architect

<user_request> $ARGUMENTS </user_request>

This triage determines the right level of process for the architecture task.

**Step 1: Detect retry attempts**

Check if this is a second (or later) attempt at the same architecture problem:
```bash
# Look for existing architecture sessions with similar topic
TOPIC_SLUG=$(echo "$ARGUMENTS" | tr '[:upper:] ' '[:lower:]-' | sed 's/[^a-z0-9-]//g' | head -c30)
ls -d ai-docs/sessions/dev-arch-*${TOPIC_SLUG}* 2>/dev/null || echo "NO_PREVIOUS_SESSIONS"
```

Also check for retry signals in the user's request:
- "again", "retry", "different approach", "rethink", "reconsider"
- "didn't work", "failed", "not happy with", "let's try"
- References to a previous design or session

**Step 2: Assess complexity signals**

| Signal | Weight | Example |
|--------|--------|---------|
| Multiple systems/services | HIGH | "design auth across API + frontend + mobile" |
| Novel domain (no existing patterns) | HIGH | "design real-time collaboration engine" |
| Cross-cutting concerns | HIGH | "security + performance + multi-tenancy" |
| Conflicting requirements | HIGH | "fast AND highly consistent AND cheap" |
| Large blast radius | MEDIUM | "redesign the entire data layer" |
| Multiple viable approaches | MEDIUM | unclear best path forward |
| Single component, clear pattern | LOW | "add caching to the user service" |
| Well-understood domain | LOW | "standard CRUD API for products" |

Score: HIGH = 2 points, MEDIUM = 1, LOW = 0; sum every signal present.
0 = Simple, 1-2 = Moderate, 3+ = Complex. So one MEDIUM or one HIGH is Moderate,
and one HIGH plus one MEDIUM, or two HIGH, is Complex.

**Step 3: Route based on assessment**

| Assessment | Is Retry? | Route |
|------------|-----------|-------|
| Simple | No | Direct architecture design (Phase 0-6) |
| Simple | Yes | Offer /team — "Previous attempt exists. Want multi-model perspectives?" |
| Moderate | No | Plan mode reasoning first, then architecture design |
| Moderate | Yes | Recommend /team — "This needs fresh perspectives" |
| Complex | No | Plan mode reasoning + offer /team before architecture |
| Complex | Yes | Strongly recommend /team — "Multi-model brainstorming before redesign" |

**When /team is recommended, present the option:**

```yaml
AskUserQuestion:
  questions:
    - question: "This looks like a {complex problem | second attempt}. How should we approach it?"
      header: "Architecture Approach"
      multiSelect: false
      options:
        - label: "Multi-model brainstorm (/team)"
          description: "Get diverse perspectives from multiple AI models before designing. Best for complex or retry scenarios."
        - label: "Plan mode reasoning"
          description: "Use Claude's extended reasoning to think through the design space, then proceed with single-model architecture."
        - label: "Direct architecture"
          description: "Skip to architecture design. Good when you already know the general direction."
```