How do engineers actually use agents?

94 real setups from their own repos and posts, each with one trick worth copying. We're building a code review tool, so we wanted to know.

Only 8 of 94 say they still read every line their agents write.
21 of 94 read the risky parts and skim the rest.
68 of 94 use two or more agents, often one checks the other.

≡Orchestrator

Peter Steinberger

OpenClaw, ex-PSPDFKit

OpenClawwebCodexappGPT-6.1 Solworktrees

8 lanes69 skills1d

Peter's trick

$autoreview, before every commit.

⚒︎Toolsmith

Simon Willison

Datasette, Django co-creator

CodexappClaude CodewebGPT-6 SolOpus 5.5phone remote

4 skillsAGENTS.md3w

Simon's trick

Showboat, agents demo their own work.

You?

Get your card, and one thing to add today.

How it works

Read-only. Nothing is shared unless you share it.

⁂Swarm Lord

Boris Cherny

Created Claude Code

Claude CodeappClaude TagFable 5.1worktrees

5 local + 5–10 web2 hooks1mo

Boris's trick

/go, test, /simplify, open PR.

◇Purist

Andrej Karpathy

nanochat, coined “vibe coding”

Claude CodecliCodexFable 5Ghostty + IDE

1 skillprogram.md3mo

Andrej's trick

autoresearch, overnight agent research.

◇Purist

Mitchell Hashimoto

Ghostty, HashiCorp co-founder

Claude appappCodex appappFableGPT-6 Astradesktop apps

2 agents2 skills1mo

Mitchell's trick

Harness engineering, after every agent mistake.

⚒︎Toolsmith

DHH

Ruby on Rails, Omarchy

CodexcliClaude CodecliOpenCodecliGPT-6.1 SolOpus 5.5

2 at once3 skills4d

DHH's trick

"Review this with claude", adversarial review from Codex.

⚒︎Toolsmith

Armin Ronacher

Flask, Earendil founder

PicliCodexSolFableGhostty + tmux

1 hook18 skills1w

Armin's trick

/answer, in place of plan mode.

⁂Swarm Lord

Theo Browne

T3 Stack, T3 Chat, T3 Code

Claude CodeappCodexappOpus 5.5Muse Wonderremote Linux

40+ threads1 hook1d

Theo's trick

T3 Code, his own GUI over every agent.

⚒︎Toolsmith

Salvatore Sanfilippo (antirez)

Redis creator, DwarfStar (ds4)

CodexcliClaude Codeclids4-agentcliGPT-6 Astra / GPT-6.1 SolFable

AGENT.mdspot checks1w

Salvatore's trick

QA_BEFORE_RELEASES.md, an agent QA pass per release.

§Rules Lawyer

ThePrimeagen

Harpoon, Vim on stream

Cursoride99ideGrok 4.6terminal + 99

19 skills6d

ThePrimeagen's trick

Field guide, agent writes its own lessons.

⁂Swarm Lord

Steve Yegge

Gas Town, Beads, Vibe Coding

Claude CodecliCodexOpus 5.5Fable 5.1Emacs + tmux

20+ agents1 MCP4d

Steve's trick

Seats, 18 long-lived Fable officers.

∿Vibe Maximalist

Pieter Levels

Photo AI, Nomads.com

Claude CodecliOpus 5.5Fable 5.1VPS + tmux

tests only3d

Pieter's trick

tm, a tmux session per site.

◇Purist

Mario Zechner

pi coding agent, libGDX

PicliGPT-5.5phone

no MCP8 skills1w

Mario's trick

Daily code limit, sized to what he can review.

≡Orchestrator

Thorsten Ball

Amp co-founder, interpreter book

AmpwebOpus 5.5GPT-6 Astraorbs

AGENT.mdspot checks2w

Thorsten's trick

“Irrefutable proof”, video proof, not line reads.

§Rules Lawyer

Jesse Vincent

Superpowers, Keyboardio, K-9 Mail

Claude CodecliEvenerMuseOpusHaiku

1 hook15 skills2w

Jesse's trick

/par, rival reviewers, 5 points.

∿Vibe Maximalist

Harper Reed

CEO 2389.ai, CTO Obama 2012

Claude CodecliCodexappFable 5.1GPT-6 Astratmux + mosh

a dozen2 hooks1w

Harper's trick

"fresh eyes" review, a subagent rereads the code.

◇Purist

Kent Beck

TDD, Extreme Programming, JUnit

Claude CodecliCodexcliGeminilive sessions

one at a time5 skills1mo

Kent's trick

TCR skill, red tests revert the code.

⚒︎Toolsmith

Kieran Klaassen

Compound Engineering, Cora

CursorClaude CodecliCodexOpus 5.5Sonnet 5.5

36 skillsAGENTS.md1w

Kieran's trick

/ce-compound, turns each fix into a lesson.

⚒︎Toolsmith

Matt Pocock

Total TypeScript, ex-Vercel

Claude CodecliOpus 5.5Opus 5Sandcastle

1 hook27 skills1w

Matt's trick

/grill-me, before every change.

⚒︎Toolsmith

Indragie Karunaratne

Context, uniprof, Sentry AI lead

Claude CodecliCodexOpus 5.5GPT-5GitHub Desktop

1 hook6 skills1d

Indragie's trick

Claudescope, transcript link in every PR.

≡Orchestrator

Addy Osmani

Claude Code at Anthropic, ex-Chrome

Claude CodeCodexFableworktrees

5 at once25 skills1w

Addy's trick

/loop PR triage, on Agent Skills, every day.

∿Vibe Maximalist

Geoffrey Huntley

Creator of the Ralph loop

OpenCodecliCursorwebLoomGPT-6.1 SolGLM / Kimi

7 skills2 MCP2w

Geoffrey's trick

The Ralph loop, an agent in a bash while loop.

≡Orchestrator

Quinn Slack

Co-founded Amp and Sourcegraph

AmpwebAmp Dial: mediumorbs

AGENTS.mdagent review3w

Quinn's trick

"I'm feeling Pucky" button, starts half their work.

◇Purist

Dax Raad

OpenCode co-founder, SST

OpenCodeAstraFablephone + server

2 MCPAGENTS.md1w

Dax's trick

Voice prompting (Hex), instead of typing.

≡Orchestrator

Nick Baumann

Codex at OpenAI, ex-Cline

CodexappGPT-6 AstraGPT-5.3-Codex-Sparkmonothread

3–6 threads1 skill3w

Nick's trick

"Chief of staff" thread, spawns every project thread.

≡Orchestrator

Adam Wathan

Tailwind CSS, Refactoring UI

AmpwebCodexappClaude CodewebSonnet 4.5cloud Orbs

AGENTS.md2w

Adam's trick

Every tedious OSS task, fed to cloud Claude Code.

⚒︎Toolsmith

Anthony Fu

Vitest, Slidev, UnoCSS, VueUse

Claude CodeCursorideSonnet 5.5Opus 5Storybook

26 skillsCLAUDE.md2w

Anthony's trick

antfubot, agents ship PRs as a bot.

≡Orchestrator

Andreas Kling

Ladybird browser, SerenityOS

Claude CodeCodexFrontier modelSSH fleet

4 computersagent review3w

Andreas's trick

Welfare checks, on workers across 4 machines.

◇Purist

Bryan Cantrill

DTrace, Oxide co-founder and CTO

Claude CodecliChatGPTOpus 4.8no closed loop

reads every line2mo

Bryan's trick

This Old Repo, reviving a 1990s game.

⚒︎Toolsmith

Birgitta Böckeler

Thoughtworks, harness engineering

CursorideClaude Code PicliSonnetOpus 4.8

1 hook6 skills2mo

Birgitta's trick

sensors check, a sidecar the agent polls.

≡Orchestrator

Burke Holland

Works on GitHub Copilot

GitHub CopilotappClaude CodewebLuna MaxOpus 5.5project leads

2 skills1 MCP2d

Burke's trick

A lead agent per project, he just files issues.

◇Purist

Andrew Gallant

ripgrep, Rust regex, Jiff

CodexcliGPT-5.6 Solsandbox

1 hook2mo

Andrew's trick

"I wrote this PR with an LLM", flagged for careful review.

⚒︎Toolsmith

Calvin French-Owen

Segment co-founder, ex-Codex

CodexcliClaude CodecliFable 5GPT-5.6 Solworktrees

3–4 overnight tasks8 skills2mo

Calvin's trick

/pr-pass, loops until CI and bots pass.

◇Purist

Cat Wu

Claude Code product, Anthropic

Claude CodeappClaude TagOpus 5.5Fable 5.1auto mode

1 hook1 skill3w

Cat's trick

Fresh config per model, reset at every model launch.

⁂Swarm Lord

Chip Huyen

AI Engineering author

CodexClaude CodeFableown runner

10 subagents per lead4 skills2mo

Chip's trick

Cross-provider runner, Codex, Claude spawn each other.

◇Purist

Chris McCord

Creator of Phoenix, Phoenix.new

Claude CodeClaudeone convo

CLAUDE.md1mo

Chris's trick

`notify msg`, in CLAUDE.md, not a hook.

⚒︎Toolsmith

David Crawshaw

exe.dev, Tailscale co-founder

ShelleywebOpenAI modelsfresh VMs

spot checks2mo

David's trick

meat, strips diffs to what matters.

≡Orchestrator

Claire Vo

ChatPRD founder, How I AI host

CodexappClaude CodeCursorwebGPT-6 AstraOpus 5.5

1 MCPagent review3w

Claire's trick

LGTM, the PR closer, a bot that clears her PR queue.

≡Orchestrator

Dan Luu

danluu.com essays, ex-Centaur

CodexcliGPT-5.6 Solone-off loops

10 agentsAGENTS.md1mo

Dan's trick

Named-persona review panels, with contrarian personas.

⚒︎Toolsmith

Drew Breunig

"How to Fix Your Context", Plumb

Claude CodeFableGLM 5.2

2 hooks4 skills4w

Drew's trick

Plumb, reviews decisions at commit.

⚒︎Toolsmith

Dex Horthy

HumanLayer CEO, 12-Factor Agents

HumanLayerideCodexClaude CodeGPT-5.6 SolFable

7 skillsCLAUDE.md5d

Dex's trick

/show-me, diagrams, not walls of prose.

∿Vibe Maximalist

Dan Shipper

CEO of Every

CodexappClaude CodeGPT-6 SolFableCodex tasks

AGENTS.mdno review3w

Dan's trick

"Want me to compound?", asked after every sweep.

≡Orchestrator

Alexander Embiricos

Codex product lead, OpenAI

CodexappCodex default modellong-lived thread

4 attempts1 skill6mo

Alexander's trick

"Using a subagent in parallel, do X", from one always-on thread.

◇Purist

Edward Z. Yang

PyTorch at Meta, ghstack

Claude CodecliCodexOpus 5.5Fable 5tmux -CC

3 work streams2 hooks1w

Edward's trick

sl follow / sl adopt, restack parallel agents.

§Rules Lawyer

Fabien Sanglard

Game Engine Black Books

Claude CodeideAntigravityidesession per feature

agent.mdreads every line2mo

Fabien's trick

"Reload agent.md", when code quality drops.

⚒︎Toolsmith

Filippo Valsorda

Go cryptography, age, mkcert

Shelley (exe.dev)webClaude CodecliGPT-6 AstraOpus 4.8exe.dev VMs

AGENTS.md2w

Filippo's trick

filippo-agent, agent's own GitHub account.

≡Orchestrator

Garry Tan

YC CEO, creator of gstack

Claude CodeappCodexcliFable 5.1Conductor

10–15 sprints2 hooks1w

Garry's trick

/plan-ceo-review, finds the 10-star product.

?Skeptic

Jeff Geerling

Ansible, Raspberry Pi YouTube

ClaudewebOpus 5.5Fable 5.1chat + files

3w

Jeff's trick

AI Disclosure, README names the exact model.

◇Purist

Geoffrey Litt

Malleable software, Notion

Claude CodeCursorideCodex CLIcliOpus 5Fable 5

2 skillsCLAUDE.md2mo

Geoffrey's trick

/explain-diff, with a quiz before he ships.

?Skeptic

George Hotz

tinygrad, comma.ai

OpenCodeKimiOpuslocal models

AGENTS.mdspot checks2d

George's trick

(kimi) in the PR title, names the model per PR.

⁂Swarm Lord

Jarred Sumner

Creator of Bun, now at Anthropic

Claude CodecliRoboBunFable 5workflows + /loop

64 Claudes2 hooks3w

Jarred's trick

Adversarial review, 2 Claudes refute every change.

⚒︎Toolsmith

Jessica Kerr

Symmathesy, Honeycomb

Claude CodeOpus 5.5Sonnet 5.5subagents

8 skills1 MCP2d

Jessica's trick

Capability Owners, standing agents per feature.

§Rules Lawyer

José Valim

Creator of Elixir, Tidewave

CodexTidewaveappSol 5.6review pane

AGENTS.mdspot checks1mo

José's trick

Adversarial review agents, on compiler & security work.

⚒︎Toolsmith

Jeremy Howard

fast.ai, Answer.AI, ULMFiT

Claude CodecliSolveitwebCodexOpus 4.5persistent kernel

8 hooks10 skills1w

Jeremy's trick

claude-air, nudges after 8 silent rounds.

≡Orchestrator

Jason Liu

Instructor creator, OpenAI DX

CodexappGPT-5.5GPT-5.3-Codex-Sparkpinned threads

17 skills3 MCP3mo

Jason's trick

$ultragoal, goal kept in an editable file.

≡Orchestrator

Kent C. Dodds

Testing Library, Epic Web

Grok BotCursorwebGrok 4.5Fablecloud agents

a dozen2 hooks2d

Kent's trick

/ship-pr, Bugbot loop, merge, Discord.

◇Purist

Kenton Varda

Cloudflare Workers, Cap'n Proto

Claude CodecliCloudflare OSwebFable 5.1Solplan first

1 skill1 MCP4w

Kenton's trick

plans/ in the repo, checked in for code review.

§Rules Lawyer

Steph Ango

CEO of Obsidian

Claude CodecliObsidian CLI

5 skillsAGENTS.md3w

Steph's trick

Fail-first fixtures, before every extraction fix.

≡Orchestrator

Lee Robinson

ML at SpaceX, ex-Cursor, Vercel

Grok BotappCursorwebOpus 4.5GPT-5.1-Codex-Maxalways-on bots

2 skillsAGENTS.md2mo

Lee's trick

Plan mode, before every long agent run.

≡Orchestrator

Will Larson

Staff Engineer, CTO at Imprint

Claude CodeAgent FleetCodexGPT 5.4~10 workspaces

five or so projects1 skill3w

Will's trick

/linear-project-loop, loops a project to its goal.

◇Purist

Maggie Appleton

Digital garden, GitHub Next

Claude CodewebCodexOpenCodeOpus 5Sonnet 4.6

23 skills1 MCP2w

Maggie's trick

maggie-design, her design taste, as a skill.

◇Purist

Alex Kladov

rust-analyzer, TigerBeetle

Claude CodecliClaude 3.7 Sonnetspec + skeleton

plan.mdspot checks5mo

Alex's trick

Hand-typed skeletons, Claude fills in the bodies.

∿Vibe Maximalist

McKay Wrigley

Founder, Takeoff AI

CodexappClaude CodecliOpus 5.5GPT-6 Astraagent swarms

9 AI tabs1 hook3w

McKay's trick

"You are AGI-pilled.", in every agent's system prompt.

⚒︎Toolsmith

Mike McQuaid

Homebrew maintainer since 2009

CodexcliClaude CodecliOpenCodecliFable 5OpenAI Codex 5.4

2 hooksAGENTS.md3w

Mike's trick

AgentIDE, his own vibe-built agent IDE.

≡Orchestrator

Mike Krieger

Instagram co-founder, Anthropic Labs

Claude DesktopappOpus 5.5Sonnet 5.5coordinator Claude

agent review2w

Mike's trick

iOS Claude pings Android Claude, to port each feature.

§Rules Lawyer

Max Woolf

Ex-BuzzFeed data scientist

Zed AgentideCodexcliClaude CodecliGPT-6 AstraGPT-5.6 Luna

7–12 subagentsAGENTS.md2w

Max's trick

codex exec Luna subagents, 7–12 cheap critics per pass.

⚒︎Toolsmith

Nathan Lambert

RLHF Book, Interconnects

Claude CodecliCodexFable 5GPT 5 ProDGX Spark

1 hook6 skills1mo

Nathan's trick

/gemini-feedback, Gemini critiques each diagram.

◇Purist

Ned Batchelder

coverage.py

Claude CodecliGitHub CopilotideLSP tool

CLAUDE.md6mo

Ned's trick

Be introspective, Claude fixes its own rules.

◇Purist

Nolan Lawson

PouchDB, emoji-picker-element

Claude CodeCodexcliCursor BugbotwebOpus 4.7GPT 5.5

2 skillsagent review2mo

Nolan's trick

code-review-turbo, three-agent PR bug hunt.

◇Purist

Orta Therox

Puzzmo, CocoaPods, Danger

Claude CodeideClaude Code on the webwebFableclones

3 clones1 skill2w

Orta's trick

/commit, human writes the headline.

≡Orchestrator

Paul Irish

Chrome DevTools, Lighthouse

CodexwebGitHub Copilot coding agentwebJuleswebGemini 3 Flash (preview)cloud PRs

12 skillsAGENTS.md1w

Paul's trick

paulirish-agent, a GitHub account for agents.

◇Purist

Sebastian Raschka

Build an LLM From Scratch author

CodexClaude CodeQwen CodecliGPT-5.5Qwen3.6-35B-A3B

4mo

Sebastian's trick

Audit the agent first, before installing any harness.

⚒︎Toolsmith

Guillermo Rauch

CEO of Vercel

Claude CodeOpus 4.6

2 skillsAGENTS.md1w

Guillermo's trick

gdp-ts, auth checks as type proofs.

⚒︎Toolsmith

Ryan Dahl

Creator of Node.js and Deno

Claude CodeCodexOpus 4.6standalone VMs

CLAUDE.md5mo

Ryan's trick

Claw Patrol, firewall between agents and prod.

≡Orchestrator

Sean Goedecke

Writing on AI and big tech

GitHub Copilot appappCopilot CLIcliGPT-6-AstraOpus 5.5Copilot app

6–7 sessions per task1 skill3w

Sean's trick

30-second triage, rejects most agent attempts.

§Rules Lawyer

Sahil Lavingia

Gumroad founder

Claude CodeCodexFable 5GPT-6issue → PR

1 skillCONTRIBUTING.md1w

Sahil's trick

Fourth-wall check, enforced in CI on every PR.

≡Orchestrator

Ryan Carson

Founder of Untangle

DevinwebCodexOpus 5.5SWE 1.7Devin + Linear

1 hook1 MCP2w

Ryan's trick

!land playbook, refreshes skills on every PR.

≡Orchestrator

Steve Klabnik

Co-author of the Rust book, jj

Claude CodecliCodexappFableGPT Soljj workspaces

90 agents2 hooks1mo

Steve's trick

rue-overnight-hunt, bugs filed only with repros.

⚒︎Toolsmith

swyx

Latent Space, AI Engineer

CodexClaude CodecliDevinwebOpus 4.6Ghostty + cmux

1 hook99 skills1w

swyx's trick

$cassandra-creator, a standing dissent agent.

?Skeptic

Terence Tao

Fields Medal mathematician, UCLA

Claude CodeOpus 5sub-agents

CLAUDE.mdtests only1w

Terence's trick

Five favorable conditions, before handing code to an agent.

≡Orchestrator

Thariq Shihipar

Claude Code team, Anthropic

Claude CodecliClaude TagFable 5Opus 5.5Claude Tag

1 + a bunch in Tag2 skills3w

Thariq's trick

Change quiz, must pass before merging.

◇Purist

Sebastián Ramírez

FastAPI, Typer, SQLModel

CodexGPT-6 Astra/goal

reads every line1w

Sebastián's trick

AI Disclaimer, names agent and model in PRs.

≡Orchestrator

Thibault Sottiaux

Codex & ChatGPT at OpenAI

CodexappGPT-5.6 SolGPT-6 Astraautomations

4 tasks2 skills1mo

Thibault's trick

Random-file bug hunt, automation, several times a day.

?Skeptic

Tom MacWright

Val Town, Placemark, ex-Mapbox

Claude Codecli PiSonnet 5GLM 5.2Neovim

1 skill2w

Tom's trick

Pi on GLM 5.2, as the security bug hunter.

⚒︎Toolsmith

Tobi Lütke

Shopify CEO, qmd

Claude Code PicliHermes AgentOpus 5.5GPT-6 Astra

1 hook2 skills1w

Tobi's trick

qmd, local search, agents too.

⚒︎Toolsmith

Nikita Prokopov

DataScript, Fira Code

Claude CodeCodexOpus 5.5live nREPL

1 skillCLAUDE.md2d

Nikita's trick

Live nREPL for the agent, instead of app restarts.

⚒︎Toolsmith

Linus Torvalds

Creator of Linux and Git

ClaudeOpusside-branch issues

CLAUDE.mdtests only2w

Linus's trick

Pedals in a ring, a bench that tests itself.

⚒︎Toolsmith

Thomas Ptacek

Matasano, Latacora, Cryptopals

Claude CodecliCodexOpus 5.5GPT-5.5spare MacBook

3 skillsCLAUDE.md3w

Thomas's trick

swiftui-app template, copied into every new app.

⚒︎Toolsmith

Robert C. Martin

Clean Code author

GrokcliCodexcliGrokno harness

5 skills.grok/rules1w

Robert's trick

uml-viewer, colored by CRAP + mutation.

≡Orchestrator

Walden Yan

Co-founder, Cognition (Devin)

DevinwebFableGPT-5.5Devin teams

agent review2mo

Walden's trick

Daily customer-report Devin, escalates top feature asks.

⚒︎Toolsmith

Wes Bos

Syntax podcast co-host

Claude CodecliCodexappOpenCodecliOpus 5.5Codex models

3 skills2 MCP3w

Wes's trick

vite-plugin-killer-instincts, hands the LLM a kill command.

⚒︎Toolsmith

Wes McKinney

pandas, Apache Arrow

Claude CodecliCodexappFableGPT-5.6 Solown tools

4-5 projects1 hook1w

Wes's trick

roborev, reviews every agent commit.

⚒︎Toolsmith

Xe Iaso

Anubis, xeiaso.net

Claude CodecliCodexOpenCodeOpus 5.5GPT-6.1 Sol

5 agents11 skills6d

Xe's trick

Assisted-by: footer, model + tool on every commit.

≡Orchestrator

Zach Lloyd

Co-founder and CEO of Warp

Warp FactorieswebWarp agentappGPT 5.6 Solfactory-as-code

2 MCP2w

Zach's trick

Scorer + self-improvement agents, agents that fix agents.

$autoreview before every commita review skill that must pass before anything lands.

Peter Steinberger 2w ↗

Showboat agents demo their own worka CLI he built so agents record the commands they ran and the real output, as proof the code works.

Simon Willison 8mo ↗

/go test, /simplify, open PRone 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.

Boris Cherny 6mo ↗

autoresearch overnight agent researchhe edits only program.md and agents run the experiments; round one found ~20 real gains.

Andrej Karpathy 7mo ↗

Harness engineering after every agent mistakehe coined the term. Every bad agent behavior gets an AGENTS.md line or a script so it never happens again.

Mitchell Hashimoto 8mo ↗

"Review this with claude" adversarial review from Codexhe drives from Codex and has Claude review the work, one agent checking the other.

DHH 6d ↗

/answer in place of plan modehe 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.

Armin Ronacher 8mo ↗

T3 Code his own GUI over every agentHe 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.

Theo Browne 1w ↗

QA_BEFORE_RELEASES.md an agent QA pass per releasea 2,088-line release gate an agent works through, which he credits with raising quality.

Salvatore Sanfilippo (antirez) 4mo ↗

Field guide agent writes its own lessonsevery session is logged with commit hashes, then the agent turns it into a project guide it must read before touching code.

ThePrimeagen 3mo ↗

Seats 18 long-lived Fable officersevery agent boots into a named office with its own memory and authority, which he says saves tokens.

Steve Yegge 4w ↗

tm a tmux session per siteA Claude-written script that drops him into a per-project tmux session with Claude Code, straight on the production server.

Pieter Levels 2mo ↗

Daily code limit sized to what he can reviewhe wrote his own agent and still rations how much code it writes in a day.

Mario Zechner 7mo ↗

“Irrefutable proof” video proof, not line readsrather than read the diff, he asks the agent in its orb to prove the work end to end, often with a video.

Thorsten Ball 1mo ↗

/par rival reviewers, 5 pointshis adversarial review command makes two subagents compete to find the most real issues, and inflated findings get a reviewer disqualified.

Jesse Vincent 5mo ↗

"fresh eyes" review a subagent rereads the codehis careful-review command sends a subagent back over new code with "fresh eyes". Jesse Vincent says he learned the trick from Harper.

Harper Reed 10mo ↗

TCR skill red tests revert the codehis own Test && Commit || Revert discipline as an always-on agent rule, so a failing step is thrown away instead of patched.

Kent Beck 6mo ↗

/ce-compound turns each fix into a lessonthe step his whole method is named for, writing every solved problem down so the next agent starts smarter.

Kieran Klaassen 5mo ↗

/grill-me before every changehis most popular skill makes the agent interview him until the design is settled, before any code is written.

Matt Pocock 1w ↗

Claudescope transcript link in every PRhe built his own hook-driven tool so reviewers read the agent session and plan behind a PR instead of the generated diff.

Indragie Karunaratne 6mo ↗

/loop PR triage on Agent Skills, every dayscheduled agent loops make the first pass on the PR queue of his 100k-star skills repo.

Addy Osmani 2mo ↗

The Ralph loop an agent in a bash while loophe invented it. One prompt file fed to an agent again and again until the work is done.

Geoffrey Huntley 1y ↗

"I'm feeling Pucky" button starts half their worka 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.

Quinn Slack 3w ↗

Voice prompting (Hex) instead of typingHe dictates prompts with fast local voice models and says voice, not OpenCode, is what replaced his editor.

Dax Raad 3mo ↗

"Chief of staff" thread spawns every project threadone persistent Codex thread runs his work. It starts each new project thread and checks in on them.

Nick Baumann 5mo ↗

Every tedious OSS task fed to cloud Claude Codesince 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.

Adam Wathan 8mo ↗

antfubot agents ship PRs as a bothis agents work under their own GitHub account and open PRs he reviews and merges, so agent work is always labelled.

Anthony Fu undated ↗

Welfare checks on workers across 4 machinesthe coordinator, not he, watches for stuck or crash-looping workers, so multi-day runs need no steering.

Andreas Kling 3w ↗

This Old Repo reviving a 1990s gamehe 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.

Bryan Cantrill 5mo ↗

sensors check a sidecar the agent pollsshe 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.

Birgitta Böckeler 3mo ↗

A lead agent per project he just files issueseach project gets a standing Copilot lead that triages, tests, merges and cleans up, so his input is the issue tracker.

Burke Holland 2w ↗

"I wrote this PR with an LLM" flagged for careful reviewhe says up front which PRs an LLM wrote, reviews them as untrusted and pares them down himself.

Andrew Gallant 4mo ↗

/pr-pass loops until CI and bots passat 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.

Calvin French-Owen 8mo ↗

Fresh config per model reset at every model launchshe tests each new model on a clean slate instead of carrying old tweaks forward.

Cat Wu 3w ↗

Cross-provider runner Codex, Claude spawn each othershe built her own runner so a strong model plans and reviews while Codex and Claude agents hand work to each other.

Chip Huyen 2mo ↗

`notify msg` in CLAUDE.md, not a hookinstead of hooks, one line in his global CLAUDE.md makes Claude ping him when work is done.

Chris McCord 8mo ↗

meat strips diffs to what mattershis own LLM tool that cuts imports, nil-checks and error handling out of a diff, so his review covers only the architecture.

David Crawshaw 2mo ↗

LGTM, the PR closer a bot that clears her PR queuea bot she built that goes through her PRs every day, merges, closes, rebases and nudges, and hands fixes to Cursor agents.

Claire Vo 1mo ↗

Named-persona review panels with contrarian personasinstead 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.

Dan Luu 3mo ↗

Plumb reviews decisions at commithis 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.

Drew Breunig 7mo ↗

/show-me diagrams, not walls of prosehis own skill makes agents explain plans and diffs as trees, call stacks and diagrams, so he can review them fast.

Dex Horthy 2mo ↗

"Want me to compound?" asked after every sweephis AGENTS.md makes the agent offer, after each run, to turn what it learned into a policy change he approves.

Dan Shipper 3mo ↗

"Using a subagent in parallel, do X" from one always-on threadinstead of opening new threads, he keeps one thread alive and hands every new task to a subagent from it.

Alexander Embiricos 6mo ↗

sl follow / sl adopt restack parallel agentstwo Sapling aliases let agents in several worktrees amend one shared stack without trampling each other.

Edward Z. Yang 7mo ↗

"Reload agent.md" when code quality dropsa one-line fix for context dilution, sent the moment the agent's style slips.

Fabien Sanglard 2mo ↗

filippo-agent agent's own GitHub accounthis agent commits from a separate, low-privilege GitHub account so it never holds his maintainer keys.

Filippo Valsorda 5mo ↗

/plan-ceo-review finds the 10-star productbefore 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.

Garry Tan 1w ↗

AI Disclosure README names the exact modelevery Claude-built repo we found ends with a section saying which model wrote what, down to the effort level.

Jeff Geerling 1mo ↗

/explain-diff with a quiz before he shipsevery agent change becomes a background-intuition-code explainer ending in five quiz questions, and he won't send code until he passes.

Geoffrey Litt 3mo ↗

(kimi) in the PR title names the model per PRhis tinygrad PRs carry the name of the model that helped, a public log of which model wrote what.

George Hotz 2w ↗

Adversarial review 2 Claudes refute every changeseparate Claude reviewers who only see the diff and try to prove it wrong let him merge a million-line port.

Jarred Sumner 5mo ↗

Capability Owners standing agents per featureher 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.

Jessica Kerr 2d ↗

Adversarial review agents on compiler & security workhe makes a second set of agents try to disprove a change before any human sees it.

José Valim 2mo ↗

claude-air nudges after 8 silent roundsa 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.

Jeremy Howard 1w ↗

$ultragoal goal kept in an editable filelong Codex runs chase a goal kept in a file, so he can add scope while the loop is still running.

Jason Liu 3mo ↗

/ship-pr Bugbot loop, merge, Discorda 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.

Kent C. Dodds 2w ↗

plans/ in the repo checked in for code reviewthe agent-written plan is committed beside the code, so reviewers and later prompts work from the same document.

Kenton Varda 2mo ↗

Fail-first fixtures before every extraction fixhis rules make the agent prove a bug with a failing fixture before fixing it, and forbid editing expectations to fit.

Steph Ango 3w ↗

Plan mode before every long agent runhe 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.

Lee Robinson 10mo ↗

/linear-project-loop loops a project to its goala software-factory skill that audits a Linear project's RFC and metrics, files missing issues, then works the unblocked ones.

Will Larson 3w ↗

maggie-design her design taste, as a skillshe says design, not build time, is always her bottleneck, so she wrote her own design philosophy down for the agent.

Maggie Appleton 6mo ↗

Hand-typed skeletons Claude fills in the bodieshe types the types and function signatures himself and has Claude complete only the bodies, so the code keeps his structure.

Alex Kladov 9mo ↗

"You are AGI-pilled." in every agent's system promptone 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.

McKay Wrigley 2mo ↗

AgentIDE his own vibe-built agent IDEhe replaced Superset, Fork and Zed with a native app he built for sandboxed agents in parallel worktrees, now in Homebrew.

Mike McQuaid 3w ↗

iOS Claude pings Android Claude to port each featurehe 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.

Mike Krieger 4mo ↗

codex exec Luna subagents 7–12 cheap critics per passinstead 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.

Max Woolf 3w ↗

/gemini-feedback Gemini critiques each diagramClaude Code draws the textbook figures, then a skill sends each one to Gemini for a second model's review.

Nathan Lambert 2mo ↗

Be introspective Claude fixes its own ruleswhen a workflow goes wrong, his CLAUDE.md has Claude work out why and propose changes to its own instructions.

Ned Batchelder 7mo ↗

code-review-turbo three-agent PR bug huntClaude, Codex and Cursor Bugbot each review the PR, then the main agent cross-checks their findings to drop hallucinated bugs.

Nolan Lawson 5mo ↗

/commit human writes the headlinehe types the first line of every commit himself and Claude only adds the body from the diff, so humans still write for humans.

Orta Therox 3mo ↗

paulirish-agent a GitHub account for agentsagent commits land under their own GitHub identity, so his history shows what he wrote and what the agent wrote.

Paul Irish undated ↗

Audit the agent first before installing any harnessbefore trying a new coding agent he clones its repo and has a trusted agent audit it for egress and blast radius.

Sebastian Raschka 4mo ↗

gdp-ts auth checks as type proofsinstead of reviewing more agent code, he wrote a library, linter and skill that make the type checker reject missing authorization checks.

Guillermo Rauch 1w ↗

Claw Patrol firewall between agents and prodrather than trust the agent, Deno routes its agents' production traffic through a gateway that parses SQL, Kubernetes and HTTP and enforces rules.

Ryan Dahl 5mo ↗

30-second triage rejects most agent attemptshe lets agents attempt every change, throws most away after a half-minute skim, and only reviews in depth the attempt worth keeping.

Sean Goedecke 5mo ↗

Fourth-wall check enforced in CI on every PRa hard rule, checked by a linter in CI, that prototype copy never calls itself a prototype or explains its changes.

Sahil Lavingia 2mo ↗

!land playbook refreshes skills on every PRhis Devin playbook for every merge starts a child session that updates skills and docs, so they never go stale.

Ryan Carson 2mo ↗

rue-overnight-hunt bugs filed only with reprosan overnight workflow that fans out bug finders across one compiler subsystem and files only findings whose repro was actually run.

Steve Klabnik 3mo ↗

$cassandra-creator a standing dissent agentnext to a CEO agent that runs a project, he sets up an independent agent whose job is to challenge it.

swyx 2mo ↗

Five favorable conditions before handing code to an agenthe delegates all the coding only when the work is non-critical, stand-alone, sandboxed, off his core skills and not competing with humans.

Terence Tao 3mo ↗

Change quiz must pass before merginginstead 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.

Thariq Shihipar 3mo ↗

AI Disclaimer names agent and model in PRseach agent-made PR states the agent, the exact model and that he reviewed it by hand.

Sebastián Ramírez 1w ↗

Random-file bug hunt automation, several times a daya scheduled Codex automation literally runs Python random to pick a file, then finds and fixes a latent bug in it.

Thibault Sottiaux 8mo ↗

Pi on GLM 5.2 as the security bug huntera cheap model that does not refuse security work, in a separate harness with Cloudflare's audit skill and a local OWASP cheatsheet checkout.

Tom MacWright 2mo ↗

qmd local search, agents toohis own all-local search engine is the base layer under the tools he builds, and he points his agents at it too.

Tobi Lütke 9mo ↗

Live nREPL for the agent instead of app restartshis CLAUDE.md points the agent at a running nREPL so it evaluates and reloads code the way a Clojure programmer does.

Nikita Prokopov 1mo ↗

Pedals in a ring a bench that tests itselfhe 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.

Linus Torvalds 2mo ↗

swiftui-app template copied into every new appone folder carries his rules, skills and no-Xcode build, so every app starts with cp -R and a sentence.

Thomas Ptacek 2mo ↗

uml-viewer colored by CRAP + mutationinstead 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.

Robert C. Martin 4w ↗

Daily customer-report Devin escalates top feature askshis favorite Devin job is not code at all, it reads every customer report each day and escalates the top requests to engineering.

Walden Yan 2mo ↗

vite-plugin-killer-instincts hands the LLM a kill commandrather 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.

Wes Bos 7mo ↗

roborev reviews every agent commithe built a background reviewer that a second model runs on every commit his agents make, and will not code without it.

Wes McKinney 8mo ↗

Assisted-by: footer model + tool on every commither 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.

Xe Iaso 2mo ↗

Scorer + self-improvement agents agents that fix agentsscorer agents grade every agent run, and other agents turn low scores into PRs that fix the factory's skills.

Zach Lloyd 2w ↗

Questions

Which coding agent do engineers use most?

Among the 94 engineers here: Claude Code (64), Codex (51), Cursor (9), OpenCode (8). 68 of 94 use more than one agent.

How do engineers review AI-written code?

31 of 94 have another model or agent review the code before it lands, 8 still read every line, 21 spot-check the risky parts, and 7 rely on tests alone or ship without a review.

Do engineers run several coding agents at once?

29 of 94 say how many agents they run in parallel, from one at a time to dozens. See each card for their own words.

How often do engineers change their AI coding setup?

Often: 79 of 94 changed their setup (agent, model, harness or how many agents they run at once) in the year to 13 October 2026. Every card shows how old its newest source is.

Where does this data come from?

From each person's own public repos (rules files like CLAUDE.md and AGENTS.md, skills, hooks), their posts, talks and podcasts. Every claim links to its source with a date. Courses and product pitches don't count. The data is open on GitHub.

Every one thing

What we found

How 94 engineers review agent code

318214327

another model or agent reviews: 31they read every line: 8spot checks: 21tests only: 4no review: 3not said: 27

Styles

Our label for how someone works, not a rating.

How to read a card

agentmodel yellowthe one thing to take from them ⚒︎Toolsmithour name for their style appwhere they run it: cli, app, ide, web 8agents at once 2d2monewest source

how it's made

94 stacks, 897 public sources. We read what each person published: rules files in their repos, posts, talks and podcasts. Every field links to where they said it, with the date.

We only use first-person sources about their own setup; courses and product pitches don't count as evidence. Anyone with fewer than five sourced facts doesn't get a card.

What this isn't: a ranking, or proof of what anyone does every day. It's what people say in public, which leans toward the interesting parts and toward people who write about agents. Values marked inferred are our reading, for example a model taken from commit co-author lines.

Archetypes and yellow keys are our picks. Setups change monthly, so cards show how old the newest source is. Last checked 12 Oct 2026.

All the data is open: github.com/prehistoriq/stacks. Wrong, outdated, or want out? Open a pull request on your card, or email [email protected], and we fix or remove it.

Who made this

Prehistoriq. We're building code review for AI-written code, where several models check each change before it ships.

prehistoriq.com

How do engineers actually use agents?

How Nick Baumann actually uses coding agents

In their words

What we found

Styles

How to read a card

How do engineers actually use agents?

You?

Questions

Every one thing

What we found

Styles

How to read a card

Who made this