Skill Architect

Analyze skill requirements and design subagents that make skills efficient and repeatable.

When to Use Subagents

Subagents excel at:

• Parallel exploration - Multiple searches/reads at once
• Isolated analysis - Focused work without context accumulation
• Repeatable operations - Same task structure, different inputs
• Research gathering - Collect info, return only findings

Analysis Process

1. Gather Skill Requirements

Identify the skill's:

• Primary workflow steps
• Input variations
• Output requirements
• External dependencies (APIs, files, tools)

2. Map Work Patterns

For each workflow step, classify:

3. Define Agents

For each identified agent, specify:

### Agent: <name>

**Purpose:** Single sentence describing what it accomplishes

**Inputs:**
- <input 1>
- <input 2>

**Returns:**
- <structured output description>

**Subagent type:** Explore | general-purpose | Bash | Plan

**Parallelizable:** Yes/No - can multiple instances run concurrently?

4. Design Agent Coordination

Map how agents interact:

Main skill flow:
1. Spawn research agents (parallel)
2. Collect results
3. Spawn processing agents (parallel)
4. Synthesize outputs
5. Present to user

Output Format

Produce an agent design document:

# Agent Design: <skill-name>

## Identified Agents

### 1. <agent-name>
[agent spec from step 3]

### 2. <agent-name>
[agent spec from step 3]

## Coordination Flow
[flow from step 4]

## Implementation Notes
- <any special considerations>

Creating the Skill

Once agent design is approved, invoke skill-creator:

"Create a skill called with the agents defined above"

The skill-creator will incorporate the agent definitions into the new skill's SKILL.md.

See references/agent-patterns.md for common agent patterns and anti-patterns.