# Discuss With Codex

> Use codex/codex-reply as a persistent second-engineer thread for non-trivial uncertainty — design choices, risky implementation decisions, debugging when stuck, review, spec ambiguity, needing a current/authoritative web_search, or any point where you'd otherwise guess. Keep one thread and iterate; don't treat the first answer as gospel.

- Skill: `yfe404/discuss-with-codex` (Agent Skill)
- Install (CLI): `npx skillmds@latest add yfe404/discuss-with-codex`
- Raw SKILL.md: https://api.skillmd.com/api/skills/yfe404/discuss-with-codex/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: yfe404 (https://skillmd.com/u/yfe404)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/yfe404/discuss-with-codex

---


Codex is a **colleague**, not an oracle. Use it to test your reasoning.

- **Gather local context first.** Read the relevant files, diff, spec, logs, or failing output before asking.
- **One thread, many turns.** Start with `codex`; continue the _same_ thread with `codex-reply` + its `threadId`.
- **Send the real artifact.** The diff, spec line, stack trace, test output, or concrete options — not a paraphrase.
- **State your current read.** What you think, what you're unsure of, and the decision you need.
- **Push back.** Add the missed constraint, challenge assumptions, ask what would change its recommendation.
- **Omit the `model` override.** Inherit the configured default; never guess a model id (a stale id errors).