Codex Second Opinion Skill

This skill enables you to leverage OpenAI Codex CLI as a second opinion source for code analysis, architectural validation, and technical reviews.

When to Use This Skill

Invoke this skill when you need to:

• Get second opinion on architectural decisions or implementation approaches
• Multi-model validation - compare OpenAI vs Anthropic perspectives
• Code review from different AI model for better coverage
• Structured JSON output with schemas for predictable parsing
• Complex analysis that benefits from consensus of multiple AI models

Do NOT use for:

• Simple tasks that don't need validation
• Time-sensitive operations where single perspective is sufficient
• Tasks already completed and validated

How This Skill Works

When invoked, use codex exec via Bash tool with these patterns:

Pattern 1: Simple Question-Answer

codex exec --output-last-message /tmp/claude/codex-answer.txt "Your question"
cat /tmp/claude/codex-answer.txt

Pattern 2: Structured Analysis (Recommended)

# Create schema
cat > /tmp/claude/schema.json << 'EOF'
{
  "type": "object",
  "properties": {
    "summary": { "type": "string" },
    "strengths": { "type": "array", "items": { "type": "string" } },
    "weaknesses": { "type": "array", "items": { "type": "string" } },
    "recommendations": { "type": "array", "items": { "type": "string" } }
  },
  "required": ["summary", "strengths", "weaknesses"]
}
EOF

# Execute with schema
codex exec --output-schema /tmp/claude/schema.json \
  --output-last-message /tmp/claude/result.json \
  "Analyze [topic]. Provide structured assessment."

# Read result
cat /tmp/claude/result.json

Pattern 3: Comparative Analysis

# Get Codex perspective
codex exec --output-last-message /tmp/claude/codex-view.txt \
  "Review this approach: [your plan]. List pros, cons, alternatives."

# Present both perspectives
cat /tmp/claude/codex-view.txt

Common Use Cases

1. Architecture Review

cat > /tmp/claude/arch-schema.json << 'EOF'
{
  "type": "object",
  "properties": {
    "assessment": { "type": "string" },
    "risks": { "type": "array", "items": { "type": "string" } },
    "alternatives": { "type": "array", "items": { "type": "string" } },
    "risk_level": { "type": "string", "enum": ["low", "medium", "high"] }
  }
}
EOF

codex exec --output-schema /tmp/claude/arch-schema.json \
  --output-last-message /tmp/claude/arch-review.json \
  "Review MCP CLI bridge pattern. Assess security, performance, maintainability."

cat /tmp/claude/arch-review.json

2. Security Review

codex exec -m gpt-5-codex --output-last-message /tmp/claude/security.txt \
  "Security review of osiris/mcp/server.py:
   - Input validation
   - Secret handling
   - Filesystem access
   Provide specific vulnerabilities and fixes."

cat /tmp/claude/security.txt

3. Code Review

codex exec --output-last-message /tmp/claude/review.txt \
  "Review osiris/mcp/tools/discovery.py focusing on:
   1. Security vulnerabilities
   2. Performance issues
   3. Code maintainability
   Provide line-level recommendations."

cat /tmp/claude/review.txt

4. Validate ADR

codex exec --output-last-message /tmp/claude/adr-review.txt \
  "Review this ADR for completeness and issues: [ADR content or file reference]"

cat /tmp/claude/adr-review.txt

Key Parameters

• Model selection: -m gpt-5-codex (for complex tasks) or -m o4-mini (faster)
• Working directory: -C /path/to/analyze (defaults to current)
• Sandbox mode: --sandbox read-only (default, safe)
• Output: --output-last-message /tmp/claude/file.txt (cleanest for text)

Best Practices

1. Always use /tmp/claude/ for outputs - respects filesystem contract
2. Prefer JSON schemas for structured, parseable responses
3. Be specific in prompts - mention file paths, exact concerns, context
4. Compare perspectives - present both Codex and your analysis
5. Use for validation - Codex complements, doesn't replace your work
6. Check authentication - ensure codex --version works before use

Output Interpretation

When presenting Codex results to user:

1. Label clearly - "Codex perspective" or "Second opinion from OpenAI"
2. Compare with your own analysis
3. Synthesize insights from both models
4. Highlight agreement and disagreement
5. Recommend based on multi-model consensus

Error Handling

Always verify Codex is available:

if ! command -v codex &> /dev/null; then
    echo "Codex CLI not found. User needs to install Codex."
    exit 1
fi

If authentication fails, inform user to run:

codex login  # ChatGPT login
# OR
printenv OPENAI_API_KEY | codex login --with-api-key

Limitations

• Codex uses OpenAI models (GPT-5, O4), not Claude
• Requires internet connection
• Different context limits than Claude
• May have different coding style/perspective

Quick Reference

# Simple question
codex exec --output-last-message /tmp/claude/out.txt "analyze X"

# Structured output
codex exec --output-schema schema.json -o /tmp/claude/result.json "analyze X"

# Different model
codex exec -m gpt-5-codex --output-last-message /tmp/claude/out.txt "complex task"

# With image
codex exec -i screenshot.png --output-last-message /tmp/claude/out.txt "explain this"

---

See reference.md for comprehensive Codex CLI documentation and advanced usage patterns.