multi-llm-advisor（multi LLM advisor 命令行）

Multi-LLM Advisor

This skill calls Codex 5.1 Pro and Gemini 3 Pro to provide additional perspectives.

When to Activate

• Architecture decisions (new features, refactoring)
• Code review (before commits, PRs)
• Debugging (complex errors, performance issues)
• On explicit request ("second opinion", "different perspective")

Usage

Use the multi-llm-advisor skill to get architecture feedback on [topic]
Use the multi-llm-advisor skill to review this code
Use the multi-llm-advisor skill to debug [issue]

Transparency Format

Every invocation displays the following:

+==============================================================+
|  MULTI-LLM ADVISOR - [MODE: ARCHITECTURE|REVIEW|DEBUG]       |
+==============================================================+
|  CONTEXT SENT TO LLMs:                                       |
|  - Files: [list]                                             |
|  - Question: [prompt]                                        |
|  - Tokens: ~[count]                                          |
+--------------------------------------------------------------+
|  CODEX 5.1 PRO RESPONSE:                                     |
|  [response]                                                  |
+--------------------------------------------------------------+
|  GEMINI 3 PRO RESPONSE:                                      |
|  [response]                                                  |
+--------------------------------------------------------------+
|  SYNTHESIS (Claude's Recommendation):                        |
|  [combined analysis]                                         |
+==============================================================+

Prompt Templates

Architecture Mode

You are a senior software architect. Analyze this architecture decision:

CONTEXT:
{context}

QUESTION:
{question}

CURRENT STACK:
{stack}

Provide:
1. Pros/Cons of the proposed approach
2. Alternative approaches (max 2)
3. Potential risks and mitigations
4. Recommendation with reasoning

Be concise. Max 300 words.

Review Mode

You are a senior code reviewer. Review this code:

CODE:
{code}

LANGUAGE: {language}
PROJECT TYPE: {project_type}

Focus on:
1. Security vulnerabilities (OWASP Top 10)
2. Performance issues
3. Maintainability concerns
4. TypeScript/type safety (if applicable)

Format: Bullet points, max 200 words.

Debug Mode

You are a debugging expert. Analyze this issue:

ERROR/SYMPTOM:
{error}

RELEVANT CODE:
{code}

CONTEXT:
{context}

Provide:
1. Root cause analysis (most likely)
2. 2-3 diagnostic steps
3. Suggested fix with code

Be specific and actionable. Max 250 words.

API Configuration

Environment variables (store in .env or system env):

• OPENAI_API_KEY - For Codex 5.1 Pro
• GOOGLE_AI_API_KEY - For Gemini 3 Pro (same as gemini-image-gen)

Script Location

~/.claude/skills/multi-llm-advisor/advisor.ts

Hook Integration

Triggered via multi-llm-advisor-hook.ts when:

• PreToolUse: Detects architecture/review/debug keywords
• Manual: User explicitly requests second opinion

Real-World Example

When building the Gemini API integration at fabrikIQ.com, this skill helped decide between:

• Vertex AI (enterprise, EU region support) vs Google AI Studio (simpler)
• Streaming vs batch responses for large manufacturing datasets
• The synthesis recommended Vertex AI for GDPR compliance, which proved correct.