compose-agent

Compose Agent — Assembly and Wiring

Take existing components (skills, workflows, SOUL.md, backend config) and wire them into a complete, runnable agent.

When to Use

• User has standalone skills and wants to bundle them into an agent
• User wants to add a skill or workflow to an existing agent
• User wants to change an agent's backend, model, or tool permissions
• User wants to fork an existing agent for a different domain

Process

Step 1: Inventory

Discover what exists:

1. Check for existing agents:

   Find directories matching: agents/*/AGENT.yaml
   

2. Check for standalone skills:

   Find files matching: skills/*/SKILL.md
   

3. Check for standalone workflows:

   Find files matching: workflows/*.yaml
   

4. Check dispatch backends:

   Read ~/.config/ai-dispatch/config.json for available backends
   

Present the inventory to the user.

Step 2: Select Components

Ask the user:

1. Base: "Start from scratch or fork an existing agent?"

   • If forking: copy the agent directory and modify

2. Skills: "Which skills should this agent include?"

   • List available skills with descriptions
   • Allow selecting existing or noting new ones to create

3. Workflows: "Does this agent need workflow enforcement?"

   • List available workflows
   • Allow selecting existing or noting new ones to create

4. Backend: "Which backend should it use?"

   • List available dispatch backends
   • Default, specific provider, or multi-backend routing

5. Identity: "Keep/modify the SOUL.md or create new?"

Step 3: Assemble

Create or update the agent directory:

If creating new:

agents/<name>/
├── SOUL.md
├── AGENT.yaml
├── skills/
│   ├── <skill-1>/SKILL.md    ← copy or symlink
│   └── <skill-2>/SKILL.md
├── workflows/
│   └── <workflow>.yaml        ← copy or symlink
└── memory/

If modifying existing:

1. Read current AGENT.yaml
2. Add/remove skills from the skills list
3. Add/remove workflows from the workflows list
4. Update configuration as needed
5. Validate any new workflow files