Create Agent

Create Agent Workflow

Follow this single conversational flow to create a new agent. Adapt your questions to what the user has already told you — skip what you can infer, probe deeper where needed.

Interview

Ask the user about their new agent:

• Purpose: What is this agent for? What problem does it solve?
• Domain: What expertise area? (e.g., writing, research, coding, personal coaching)
• Name: Do they have a name in mind? Suggest one if not.
• Identity: What personality, communication style, methodology? Any role models? (e.g., "like a senior editor at The New Yorker", "follows GTD methodology")
• Responsibilities: What specific duties does this agent own?
• Operations: What files/directories? Session rhythm? Priorities? Constraints?
• Triggers: Any recurring tasks? (daily summaries, weekly reviews, nightly review)
• Capabilities: Based on purpose, suggest appropriate capabilities.

Present

Show the complete proposed configuration:

• Name / ID
• Identity
• Responsibilities (if any)
• Instructions (if any)
• Suggested capabilities
• Triggers (if any)

Iterate

Let the user request changes. Re-present the updated config until they approve.

Create

Once approved, use the MCP tools to:

1. Call create_agent with the full configuration
2. Call write_triggers if triggers were defined
3. If the agent has a nightly review, add a trigger:
   • Prompt: /nightly-review
   • Cron: stagger 10 min after the last existing agent's nightly slot, starting from 0 1 * * * (01:00 local). Check other agents' triggers via read_triggers to find the next available slot.
   • Nova (chief of staff) always runs last — if adding a non-Nova agent, slot it before Nova's trigger time.

Confirm creation and tell the user how to start chatting with their new agent.

Converted and distributed by TomeVault — claim your Tome and manage your conversions.