/cf-review-out

Context footprint: ⚡ (low) — what does this mean?

The /cf-review-out skill generates a complete, self-contained review document that any external AI agent (Gemini, Codex, ChatGPT, or even a human) can use to review your code changes. Pair with /cf-review-in to collect results.

Usage

/cf-review-out [label]

• No arguments — Claude auto-generates a label based on your changes (no confirmation needed)
• With label — uses the provided label directly (must be snake-case with prefix like fix-, feature-, improve-)

Examples

# Let Claude suggest a label
/cf-review-out

# Provide your own label
/cf-review-out fix-auth-bypass

# Feature review
/cf-review-out feature-cross-agent-review

Workflow

1. Determine label — Auto-generate a descriptive snake-case label or use the one provided (no confirmation step)
2. Gather diff — Collect all uncommitted changes via git diff
3. Build prompt — Create a self-contained review document with:
   • Context section — project name, branch, date, files/lines changed
   • Full diff of changes
   • 5-layer review criteria (project rules, plan alignment, code quality, security, testing)
   • Exact output format specification
   • Where the external agent should save results
   • Output checklist — self-verification checklist for the external reviewer
4. Save prompt — Write to docs/reviews/<label>-prompt.md
5. Guide user — Show next steps for running the external agent

What the Prompt Contains

The generated prompt is a complete, standalone document:

• Frontmatter — label, date, status (pending)
• Context — project name, branch, date, files changed, lines changed
• Review criteria — 5-layer methodology matching /cf-review
• Output format — exact markdown format the external agent must follow (Critical/Important/Suggestion)
• Save location — tells the agent to write results to docs/reviews/<label>-result-<service>.md
• Output checklist — self-verification checklist (all sections present, confidence scores, no false positives)
• Full diff — all code changes to review (truncated at 5000 lines with a note)
• Security boundary — diff marked as untrusted input

The Complete Flow

/cf-review-out fix-login     → generates docs/reviews/fix-login-prompt.md
(user runs any external agent on the prompt file)
external agent writes        → docs/reviews/fix-login-result-gemini.md
/cf-review-in fix-login      → reads results, presents findings, offers to fix

Any Agent, Any Time

The prompt file is a plain markdown document. You can use it with:

• Gemini CLI — gemini < docs/reviews/fix-login-prompt.md
• Codex CLI — codex -q --full-auto < docs/reviews/fix-login-prompt.md
• ChatGPT — copy-paste the prompt into a chat
• Human reviewer — read it and write findings in the specified format
• Any other AI — any tool that can read markdown

Configuration

Reviews are stored in {docsDir}/reviews/ (default: docs/reviews/). The docsDir is read from .coding-friend/config.json.

When to Use

• Second opinion — when you want cross-validation from a different AI model
• Before merging critical code — auth, payments, security-sensitive changes
• Async review — dispatch now, collect results later (even days later)
• Multiple reviewers — send the same prompt to multiple agents, collect all results