multi

# Multi - Multi-Provider Override

**Forces multi-provider execution for any task using all available AI providers.**

## 🤖 INSTRUCTIONS FOR CLAUDE

### EXECUTION MECHANISM — NON-NEGOTIABLE

**You MUST dispatch work to external providers (Codex, Gemini, etc.) for this command. You are PROHIBITED from:**
- ❌ Executing the entire task using only Claude-native tools
- ❌ Using a single Agent subagent instead of multi-provider dispatch
- ❌ Skipping provider dispatch because "I can handle this alone"

**Multi-LLM orchestration is the purpose of this command.** Single-model execution defeats its purpose.

When the user invokes this command (e.g., `/octo:multi <task>`):

### Step 1: Cost Awareness & Intent Confirmation

**CRITICAL: Before forcing multi-provider execution, use AskUserQuestion to confirm intent and cost awareness:**

```javascript
AskUserQuestion({
  questions: [
    {
      question: "Why do you need multiple AI perspectives?",
      header: "Intent",
      multiSelect: false,
      options: [
        {label: "High-stakes decision", description: "Critical choice requiring comprehensive analysis"},
        {label: "Quality validation", description: "Cross-check important work for accuracy"},
        {label: "Learning different approaches", description: "See how different models think"},
        {label: "Comparing perspectives", description: "Want to see model-specific insights"},
        {label: "Just exploring", description: "Curious about multi-AI capabilities"}
      ]
    },
    {
      question: "Are you aware this uses external API credits?",
      header: "Cost",
      multiSelect: false,
      options: [
        {label: "Yes, proceed (~$0.02-0.08/query)", description: "I understand this costs money via Codex + Gemini APIs"},
        {label: "Tell me more about costs", description: "Explain what I'll be charged"},
        {label: "Use free providers only", description: "Skip Codex/Gemini, use Claude only"}
      ]
    }
  ]
})
```

**WAIT for the user's answers before proceeding.**

**After receiving answers:**

- **If user selected "Tell me more about costs"**: Explain the cost breakdown, then ask question 2 again
- **If user selected "Use free providers only"**: Explain that multi requires external providers, suggest using Claude directly instead
- **If user selected "Yes, proceed"**: Continue with multi-provider execution
- **Store intent** in context for provider selection (high-stakes → use all 3, exploring → maybe skip one if unavailable)

### Step 2: Check Provider Availability & Execute

Check which AI providers are available:

```javascript
const codexAvailable = await checkCommandAvailable('codex');
const geminiAvailable = await checkCommandAvailable('gemini');

if (!codexAvailable && !geminiAvailable) {
  console.log("⚠️ No external providers available. Multi-provider mode requires Codex and/or Gemini.");
  console.log("Run `/octo:setup` to configure external providers, or use Claude directly.");
  return;
}

// Proceed with available providers
```

Execute the task with all available providers, incorporating user intent from Step 1.

### Step 3: Adversarial Synthesis (MANDATORY — do NOT skip)

**After collecting responses from all providers, you MUST synthesize — not concatenate.**

The whole point of multi-provider execution is diversity of opinion. Presenting provider responses back-to-back without synthesis defeats this purpose. You MUST produce a structured synthesis with these sections:

1. **Consensus** — Where providers agree. State the shared conclusion and note how many providers converged on it.
2. **Divergence** — Where providers disagree. Quote each provider's position verbatim so the user sees the actual disagreement, not a smoothed-over summary.
3. **Resolution** — Your confidence-weighted recommendation resolving the conflicts. State your confidence level (high/medium/low) and WHY you side with one position over another. If you cannot resolve, say so — an honest "this depends on your context" is better than a false consensus.
4. **Minority Opinions** — Dissenting views that may be valuable even if outnumbered. A single provider flagging a security risk or edge case that others missed is often the most valuable insight.

**Present this synthesis in the chat using this format:**

```markdown
## Multi-Provider Synthesis

### Consensus
[Where all providers agree]

### Divergence
- 🔴 **Codex**: "[direct quote or close paraphrase]"
- 🟡 **Gemini**: "[direct quote or close paraphrase]"
- 🔵 **Claude**: "[direct quote or close paraphrase]"

### Resolution
[Your recommendation with confidence level and reasoning]

### Minority Opinions
[Dissenting views worth preserving — or "None: all providers converged"]
```

**Skip this step with `--fast` or when user explicitly requests speed over thoroughness.** In fast mode, present raw provider responses without synthesis.

---

## Quick Usage

Just use natural language:
```
"Run this with all providers: What is Redis?"
"I want all three AI models to look at this architecture"
"Get multiple perspectives on whether to use TypeScript"
"Force multi-provider analysis of this design decision"
```

Or use explicit commands:
```
/octo:multi "Explain how OAuth works"
/octo:multi "Review this simple function"
/octo:multi "What is JWT?"
```

## How It Works

This command activates the multi-provider skill in **forced mode**, which:
- Executes multi-provider analysis even for simple tasks
- Uses Codex CLI + Gemini CLI + Claude simultaneously
- Provides multiple perspectives when you need comprehensive analysis
- Bypasses automatic routing that might use only Claude

## What This Does

Normal Claude Octopus workflows automatically decide when to use multiple providers:
- "octo research OAuth" → automatically triggers multi-provider (probe workflow)
- "What is OAuth?" → uses Claude only (simple question)

**The multi command forces multi-provider mode even for simple tasks:**
- `/octo:multi "What is OAuth?"` → forces Codex +