name: llm-external-review description: Invoke EXTERNAL LLM models (Codex, Gemini) for reviews. This skill calls external CLIs - Claude must NOT do the review itself. Use for second opinions, code review, architecture review, brainstorming.

LLM External Review

⚠️ CRITICAL: This skill invokes EXTERNAL AI models via CLI tools.

You MUST run the review.py script. Do NOT perform this review yourself. The entire purpose is to get a perspective from a DIFFERENT AI (Codex/Gemini). If you do the review yourself, you have FAILED to execute this skill correctly.

What This Skill Does

1. Prepares context about the task
2. Calls review.py which invokes Codex CLI or Gemini CLI
3. Returns the EXTERNAL model's response
4. You synthesize and present their feedback

Required Execution Steps

Step 1: Write context to temp file

cat > /tmp/external_review_context.md << 'EOF'
# Context
{Brief description of goal, constraints, current state}

# Task
{What to review: code, architecture, plan, etc.}

# Files
{List key files the external model should examine}
EOF

Step 2: Run the external review script (MANDATORY)

python3 ${CLAUDE_PLUGIN_ROOT}/scripts/review.py \
  --model codex \
  --task review \
  --context /tmp/external_review_context.md \
  --workdir .

Step 3: Present the external model's response

Parse the JSON output and present findings to the user.

Model Selection

Task Types

CLI Options

Example: Architecture Review

# 1. Prepare context
cat > /tmp/external_review_context.md << 'EOF'
# Context
Reviewing the architecture of a Python ML pipeline project.

# Task
Review the overall architecture for scalability and maintainability.

# Files
- src/fund_rating/ - main package
- docs/architecture.md - current architecture docs
EOF

# 2. Call external model (REQUIRED)
python3 ${CLAUDE_PLUGIN_ROOT}/scripts/review.py \
  --model gemini \
  --task plan \
  --context /tmp/external_review_context.md \
  --workdir .

Response Format

The script returns JSON:

{
  "model": "codex",
  "task_type": "review",
  "status": "success",
  "response": {
    "summary": "One-line finding",
    "key_points": ["..."],
    "concerns": ["..."],
    "suggestions": ["..."]
  }
}

Error Handling

• CLI not found: User needs to install:
  • Codex: npm i -g @openai/codex
  • Gemini: https://github.com/google-gemini/gemini-cli
• Auth error: Run codex login or gemini /auth
• Timeout: Increase with --timeout 300

Remember

⚠️ You are Claude. This skill exists to get opinions from OTHER AI models.

CORRECT: Run review.py → Get Codex/Gemini response → Present to user

WRONG: Read the codebase yourself and provide your own review

If Codex/Gemini CLIs are not installed, tell the user how to install them. Do NOT fall back to doing the review yourself.