archflow-autopilot

Runs queued release stories unattended on one branch after a blocker interview.

28|4|Updated Mar 12, 2026
One-click install
npx skills add https://github.com/AZidan/archflow --skill archflow-autopilot-azidan
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: archflow-autopilot
Source: https://github.com/AZidan/archflow/tree/main/adapters/copilot/.github/skills/archflow-autopilot
Command: npx skills add https://github.com/AZidan/archflow --skill archflow-autopilot-azidan

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Overnight or unattended coding sessions stall whenever the agent hits a decision only the user can make, wasting hours of potential progress. This Skill batches every blocking question into one upfront interview, records the answers in a committed ledger, then builds story after story on a single branch without further interaction. ## Core Features & Use Cases - Batched blocker interview: Collects run-policy and per-story decisions up front using structured multiple-choice questions, writing every answer to a YAML ledger at .archflow/autopilot/{run-id}.yaml. - Unattended story loop: Builds each queued story on its own task branch, runs qa-engineer and pm-reviewer, walks statuses (in_progress → review → done), and merges accepted work into the run branch. - Park/fail handling and reporting: Parks stories blocked on undecided questions with candidate answers, fails broken stories after QA retries, and emits one end-of-run report listing done, parked, and failed work. - Use Case: Before leaving for the night, run /archflow-autopilot to answer a handful of questions, then wake up to a branch of reviewed, merged stories plus a report of anything that needs your decision. ## Quick Start Ask the agent to run /archflow-autopilot to interview you and then build every eligible story in the active release on a single branch.

Frequently Asked Questions about archflow-autopilot

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run coding stories unattended overnight with an AI agent?

Run /archflow-autopilot, which interviews you once to batch every blocking decision, writes the answers to a ledger, then builds each queued story on task branches merged into a single run branch. You review and merge that branch yourself in the morning.

How do I run only specific stories with archflow-autopilot?

Pass story ids after the command, such as /archflow-autopilot S3-04 S3-05, and only those stories run in the given order. Ids not in the active release halt the run, since autopilot never promotes backlog items on its own.

What happens when an unattended run hits a decision it cannot make?

The story is parked: work-in-progress is committed with a wip: prefix on its task branch, the story gets a parked block containing the question and 2-4 candidate answers, and the run continues with the next story. Parked stories block the release by default until you answer.

Can archflow-autopilot merge to main or open pull requests?

No. Autopilot may only merge task branches into its own run branch. It must never merge or push to main, open pull requests, force-push, stash user work, ship releases, or perform irreversible operations like deployments or production migrations.

How do I resume an interrupted autopilot run?

Use /archflow-autopilot resume, which reads the newest ledger with status preflight or running, verifies the run branch still matches the ledger, re-asks only questions for parked stories, and continues the queue without repeating the full interview.

Why does autopilot halt on a dirty git working tree?

A dirty tree halts the run because uncommitted changes belong to the user and an unattended run must never stash or bury them. Autopilot treats any failed precondition as a hard stop rather than working around it.