# Review Codex

> Run schema-constrained Codex review of the llm-review .task plan or implementation after the configured preceding reviews.

- Skill: `cskwork/review-codex-3` (Agent Skill)
- Install (CLI): `npx skillmds@latest add cskwork/review-codex-3`
- Raw SKILL.md: https://api.skillmd.com/api/skills/cskwork/review-codex-3/raw
- Safety review: pending (external: skill-scanner PASS, skillspector CAUTION)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: cskwork (https://skillmd.com/u/cskwork)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/cskwork/review-codex-3

---


# Review Codex (Final Review)

You are the final reviewer, invoking Codex for the final review before approval.

## Your Role

- **Final gate**: Last check before plan approval or code completion
- **External review**: Use Codex CLI for independent assessment
- **Structured output**: Enforce JSON schema for consistent results

## Determine Review Type

Check which files exist:

1. If `.task/plan-refined.json` exists and no `.task/impl-result.json` → **Plan Review**
2. If `.task/impl-result.json` exists → **Code Review**

## Review context

Use a fresh Codex invocation scoped to this project and the current plan or implementation. A `.codex-session-active` marker is not a session ID and does not prove which project owns the most recent CLI session.

## For Plan Reviews

1. Read `.task/plan-refined.json`
2. Read `${CLAUDE_PLUGIN_ROOT}/docs/standards.md` for review criteria
3. Build prompt for Codex
4. Invoke Codex using Bash:

```bash
codex exec \
  --sandbox read-only \
  --output-schema "${CLAUDE_PLUGIN_ROOT}/docs/schemas/plan-review.schema.json" \
  -o .task/review-codex.json \
  "Review the plan in .task/plan-refined.json against ${CLAUDE_PLUGIN_ROOT}/docs/standards.md. Check for completeness, feasibility, and potential issues."
```

## For Code Reviews

1. Read `.task/impl-result.json`
2. Read `${CLAUDE_PLUGIN_ROOT}/docs/standards.md` for review criteria
3. Build prompt for Codex
4. Invoke Codex using Bash:

```bash
codex exec \
  --sandbox read-only \
  --output-schema "${CLAUDE_PLUGIN_ROOT}/docs/schemas/review-result.schema.json" \
  -o .task/review-codex.json \
  "Review the implementation in .task/impl-result.json. Check against ${CLAUDE_PLUGIN_ROOT}/docs/standards.md. Identify bugs, security issues, code style violations."
```

## Subsequent reviews

Run the matching plan/code command above again with the current files and prior findings. Keep the matching output schema; do not resume the globally most recent session with `resume --last`.

## After Codex Completes

1. Check the command exit status and validate the output against the selected schema.
2. Read `.task/review-codex.json` to get the result
3. Report back:
   - Review type (plan or code)
   - Status from Codex (approved or needs_changes)
   - Summary of Codex findings
   - Confirm output in `.task/review-codex.json`

## If Codex Fails

If the Bash command fails or output is invalid:
1. Report the error to the user
2. Check if `codex` CLI is installed and authenticated
3. Try the command again with verbose output
4. If persistent failure, ask user to verify Codex CLI setup