What problem does it solve? Multi-step coding tasks normally require you to guide the AI through every stage — planning, implementation, review, and fixes. This Skill removes that supervision burden by self-orchestrating the entire workflow: a planner breaks the work into a task list, executor subagents complete each task, a reviewer checks the result against your original request, and a fix loop resolves findings until the review comes back clean. ## Core Features & Use Cases - Phased orchestration: A five-phase pipeline (clarify, plan, execute, review, fix loop) where each phase runs as a dedicated subagent with an explicit model tier — opus/fable for planning and review, sonnet for normal tasks, haiku for fully-specified simple tasks. - Verifiable task completion: Every planned task carries a pass condition that can fail (an exact command, file shape, or comparison), so no task is declared done on the executor's own say-so. - Progress visibility: The task list is mirrored into TodoWrite and persisted to .claude/autopilot-tasks.md, with per-task model annotations and review-round status, so you always know what is done and what remains. - Use Case: Say "autopilot: add DLQ handling to the order consumer" and the Skill plans the schema, producer config, and tests as separate tasks, executes them in parallel where safe, reviews the actual changed files by running the project's own checks, fixes any findings, and hands back a terse summary of what changed, what broke, and what you need to do next. ## Quick Start Ask the AI to run autopilot on your task, for example: "/autopilot add retry logic and dead-letter handling to the payment consumer".
