Plan Review

/plan-review — GPT Review of Execution Plans

Usage

/plan-review <plan-file-or-description>

Description

Sends a Claude-authored execution plan to GPT via codemoot plan review for structured critique. GPT returns ISSUE (HIGH/MEDIUM/LOW) and SUGGEST lines with file-level references. Uses session resume for iterative refinement.

Instructions

When the user invokes /plan-review, follow these steps:

Step 1: Prepare the plan

If the user provides a file path: Use it directly.

If the user describes what to review:

1. Write the plan to a temp file (e.g., /tmp/plan-review.md)
2. Use that path

If reviewing the current conversation's plan:

1. Extract the plan from context
2. Write to a temp file

Step 2: Run plan review

cd C:\Users\ramka\Desktop\cowork\codemoot\packages\cli && npx codemoot plan review <plan-file> [options]

Options:

• --phase <N> — Review only a specific phase
• --build <buildId> — Link to a build run
• --timeout <ms> — Override default timeout (default: 300000)
• --output <file> — Save full review to file
• - — Read plan from stdin

Step 3: Parse and present output

The command outputs JSON:

{
  "planFile": "path",
  "verdict": "needs_revision",
  "score": 5,
  "issues": [
    { "severity": "high", "message": "..." },
    { "severity": "medium", "message": "..." }
  ],
  "suggestions": ["..."],
  "review": "truncated to 2KB...",
  "sessionId": "...",
  "usage": { "inputTokens": ..., "outputTokens": ..., "totalTokens": ... }
}

Present as:

## GPT Plan Review

**Score**: X/10 | **Verdict**: APPROVED/NEEDS_REVISION

### Issues
- [HIGH] description with file references
- [MEDIUM] description

### Suggestions
- suggestion text

Session: <sessionId> | Tokens: <totalTokens> | Duration: <durationMs>ms

Step 4: If NEEDS_REVISION

Ask the user if they want to revise and re-submit:

1. Address the HIGH issues first
2. Revise the plan
3. Write updated plan to the same file
4. Re-run codemoot plan review — session resume gives GPT context of prior review

Important Notes

• Session resume: GPT remembers prior reviews via thread resume
• Codebase access: GPT reads actual project files to verify plan references
• Output capped: JSON review field is capped to 2KB. Use --output for full text
• Score-based convergence: Plans scoring >= 8/10 on 2nd+ iteration auto-converge in plan generate
• Zero API cost: Uses ChatGPT subscription via Codex CLI

Converted and distributed by TomeVault — claim your Tome and manage your conversions.