Task Router

Intelligent routing skill that analyzes user requests and determines optimal agent dispatch.

Purpose

Reduce orchestrator cognitive load by automatically:

1. Inferring user intent (via /infer-intent)
2. Classifying task type and complexity
3. Identifying required agent(s)
4. Detecting parallelization opportunities
5. Recommending model tier per agent

Integrated Workflow

User Request
     │
     ▼
┌─────────────────┐
│  /infer-intent  │ ← Step 0: Understand what user wants
└────────┬────────┘
         │ Intent Summary (YAML)
         ▼
┌─────────────────┐
│  /task-router   │ ← This skill: Route to agents
└────────┬────────┘
         │ Dispatch Plan
         ▼
    Execute Plan

When to Use

• At the start of any non-trivial task
• When uncertain which agent(s) to spawn
• Before spawning multiple agents (to optimize parallelization)
• Automatically invoked by /infer-intent for complex tasks

Routing Process

Step 1: Task Classification

Classify the request into categories:

Step 2: Complexity Assessment

Score task complexity (1-10):

Complexity Levels:

• 1-3: Simple (may not need agent)
• 4-6: Moderate (single agent)
• 7-9: Complex (multiple agents, sequential)
• 10+: Critical (multiple agents, Doc Agent required)

Step 3: Parallelization Detection

Tasks can run in parallel when:

• No data dependencies between tasks
• Different domains (frontend vs backend)
• Independent bug fixes
• Research across different features

Tasks must be sequential when:

• API contract needed before frontend
• Schema needed before server actions
• Tests depend on implementation
• Security review before deployment

Step 4: Model Recommendation

Output Format

## Task Routing Analysis

### Classification
- **Primary Type**: [category]
- **Secondary Types**: [if multi-domain]
- **Complexity Score**: [N]/10 ([level])

### Agent Dispatch Plan

| Order | Agent | Model | Task |
|-------|-------|-------|------|
| 1 | [agent] | [model] | [specific task] |
| 2 | [agent] | [model] | [specific task] |

### Parallelization
- **Can Parallelize**: [Yes/No]
- **Parallel Groups**: [if yes, which agents can run together]
- **Dependencies**: [what must complete before what]

### Recommendation
[1-2 sentence summary of recommended approach]

Quick Routing Table

Integration

After routing, spawn agents using:

• /spawn-frontend-agent
• /spawn-backend-agent
• /spawn-research-agent
• /spawn-doc-agent
• /spawn-test-agent
• /spawn-architect-agent
• /spawn-security-agent
• /spawn-ops-agent

For parallel dispatch, use /parallel-agents with the routing analysis.

Intent-Aware Routing

When /infer-intent provides an intent summary, use it to enhance routing:

Using Intent Summary

If intent summary is available:

# From /infer-intent
intent:
  primary: create
  secondary: [modify]
scope:
  level: cross-cutting
  domains: [frontend, backend]
complexity:
  score: 7
agents:
  primary: architect-agent
  supporting: [backend-agent, frontend-agent]
  parallel_possible: false
routing:
  recommended_flow: 2-gate

Map directly to dispatch plan:

### Agent Dispatch Plan

| Order | Agent | Model | Task |
|-------|-------|-------|------|
| 1 | architect-agent | opus | Design system architecture |
| 2 | backend-agent | opus | Implement server-side |
| 2 | frontend-agent | sonnet | Implement UI (parallel with backend if disjoint) |
| 3 | test-agent | sonnet | Verify complete flow |

Without Intent Summary

If no intent summary provided, run the full routing process (Steps 1-4 above).

Quick Reference

See .claude/agents/AGENT_INVENTORY.md for:

• Complete agent profiles and capabilities
• Model selection guide
• Parallel agent patterns
• Decision tree for agent selection
• Cost optimization strategies

Related Skills