autopilot-codex

Run the full Codex-as-Manager autopilot — Codex plans, you implement, Codex reviews each diff, the Stop hook drives the loop until done. Safety rails on Bash; iteration cap on the Stop loop. Confirm before risky actions.

Show the current task contract from state.json without spawning a Codex turn. Read-only — useful when you need a quick reminder of what you should be implementing.

Ask OpenAI Codex (manager / architect) to decompose a goal into a structured plan with ordered ~5-30min tasks. Saves a plan markdown to docs/autopilot/plans/. Read-only — no code changes.

Manually trigger Codex review of the current working tree against the active task's acceptance contract. Useful when the autopilot is paused, when re-reviewing a specific task, or when debugging review behavior. Read-only.

Show the current AutoPilot Codex run state — goal, plan path, current task, base SHA, last review, iteration count.

USE PROACTIVELY whenever the user states a fresh engineering goal in this repo and AutoPilot Codex is enabled. Consults Codex (the manager) via the bridge, persists a structured plan, and hands the FIRST task back to the main thread. Read-only — never writes code itself.

Use ONCE for a focused architectural question (data-model trade-off, library choice, API shape, failure-mode analysis) where you want Codex's opinion without committing to a full plan. Returns Codex's plain-text answer; does NOT mutate plugin state. Read-only.

Use AFTER any non-trivial Edit/Write/MultiEdit during an active autopilot run to get a Codex verdict on the change. Reads the current task from state.json, computes the diff vs base_sha, and asks Codex (the manager) whether the diff meets the task's acceptance criteria. Returns the verdict JSON verbatim. Read-only — never edits files.

Use when running an active autopilot-codex session — explains the plan→implement→review→verdict cycle, state machine, and termination conditions. Required reading when status='running' and the implementer needs to act inside the loop.

Use when interpreting state mutations during plan/review/verdict transitions — explains exactly which fields the bridge writes when, so the implementer can debug 'why is current_task suddenly T3' or 'why is base_sha old'.

Use when deciding whether a question warrants spawning a Codex turn or you can answer locally — covers the cost/value matrix and which subagent or command is the right entry point for each kind of consultation.

Use whenever you're about to run a destructive Bash command, push code, or spend Codex tokens — explains every cap and denylist this plugin enforces and how to extend them.

Use whenever the user invokes /autopilot-codex:* in this repo, OR mentions "autopilot", "Codex manager", "Codex architect", or asks Codex to plan/review for them. Establishes the role split — OpenAI Codex is manager/architect, Claude Code is implementer — and the operating contract.

6 hooks across 6 events

Surfaces autopilot hook log lines (review verdicts, budget exhaustion, iteration cap) as session notifications so the user sees state transitions without polling /codex-status.