Agents - Agent Composition System

Overview

The Agents skill manages agent composition and orchestration:

• Compose dynamic agents from traits (expertise + personality + approach)
• Spawn parallel agents for divide-and-conquer work
• Select appropriate models for task complexity

Agent Types

Type	Definition	Best For
Named Agents	Pre-configured specialists (Engineer, Architect, QATester)	Recurring work, specialized expertise
Dynamic Agents	Task-specific agents composed from traits	One-off tasks, novel combinations

Using Named Agents

Task({
  subagent_type: "Engineer",
  prompt: "Implement the user authentication feature...",
  model: "sonnet"
})

Available Agents:

• Engineer - TDD implementation, battle-tested patterns
• Architect - System design, trade-off analysis
• QATester - Browser validation, quality gates
• backend-agent - API design, services
• frontend-agent - UI components, state
• security-agent - AppSec, OWASP Top 10
• data-agent - Database design, migrations
• sre-agent - Infrastructure, reliability

Composing Dynamic Agents

When you need a specific expertise combination:

Traits Available:

Expertise (domain knowledge):

• security, legal, finance, medical, technical
• research, creative, business, data, communications

Personality (behavior style):

• skeptical, enthusiastic, cautious, bold, analytical
• creative, empathetic, contrarian, pragmatic, meticulous

Approach (work style):

• thorough, rapid, systematic, exploratory
• comparative, synthesizing, adversarial, consultative

Example Compositions:

Need	Traits
Security architecture review	security, skeptical, thorough, adversarial
Legal contract review	legal, cautious, meticulous, systematic
Creative content development	creative, enthusiastic, exploratory
Red team critique	contrarian, skeptical, adversarial, bold
Quick business assessment	business, pragmatic, rapid, comparative

Model Selection (CRITICAL)

Task Type	Model	Speed
Grunt work, verification	haiku	10-20x faster
Standard implementation	sonnet	Balanced
Deep reasoning, architecture	opus	Maximum intelligence

Rule: Parallel agents especially benefit from haiku for speed.

Parallel Execution

Spawn multiple agents in ONE message for parallel work:

// All run simultaneously
Task({ subagent_type: "Intern", prompt: "Research company A...", model: "haiku" })
Task({ subagent_type: "Intern", prompt: "Research company B...", model: "haiku" })
Task({ subagent_type: "Intern", prompt: "Research company C...", model: "haiku" })

Spotcheck Pattern: Always launch a spotcheck agent after parallel work to verify consistency.

Task Prompt Requirements

When delegating, ALWAYS include:

1. WHY - Business context, why this matters
2. WHAT - Current state, existing implementation
3. EXACTLY - Precise actions, file paths, patterns
4. SUCCESS CRITERIA - What good output looks like

Task({
  subagent_type: "Engineer",
  prompt: `
    ## Context
    We're adding dark mode. Users have requested this feature.

    ## Current State
    Theme is in src/theme/, uses CSS variables.

    ## Task
    1. Add dark mode toggle to settings page
    2. Persist preference to localStorage
    3. Apply theme on page load

    ## Success Criteria
    - Toggle works in browser
    - Theme persists across refresh
    - No flash of wrong theme on load
  `,
  model: "sonnet"
})