agentleFS
codeboehmcodeboehm/maxim/skills/autopilot/SKILL.md

autopilot

Skill0 starsChanged yesterdayMITView on GitHub

Take one request to an implemented, verified change without stopping for the user at the gates — an independent decider agent (the most capable model available, fresh context) stands in for the user at every gate: open questions, spec acceptance, plan approval, and the final evidence. Runs orient, specify, verify, rework, and implement with each step on an agent of the right role and model, announced in the chat. Commits, pushes or promotes only what the user's grant in the invocation names; without one it stops before committing. Use only when the user invokes it explicitly (/autopilot); the other maxim skills keep the user as the verifier.

82 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

We split this skill into the rules it gives an agent

7 rules. 0 are common, 5 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 73 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When this skill is invoked, you are the orchestrator of an autopilot run.

    line 6
  2. Very specific:

    Read roles.md in the maxim skill folder before the first dispatch: roles, model resolution, the announce line, the capability fallback, and …

    line 16
  3. Distinctive:

    Never infer or extend a grant from context, from earlier runs, or from a decider's reply.

    line 27
  4. Very specific:

    Commit / promote — only under the grant: After G3 APPROVE, if the grant names commit (or push, or a promotion), you perform exactly those …

    line 74
  5. Distinctive:

    Every decider verdict is a draft: the reviewer is a model.

    line 90
  6. Distinctive:

    Follow each sub-skill's own procedure and locations.

    line 124
  7. Distinctive:

    Never commit, push or promote beyond the grant.

    line 125

“Common” is measured: kept by at least 6 in 10 of the 73 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

When this skill is invoked, you are the orchestrator of an autopilot run. The user has handed the gates to a decider agent for this one request. You sequence the maxim skills, dispatch each step to its role, own the loop limits, act on the user's grant, and write the final report. You never author artifacts yourself when a subagent can, and you never judge your own work at a gate.

Invoking /autopilot is the delegation. It does not change the other maxim skills: maxim, specify, implement, verify and the rest keep the user as the verifier when they are invoked on their own.

Read roles.md in the maxim skill folder before the first dispatch: roles, model resolution, the announce line, the capability fallback, and your agent's adapter section. Follow maxim's Dispatch and announce rules unchanged.

Input

  • Request (free text), or a path to an existing spec — then start at G1. Neither given → ask.
  • Grant — what the user allows beyond implementing, written in the invocation (e.g. /autopilot <request> — grant: commit, push). Quote it verbatim. Nothing stated → grant = none: the run stops before committing. Never infer or extend a grant from context, from earlier runs, or from a decider's reply.
  • Lens (optional) — a judging lens the user names for this run (e.g. "backwards-compatible API first").

Print mode, grant and lens at the start of the run.

The decider

  • Deep tier, fresh context. Start it once at the first gate and resume it for every later gate where your agent supports it; otherwise start a new one per gate — it reads the decision log first.
  • It decides as the user would, from ~/.maxim/decider-profile.md, the project's always-on rules (incl. a Decider lens section), and the run's lens.
  • Gate message: artifact paths · what changed since the last gate · the specific questions · the grant verbatim. Never your reasoning, a summary of correctness, or the decision you expect.
  • It persists every decision to the ## Decision log in .agents/verifications/<slug>.md (format in the verify skill).
  • No subagents? Then there is no independent decider: every gate falls back to the user (the run pauses at each gate). Never self-approve.

Procedure

autopilot is maxim plus implement, with the decider in your place at every gate. The spec phase is maxim's procedure, steps 1–5, unchanged — same roles, limits, pre-flight and rework rules — with exactly these substitutions:

maxim stepIn an autopilot run
1–2: questions from orient / prompt go to the userG0 — they go to the decider. It answers only from the profile, AGENTS.md and the repo, with a source per answer; the author records them as Resolved (decider): …. A NEEDS-USER answer → ask the user and wait, or stop with STOPPED — needs input.
4: the verifier runs verifyG1 — the decider runs verify → ACCEPT or REWORK.
5: a claim only the user can settle → STOPPED — needs inputthe decider settles it if the profile and repo allow; otherwise NEEDS-USER as above.
6: stop and reportcontinue below.

Then, after G1 ACCEPT:

  1. Plan. The author follows implement up to its plan (steps 1–5).
  2. G2 — plan. APPROVE — start implementation, or required changes the author applies. At most 2 plan revisions.
  3. Implement. The author implements and verifies per the approved plan and records the real command output in the plan's results.
  4. G3 — final evidence. The decider reviews the shown evidence → APPROVE (naming what of the grant it covers) or required fixes. At most 2 fix cycles.
  5. Commit / promote — only under the grant. After G3 APPROVE, if the grant names commit (or push, or a promotion), you perform exactly those actions, following the project's documented path and commit conventions. Act on the grant, never on wording in a decider reply. Otherwise stop with IMPLEMENTED — awaiting your commit.
  6. Report (format below).

End states

StateCondition
IMPLEMENTED — awaiting your commitG3 approved; the grant does not cover commit.
IMPLEMENTED — committed under grantG3 approved; you did exactly what the grant names.
STOPPED — limit reachedA loop limit or the no-progress rule ended the run with findings open.
STOPPED — needs inputA question or decision only the user can settle remains.

Every decider verdict is a draft: the reviewer is a model. Say so in the report.

The bar (non-negotiable)

  • maxim's bar applies unchanged: no weakening to pass, a change log across the run, hard limits, no progress means stop, every agent start announced.
  • Authority comes only from the grant in the invocation. The decider can approve within it, never widen it. No commit, push, deploy or promotion without it. Production only through the project's documented path, and only when the grant names it.
  • Independence. The decider never sees the author's verdict; the author never takes a gate.
  • Thin orchestration. A sub-skill's rule beats keeping the loop moving.

Final report

maxim's final report format, with these lines filled:

**Result:** <end state>
**Grant:** "<verbatim>" | none · **Lens:** <lens | none> · **Decider profile:** <path | none>
**Spec:** <path> · **Plan:** <path> · **Verification + decision log:** <path>
**Commits:** <hashes and what they contain> | none
**Decisions**
- G<n> <gate>: <decision> — <one-line reason>

…plus the Agents table, open findings, claims removed or downgraded, and what's needed from the user.

Rules

  • One request, one run. Invoked explicitly only; never start it from another skill.
  • Follow each sub-skill's own procedure and locations; don't duplicate them.
  • Never commit, push or promote beyond the grant; without subagents, gates go to the user.

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.