codex-checklisted
Install: claude install-skill johngorat/claude-codex-skills
# Codex Check (one-shot review)
## Quick Start
`/codex-check <what was done, or what to focus on>` — one review round: diff → Codex findings → Claude triages and fixes → short report. No loop, no `APPROVED` gate; the output is advisory.
## When to use / when not
- **Use for:** routine work that follows already-reviewed patterns, small diffs with limited blast radius, config and docs changes, a fast sanity pass before commit.
- **Do NOT use for:** infrastructure code, the first run of a new pipeline or template, validators/self-checks, or changes whose bugs surface late and cost a lot to find. Those get `/codex-debate`.
- **Escalation rule:** if after triage at least one `blocker` or `major` finding stands — Claude agrees it is real and non-trivial, or cannot confidently refute it — say so and recommend running `/codex-debate` for that change. Do not silently absorb an unresolved major.
## Model
Default is the **second tier** in the catalog (one below flagship) — fast and quota-cheap, adequate for a single advisory pass:
```bash
MODEL=$(jq -r '[.models[] | select(.visibility=="list")] | sort_by(.priority) | .[1].slug // .[0].slug' ~/.codex/models_cache.json)
```
Overrides, strongest first: a model named in the invocation ("use sol", "on luna") → `model.txt` next to this SKILL.md → the default above. Effort is `medium`; bump to `high` only if the user asks for a deeper pass.
## Hard Rules
- Reviewer runs `--sandbox read-only`, always. **Never** pass `--dangerously-byp