菜单
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.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
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 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.
| name | consulting-codex |
| description | 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. |
| effort | high |
| allowed-tools | Bash(autopilot-codex:*), Read |
Every Codex turn costs tokens. Per the May 2026 rate card, a single
gpt-5.4 call on autopilot averages 25–40k input + 5–10k output =
$0.10–$0.30. Multiply by 30 iterations and a runaway loop is real money.
This skill is the cost/value matrix for deciding "do I ask Codex, or do
I just answer the user".
current_task, plan,
last_review, last_verdict are local. Use /codex-status or
/codex-next first.approve and you want to "double-check", don't. The verdict is the
contract.| Kind of question | Entry point |
|---|---|
| "Decompose this engineering goal into tasks." | codex-manager subagent → mcp__autopilot-bridge__plan |
| "Does this diff meet the current task's acceptance contract?" | codex-reviewer subagent → mcp__autopilot-bridge__review |
| "Is the goal achieved? If not, what next?" | Stop hook (automatic) → mcp__autopilot-bridge__verdict |
| "Should I use library X or Y for this?" (one-shot architectural) | codex-planner subagent (haiku, no state mutation) |
| "Re-plan from scratch — original plan exhausted but goal not met." | /codex-plan <revised goal> |
Codex's prompts (in lib/prompts.mjs) embed:
That's already enough for most calls. Don't add to the prompt unless you're sure — extra context is extra tokens, and Codex's role-specific prompts are tuned to deny shell / MCP / file I/O. Adding hints can break that framing.
gpt-5.4 (default) for plan / review / verdict. gpt-5.5 is
twice the price for marginal quality on these structured tasks.gpt-5.4-mini or haiku for codex-planner one-shot questions
— they're $0.50/$2 per 1M tokens; an architectural opinion costs
pennies.--budget USD on /autopilot to cap a session. The
budget-guard.mjs check hook flips status to stopped_budget once
spend ≥ cap. See safety-rails skill.The bridge prompts explicitly forbid Codex from:
Codex's only output is the structured JSON for plan / review / verdict.
If a Codex response includes shell commands or "I'll just run X" — that's
prompt drift; the bridge's validateXxxShape will reject the output and
the autopilot will surface a clean error to the user.
The bridge does not use codex-reply or --resume. Each call
spawns a fresh codex exec process. Continuity is achieved by embedding
state.json in the prompt — Codex sees the running plan, last_review,
last_verdict, recent commits in every prompt. This trades token cost
for reproducibility: every turn is replayable from the prompt log.