agentleFS
jaroslawherodjaroslawherod/micold-ai-ide/.claude/skills/speckit-autopilot/SKILL.md

speckit-autopilot

Skill1 starsChanged 3 days agoApache-2.0View on GitHub

Use when the user hands over a feature idea, a bug report, a chore or a GitHub issue and wants it run end to end with as little of their involvement as possible — "autopilot", "run it autonomously", "take it all the way to main", "only ask me when you must" — or says to resume or continue an interrupted autopilot run.

The file

Files
SKILL.md

Spec Kit autopilot

One prompt or one issue in; reviewed work merged to main out. Ask the human only for decisions that are theirs. No human reviews anything here: every artifact and diff is reviewed by a fresh-context subagent, never the context that wrote it. Green CI is not a review.

This file only routes. Each step is a small file: read the one the table names when its moment comes, and no other. Follow each to the letter; a rule that looks slow is what stops an unattended flow shipping the wrong thing.

You are the orchestrator

This session chooses the flow, keeps the issue, dispatches units, asks the human, waits on CI and merges. It never runs a Spec Kit skill and never reads a unit's task files or transcript. Everything it reads is re-read on every later call: batch independent probes into one message.

Each unit is a fresh subagent that reads rules/unit.md and the task files its flow lists. Overview for humans: README.md.

Route

First scripts/autopilot/preflight.sh, before any fetch or claim. PREFLIGHT-FAILED: stop and tell the user what it names (it leaves no label, comment or claim on the issue). Then git fetch origin, and check the worktree is clean.

WhenRead
The argument is resumetasks/resume.md
Any other argument (a trailing supervised also: rules/supervised.md): #<n> or an issue URL (gh issue view <n> --json title,body,labels), <flow> [high|low] #<n>, or text. Choose the flow and the effort from the command, the labels and the textflows/choose.md, then only the chosen one of flows/bug.md, flows/bugfix.md, flows/feature.md, flows/chore.md
The flow is chosen: claim or open the issue, write the labels backtasks/issue.md
The session is a cloud threadrules/cloud.md
Starting a unit, and when it returnstasks/dispatch.md
A unit returned a PRtasks/merge.md
CI red outside this flow, a PR with no checks, a refused mergetasks/ci.md
A unit returned NEXT:, or passed its flow's capflows/switch.md
A unit returned ESCALATE, or the flow is not plaintasks/ask.md
The run's last PR mergedtasks/handoff.md
Unsure whether something is this flow's to touchrules/ownership.md

Red flags: stop and re-read the file the table names

ThoughtReality
"This step is small, I'll do it here"Everything the orchestrator reads is re-read on every later call. Dispatch the unit.
"It's labelled bug, but I'd call it a feature"Labels win over your reading. They disagree with the text: ask once.
"I'll run the full flow to be safe"The lighter flows exist because the heavy steps found nothing on small work. Follow the flow chosen.
"I'll check the unit's work in its transcript"Read the ledger and the five-line return.
"CI is green, so it's reviewed"Units run the reviews. CI checks only that the code works.
"That green PR from another session is ready, I'll merge it"Not in the ledger, not yours. No --delete-branch either.
"main is red, I'll wait for it to recover"Silent waiting stalls the flow. Escalate as blocked by work outside my flow.
"Quick question for the user…"Batch it with a recommendation under the banner, or resolve it from evidence.
"I remember where I was"The ledger and gh pr view say where you are. Memory does not.
"Everything merged. Done!"Run the handoff task: check, token report, close the issue, WORK COMPLETE.

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.