agentleFS
BoredHFBoredHF/albertskills/skills/autopilot/SKILL.md

autopilot

Skill2 starsChanged yesterdayMITView on GitHub

Hands-off mode where Claude prompts itself until a mission is done, then reports back. Asks one round of intake questions, scouts the project, writes a mission plan, and a Stop hook keeps re-prompting Claude step by step (with stuck detection, budget limits and a final report). Use when the user types /autopilot, says "autopilot", "go autopilot", "run on your own", "keep going without me", "do it yourself and get back to me", or wants Claude to stop needing prompts. Also "/autopilot off" or "status".

  • Commits and pushes

81 other projects use a version of this.

Adopted by

Most recently taken up by

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.
  • 0mlkrizzz655335vSkill

    review-bugbot

    缺陷导向代码审查。用 Bugbot 子代理找 bug。

    Also in 3 projects that use this

  • 0mlkrizzz655335vSkill

    rename-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

  • Tsumugii24Skill

    split-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

  • PandelisZSkill

    update-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

  • 32103210Skill

    migrate-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

  • 32103210Skill

    update-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

The file

Show its rules and text

We split this skill into the rules it gives an agent

13 rules. 0 are common, 5 are distinctive, and 8 are very specific to this project.

  • 0

    Common

    Most of the 72 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 8

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Never edit ~/.claude/settings.json without an explicit yes (see First run).

    line 16
  2. Distinctive:

    Never ask more than the one intake round before the final report, except to report a real blocker.

    line 17
  3. Distinctive:

    Never check off a step you haven't verified by running something.

    line 18
  4. Distinctive:

    Never run git push, delete files, add dependencies or change schemas/APIs when Check-ins is "Before anything risky".

    line 19
  5. Very specific:

    Keep state in .work/ of the project, never elsewhere.

    line 20
  6. Very specific:

    Run node ${CLAUDE_SKILL_DIR}/scripts/toggle.mjs off (or status) in the project folder.

    line 23
  7. Very specific:

    Before starting, read ~/.claude/settings.json and look for autopilot in its hooks.

    line 26
  8. Very specific:

    Stop hook (stop.mjs) and SessionStart hook (brief.mjs) are required.

    line 27
  9. Distinctive:

    If the user says no, say autopilot cannot loop without the Stop hook, and stop.

    line 30
  10. Very specific:

    Engage: Run node ${CLAUDE_SKILL_DIR}/scripts/toggle.mjs on <cards|mission> <steps> <goal>.

    line 53
  11. Very specific:

    Do one step, verify it for real (run the test or build), check it off, and add one ## Log line: N · what changed · how verified.

    line 56
  12. Distinctive:

    Keep replies to 2 to 4 lines while running.

    line 61
  13. Very specific:

    If all steps are done, run the Done when check one last time.

    line 65

“Common” is measured: kept by at least 6 in 10 of the 72 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/autopilot

The user says "autopilot", answers one question round, and walks away. You work until the mission is done, then report back. A Stop hook (stop.mjs) re-prompts you after every reply while .work/autopilot.json is on. Scripts live in ${CLAUDE_SKILL_DIR}/scripts/.

When not to use it

  • The task is one small edit or one question: just do it.
  • The goal can't be checked by a command you can run: ask the user to define "done" first.
  • The user wants to review each step: work normally.

Hard rules

  • Never edit ~/.claude/settings.json without an explicit yes (see First run).
  • Never ask more than the one intake round before the final report, except to report a real blocker.
  • Never check off a step you haven't verified by running something.
  • Never run git push, delete files, add dependencies or change schemas/APIs when Check-ins is "Before anything risky".
  • Keep state in .work/ of the project, never elsewhere.

off / status

Run node ${CLAUDE_SKILL_DIR}/scripts/toggle.mjs off (or status) in the project folder. Print the line it outputs. Stop.

First run (once per machine)

Before starting, read ~/.claude/settings.json and look for autopilot in its hooks.

  • Stop hook (stop.mjs) and SessionStart hook (brief.mjs) are required. Without the Stop hook nothing re-prompts you.
  • The statusLine entry (statusline.mjs) is optional: it only adds a badge. If the user already has a statusLine, skip it.
  • If any required entry is missing, show the user the entries from ${CLAUDE_SKILL_DIR}/references/hooks.json with every SKILL_DIR replaced by the absolute path ${CLAUDE_SKILL_DIR}, and ask. Merge them only after a yes, keeping existing hooks. Tell the user to restart Claude Code so the hooks load, then stop. If the hook commands point at a folder that no longer exists (the skill moved or a plugin updated), repeat this step.
  • If the user says no, say autopilot cannot loop without the Stop hook, and stop.

Start (anything else)

  1. Scout quietly (5 cheap calls at most). Use the [project brief] from session start if there is one. Otherwise check: ls, the README's first 40 lines, .work/cards.md and state.md if present, git status --short | head, and how tests and builds run (package.json scripts, gradle, cargo and so on). Don't read source files yet.
  2. Intake: one AskUserQuestion call, 3 questions. Fit the options to what you found, best fit first, marked (Recommended):
    • Mission, e.g. "Finish my /work cards" (only if cards.md has open delegate or pair cards), "Get build + tests green", "Build a feature" (they type it in Other), "Clean up / polish". Skip this question if the user already stated the goal.
    • Run length: Quick (5 steps) · Normal (12) · Long (25).
    • Check-ins: "Only real decisions" (you pick sensible defaults and log them) · "Before anything risky" (deleting, deps, schema/API changes, git push).
  3. Write .work/autopilot.md. For "Finish my /work cards", skip this and use mode cards.
    # Autopilot mission
    Goal: <one sentence>
    Done when: <a check you can run, e.g. `npm test` passes and `npm run build` exits 0>
    Check-ins: <their answer>
    ## Steps
    - [ ] <3-10 small, verifiable steps, in order>
    ## Log
    ## Report
  4. Engage. Run node ${CLAUDE_SKILL_DIR}/scripts/toggle.mjs on <cards|mission> <steps> <goal>. Then go straight into step 1 in the same reply. Don't stop to confirm.

Each step (the hook prompts you with [autopilot step N/M])

  • Do one step, verify it for real (run the test or build), check it off, and add one ## Log line: N · what changed · how verified.
  • If the delegate skill is installed, push each step down to a cheaper model through it. Send a small, self-contained step through the quick lane (one haiku agent). Send bulk work, or a step that splits into 2+ pieces, through waves. Keep inline only steps of 2 tool calls or fewer, and judgment calls. Check off a delegated step only after you've verified it. Without delegate, do the steps yourself.
  • Found more work? Add steps. A step failing twice? Change approach or split it. The hook ends the run if you sit on one step for 3 rounds.
  • Decide small things yourself and log them as decided: …. Respect the Check-ins rule.
  • Need the user? End the reply with the question, which stops the loop, and add blocked: <why> to the Log.
  • Keep replies to 2 to 4 lines while running. The work goes in files, not chat.
  • Obey the [pace] tier if one is shown. At RED, the hook asks you to wrap up.

Wrap up (the hook says "Stopping" or "Every step is checked")

  1. If all steps are done, run the Done when check one last time. Fix failures before reporting.
  2. Fill ## Report in .work/autopilot.md: outcome (done / partial / stuck), files changed, decisions made, what's left or needs the user.
  3. Send a PushNotification if the tool is available: Autopilot done: <outcome in ~10 words>.
  4. Reply with the report in 5 lines or fewer, then TL;DR: <outcome>, and end with [autopilot:done].

Anti-patterns

  • Re-asking intake questions after engaging.
  • Writing steps like "improve code" that no command can verify.
  • Checking off a step from memory or because the edit looked right.
  • Long chat replies mid-run, or restating the plan each step.
  • Doing a big step inline when delegate is available.
  • Continuing past a failing Done when check to write a cheerful report.
  • Leaving the flag on after finishing: run toggle.mjs off if the hook did not.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.