claude-review

Claude Review

Delegate a code review to Claude Code and get back structured findings.

When to Use

• After writing or modifying code and wanting a second opinion
• When asked to "get Claude to review this"
• When a review requires deep context that would benefit from a fresh Claude session

Call Pattern

Step 1: Start a review

Call claude_code with a structured review prompt:

mcp__claude-code__claude_code:
  prompt: |
    Review the following code for correctness, security, and quality.

    SCOPE: {files or description of what to review}

    Evaluate:
    1. Correctness — logic errors, edge cases, off-by-one, race conditions
    2. Security — injection, auth bypass, data exposure, input validation
    3. Quality — readability, maintainability, test coverage, naming
    4. Architecture — coupling, abstraction leaks, layer violations

    For each finding report:
    - File:line
    - Severity: Critical / High / Medium / Low
    - Category
    - Issue description
    - Suggested fix

    PROVENANCE NOTE: The code being reviewed was written by OpenAI Codex. Evaluate
    with full rigor — do not defer to it. Apply independent judgment on every finding.
  cwd: {project working directory}
  effort: high
  permissionMode: plan

Save the returned session_id as {review_session_id}.

Step 2: Follow up (optional)

To ask about specific findings or request deeper analysis on a section:

mcp__claude-code__claude_code_reply:
  session_id: {review_session_id}
  prompt: "Expand on finding #3 — what's the exact attack vector and what's the minimal fix?"

Output Format

Display the review findings in a structured table:

File:Line	Severity	Category	Issue	Fix
...	Critical	Security	...	...

Follow with a summary count by severity and a recommended action (fix now vs fix later vs acceptable risk).

Notes

• permissionMode: plan prevents Claude from modifying files — review is read-only
• Set effort: high for thorough analysis; use medium for quick spot-checks
• Pass cwd to anchor relative file paths in findings
• If the project has a test suite, mention it in the prompt so Claude can assess coverage gaps