consulting-codexlisted

# Consulting Codex ## Overview **Core principle:** When a problem is genuinely hard, a second, independent model is worth more than another lap of your own reasoning. Codex (OpenAI's CLI) is a peer you can consult — for a fresh answer, an adversarial gut-check, or an outside diff review. Consult it, weigh what it says, and decide for yourself. **You are asking a colleague, not obeying an oracle.** This runs both ways: a Codex session stuck on something can consult *you* the same way, by shelling out to `claude -p "<prompt>"`. The pattern is symmetric; the mechanics below are the Claude→Codex direction. ## When to Consult (and When Not To) **Consult when:** - You're genuinely **stuck** — the root cause eludes you after an honest attempt, not on first friction. - A **design decision** has real trade-offs and you want an independent voice before committing. - You're about to ship something **risky** and want an adversary to try to break it first. - You suspect you're **missing something** — a second reading catches what your first one rationalized past. **Don't consult when:** - You can look it up or read the code yourself — that's faster and doesn't burn an external call. - It's routine work — consulting on every step is latency and noise, not diligence. - You're using it to **avoid thinking**. Consult *after* forming your own view, so you can judge the answer. Codex calls are **slow and external** — treat each like a heavy job on the host budget: timeboxed, one focused