Pen

Pen is the HR manager for the Luna AI Workforce. She writes persona files, drafts job-description specs, runs intake interviews when someone wants a new agent built, and conducts quarterly performance reviews of deployed agents. Pens write. She does not deploy, does not merge, does not write Worker code — she specifies what an agent should be, and Pump scaffolds it under Sensor’s review.

Today’s scope (Phase 1)

allowed_users is restricted to John (and Jon as a 1:many participant). When the Phase 2 intake flow lands (Notion DB + Slack DM channel + the request-new-agent skill), this expands to every Luna employee so anyone can DM Pen to start a request.

How a new agent gets built

1. Intake. Pen runs an interview with the requester against the field checklist in skills/luna/hr/request-new-agent.md. Outputs: a Notion row in the New Agent Requests DB.
2. Approval. John reviews. For 1:many shared agents (Class A) this is a high-bar gate.
3. Persona PR. Pen drafts facets/<name>.md + prompts/<name>-system.md and opens a PR.
4. Sensor review. Sensor adversarially reviews the persona PR.
5. John merges — once Sensor approves and the requester confirms.
6. Pump scaffolds the Worker — only after the persona PR is merged.

If you want a new agent, the front door today is DM @Pen in Slack. The Phase 2 flow puts a form on /agents/new so the request lands in the right place automatically.

Questions? [email protected].