autopilot
Skill9 starsChanged 16 days agoMITView on GitHub
The whole coding loop on one deliberate yes: map, spec, critic-hardened plan, GitHub issues, parallel worktree implementers, fresh-context review loops, one PR with the evidence. Acts only on its itemized charter; parks what it can't settle honestly. Commands: /autopilot <task> [--merge], resume, status. It signs a charter and pushes branches, so it never auto-triggers — explicit invocation only.
76 other projects use a version of this.
Adopted by
Most recently taken up by
- mkritter3/codex-paired-superpowerstheir own version, first seen Nov 15, 2026
- stericishere/skillstheir own version, first seen Nov 15, 2026
- yan-labs/yan-skillstheir own version, first seen Nov 10, 2026
- Yeachan-Heo/oh-my-claudecodetheir own version, first seen Nov 9, 2026
- zahidakhyar/skillstheir own version, first seen Nov 8, 2026
- kimkon1011/orgiast-claude-rulestheir own version, first seen Nov 8, 2026
Used with
Projects with it also use review-bugbot, rename-chat and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
0mlkrizzz655335vSkillreview-bugbot
缺陷导向代码审查。用 Bugbot 子代理找 bug。
Also in 3 projects that use this
0mlkrizzz655335vSkillrename-chat
重命名当前聊天。按聊天内容改标题。"disable-model-invocation: true
Also in 3 projects that use this
Tsumugii24Skillsplit-to-prs
Split current work into small reviewable PRs. Use when the user asks to split a chat, set of changes, branch, or PR.
Also in 4 projects that use this
PandelisZSkillupdate-cli-config
View and modify Cursor CLI configuration settings in ~/.cursor/cli-config.json. Use when the user wants to change CLI settings, configure permissions, switch approval mode, enable vim mode, toggle display options, configure sandbox, or manage any CLI preferences.
Also in 3 projects that use this
32103210Skillmigrate-to-skills
Convert 'Applied intelligently' Cursor rules (.cursor/rules/*.mdc) and slash commands (.cursor/commands/*.md) to Agent Skills format (.cursor/skills/). Use when you want to migrate rules or commands to skills, convert .mdc rules to SKILL.md format, or consolidate commands into the skills directory.
Also in 3 projects that use this
32103210Skillupdate-cursor-settings
Modify Cursor/VSCode user settings in settings.json. Use when you want to change editor settings, preferences, configuration, themes, font size, tab size, format on save, auto save, keybindings, or any settings.json values.
Also in 3 projects that use this
- +1 more from 32103210
- +1 more from 0mlkrizzz655335v
Other versions76 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 76
- Yeachan-Heo/oh-my-claudecode★ 41ksame name
- Yeachan-Heo/oh-my-claude-sisyphus★ 41ksame name
- Yeachan-Heo/oh-my-codex★ 34ksame name
- SethGammon/Citadel★ 927same name
- H-mmer/pentest-agents★ 815same name
- Harry-kp/vortix★ 757same name
- nick-vels/skills★ 599same name
- yan-labs/yan-skills★ 307same name
- kentcdodds/kcd-skills★ 64same name
- Kanevry/session-orchestrator★ 53same name
66 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository21 other files it gives its agents.
The file
We split this skill into the rules it gives an agent
8 rules. 0 are common, 5 are distinctive, and 3 are very specific to this project.
0
Common
Most of the 68 edited versions we read keep these.
5
Distinctive
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 32
WILL: read anything in the repo.
- Distinctive: line 69
Run map in auto mode: no AGENTS.md → generate one.
- Very specific: line 84
Write plan.md per feature Step 2 — 2–5-minute tasks, exact files, own verification step, dependency order.
- Distinctive: line 136
Run done's fake-green tripwires on every diff: tests modified or deleted in the diff that makes them pass, new skips, hardcoded …
- Distinctive: line 156
Run done in auto mode on the integrated diff — the full evidence checklist, including *actually running the change*, with output quoted.
- Very specific: line 177
Abort: create out/dev/<slug>/ABORT (empty file) and the orchestrator stops cleanly at the next wave boundary — worktrees intact, state …
- Distinctive: line 190
Tests are the arbiter: No LLM verdict — critic, reviewer, or implementer, local model or cross — outranks the suite.
- Distinctive: line 196
Inherited hard rules hold: Never weaken a test, no fake-green, scope contract, append-only honest logging — every rule from feature/done …
“Common” is measured: kept by at least 6 in 10 of the 68 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
autopilot — the whole loop, one yes
/autopilot <task> — run the full pipeline, stop at an open PR
/autopilot <task> --merge — also merge, once every gate is green
/autopilot resume <change> — pick up a crashed, aborted, or parked run
/autopilot status — report runs in flight under out/dev/
Answer one question: what would come back if this task were handed to a careful team and nobody watched over their shoulder? The other coding skills put a human at every gate; this one replaces those gates with the things that made them work — fresh-context review, diverse critique, and a test suite as the arbiter — and reserves the human for exceptions. Human on the loop, not in it: you sign once, you get a PR, and anything the pipeline couldn't settle honestly comes back parked, not guessed.
The charter — one yes, spelled out
The invocation is the consent. Echo this before running, then don't ask again:
- WILL: read anything in the repo; write state under
out/dev/<slug>/; create GitHub issues labeledautopilot; create and push branchesautopilot/<slug>/*; open one PR; comment on and close its own issues; with--merge, merge that PR once every gate is green. - NEVER: push to the default branch directly, force-push, rewrite history, delete branches or issues it didn't create, weaken a test to get green, or touch anything outside this repo.
Auto mode — how the other skills run here
Autopilot drives map, feature, and done in auto mode, a shared
convention: every approval gate becomes a self-approval on the record —
the decision and its reason are appended to log.md instead of asked — and
every verdict that would need a human becomes a PARK. The skills' own
rules (task sizing, TDD, tripwires, never-weaken-tests) are unchanged; only
the waiting is removed. In auto mode the run never calls a tool that
stops to wait on a human — AskUserQuestion first among them (waiting on
sub-agents or commands is fine): with nobody watching, such a call isn't a
question, it's a deadlock. A decision only a human can settle is parked
with the decision spelled out — never waited on, never guessed.
Step 0 — Preflight
- Confirm: a git repo, a clean tree,
ghauthenticated. Withoutgh, degrade honestly — local task files instead of issues, stop at the branch instead of a PR — and say so in the run header. - Find the test command (
AGENTS.md,package.json,Makefile,pyproject.toml, CI config). No test suite? Don't refuse — schedule a testing-bootstrap phase as the plan's first wave: stand up the runner, then write characterization tests around exactly the code the plan will touch. Without an objective test to pass, every review loop below is just one model agreeing with another; the bootstrap fixes that. - Open
out/dev/<slug>/log.mdwith the run header: charter, stop point (PR / merge), test command, start commit.
Step 1 — Orient
Run map in auto mode: no AGENTS.md → generate one; one exists → refresh
only if stale. Every downstream agent starts oriented instead of cold.
Step 2 — Spec: interview the codebase, not the user
feature Step 1 with the questions turned inward — the facts-vs-decisions
split holds with no user present. Every fact is answered by reading code,
docs, tests, and git log; whatever evidence can't settle — an open
decision or an unverifiable fact — is never resolved silently: it becomes a
numbered assumption (A1…) in spec.md, visible and revisable at the
PR, where wrong assumptions are cheap to catch. The
spec remains the contract every review below is judged against.
Step 3 — Plan, hardened by critique instead of approval
Write plan.md per feature Step 2 — 2–5-minute tasks, exact files, own
verification step, dependency order. Then replace the human approval with a
bounded critique loop: parallel critic sub-agents, each with a different
lens — feasibility, hidden dependencies & ordering, test coverage, scope
creep — each returning blocking findings or an approval. If another
model's agent CLI is installed (detect per done Step 2), give one critic
seat to it: fresh context varies the perspective, a different model varies
the blind spots. Fix the blocking
findings, re-run the critics once. Two rounds maximum: past that,
critique stops finding real problems and starts agreeing with itself. Surviving blocking
findings park the run at the plan stage — a plan the critics can't pass is
the user's call, not a coin flip.
Step 4 — Tasks become GitHub issues
One issue per plan task, created in dependency order so every
depends-on: #n names a real number: the full briefing (a fresh agent must
be able to execute from the issue text alone), the files it touches, its
verification command, and the spec IDs it implements ([D2, A1], where
one applies). Label
autopilot, title [<slug>] <task name>. Record issue numbers back into
plan.md. Issue prose describes behavior, not code — "the sync service
drops the patch", never applyPatch() at line 42, no file paths or line
numbers in prose; analysis artifacts cite
file:line because they're verified now, tracker artifacts must outlive the
refactor. (The files-touched list stays — it's the worker's briefing, not
durable prose.)
Local files stay canonical; the issues are the published mirror — and the
hand-off that actually reaches the workers, since the run's out/ state is
uncommitted mid-run and invisible inside the worktrees below.
Step 5 — Dispatch in dependency waves
- One sub-agent per task, in an isolated git worktree, on branch
autopilot/<slug>/task-<n>, with the full task briefing in its prompt. - Independent tasks run concurrently; a dependent wave starts only when the tasks it depends on have merged (Step 7).
- Each agent works by feature Step 3, verbatim: failing test first, then the minimal green, then a checkpoint commit — never weaken an existing test, no TODOs or skips as a path to green, two failed attempts at the same task → stop and park, don't grind the wall.
Step 6 — Review loop, per task, until clean
- Fresh-context reviewers per done Step 2: they see only the diff, the
task briefing, and the spec — never the implementer's conversation. A
reviewer with no memory of writing the code has no loyalty to it.
Spec-compliance first, correctness second, style never. That includes
done's optional cross-model second opinion when a
codex/geminiCLI is on the PATH — it earns the most here, where no human sits at any gate. - Run done's fake-green tripwires on every diff: tests modified or deleted in the diff that makes them pass, new skips, hardcoded expectations, TODO stand-ins.
- Findings go back to the implementer to fix, then re-review. Clean → the task is merge-ready. Three rounds without clean → PARK the task with the findings on the record; the rest of the run keeps flowing.
- The arbiter is the test suite: a reviewer's approval doesn't override a red suite, and a green suite doesn't excuse a spec violation.
Step 7 — Integrate wave by wave
The orchestrator merges merge-ready task branches into
autopilot/<slug>/main in dependency order, running the full suite after
every merge. A red integration points at the merge that broke it — that
task returns to Step 6 with the failure attached; independent tasks keep
flowing. Only the orchestrator merges; workers never touch each other's
branches.
Step 8 — Ship
- Run done in auto mode on the integrated diff — the full evidence checklist, including actually running the change, with output quoted.
- Push the integration branch; open one PR: the spec's summary, the
evidence checklist,
Closes #nfor every completed issue — and parked tasks listed plainly, not swept. - Comment each issue with its outcome; parked issues stay open with the park reason.
--mergeand every gate green → merge (never force, branch protection respected). Otherwise the PR is the deliverable — merging is yours.- Report in chat and
log.md: per task — SHIPPED / AT PR / PARKED (why) — and where human attention is actually needed. Clean up worktrees; archiveout/dev/<slug>/once merged.
State and the escape hatch
out/dev/<slug>/: spec.md, plan.md (tasks + issue #s + status),
log.md (append-only; the orchestrator is its only writer — parallel
writers corrupt logs), tasks/<n>.md, .work/. Resume uses feature's
ritual: log, then plan, then git log, then continue at the first
unfinished task — never trust residual conversation memory.
Abort: create out/dev/<slug>/ABORT (empty file) and the orchestrator
stops cleanly at the next wave boundary — worktrees intact, state saved,
resume picks it back up.
Guardrails
- The charter is the whole consent. Nothing outside the WILL list, ever — scope discovered mid-run becomes a follow-up issue, not a silent expansion of the mandate.
- Park, don't guess. Ambiguity a critic loop can't settle, a task two attempts couldn't crack, a review three rounds couldn't clean — parked with the evidence, run continues. A guessed answer costs more than a parked task.
- Tests are the arbiter. No LLM verdict — critic, reviewer, or implementer, local model or cross — outranks the suite. No suite → bootstrap one before implementing.
- Bounded loops everywhere. Two critique rounds, two implementation attempts, three review rounds. Unbounded self-correction diverges or rubber-stamps; the bounds are what keep it honest.
- Inherited hard rules hold. Never weaken a test, no fake-green, scope contract, append-only honest logging — every rule from feature/done applies verbatim; autonomy removed the waiting, not the rules.
- Honest report or no report. Parked tasks appear in the PR body and the final report with reasons — a run that hides its failures has failed twice.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

