autopilot

Orchestrates an autonomous interview-to-QA delivery loop with gated review cycles.

Updated Feb 1, 2024
One-click install
npx skills add https://github.com/ogiboy/portfolio-app --skill autopilot-ogiboy
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: autopilot
Source: https://github.com/ogiboy/portfolio-app/tree/main/.codex/skills/autopilot
Command: npx skills add https://github.com/ogiboy/portfolio-app --skill autopilot-ogiboy

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Non-trivial coding tasks often fail because requirements are vague, planning is skipped, and code ships without review or QA. Autopilot enforces a strict autonomous loop that takes a raw idea or issue through clarification, consensus planning, durable implementation, code review, and adversarial QA before declaring completion. ## Core Features & Use Cases - Strict Phase Pipeline: Runs deep-interview, ralplan, ultragoal, code-review, and ultraqa in order, with automatic return-to-planning when review or QA gates are not clean. - Durable State Management: Persists phase, iteration, review verdicts, and handoff artifacts via the omx state CLI so work can be resumed with "continue" after interruption. - Consensus Gating: Requires sequential Architect and Critic approvals before implementation begins, preventing unreviewed plans from reaching execution. - Use Case: A user says "$autopilot implement GitHub issue #42" and the loop clarifies requirements, produces an approved plan, implements and verifies the change, then iterates through code review and QA until both gates are clean. ## Quick Start Ask the assistant to run autopilot on your task, for example by saying "use autopilot to build the password reset feature from issue #42".

Frequently Asked Questions about autopilot

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

FAQPage Schema
How do I run an autonomous coding workflow from a GitHub issue?

Invoke autopilot with the issue reference, such as "$autopilot implement GitHub issue #42". The loop clarifies requirements through deep-interview, plans via ralplan, implements with ultragoal, then gates completion on code review and QA.

What phases does the autopilot loop run in order?

The fixed order is deep-interview, ralplan, ultragoal, code-review, then ultraqa. If code review or QA is not clean, the loop returns to ralplan with the findings and repeats until gates pass or a blocker is reported.

Can I resume an interrupted autopilot session?

Yes. Saying "continue", "resume", or "keep going" reads the persisted autopilot state and picks up from the recorded current_phase without restarting discovery or discarding handoff artifacts.

When should I not use an autonomous delivery loop?

Avoid it for brainstorming, single focused code changes, or review-only requests. Use plan/ralplan for exploration, ultragoal for one targeted change, and code-review when you only want critique of existing code.

What happens when code review finds problems in the implementation?

A non-clean review increments the review cycle and enters a rework phase scoped to the findings. If the plan itself is wrong, the loop returns to ralplan with a recorded return_to_ralplan_reason.