second-opinionlisted

# /second-opinion — Cross-LLM Review via Codex CLI Get an independent review from OpenAI via the Codex CLI. Three modes for different needs: quick pass/fail review, adversarial break-it pass, or multi-turn consultation. ## Authentication Codex CLI uses `chatgpt` auth mode (ChatGPT subscription), not an API key. API-key auth also works but burns tokens against your OpenAI bill; the subscription path is free at the margin and is what this skill assumes. Verify before running: ```bash grep -q '"auth_mode": "chatgpt"' ~/.codex/auth.json && echo "OK: Using ChatGPT subscription" || echo "ERROR: Not using ChatGPT auth" ``` If not authenticated, tell the user to run `! codex login` (interactive). ## Mode Detection Parse user input to determine mode: | Input | Mode | |-------|------| | `/second-opinion review` | Review | | `/second-opinion challenge` or `/second-opinion challenge security` | Challenge | | `/second-opinion <any question>` | Consult | | `/second-opinion` (no args) | Auto-detect: if uncommitted changes exist → Review, else → Consult | ## Mode 1: Review (pass/fail code review) Reviews the current diff against the base branch. ```bash cd <project-root> codex review --base HEAD~1 -c 'model_reasoning_effort="xhigh"' 2>/dev/null ``` Or against a specific base: ```bash codex review --base main -c 'model_reasoning_effort="xhigh"' 2>/dev/null ``` **Output handling:** - Present Codex's output verbatim inside a `### CODEX SAYS` section - Look for `[P1]` markers in the