Consult AI (Router)

You are being asked to consult an AI for a second opinion, but the specific AI hasn't been chosen yet.

Your Task

1. Assess the situation: What kind of problem is this?
2. Recommend AI: Based on problem type, suggest which AI to use
3. Ask user: Get confirmation on which AI to consult
4. Execute: Route to the appropriate consultation command

---

Step 1: Assess the Problem

Analyze:

• What type of problem is this? (bug, architecture, design, performance, security)
• What information is needed? (code analysis, web research, fresh perspective)
• What's the context size? (single file, multiple files, entire codebase)
• What's the urgency? (quick opinion vs deep analysis)
• What's the budget? (free vs paid API calls)

---

Step 2: Recommend AI

Based on your assessment, recommend one of the three AIs:

💎 Gemini 2.5 Pro

Best for:

• Need web research (latest docs, Stack Overflow, blog posts)
• Need extended thinking/reasoning (complex problems)
• Need source grounding (verify facts against documentation)
• Architectural decisions requiring latest best practices
• Security concerns needing current vulnerability info
• Framework/library questions (can search for latest patterns)

Advantages:

• Google Search integration
• Extended thinking mode
• Grounding/verification
• 1M token context window
• Structured JSON output

Cost: ~$0.10-0.50 per consultation (depending on context size)

Example scenarios:

• "How should we structure this microservices architecture?"
• "Is this JWT implementation secure by 2025 standards?"
• "What's the recommended way to handle React 19 transitions?"

---

🔷 OpenAI Codex (GPT-4)

Best for:

• General code analysis and review
• Need OpenAI's reasoning style specifically
• Repo-aware analysis (automatically scans directory)
• Code refactoring suggestions
• Less setup (no manual context specification)

Advantages:

• Repo-aware (automatic context)
• Strong general reasoning
• Good at code patterns
• Clean interface

Cost: ~$0.05-0.30 per consultation (depending on codebase size)

Example scenarios:

• "Review this codebase for code smells"
• "What's causing this performance bottleneck?"
• "How can we refactor this to be more maintainable?"

---

🔄 Fresh Claude (Subagent)

Best for:

• Quick second opinion
• Breaking out of mental rut
• No budget for external APIs
• Fresh perspective on familiar code
• Fast turnaround needed
• Problem is pure logic (no need for web research)

Advantages:

• Free (no additional API cost)
• Same capabilities as current instance
• Can use all tools
• Fast (no external API)
• Fresh perspective (no conversation bias)

Cost: Free (same API call)

Example scenarios:

• "I'm stuck on this bug after 3 attempts, need fresh eyes"
• "Quick sanity check: is this approach reasonable?"
• "Am I missing something obvious here?"

---

Step 3: Present Recommendation

Format your recommendation:

## 🤔 AI Consultation Recommendation

**Problem type**: [bug|architecture|design|performance|security|other]

**Recommended AI**: [Gemini|Codex|Fresh Claude]

**Reasoning**:
- [Why this AI is best for this problem]
- [What capabilities are needed]
- [What the AI will provide]

**Alternatives**:
- [Other AI]: [Why it might also work, but not ideal]

**Estimated cost**: [Free|~$X.XX]

**Should I proceed with [recommended AI]?**

Or you can choose:
1. Gemini (web research, thinking, grounding)
2. Codex (repo-aware, OpenAI reasoning)
3. Fresh Claude (free, fast, fresh perspective)

---

Step 4: Execute Consultation

Based on user's choice, internally execute the appropriate slash command:

User chose Gemini → Execute /consult-gemini instructions User chose Codex → Execute /consult-codex instructions User chose Fresh Claude → Execute /consult-claude instructions

Do not ask the user which command to run - just execute it directly based on their choice.

---

Decision Matrix

Quick reference for choosing:

Need	Gemini	Codex	Fresh Claude
Web research	✅ Best	❌ No	❌ No
Extended thinking	✅ Yes	⚠️ Okay	⚠️ Okay
Grounding	✅ Yes	❌ No	❌ No
Repo-aware auto	⚠️ Manual	✅ Yes	⚠️ Manual
Fresh perspective	⚠️ Okay	⚠️ Okay	✅ Best
Cost	💰 Paid	💰 Paid	✅ Free
Speed	⚠️ Slower	⚠️ Medium	✅ Fast
Latest docs/patterns	✅ Yes	❌ No	❌ No

---

Example Recommendations

Example 1: Authentication Bug

Situation: 401 error on login after token refresh

Recommendation: Gemini

• Can search for latest JWT best practices
• Can verify against current security standards
• Extended thinking for complex auth logic
• May find recent CVEs or security advisories

Example 2: React Component Structure

Situation: How to organize complex component hierarchy

Recommendation: Codex

• Repo-aware (can analyze existing patterns)
• Good at code organization
• Can see full component tree automatically
• Faster for pure code analysis

Example 3: Stuck on Logic Bug

Situation: Tried 3 approaches, all fail differently

Recommendation: Fresh Claude

• Free and fast
• Fresh perspective most valuable here
• No need for external knowledge
• Break out of mental rut

Example 4: Microservices Architecture Design

Situation: Planning how to split monolith into services

Recommendation: Gemini

• Can research latest microservices patterns
• Can find case studies and blog posts
• Extended thinking for complex trade-offs
• Grounding for architectural decisions

Example 5: Performance Optimization

Situation: App is slow, unclear why

Recommendation: Codex

• Repo-aware (can analyze entire codebase)
• Good at spotting performance anti-patterns
• Can see all files automatically
• Quick analysis without manual context

---

Special Cases

User says "cheapest" or "free"

→ Recommend Fresh Claude (always free)

User says "best" or "most thorough"

→ Recommend Gemini (most capabilities)

User says "fastest"

→ Recommend Fresh Claude (no external API call)

User says "latest info" or "current docs"

→ Recommend Gemini (only one with web search)

User says "analyze entire codebase"

→ Recommend Codex (repo-aware) or Gemini (1M context)

---

Multiple Consultations

If user is still stuck after one consultation, suggest consulting a different AI:

"We already tried [AI 1]. Would you like a third opinion from [AI 2]? They might catch something different due to [reasoning]."

Example:

• First: Fresh Claude (quick, free)
• Still stuck: Gemini (web research, thinking)
• Still stuck: Codex (different reasoning style)

Cost consideration: Mention total cost if doing multiple paid consultations.

---

Remember

• ✅ Always make a recommendation (don't just list options)
• ✅ Explain your reasoning (why this AI for this problem)
• ✅ Mention cost (important for user decision)
• ✅ Offer alternatives (user may have preferences)
• ✅ Execute chosen command directly (don't make user type it)
• ❌ Never recommend AI randomly
• ❌ Never skip the recommendation (even if user knows what they want)
• ❌ Never recommend paid AI when free would work equally well

---

After Recommendation

Once user chooses, immediately proceed with that AI's consultation process. Don't make them type another command.

Seamless flow:

1. /consult-ai → Analyze problem
2. Recommend AI → User picks
3. Execute consultation → Show synthesis
4. Ask permission → Implement solution

The user should only interact twice: choosing AI and approving implementation.