coderabbit-request

Requesting Review

Overview

This skill dispatches a CodeRabbit reviewer subagent to analyze your uncommitted changes and return a structured list of issues. It's the first step in the change → review → triage → fix pipeline.

When to use: After making ANY file changes (code, skills, documentation, config), before considering work complete.

CRITICAL: Always Request Review

After completing changes, invoke this skill:

• ✅ Code files (.ts, .vue, .js, etc.) - strongest support (AST-aware linting, security, auto-tests)
• ✅ Test files - strongest support
• ✅ Skill files (SKILL.md) - lighter support (markdownlint)
• ✅ Documentation (.md files) - lighter support (markdownlint)
• ✅ Configuration files - lighter support (generic config linting)

Do NOT skip review because "it's just documentation" or "it's a small change". Even lighter analysis catches issues.

Outputs: Structured JSON with categorized issues (Critical/Important/Minor).

Process

Step 1: Prepare Context

Stage all changes before review. The --type uncommitted flag analyzes both staged and unstaged changes, but staging ensures nothing is missed:

# Stage all changes
git add -A

# Verify staged changes
git diff --cached --stat

Document in your mind what was just implemented. This context helps the reviewer subagent understand scope.

Step 2: Run CodeRabbit via gob

CRITICAL: Always use gob add for CodeRabbit commands - they take 1-3 minutes.

# Start CodeRabbit in background
gob add coderabbit --prompt-only --type uncommitted

# Continue working or wait for results
gob await-any

The --prompt-only flag makes output succinct and token-efficient.

Step 3: Parse Results

When CodeRabbit completes, parse the output into this JSON structure:

{
  "critical": [
    {"file": "...", "line": NNN, "issue": "...", "suggestion": "..."}
  ],
  "important": [...],
  "minor": [...]
}

Include only genuine issues CodeRabbit found. Do not invent.