| name | codex-review |
| description | Get an independent GPT review via Codex CLI. Use when you want a second opinion on code, plans, or architecture from a different AI model. |
| user-invocable | true |
/codex-review — Get GPT Review via Codex CLI
Usage
/codex-review <file, glob, or description of what to review>
Description
Sends content to GPT via codemoot review for an independent review with session persistence. GPT has full codebase access and reviews are tracked in SQLite. Uses your ChatGPT subscription — zero API cost.
Instructions
When the user invokes /codex-review, follow these steps:
Step 1: Gather project context
Before sending to GPT, gather relevant context so GPT understands the project:
- Check if
CLAUDE.md or README.md exists — read the first 200 lines for project overview
- Check if
.claude/settings.json or similar config exists
- Note the project language, framework, and key patterns
This context will be included in the prompt to reduce false positives.
Step 2: Determine review mode
If the user specifies a file or glob:
codemoot review <file-or-glob> --focus all
If the user specifies a diff:
codemoot review --diff HEAD~3..HEAD
If the user gives a freeform description:
codemoot review --prompt "PROJECT CONTEXT: <context from step 1>
REVIEW TASK: <user's description>
Evaluate on: Correctness, Completeness, Quality, Security, Feasibility.
For security findings, verify by reading the actual code before flagging.
Provide SCORE: X/10 and VERDICT: APPROVED or NEEDS_REVISION"
For presets:
codemoot review <target> --preset security-audit
codemoot review <target> --preset quick-scan
codemoot review <target> --preset performance
Step 3: Parse and present the output
The command outputs JSON to stdout. Parse it and present as clean markdown:
## GPT Review Results
**Score**: X/10 | **Verdict**: APPROVED/NEEDS_REVISION
### Findings
- [CRITICAL] file:line — description
- [WARNING] file:line — description
### GPT's Full Analysis
<review text>
Session: <sessionId> | Tokens: <usage> | Duration: <durationMs>ms
Step 4: If NEEDS_REVISION
Ask if user wants to fix and re-review. If yes:
- Fix the issues
- Run
codemoot review again — session resume gives GPT context of prior review
- GPT will check if previous issues were addressed
Important Notes
- Session resume: Each review builds on prior context. GPT remembers what it reviewed before.
- Codebase access: GPT can read project files during review via codex tools.
- No arg size limits: Content is piped via stdin, not passed as CLI args.
- Presets: Use --preset for specialized reviews (security-audit, performance, quick-scan, pre-commit, api-review).
- Background mode: Add --background to enqueue and continue working.
- Output capped: JSON
review field is capped to 2KB to prevent terminal crashes. Full text is stored in session_events.
- Progress: Heartbeat every 60s, throttled to 30s intervals. No agent_message dumps.
