: it isn't satisfying to use, it has all these ethical problems, but it works really well.", "more": "Only the highest-level models work well for writing code", "source": "https://macwright.com/2026/01/13/mind-changing", "seen": "2026-01-13" } }, { "schema": 2, "handle": "tobi", "status": "ok", "checked": "2026-10-07", "name": { "value": "Tobi Lütke", "source": "https://github.com/tobi" }, "x": { "value": "tobi", "source": "https://x.com/tobi" }, "github": { "value": "tobi", "source": "https://github.com/tobi" }, "site": { "value": "https://tobi.lutke.com", "source": "https://github.com/tobi" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/347?v=4", "source": "https://github.com/tobi" }, "known_for": { "value": "Shopify CEO, qmd", "source": "https://github.com/tobi/qmd" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "most recent commits", "surface": null, "source": "https://github.com/tobi/track-atlas/commit/8c12ac414ea74ffe4a3ef74b294a2eafab7ee60d", "seen": "2026-10-06", "quote": "Co-authored-by: Claude Opus 5.5 (1M context) ", "inferred": true, "also": [ { "source": "https://github.com/tobi/liquid-il", "seen": null, "undated": true, "quote": "A Liquid template engine, 100% vibe coded by Claude Code against liquid-spec" }, { "source": "https://x.com/tobi/status/2018506396321419760", "seen": "2026-02-03", "quote": "I was missing cc’s tasks system and told it to spawn clause in tmux and interrogate it about it and make an implementation for itself." } ] }, { "name": "Pi", "icon": "pi", "role": "self-extending harness", "surface": "cli", "surface_source": { "source": "https://github.com/tobi/pi-herdr-workers/blob/e540d099cff8415b72bde553fdaa6b4116b07697/README.md#L23", "seen": "2026-09-01", "quote": "Turns this pane into an orchestrator and puts other Pi agents in neighboring Herdr panes." }, "source": "https://x.com/tobi/status/2056198717225464307", "seen": "2026-05-18", "quote": "I’ve had very good results running autoresearch with local qwen 3.6 26b model as long as I had a simple vibed pi “advisor” extension that allowed it to periodically ask GPT 5.5 for ideas.", "also": [ { "source": "https://x.com/tobi/status/2030771823151853938", "seen": "2026-03-08", "quote": "told my pi to read this github repo and make a version of that for the qmd query-expansion model with the goal of highest quality score and speed." }, { "source": "https://x.com/tobi/status/2018506396321419760", "seen": "2026-02-03", "quote": "Pi is the most interesting agent harness." } ] }, { "name": "Hermes Agent", "icon": null, "role": "personal agent, skills", "surface": null, "source": "https://x.com/tobi/status/2072368961795039694", "seen": "2026-07-01", "quote": "If you use Hermes, just go “/learn duckdb and ducklake and use it” and all your data asks will get better.", "also": [ { "source": "https://x.com/tobi/status/2041719844920283326", "seen": "2026-04-08", "quote": "Hermes agent ships with this nifty /manim_video skill so I asked it to explain how a QMD query works:" } ] } ], "models": [ { "name": "Claude Opus 5.5", "icon": "claude", "detail": "1M context", "source": "https://github.com/tobi/track-atlas/commit/8c12ac414ea74ffe4a3ef74b294a2eafab7ee60d", "seen": "2026-10-06", "quote": "Co-authored-by: Claude Opus 5.5 (1M context) ", "inferred": true, "history": [ { "value": "Claude Opus 4.5", "source": "https://github.com/tobi/try/commit/bde30a45e5cc633158a17f34a3334d18793db924", "seen": "2026-03-07", "quote": "Co-Authored-By: Claude Opus 4.5 ", "inferred": true }, { "value": "Claude Opus 4.8", "source": "https://github.com/tobi/liquid-il/commit/35dda8282a218996e4d840bd166a931e6b4a15ce", "seen": "2026-07-06", "quote": "Co-Authored-By: Claude Opus 4.8 ", "inferred": true } ] }, { "name": "GPT-6 Astra", "icon": "openai", "source": "https://x.com/tobi/status/2096431707079454874", "seen": "2026-09-06", "quote": "GPT6 Astra made a headless version for me.", "history": [ { "value": "GPT 5.5 (as advisor)", "source": "https://x.com/tobi/status/2056198717225464307", "seen": "2026-05-18", "quote": "a simple vibed pi “advisor” extension that allowed it to periodically ask GPT 5.5 for ideas." }, { "value": "GPT-5.6-sol", "source": "https://x.com/tobi/status/2078200377309090285", "seen": "2026-07-17", "quote": "I feel like GPT-5.6-sol is the first model that just doesn't need it anymore? It just keeps on going until the thing is done." } ] } ], "review_style": { "value": "Spec suite the agent must pass", "source": "https://x.com/tobi/status/1995316357060063684", "seen": "2025-12-01", "quote": "5. Instruct Claude to write the C code using those libraries and not stop until all specs pass", "kind": "unknown", "inferred": true, "also": [ { "source": "https://x.com/tobi/status/2101817922607853708", "seen": "2026-09-20", "quote": "Accept code on merit and ensure that a person takes accountability for it." } ] }, "harness": { "value": "herdr agent panes, Pi /team workers", "short": "herdr", "source": "https://x.com/tobi/status/2086488457132626052", "seen": "2026-08-09", "quote": "my favorite way to explain people herdr is to just open a new workspace, launch an agent and tell it to:", "history": [ { "value": "tmux", "source": "https://x.com/tobi/status/2018506396321419760", "seen": "2026-02-03", "quote": "I was missing cc’s tasks system and told it to spawn clause in tmux and interrogate it about it and make an implementation for itself." }, { "value": "cmux", "source": "https://x.com/tobi/status/2057108476560642402", "seen": "2026-05-20", "quote": "Cmux is great." } ] }, "parallelism": null, "rules_files": { "value": "AGENTS.md", "short": "AGENTS.md", "url": "https://github.com/tobi/dotfiles/blob/c5117778bebdb63d75a0469dd653e3a2566e825c/AGENTS.md", "source": "https://github.com/tobi/dotfiles/blob/c5117778bebdb63d75a0469dd653e3a2566e825c/AGENTS.md#L7", "seen": "2026-07-25", "quote": "- This repository is public. Treat every tracked file and every commit as permanently public." }, "skills": { "value": [ "layer-tools", "surface-tools" ], "count": 2, "featured": [ "layer-tools", "surface-tools" ], "source": "https://github.com/tobi/track-atlas/tree/8c12ac414ea74ffe4a3ef74b294a2eafab7ee60d/skills", "seen": "2026-10-06" }, "mcp_servers": { "value": [ "context7" ], "source": "https://github.com/tobi/ac-tracer/blob/880e9e4edccb1de39b162a2056fabbc380e5ed57/.claude/settings.local.json#L8", "seen": "2026-02-16", "quote": "\"mcp__plugin_context7_context7__resolve-library-id\"," }, "hooks": { "value": [ "gitleaks pre-commit secret scan" ], "source": "https://github.com/tobi/dotfiles/blob/5648d49fa4aea504423670a22def661163714653/.githooks/pre-commit#L16", "seen": "2026-08-08", "quote": "exec \"$gitleaks\" git --staged --redact --no-banner" }, "signature": { "value": "qmd", "kind": "tool", "caption": "local search, agents too", "why": "Our pick: his own all-local search engine is the base layer under the tools he builds, and he points his agents at it too.", "editorial": true, "source": "https://x.com/tobi/status/2013217570912919575", "seen": "2026-01-19", "quote": "I think QMD is one of my finest tools. I use it every day because it’s the foundation of all the other tools I build for myself." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "He builds his own agent tooling: qmd, Pi extensions, herdr workers, microVM sandboxes." }, "signature_quote": { "value": "I have my agent now set to randomize the main chat AI every day.", "more": "I could just start the day on GPT Sol or on Opus 5 or something.", "source": "https://fs.blog/knowledge-project-podcast/tobi-lutke-3/", "seen": "2026-09-10" } }, { "schema": 2, "handle": "tonsky", "status": "ok", "checked": "2026-10-12", "name": { "value": "Nikita Prokopov", "source": "https://github.com/tonsky" }, "x": { "value": "nikitonsky", "source": "https://x.com/nikitonsky" }, "github": { "value": "tonsky", "source": "https://github.com/tonsky" }, "site": { "value": "https://tonsky.me", "source": "https://github.com/tonsky" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/285292?v=4", "source": "https://github.com/tonsky" }, "known_for": { "value": "DataScript, Fira Code", "source": "https://github.com/tonsky" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary", "surface": null, "source": "https://x.com/nikitonsky/status/2098164952120832133", "seen": "2026-09-10", "quote": "Over the last 6 months, I was using Codex and Claude Code on open-source grants.", "also": [ { "source": "https://github.com/tonsky/datascript/blob/ca4dec879d02ac919c8a791d3f9258f0a6c872c4/CLAUDE.md#L3", "seen": "2026-08-13", "quote": "This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository." } ], "inferred": true, "history": [] }, { "name": "Codex", "icon": "codex", "role": "OSS grant, expired 2026-09-07", "surface": null, "source": "https://x.com/nikitonsky/status/2096861562585645119", "seen": "2026-09-07", "quote": "So my OSS @openai Codex license expired today. Lots of work done with it over the last 6 months, including long-standing issues in Fira Code, DataScript, Humble UI.", "history": [] } ], "models": [ { "name": "Claude Opus 5.5", "icon": "claude", "source": "https://github.com/tonsky/Clojure-Sublimed/commit/a86c4a4083086371d673ef72f8a5b06b182a2dee", "seen": "2026-10-11", "quote": "Co-Authored-By: Claude Opus 5.5 ", "inferred": true, "history": [ { "value": "Claude Fable 5.1", "source": "https://github.com/tonsky/clojure-plus/commit/662df2562f9b2deb0de1d29232afc89c480622d7", "seen": "2026-10-05", "quote": "Co-Authored-By: Claude Fable 5.1" } ] } ], "review_style": null, "harness": { "value": "Agent works against a live nREPL (port 5556) and reloads namespaces instead of restarting the app", "short": "live nREPL", "source": "https://github.com/tonsky/chillalert/blob/6bd46030dbbe911e60c2b0b6b04778b9dfba9db1/CLAUDE.md#L3", "seen": "2026-09-08", "quote": "- An nREPL server is running on port 5556 (started via `script/nrepl.sh`, port also in `.nrepl-port`). Use the `clojure-eval` skill to evaluate code, reload edited namespaces and check that they compile, instead of restarting the app", "history": [] }, "parallelism": null, "rules_files": { "value": "CLAUDE.md (five hand-written lines)", "short": "CLAUDE.md", "url": "https://github.com/tonsky/chillalert/blob/6bd46030dbbe911e60c2b0b6b04778b9dfba9db1/CLAUDE.md", "source": "https://github.com/tonsky/chillalert/blob/6bd46030dbbe911e60c2b0b6b04778b9dfba9db1/CLAUDE.md#L1", "seen": "2026-09-08", "quote": "- Use as little dependencies as possible" }, "skills": { "value": [ "clojure-eval" ], "count": 1, "featured": [ "clojure-eval" ], "source": "https://github.com/tonsky/chillalert/blob/6bd46030dbbe911e60c2b0b6b04778b9dfba9db1/CLAUDE.md#L3", "seen": "2026-09-08" }, "mcp_servers": null, "hooks": null, "signature": { "value": "Live nREPL for the agent", "kind": "rule", "caption": "instead of app restarts", "why": "Our pick: his CLAUDE.md points the agent at a running nREPL so it evaluates and reloads code the way a Clojure programmer does.", "editorial": true, "source": "https://github.com/tonsky/chillalert/blob/6bd46030dbbe911e60c2b0b6b04778b9dfba9db1/CLAUDE.md#L3", "seen": "2026-09-08", "quote": "Use the `clojure-eval` skill to evaluate code, reload edited namespaces and check that they compile, instead of restarting the app" }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "He has agents build small tools he needs, inside a REPL-first Clojure setup." }, "signature_quote": { "value": "At this point it’s easier to ask LLM to build exactly what you need than searching for an existing project that might or might not do what you want.", "source": "https://mastodon.online/@nikitonsky/117100554251443176", "seen": "2026-08-15" } }, { "schema": 2, "handle": "torvalds", "status": "ok", "checked": "2026-10-06", "name": { "value": "Linus Torvalds", "source": "https://github.com/torvalds" }, "x": null, "github": { "value": "torvalds", "source": "https://github.com/torvalds" }, "site": null, "avatar": { "value": "https://avatars.githubusercontent.com/u/1024025?v=4", "source": "https://github.com/torvalds" }, "known_for": { "value": "Creator of Linux and Git", "source": "https://lwn.net/Articles/1073761/" }, "agents": [ { "name": "Claude", "icon": "claude", "role": "primary, side projects (Opus)", "surface": null, "source": "https://github.com/torvalds/GuitarPedal/commit/10a96b1bf60a8d0f33bdae7af0a410bb78bd0fa7", "seen": "2026-09-30", "quote": "And honestly, the other part of that was definitely Claude. It's just much easier to experiment with different approaches", "also": [ { "source": "https://github.com/torvalds/GuitarPedal/commit/42983ff047a56e5d8aeae8cc98665febe3c20d28", "seen": "2026-08-10", "quote": "I'm back to doing some sanity checking of the guitar pedal using Claude Opus" } ], "history": [ { "value": "Google Antigravity (Python visualizer)", "surface": "ide", "source": "https://github.com/torvalds/AudioNoise/blob/71b256a7fcb0aa1250625f79838ab71b2b77b9ff/README.md#L37-L42", "seen": "2026-01-09", "quote": "It started out as my typical \"google and do the monkey-see-monkey-do\" kind of programming, but then I cut out the middle-man -- me -- and just used Google Antigravity to do the audio sample visualizer." }, { "value": "Google Antigravity (WebMIDI JavaScript)", "surface": "ide", "source": "https://github.com/torvalds/GuitarPedal/commit/9c945be5ea6e6293001ee9394214a68de68ae1eb", "seen": "2026-06-11", "quote": "As always, any JavaScript and style sheet stuff is all generated by antigravity. I lack both the competence and interest." }, { "value": "Switch from Antigravity to Opus", "surface": null, "source": "https://github.com/torvalds/GuitarPedal/commit/b31bcb13212bf781be560b17888f499823bda8a4", "seen": "2026-07-28", "quote": "I've done all of the WebMIDI development using antigravity, because I obviously don't do JS (and let's be honest - I'm not exactly a UI person even in other languages, so let's not blame just JS here...). And today I decided to try Opus just to see how that works." } ] } ], "models": [ { "name": "Claude Opus", "icon": "claude", "source": "https://github.com/torvalds/GuitarPedal/commit/42983ff047a56e5d8aeae8cc98665febe3c20d28", "seen": "2026-08-10", "quote": "I'm back to doing some sanity checking of the guitar pedal using Claude Opus", "history": [] } ], "review_style": { "value": "Opus writes verification scripts; the AI checks its own output", "source": "https://github.com/torvalds/GuitarPedal/commit/d1b95dc5e71e0a2d42912001c5fa4469bbc88766", "seen": "2026-08-04", "quote": "And, incidentally, it also ends up being really useful to have the AI validate its own output. Win-win.", "kind": "tests-only", "inferred": true, "also": [ { "source": "https://github.com/torvalds/uemacs/commit/51745a3585981213fcc6f5f56ded5b5745458f69", "seen": "2026-08-05", "quote": "This was all done with more of that \"Let's just write verification scripts to track that it all acts the same\" model of using an AI helper to change how I usually work." }, { "source": "https://github.com/torvalds/GuitarPedal/commit/dab964e5a1c3ffa5ab19cc516edcdb6232c6419b", "seen": "2026-09-19", "quote": "And by \"me\" I obviously mean that I still don't really do python, so it's really just Opus." }, { "source": "https://lwn.net/Articles/1073761/", "seen": "2026-05-25", "quote": "you need to understand not just your prompts, but you need to understand your end result, because that's the only way you can maintain it long term" } ] }, "harness": { "value": "Issue tracker on a separate git side branch, plus a hardware validation bench Opus drives", "short": "side-branch issues", "source": "https://github.com/torvalds/uemacs/commit/a66fbc3243a971f3e78196805ab013ce762483b8", "seen": "2026-08-04", "quote": "I copied over my issue tracking infrastructure that keeps track of things in a separate side branch, and the general rules, and hey presto - productive AI sessions with Opus.", "history": [] }, "parallelism": null, "rules_files": { "value": "CLAUDE.md, not in the public tree", "short": "CLAUDE.md", "url": null, "source": "https://github.com/torvalds/GuitarPedal/commit/e8cfc09a0495f365c708467d5598aea7515eddfc", "seen": "2026-09-19", "quote": "CLAUDE.md calls capture-rat.py the worked example for bench captures." }, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "Pedals in a ring", "kind": "tool", "caption": "a bench that tests itself", "why": "Our pick: he had Opus turn three pedals and a signal generator into an automated test bench, so agent changes get measured on real hardware instead of eyeballed on a scope.", "editorial": true, "source": "https://github.com/torvalds/GuitarPedal/commit/d1b95dc5e71e0a2d42912001c5fa4469bbc88766", "seen": "2026-08-04", "quote": "I started the day with \"I want to connect two pedals together and have them validate each other\", and ended the day with three pedals, a signal generator, and a USB MIDI dongle all connected together as a validation bench with automated checks." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Has Opus build verification scripts and a self-testing pedal bench before he trusts a change." }, "signature_quote": { "value": "I'm finding that I do things differently when developing with an AI than I do by hand", "more": "Having a test suite is no longer about writing the boring parts when Opus writes it for me, but instead about what those tests find.", "source": "https://github.com/torvalds/GuitarPedal/commit/d1b95dc5e71e0a2d42912001c5fa4469bbc88766", "seen": "2026-08-04" } }, { "schema": 2, "handle": "tqbf", "status": "ok", "checked": "2026-10-12", "name": { "value": "Thomas Ptacek", "source": "https://fly.io/blog/youre-all-nuts/" }, "x": { "value": "tqbf", "source": "https://fly.io/blog/youre-all-nuts/" }, "github": { "value": "tqbf", "source": "https://github.com/tqbf" }, "site": { "value": "https://sockpuppet.org", "source": "https://github.com/tqbf" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/13419?v=4", "source": "https://github.com/tqbf" }, "known_for": { "value": "Matasano, Latacora, Cryptopals", "source": "https://sockpuppet.org/me/" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary, for code", "surface": "cli", "surface_source": { "source": "https://sockpuppet.org/blog/2026/09/17/how-to-write-with-an-llm/", "seen": "2026-09-17", "quote": "Then, give the thing the list of editing prompts you came up with, and have it run each through the Codex, Claude, or Antigravity CLIs.", "inferred": true }, "source": "https://sockpuppet.org/blog/2026/08/20/stop-making-tuis/", "seen": "2026-08-20", "quote": "In fact, that’s my entire process at this point: I copy a template app directory, open Claude or Codex in it, and tell it what I want to build.", "also": [ { "source": "https://news.ycombinator.com/item?id=47907285", "seen": "2026-04-26", "quote": "I have a standard OpenAI/ChatGPT Pro account; GPT5 is my daily driver for math, and Claude for code." } ], "history": [ { "value": "Zed agent mode as primary interface", "surface": "ide", "source": "https://news.ycombinator.com/item?id=44163292", "seen": "2025-06-02", "quote": "I use Zed as my primary interface to \"actually doing project work\" LLM stuff, because it front-ends both OpenAI and Google/Gemini models, and because I really like the interface." } ] }, { "name": "Codex", "icon": "codex", "role": "alternates with Claude", "surface": null, "source": "https://sockpuppet.org/blog/2026/08/20/stop-making-tuis/", "seen": "2026-08-20", "quote": "You want to make sure you’ve got computer-use, or whatever Codex calls it, enabled.", "history": [ { "value": "Codex CLI for casual stuff, starting Codex Web", "surface": "cli", "source": "https://news.ycombinator.com/item?id=44163292", "seen": "2025-06-02", "quote": "I use Codex CLI for casual stuff, because of the ergonomics of just popping open another terminal tab." } ] } ], "models": [ { "name": "Claude Opus 5.5", "icon": "claude", "inferred": true, "source": "https://github.com/tqbf/subpanel/commit/4fbc956835992ff0d45b2717789b0fe78c810e9f", "seen": "2026-09-23", "quote": "Co-Authored-By: Claude Opus 5.5 ", "history": [ { "value": "Gemini 2.5", "source": "https://fly.io/blog/youre-all-nuts/", "seen": "2025-06-02", "quote": "For the last month or so, Gemini 2.5 has been my go-to †." }, { "value": "Claude Opus 4.7", "source": "https://github.com/tqbf/djroomba/commit/ff3294f0d44a8921c97baaef6fa692c8e7a48db3", "seen": "2026-05-15", "quote": "Co-Authored-By: Claude Opus 4.7 (1M context) ", "inferred": true }, { "value": "Claude Fable 5", "source": "https://github.com/tqbf/casette/commit/768fa64cf3700f6237d707baa23cb78fa635cbd7", "seen": "2026-06-13", "quote": "Co-Authored-By: Claude Fable 5 ", "inferred": true } ] }, { "name": "GPT-5.5", "icon": "openai", "detail": "math, not code", "source": "https://news.ycombinator.com/item?id=47912199", "seen": "2026-04-26", "quote": "As I said elsewhere: I've fed thousands of math problems through ChatGPT (starting with 4o and now with 5.5).", "history": [] } ], "review_style": { "value": "Agents push PRs, only he merges", "source": "https://github.com/tqbf/swiftui-app/blob/3ca1cc4b8ffdbd3b628cc847574c9951b036fe12/CLAUDE.md#L17", "seen": "2026-08-20", "quote": "* You may push PRs, but you MUST NOT ever merge them youself to main.", "kind": "unknown", "also": [ { "source": "https://sockpuppet.org/blog/2026/05/12/emacsification/", "seen": "2026-05-12", "quote": "If an agent wrote all the SwiftUI code in my project, what do you have to gain from closely reading it?" }, { "source": "https://fly.io/blog/youre-all-nuts/", "seen": "2025-06-02", "quote": "I have to read the code line-by-line anyways." }, { "source": "https://x.com/tqbf/status/2108598079142125962", "seen": "2026-10-09", "quote": "I read \\infty more lines of code now than I write; being stuck in my editor (or worse, the Github PR viewer) is silly." } ] }, "harness": { "value": "A spare MacBook running Claude, Makefile builds, never Xcode", "short": "spare MacBook", "source": "https://sockpuppet.org/blog/2026/05/12/emacsification/", "seen": "2026-05-12", "quote": "I recruited an old Macbook to run Claude on. I set up Xcode and git. I got Claude configured, and tracked down some Swift and macOS design skills.", "also": [ { "source": "https://sockpuppet.org/blog/2026/08/20/stop-making-tuis/", "seen": "2026-08-20", "quote": "My biggest quality-of-life win is never having to open Xcode. Thankfully, my friend Josh built a purely Makefile-driven build process after trying to compile MDV for himself. I’ve just had Claude copy it to every new project I do." } ], "history": [] }, "parallelism": null, "rules_files": { "value": "CLAUDE.md = AGENTS.md (17 lines), copied into every app", "short": "CLAUDE.md", "url": "https://github.com/tqbf/swiftui-app/blob/3ca1cc4b8ffdbd3b628cc847574c9951b036fe12/CLAUDE.md", "source": "https://github.com/tqbf/swiftui-app/blob/3ca1cc4b8ffdbd3b628cc847574c9951b036fe12/CLAUDE.md#L5-L6", "seen": "2026-08-20", "quote": "* Record progress in PROGRESS.md. I should be able to tell a future agent \"read PLAN.md and\nPROGRESS.md\" and trust it's up to speed with this codebase." }, "skills": { "value": [ "swiftui-pro", "typography-designer", "macos-design" ], "count": 3, "featured": [ "swiftui-pro", "macos-design", "typography-designer" ], "source": "https://github.com/tqbf/swiftui-app/blob/3ca1cc4b8ffdbd3b628cc847574c9951b036fe12/CLAUDE.md#L8-L15", "seen": "2026-08-20" }, "mcp_servers": null, "hooks": null, "signature": { "value": "swiftui-app template", "kind": "tool", "caption": "copied into every new app", "why": "Our pick: one folder carries his rules, skills and no-Xcode build, so every app starts with cp -R and a sentence.", "editorial": true, "source": "https://sockpuppet.org/blog/2026/08/20/stop-making-tuis/", "seen": "2026-08-20", "quote": "In fact, that’s my entire process at this point: I copy a template app directory, open Claude or Codex in it, and tell it what I want to build.", "also": [ { "source": "https://github.com/tqbf/subpanel/commit/1d8232bad680798cbb93389f109d728cf6189ee7", "seen": "2026-09-23", "quote": "Starter template baseline (cp -R swiftui-app)" } ] }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Builds personal macOS apps from one reusable template and tells people to write their own agents." }, "signature_quote": { "value": "AI today is most effective when it's not vibing, but rather copiloting a skilled operator.", "more": "I don't necessarily want my agent to build an entire system for me, even if it's ultimately the author of almost every line of code; I'm actively making decisions throughout.", "source": "https://news.ycombinator.com/item?id=49513668", "seen": "2026-08-31" } }, { "schema": 2, "handle": "unclebob", "status": "ok", "checked": "2026-10-07", "name": { "value": "Robert C. Martin", "source": "https://github.com/unclebob" }, "x": { "value": "unclebobmartin", "source": "https://x.com/unclebobmartin" }, "github": { "value": "unclebob", "source": "https://github.com/unclebob" }, "site": { "value": "http://cleancoder.com", "source": "https://github.com/unclebob" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/36901?v=4", "source": "https://github.com/unclebob" }, "known_for": { "value": "Clean Code author", "source": "https://x.com/unclebobmartin/status/2105647907705868659" }, "agents": [ { "name": "Grok", "icon": "grok", "role": "primary, writes the code", "surface": "cli", "surface_source": { "source": "https://github.com/unclebob/uml-viewer/commit/a2a3fd5ee91618477c6471e8896586daae8922ed", "seen": "2026-09-14", "quote": "Add Grok tmux session, whole-file source viewer, and +/− member marks." }, "source": "https://x.com/unclebobmartin/status/2107162434892959990", "seen": "2026-10-05", "quote": "However, I just had grok build me a script editor. It took it 30 minutes.", "also": [ { "source": "https://x.com/unclebobmartin/status/2098883971564359680", "seen": "2026-09-12", "quote": "I had @grok implement this from a single prompt. It took 40 minutes." } ], "history": [ { "value": "grok-cli", "surface": "cli", "source": "https://x.com/unclebobmartin/status/1996734975899504677", "seen": "2025-12-05", "quote": "grok-cli did a splendid job of modifying my app to convert synchronous http/get calls in the drawing loop into an asynchronous process." }, { "value": "Claude", "surface": "cli", "inferred": true, "source": "https://x.com/unclebobmartin/status/2009274358087250227", "seen": "2026-01-08", "quote": "I am impressed that it has a much better grasp of the code than grok." }, { "value": "Claude as architect, Codex as coder and reviewer", "surface": null, "source": "https://x.com/unclebobmartin/status/2045477573535142212", "seen": "2026-04-18", "quote": "I found that the best agent to use as the architect was Claude." }, { "value": "Codex", "surface": "cli", "source": "https://x.com/unclebobmartin/status/2054970327592042661", "seen": "2026-05-14", "quote": "I just cancelled my Claude account. I've been using codex, and haven't used Claude in several weeks." } ] }, { "name": "Codex", "icon": "codex", "role": "reviews Grok's code", "surface": "cli", "surface_source": { "source": "https://github.com/unclebob/swarm-forge/commit/e77d596e0405575d23e32a45d5c773f0b5776bdc", "seen": "2026-08-24", "quote": "Launch Codex and Copilot with --no-alt-screen", "inferred": true }, "source": "https://x.com/unclebobmartin/status/2107226963458810345", "seen": "2026-10-05", "quote": "I'm using codex to review the code that grok wrote. Good results.", "history": [] } ], "models": [ { "name": "Grok", "icon": "grok", "source": "https://x.com/unclebobmartin/status/2101018229976043885", "seen": "2026-09-18", "quote": "It is eerie just how good @grok is now. I've stopped using any other model for the time being.", "history": [ { "value": "Claude", "source": "https://x.com/unclebobmartin/status/2009274358087250227", "seen": "2026-01-08", "quote": "After one day of using Claude (Yes, I heard all of you out there in the peanut gallery) I am both impressed and horrified." } ] } ], "review_style": { "value": "Does not read agent code: Codex reviews Grok's output, he inspects a UML diagram colored by CRAP and mutation scores", "source": "https://x.com/unclebobmartin/status/2105647907705868659", "seen": "2026-10-01", "quote": "People seem shocked that the author of Clean Code doesn't read code anymore.", "kind": "agent-reviews", "also": [ { "source": "https://x.com/unclebobmartin/status/2107226963458810345", "seen": "2026-10-05", "quote": "I'm using codex to review the code that grok wrote. Good results." }, { "source": "https://x.com/unclebobmartin/status/2102079838639038629", "seen": "2026-09-21", "quote": "A simple demo of uml-viewer. This is how I review the projects I create with agents nowadays." }, { "source": "https://x.com/unclebobmartin/status/2080257779395154409", "seen": "2026-07-23", "quote": "My current strategy is to not read any of the code written by my agents. That’s the only way I can take advantage of their productivity. What I do instead is to surround the agents with extreme constraints." } ] }, "harness": { "value": "A few agents on isolated tasks, prompted by hand to run crap, dry and mutate; no harness", "short": "no harness", "source": "https://x.com/unclebobmartin/status/2105668991897448890", "seen": "2026-10-01", "quote": "I'm back to working the way I did in March. A few agents open doing isolated tasks. Manually prompting them to do crap, dry, and mutate. No harness.", "history": [ { "value": "swarm-forge: role agents (specifier, coder, cleaner, architect, hardener, QA) in tmux and git worktrees", "source": "https://x.com/unclebobmartin/status/2062557016086786435", "seen": "2026-06-04", "quote": "I just ran a swarm-forge (branch six-pack) with six agents. Specifier, coder, cleaner, architect, hardender, QA." }, { "value": "Drops the tight harness", "source": "https://x.com/unclebobmartin/status/2098432570887217520", "seen": "2026-09-11", "quote": "And while I was heads-down getting that to work, the agents got a LOT better. So much so that when I came up for air, the need for my harness was obviated." } ] }, "parallelism": null, "rules_files": { "value": ".grok/rules/uml-viewer.md (Grok rules); AGENTS.md in older repos", "short": ".grok/rules", "url": "https://github.com/unclebob/uml-viewer/blob/0f54dd782525a49d7ba6ab1b7e2d6649a0a59078/.grok/rules/uml-viewer.md", "source": "https://github.com/unclebob/uml-viewer/blob/0f54dd782525a49d7ba6ab1b7e2d6649a0a59078/.grok/rules/uml-viewer.md#L50", "seen": "2026-10-01", "quote": "Do not commit or push unless asked. Esc interrupts a turn in this terminal;" }, "skills": { "value": [ "crap4clj", "using-clj-mutate", "speclj-structure-check", "crap4go", "using-mutate4go" ], "count": 5, "featured": [ "crap4clj", "using-clj-mutate", "speclj-structure-check" ], "source": "https://github.com/unclebob/crap4clj/blob/e90be2e7aa02c2ec5d6544e9cecddc6b73ad3e4d/SKILL.md#L2", "seen": "2026-09-17" }, "mcp_servers": null, "hooks": null, "signature": { "value": "uml-viewer", "kind": "tool", "caption": "colored by CRAP + mutation", "why": "Our pick: instead of reading agent code, he drills through a live UML diagram of the system, colored by CRAP and mutation scores, and tells a companion agent what to change.", "editorial": true, "source": "https://x.com/unclebobmartin/status/2100938985543446954", "seen": "2026-09-18", "quote": "The tool displays the structure of the system in a UML-like diagram, color coded with crap and mutation scores." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Has agents build his own quality tools (crapper, mutator, dryer, uml-viewer) and supervises through them instead of reading code." }, "signature_quote": { "value": "People seem shocked that the author of Clean Code doesn't read code anymore.", "more": "But that should not be surprising.", "source": "https://x.com/unclebobmartin/status/2105647907705868659", "seen": "2026-10-01" } }, { "schema": 2, "handle": "walnutwaldo", "status": "ok", "checked": "2026-10-07", "name": { "value": "Walden Yan", "source": "https://github.com/walnutwaldo" }, "x": { "value": "walden_yan", "source": "https://x.com/walden_yan" }, "github": { "value": "walnutwaldo", "source": "https://github.com/walnutwaldo" }, "site": { "value": "https://waldenyan.com", "source": "https://github.com/walnutwaldo" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/22627487?v=4", "source": "https://github.com/walnutwaldo" }, "known_for": { "value": "Co-founder, Cognition (Devin)", "source": "https://x.com/walden_yan/status/2104260390230323451" }, "agents": [ { "name": "Devin", "icon": null, "role": "primary, runs a team of Devins", "surface": "web", "inferred": true, "surface_source": { "source": "https://github.com/walnutwaldo/Personal-Site/pull/16", "seen": "2026-08-05", "quote": "Link to Devin session: https://staging.itsdev.in/sessions/5f04acb5c60c45e0b7a9a528dfa0c5df\nRequested by: @walnutwaldo" }, "source": "https://x.com/walden_yan/status/2034689901892084009", "seen": "2026-03-19", "quote": "I’ve had a lot of fun pointing Devin to look at everything in a slack channel, split up the tasks automatically, and then start a team of Devin’s to tackle them.", "history": [ { "value": "Windsurf agent tabs", "surface": "ide", "source": "https://x.com/walden_yan/status/2003929796762652721", "seen": "2025-12-24", "quote": "Happy holidays 🎁 The agent tabs in Windsurf are now the most natural way to work with parallel agents.\n\nFar more flexible in my experience than a dedicated agent view or being constrained to just a sidebar." } ] } ], "models": [ { "name": "Fable", "icon": "claude", "detail": "via Devin Fusion", "source": "https://x.com/walden_yan/status/2076715678242124154", "seen": "2026-07-13", "quote": "But then I tried it, and on one of my tasks, Fusion w/ Fable was 4x cheaper and gave a better PR.", "inferred": true, "history": [ { "value": "Opus 4.5, plan first", "source": "https://x.com/walden_yan/status/2001438759590736091", "seen": "2025-12-17", "quote": "Opus 4.5 is the first model where asking it to plan first & get my approval has noticeably multiplied my productivity. No fancy modes needed" } ] }, { "name": "GPT-5.5", "icon": "openai", "source": "https://x.com/walden_yan/status/2047783217370152988", "seen": "2026-04-24", "quote": "I find myself really reaching for 5.5 nowadays. Seems like a model built for cloud.", "history": [] } ], "review_style": { "value": "Devin Review first, then asks Devin what to look at first", "kind": "agent-reviews", "source": "https://x.com/walden_yan/status/2034689901892084009", "seen": "2026-03-19", "quote": "Then when I go to review, I ask it to tell me what I should look at first", "also": [ { "source": "https://x.com/walden_yan/status/2064755974548902006", "seen": "2026-06-10", "quote": "Humans should only look at PRs after an AI has already reviewed it." }, { "source": "https://x.com/walden_yan/status/2016667442425319828", "seen": "2026-01-29", "quote": "Me: isn't this code hacky?\nDevin Review: no ur code is hacky" }, { "source": "https://x.com/walden_yan/status/2082874473976639994", "seen": "2026-07-30", "quote": "Breaking PRs into stacks has been genuinely only of the greatest life hacks I learned (thanks to @ido_pesok)" } ], "inferred": true }, "harness": { "value": "Slack channel feeding a team of cloud Devins; stacked PRs", "short": "Devin teams", "source": "https://x.com/walden_yan/status/2034689901892084009", "seen": "2026-03-19", "quote": "I’ve had a lot of fun pointing Devin to look at everything in a slack channel, split up the tasks automatically, and then start a team of Devin’s to tackle them.", "history": [ { "value": "Windsurf agent tabs for parallel agents", "source": "https://x.com/walden_yan/status/2003929796762652721", "seen": "2025-12-24", "quote": "The agent tabs in Windsurf are now the most natural way to work with parallel agents." } ] }, "parallelism": null, "rules_files": null, "skills": null, "mcp_servers": null, "hooks": null, "signature": { "value": "Daily customer-report Devin", "kind": "habit", "caption": "escalates top feature asks", "why": "Our pick: his favorite Devin job is not code at all, it reads every customer report each day and escalates the top requests to engineering.", "editorial": true, "source": "https://x.com/walden_yan/status/2090158861974163826", "seen": "2026-08-19", "quote": "My favorite use for Devin isn’t even writing code. Every day, my Devin reads every customer report and escalates all the top feature requests to our eng team. Eyes everywhere." }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "He points Devin at a Slack channel, lets it split the work and launches a team of Devins." }, "signature_quote": { "value": "I spend so much on tokens nowadays but have way more trust in my PRs.", "more": "Self-fixing autonomous agents is the future and its here", "source": "https://x.com/walden_yan/status/2026433717061984336", "seen": "2026-02-24" } }, { "schema": 2, "handle": "wesbos", "status": "ok", "checked": "2026-10-07", "name": { "value": "Wes Bos", "source": "https://github.com/wesbos" }, "x": { "value": "wesbos", "source": "https://x.com/wesbos" }, "github": { "value": "wesbos", "source": "https://github.com/wesbos" }, "site": { "value": "https://wesbos.com", "source": "https://github.com/wesbos" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/176013?v=4", "source": "https://github.com/wesbos" }, "known_for": { "value": "Syntax podcast co-host", "source": "https://syntax.fm/show/980/ai-coding-explained/transcript" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "most commits in his newest repos", "surface": "cli", "surface_source": { "source": "https://syntax.fm/show/1023/mosh-caddy-and-tailscale-a-remote-dev-deep-dive/transcript", "seen": "2026-07-22", "quote": "I've been using Tmox for about a month since we've had Ben on, and I absolutely love it because it makes, having multiple, like, terminals open and having your agent be able to do it." }, "source": "https://x.com/wesbos/status/2054600706644914681", "seen": "2026-05-13", "quote": "Burned $91.34 with Claude Code /goal in 3.5 hours", "also": [ { "source": "https://syntax.fm/show/1009/54-ai-generated-and-climbing-state-of-ai/transcript", "seen": "2026-06-01", "quote": "I'm probably fifty fifty split between Claude and Cursor." }, { "source": "https://github.com/wesbos/gym-equipment/commit/d7995bdc6c2e1ced4101fa95f95607f6a640eb8e", "seen": "2026-09-22", "quote": "Co-Authored-By: Claude Opus 5.5 (1M context) " } ], "inferred": true, "history": [] }, { "name": "Codex", "icon": "codex", "role": "a third of his time, desktop app", "surface": "app", "source": "https://syntax.fm/show/1036/cursor-and-openai-break-up/transcript", "seen": "2026-09-07", "quote": "I I split my time probably right now 33% each between codex desktop, I've been really enjoying.", "history": [ { "value": "Cursor", "surface": "ide", "source": "https://syntax.fm/show/980/ai-coding-explained/transcript", "seen": "2026-02-18", "quote": "I find myself migrating to when I'm, like, working on existing application, I find myself migrating to Cursor and and mostly agents." }, { "value": "Cursor", "surface": "ide", "source": "https://syntax.fm/show/1009/54-ai-generated-and-climbing-state-of-ai/transcript", "seen": "2026-06-01", "quote": "I'm probably fifty fifty split between Claude and Cursor." } ] }, { "name": "OpenCode", "icon": "opencode", "role": "testing new and cheap models", "surface": "cli", "source": "https://syntax.fm/show/1036/cursor-and-openai-break-up/transcript", "seen": "2026-09-07", "quote": "Open code CLI with I've been testing all of these new models that that have been coming out, especially a lot of the, like, free or cheap Chinese ones.", "history": [] } ], "models": [ { "name": "Claude Opus 5.5", "icon": "claude", "source": "https://x.com/wesbos/status/2102450119975277027", "seen": "2026-09-22", "quote": "I've been using Opus 5.5 over the last week and ran a TON of tokens through it.", "history": [], "inferred": true }, { "name": "Codex models", "icon": "openai", "source": "https://syntax.fm/show/1036/cursor-and-openai-break-up/transcript", "seen": "2026-09-07", "quote": "Yeah. I've been using codex models quite a bit.", "history": [] } ], "review_style": { "value": "Reviews after the fact: tests it works, checks for duplicated messes", "source": "https://syntax.fm/show/978/should-a-new-coder-use-ai/transcript", "seen": "2026-02-11", "quote": "And then the second last second longest part is is actually reviewing it after the fact. Right? Like, testing that it actually works, that it didn't make a mess somewhere, that it didn't just copy paste and duplicate a whole bunch of styles and and make a problem for for down the road.", "kind": "spot-checks", "inferred": true, "also": [] }, "harness": { "value": "tmux (trying Herder), Termius on mobile, remote box over Cloudflare Tunnel", "short": "tmux", "source": "https://syntax.fm/show/1023/mosh-caddy-and-tailscale-a-remote-dev-deep-dive/transcript", "seen": "2026-07-22", "quote": "I've been using Tmox for about a month since we've had Ben on, and I absolutely love it because it makes, having multiple, like, terminals open and having your agent be able to do it.", "also": [ { "source": "https://syntax.fm/show/1023/mosh-caddy-and-tailscale-a-remote-dev-deep-dive/transcript", "seen": "2026-07-22", "quote": "So for mobile, I do Termius." }, { "source": "https://syntax.fm/show/987/remote-coding-agents/transcript", "seen": "2026-03-16", "quote": "I use Cloudflare's, Cloudflare Tunnels with their, what is it called, their access product, which is similar to Tailscale, but probably way harder to set up." } ], "history": [ { "value": "Editor remote SSH into a second laptop", "source": "https://syntax.fm/show/987/remote-coding-agents/transcript", "seen": "2026-03-16", "quote": "I had that the second laptop, and then I use a SSH plug in to remote into it." } ] }, "parallelism": null, "rules_files": null, "skills": { "value": [ "tip", "blog-post", "wes-bos-writing-style" ], "count": 3, "featured": [ "tip", "wes-bos-writing-style" ], "source": "https://github.com/wesbos/wesbos/tree/c8a8bff46d4ccbddae81d42abb752f70af8e0863/.agent/skills", "seen": "2026-03-02", "also": [ { "source": "https://syntax.fm/show/1003/skills-skills-skills/transcript", "seen": "2026-05-11", "quote": "Alright. So I'm gonna start with a skill that I've I've built myself, and I call this my hot tip skill." } ] }, "mcp_servers": { "value": [ "cloudflare", "davinci-resolve" ], "source": "https://syntax.fm/show/1031/my-coding-skills-are-atrophying/transcript", "seen": "2026-08-19", "quote": "Like, the one I use a lot is the the CloudFlare MCP, and I have to, like, log in with all of them and whatnot.", "also": [ { "source": "https://syntax.fm/show/1037/webmcp-is-here-and-you-should-care/transcript", "seen": "2026-09-09", "quote": "But, like, we all of us on Syntax, we use DaVinci Resolve, and we all have the DaVinci Resolve MCP running because in some cases, it's way easier to just type in the box what you want it to do than to figure out what the crazy clicks and and whatever that you need to do." } ] }, "hooks": null, "signature": { "value": "vite-plugin-killer-instincts", "kind": "tool", "caption": "hands the LLM a kill command", "why": "Our pick: rather than a rule the agent can ignore, he ships a deterministic Vite plugin whose error tells the agent which process holds the port and how to kill it.", "editorial": true, "source": "https://github.com/wesbos/vite-plugin-killer-instincts/blob/26b8105bd6008565fc107ba799ef706d38711adc/README.md#L34", "seen": "2026-03-10", "quote": "When `autoKill` is `false` (the default), the plugin throws an error with the PID and a `kill` command you (the llm) can run manually.", "also": [ { "source": "https://syntax.fm/show/998/how-to-fix-vibe-coding/transcript", "seen": "2026-04-22", "quote": "And the beautiful thing about that is that it is so easy to get the AI to write a deterministic plug in." } ] }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Writes his own Vite plugins and skills so agents hit deterministic checks, and rotates agents to compare them." }, "signature_quote": { "value": "I am not at a spot where I'm running five of them at a time because simply, like, the actual generation of the code is not the the slow part right now.", "more": "So much of software development is not just writing code.", "source": "https://syntax.fm/show/978/should-a-new-coder-use-ai/transcript", "seen": "2026-02-11" } }, { "schema": 2, "handle": "wesm", "status": "ok", "checked": "2026-10-06", "name": { "value": "Wes McKinney", "source": "https://github.com/wesm" }, "x": { "value": "wesmckinn", "source": "https://github.com/wesm" }, "github": { "value": "wesm", "source": "https://github.com/wesm" }, "site": { "value": "https://wesmckinney.com", "source": "https://github.com/wesm" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/329591?v=4", "source": "https://github.com/wesm" }, "known_for": { "value": "pandas, Apache Arrow", "source": "https://github.com/wesm" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "implements via subagents", "surface": "cli", "surface_source": { "source": "https://wesmckinney.com/transcripts/2026-05-08-show-us-your-agent-skills.html", "seen": "2026-05-08", "quote": "Every Git worktree has a stack of terminals — typically my stack is Claude Code and/or Codex plus roborev running in every worktree." }, "source": "https://wesmckinney.com/blog/agentic-engineering-aug-2026/", "seen": "2026-08-12", "quote": "Superpowers turns the reviewed spec into an implementation plan, which we implement either with subagent-driven development (mostly with Claude) or inline execution (mostly with Codex, because Codex’s subagents are… not great)." }, { "name": "Codex", "icon": "codex", "role": "inline execution and reviews", "surface": "app", "surface_source": { "source": "https://wesmckinney.com/blog/agentic-engineering-aug-2026/", "seen": "2026-08-12", "quote": "Outside of Forge, where we mostly do maintenance and last-mile development work to land changes, we still use terminal applications (like Kitty and Ghostty) and desktop apps (like Codex/ChatGPT and T3Code) for larger, longer-running projects or things that need browser/computer user." }, "source": "https://wesmckinney.com/transcripts/2026-05-08-show-us-your-agent-skills.html", "seen": "2026-05-08", "quote": "I use both Codex and Claude to do development.", "history": [ { "value": "Codex as reviewer of Claude Code's work", "surface": null, "source": "https://wesmckinney.com/transcripts/2026-04-08-joe-reis-ai-agents-mythical-agent-month.html", "seen": "2026-04-08", "quote": "I use Codex to review Claude’s work. That’s my standard workflow." } ] } ], "models": [ { "name": "Claude Fable", "icon": "claude", "detail": null, "inferred": true, "source": "https://wesmckinney.com/blog/agentic-engineering-aug-2026/", "seen": "2026-08-12", "quote": "The work produced by the latest frontier models (5.6-Sol and Fable) is extremely sloppy and almost never suitable for production without substantial hardening.", "history": [ { "value": "Opus 4.6", "source": "https://wesmckinney.com/transcripts/2026-04-08-joe-reis-ai-agents-mythical-agent-month.html", "seen": "2026-04-08", "quote": "Even Opus 4.6 – if you’re just committing and shipping the code that’s coming out of Opus 4.6, that code is a bunch of hot garbage." } ] }, { "name": "GPT-5.6 Sol", "icon": "openai", "detail": null, "inferred": true, "source": "https://wesmckinney.com/blog/agentic-engineering-aug-2026/", "seen": "2026-08-12", "quote": "The work produced by the latest frontier models (5.6-Sol and Fable) is extremely sloppy and almost never suitable for production without substantial hardening.", "history": [ { "value": "GPT-5.5, extra-high, as reviewer", "source": "https://wesmckinney.com/transcripts/2026-05-08-show-us-your-agent-skills.html", "seen": "2026-05-08", "quote": "I’ve found that Codex is the strongest code reviewer out there, at least GPT-5.5, that model specifically." } ] } ], "review_style": { "value": "roborev reviews every commit; he checks structure", "source": "https://wesmckinney.com/transcripts/2026-05-08-show-us-your-agent-skills.html", "seen": "2026-05-08", "quote": "I almost don’t read code now. My approach with roborev is that it’s my code reader. The mantra is that roborev reads every line of code that is generated.", "kind": "agent-reviews", "also": [ { "source": "https://wesmckinney.com/transcripts/2026-05-08-show-us-your-agent-skills.html", "seen": "2026-05-08", "quote": "And it gets read multiple times — whenever I push up a pull request the branch gets re-reviewed, so by the time I’m merging a pull request into a repository the code has been read by agents four or five times, minimum." }, { "source": "https://wesmckinney.com/blog/agentic-engineering-aug-2026/", "seen": "2026-08-12", "quote": "Agents can do the typing and checking, but the human remains accountable for the result." }, { "source": "https://wesmckinney.com/blog/agentic-engineering-aug-2026/", "seen": "2026-08-12", "quote": "Design and taste are not delegated to an agent." } ] }, "harness": { "value": "Own stack: Ghosthub terminal over tmux/Herdr/Zellij, Kenn Forge worktree workspaces, Kata issues, roborev, AgentsView", "short": "own tools", "source": "https://wesmckinney.com/blog/agentic-engineering-aug-2026/", "seen": "2026-08-12", "quote": "The result is a stack where each tool owns one layer of the problem: Kenn Forge is the workspace where changes get reviewed and landed, Ghosthub is a multiplexer-native terminal for local and remote agent sessions, Kata is the system of record for intent, and AgentsView and roborev keep us accountable about what the agents are doing and whether their work is actually correct.", "history": [ { "value": "Three columns of Kitty terminals", "source": "https://wesmckinney.com/transcripts/2026-04-08-joe-reis-ai-agents-mythical-agent-month.html", "seen": "2026-04-08", "quote": "I generally have three columns of terminals on my monitor. Three Kitty windows and every Kitty window has a stack of terminals with horizontal splits." }, { "value": "Superset worktree manager", "source": "https://wesmckinney.com/transcripts/2026-04-08-joe-reis-ai-agents-mythical-agent-month.html", "seen": "2026-04-08", "quote": "About a week ago, I started using a new system called Superset – not Apache Superset, it’s superset.sh." } ] }, "parallelism": { "value": 5, "short": "4-5 projects", "inferred": true, "source": "https://wesmckinney.com/transcripts/2026-05-08-show-us-your-agent-skills.html", "seen": "2026-05-08", "quote": "I work on four or five projects during the day, I’ll run parallel spec interviews with superpowers.", "history": [ { "value": 3, "short": "2-3 projects", "source": "https://wesmckinney.com/transcripts/2026-04-08-joe-reis-ai-agents-mythical-agent-month.html", "seen": "2026-04-08", "quote": "I’ll work on two or three projects simultaneously, multiple sessions per project." } ] }, "rules_files": { "value": "Clanker Constitution, linked into ~/.codex/AGENTS.md and ~/.claude/rules", "short": "Constitution", "url": "https://github.com/kenn-io/constitution/blob/48df1551c31fbbb783560e11006b61a43ef897fe/CONSTITUTION.md", "source": "https://github.com/kenn-io/constitution/blob/48df1551c31fbbb783560e11006b61a43ef897fe/README.md#L12-L14", "seen": "2026-08-11", "quote": "Keep a persistent checkout of this repository on trusted `main`. Link\n`CONSTITUTION.md` to `~/.codex/AGENTS.md` and to\n`~/.claude/rules/clanker-constitution.md`." }, "skills": { "value": [ "roborev-design-review", "roborev-design-review-branch", "roborev-fix", "roborev-lookahead-review", "roborev-lookahead-review-branch", "roborev-refine", "roborev-respond", "roborev-review", "roborev-review-branch", "roborev-snooze", "roborev-superpowers" ], "count": 11, "featured": [ "roborev-fix", "roborev-review-branch", "roborev-superpowers" ], "source": "https://github.com/kenn-io/roborev/tree/4aa94982c61b31f24707d3740e9c7d62326bb42e/internal/skills/claude", "seen": "2026-10-06" }, "mcp_servers": null, "hooks": { "value": [ "roborev post-commit review hook" ], "source": "https://wesmckinney.com/transcripts/2026-04-08-joe-reis-ai-agents-mythical-agent-month.html", "seen": "2026-04-08", "quote": "This is how the RoboRev code review system works – every time the agents commit, it fires a post-commit hook that triggers a code review." }, "signature": { "value": "roborev", "kind": "tool", "caption": "reviews every agent commit", "why": "Our pick: he built a background reviewer that a second model runs on every commit his agents make, and will not code without it.", "editorial": true, "source": "https://wesmckinney.com/transcripts/2026-02-10-rill-data-podcast.html", "seen": "2026-02-10", "quote": "And so basically I do not code with agents without a continuous reviewer running in the background." }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Built his own reviewer, session analytics, issue tracker, forge and terminal to keep agents accountable." }, "signature_quote": { "value": "The clankers (what we call the coding agents, since “agent” gives them too much credit) are not in charge, we are.", "source": "https://wesmckinney.com/blog/agentic-engineering-aug-2026/", "seen": "2026-08-12" } }, { "schema": 2, "handle": "xe", "status": "ok", "checked": "2026-10-12", "name": { "value": "Xe Iaso", "source": "https://github.com/Xe" }, "x": null, "github": { "value": "Xe", "source": "https://github.com/Xe" }, "site": { "value": "https://xeiaso.net", "source": "https://github.com/Xe" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/529003?v=4", "source": "https://github.com/Xe" }, "known_for": { "value": "Anubis, xeiaso.net", "source": "https://anubis.techaro.lol/blog/2026/anubis-wasm/" }, "agents": [ { "name": "Claude Code", "icon": "claudecode", "role": "primary", "surface": "cli", "surface_source": { "source": "https://xeiaso.net/notes/2026/claude-code-wins-april-fools/", "seen": "2026-04-01", "quote": "When you run /buddy, it hatches a coding companion that hangs out in your Claude Code interface like a tamagochi.", "inferred": true }, "source": "https://github.com/Xe/soteria/commit/f8d523eda908cbcba11ec3abd584f0abe2b8e09d", "seen": "2026-09-28", "quote": "Assisted-by: Claude Opus 5.5 via Claude Code", "also": [ { "source": "https://www.tigrisdata.com/blog/agent-sandbox-go/", "seen": "2026-05-28", "quote": "I wanted to see how far I could get by feeding the POSIX spec to Claude Code." }, { "source": "https://bsky.app/profile/xeiaso.net/post/3mhd6mqdky22o", "seen": "2026-03-18", "quote": "TBH i am getting about $200 per month of utility out of a claude max 20x plan" } ] }, { "name": "Codex", "icon": "codex", "role": "GPT-6, Sept 2026 repos", "surface": null, "source": "https://github.com/Xe/kefka/commit/f2bd6efaa5daa7959102d121c078fd02030231dc", "seen": "2026-09-24", "quote": "Assisted-by: GPT-6 via Codex", "inferred": true }, { "name": "OpenCode", "icon": "opencode", "role": "GLM 5.2, homelab infra", "surface": null, "source": "https://github.com/Xe/x/commit/a3b2abc13d283aeac76c11a24dea4a7b9105d4a0", "seen": "2026-07-22", "quote": "Assisted-by: GLM 5.2 via opencode", "inferred": true } ], "models": [ { "name": "Claude Opus 5.5", "icon": "claude", "source": "https://github.com/Xe/soteria/commit/f8d523eda908cbcba11ec3abd584f0abe2b8e09d", "seen": "2026-09-28", "quote": "Assisted-by: Claude Opus 5.5 via Claude Code", "inferred": true, "history": [ { "value": "Claude Opus 4.8", "source": "https://github.com/Xe/kefka/commit/94304df23c2133972d891034490f0b9f7d702e0b", "seen": "2026-06-02", "quote": "Assisted-by: Claude Opus 4.8 via Claude Code" }, { "value": "Claude Opus 5", "source": "https://github.com/Xe/site/commit/374111799075960d5f670ce5fec925118a6b0b5c", "seen": "2026-08-11", "quote": "Assisted-by: Claude Opus 5 via Claude Code" } ] }, { "name": "GPT-6.1 Sol", "icon": "openai", "detail": "via Codex", "source": "https://github.com/Xe/x/commit/3c2a40778dc3a8b9467f33ba4ed71fd38c7c47d5", "seen": "2026-10-07", "quote": "Assisted-by: GPT-6.1 Sol via Codex", "history": [ { "value": "GPT-6", "source": "https://github.com/Xe/kefka/commit/f2bd6efaa5daa7959102d121c078fd02030231dc", "seen": "2026-09-24", "quote": "Assisted-by: GPT-6 via Codex" } ] } ], "review_style": { "value": "A second agent team reviews; /reviewbot on PRs", "kind": "agent-reviews", "source": "https://xeiaso.net/blog/2026/vibe-coding-sponsor-panel/", "seen": "2026-03-09", "quote": "One agent team split the spec into tasks and started building. A second reviewed output and flagged issues.", "also": [ { "source": "https://github.com/Xe/x/blob/ebe67ac2f236c24d7d714dfd3eb39a9d1af7bfaa/.claude/git-workflow.md#L9", "seen": "2026-03-21", "quote": "Comment `/reviewbot` to trigger automated code review" }, { "source": "https://xeiaso.net/blog/2026/markdownlang/", "seen": "2026-02-10", "quote": "I haven't looked at basically any of the code (intentionally, it's part of The Bit™️)" } ], "inferred": true }, "harness": { "value": "Claude Code agents sharing a single git checkout, on a Claude Max 20x plan", "short": "one checkout", "source": "https://www.tigrisdata.com/blog/agent-sandbox-go/", "seen": "2026-05-28", "quote": "The porting process was mostly autonomous with each of the agents fighting for supremacy in a single git checkout.", "history": [] }, "parallelism": { "value": 5, "short": "5 agents", "source": "https://www.tigrisdata.com/blog/agent-sandbox-go/", "seen": "2026-05-28", "quote": "With 5 agents going at once, I managed to cover the critical set of coreutils", "history": [] }, "rules_files": { "value": "Global CLAUDE.md shipped as a system package; per-repo AGENTS.md, with CLAUDE.md as '@AGENTS.md'", "short": "AGENTS.md", "url": "https://github.com/Xe/x/blob/de9b63d24e5216518495966b5816156cfdde1c56/AGENTS.md", "source": "https://github.com/Xe/x/blob/de9b63d24e5216518495966b5816156cfdde1c56/AGENTS.md#L7", "seen": "2026-08-26", "quote": "Always ask the user for intent before writing code." }, "skills": { "value": [ "excalidraw-diagrams", "go-table-driven-tests", "gorm-dao", "sound-like-a-person", "templ-components", "templ-htmx", "templ-http", "templ-syntax", "xe-design-system", "xe-go-style", "xe-writing-style" ], "count": 11, "featured": [ "templ-syntax", "xe-writing-style", "gorm-dao" ], "source": "https://github.com/Xe/agent-plugins/tree/12dc8172511e0b560764c0e2508b6725529e9ef4/plugins", "seen": "2026-07-06" }, "mcp_servers": { "value": [ "python-wasm-mcp" ], "source": "https://github.com/Xe/x/blob/85e341b7fc1e8d5d78e16f336966f198fc97a74c/.mcp.json#L6", "seen": "2026-01-10", "quote": "\"args\": [\"run\", \"./cmd/python-wasm-mcp\"]," }, "hooks": null, "signature": { "value": "Assisted-by: footer", "kind": "rule", "caption": "model + tool on every commit", "why": "Our pick: her global rules make every agent sign its commits with the model and tool, so her git log is an honest record of who wrote what.", "editorial": true, "source": "https://github.com/Xe/x/blob/de9b63d24e5216518495966b5816156cfdde1c56/pkgs/claude-code-cfg/CLAUDE.md#L7-L10", "seen": "2026-08-26", "quote": "AI agents must disclose what tool and model they are using in the \"Assisted-by\" commit footer:" }, "archetype": { "value": "Toolsmith", "editorial": true, "why": "Builds her own agent tooling (a skills marketplace with evals, a PR review bot, a sandbox) and makes every agent sign its commits." }, "signature_quote": { "value": "I use them every day and I'm telling you: the default output has no soul.", "more": "It's correct. It's competent. It's fine.", "source": "https://xeiaso.net/blog/2026/ai-abstraction/", "seen": "2026-03-11" } }, { "schema": 2, "handle": "zachlloyd", "status": "ok", "checked": "2026-10-06", "name": { "value": "Zach Lloyd", "source": "https://github.com/zachlloyd" }, "x": { "value": "zachlloydtweets", "source": "https://x.com/zachlloydtweets" }, "github": { "value": "zachlloyd", "source": "https://github.com/zachlloyd" }, "site": { "value": "https://www.warp.dev", "source": "https://www.lennysnewsletter.com/p/how-warp-ships-2000-prs-a-month-with" }, "avatar": { "value": "https://avatars.githubusercontent.com/u/8727141?v=4", "source": "https://github.com/zachlloyd" }, "known_for": { "value": "Co-founder and CEO of Warp", "source": "https://www.lennysnewsletter.com/p/how-warp-ships-2000-prs-a-month-with" }, "agents": [ { "name": "Warp Factories", "icon": null, "role": "primary, his company's cloud factory", "surface": "web", "source": "https://x.com/zachlloydtweets/status/2090900505300177181", "seen": "2026-08-21", "quote": "Here, we use the Warp Terminal to kick off a run in a Factory and see it update on our web app alongside hundreds of other agent runs from our team." }, { "name": "Warp agent", "icon": null, "role": "in the Warp terminal, for CEO tasks", "surface": "app", "source": "https://www.chatprd.ai/how-i-ai/inside-warps-software-factory", "seen": "2026-09-18", "quote": "So, instead of fumbling with the UI, he uses a coding agent in Warp that has access to the Figma MCP (Multi-modal Command Palette)." } ], "models": [ { "name": "GPT 5.6 Sol", "icon": "openai", "detail": "high, team default", "inferred": true, "source": "https://x.com/zachlloydtweets/status/2097746902418948468", "seen": "2026-09-09", "quote": "- Benchmarking models on our own data to find the most efficient cost/performance pick. GPT 5.6 Sol (high) today", "history": [ { "value": "Claude Opus 5", "source": "https://x.com/zachlloydtweets/status/2095905220698280388", "seen": "2026-09-04", "quote": "All 50%+ cheaper than our previous default (Opus 5)" } ] } ], "review_style": { "value": "Engineer reviews the agent PR and ships it; QA video attached", "source": "https://x.com/zachlloydtweets/status/2104565422439932412", "seen": "2026-09-28", "quote": "One big change is how we approach code review - for low risk changes, we now let engineers review agent PRs and ship themselves, rather than requiring a second dev’s review.", "kind": "unknown", "inferred": true, "also": [ { "source": "https://www.chatprd.ai/how-i-ai/inside-warps-software-factory", "seen": "2026-09-18", "quote": "This video is attached to the PR, giving the human reviewer immediate, visual evidence that the feature works as intended." } ] }, "harness": { "value": "Warp Factories defined in one config file: Slack request, triage, spec, implement, review, QA", "short": "factory-as-code", "source": "https://x.com/zachlloydtweets/status/2105364556482511042", "seen": "2026-09-30", "quote": "We run all of our company's agents out of this configuration file. It's Terraform for agents", "history": [] }, "parallelism": null, "rules_files": null, "skills": null, "mcp_servers": { "value": [ "figma", "granola" ], "source": "https://www.chatprd.ai/how-i-ai/inside-warps-software-factory", "seen": "2026-09-18", "quote": "Can you use the Granola MCP to look back over my last four weeks of sales meetings and try to build up a list of the top 10 frequently most asked questions in these meetings as they pertain to Warp software factories?" }, "hooks": null, "signature": { "value": "Scorer + self-improvement agents", "kind": "tool", "caption": "agents that fix agents", "why": "Our pick: scorer agents grade every agent run, and other agents turn low scores into PRs that fix the factory's skills.", "editorial": true, "source": "https://x.com/zachlloydtweets/status/2105659008577060915", "seen": "2026-10-01", "quote": "- Self-improvement agents turn low-scoring conversations into improvements to your agent set up (skills, AGENTS.md, etc)" }, "archetype": { "value": "Orchestrator", "editorial": true, "why": "Designs the agent system and its metrics rather than steering single agents." }, "signature_quote": { "value": "I'm spending most of my real engineering brainpower at this point architecting how our agent system runs, not guiding individual agents,", "source": "https://x.com/zachlloydtweets/status/2093742126630654213", "seen": "2026-08-29" } } ] }