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.

Verify:

• All issues have file, line, issue, and suggestion
• Issues are correctly categorized by severity
• No invented issues
• CodeRabbit's exact wording preserved

Step 4: Return Results

Pass structured issue list to next skill (coderabbit-triage).

Example Output

{
  "critical": [
    {
      "file": "src/webhooks.ts",
      "line": 42,
      "issue": "Missing HMAC signature verification for webhook authenticity",
      "suggestion": "Add HMAC-SHA256 signature verification using crypto module. Compare against timestamp to prevent replay attacks."
    }
  ],
  "important": [
    {
      "file": "src/webhooks.ts",
      "line": 89,
      "issue": "Race condition in concurrent webhook handling. Multiple async calls can update state simultaneously.",
      "suggestion": "Implement idempotency key handling. Store processed keys in cache with TTL to deduplicate retries."
    },
    {
      "file": "src/webhooks.ts",
      "line": 156,
      "issue": "Missing error logging for webhook failures. Makes debugging production issues difficult.",
      "suggestion": "Add structured logging with timestamp, error type, and webhook ID. Include full error stack."
    }
  ],
  "minor": []
}

Key Principles

Objectivity: CodeRabbit is the source of truth. Don't filter or reinterpret findings.

Completeness: Return all issues CodeRabbit found, regardless of difficulty.

Clarity: Preserve CodeRabbit's exact issue descriptions and suggestions.

Structure: Always return JSON with three severity levels, even if some are empty arrays.

Common Patterns

When CodeRabbit finds nothing: Return empty arrays for all severity levels.

{
  "critical": [],
  "important": [],
  "minor": []
}

When issues span multiple files: Group by file but preserve line numbers.

When suggestion is vague: Include CodeRabbit's exact wording. Don't interpret or improve it.

Constraints

• DO: Use --prompt-only flag for efficiency
• DO: Let CodeRabbit run to completion
• DO: Preserve exact issue descriptions from CodeRabbit output
• DON'T: Filter out any findings
• DON'T: Add your own issues
• DON'T: Modify or improve CodeRabbit's suggestions

Integration Points

This skill feeds directly into coderabbit-triage:

coderabbit-request
        ↓ (outputs JSON)
coderabbit-triage
        ↓ (outputs task plan)
coderabbit-fix (parallel or sequential)

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: kellykampen
• Source: kellykampen/agent-skills
• License: MIT

Install and usage instructions live in the source repository linked above.