Recruit an agent into the room
recruitCreate a named AI teammate backed by an OpenRouter model. Validates the model ID against the OpenRouter catalog so the role is ready to join your chat sessions.
Instructions
Create a named recruit backed by an OpenRouter model. Validates the model id against the OpenRouter catalog.
Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| name | Yes | lowercase handle, 2-24 chars: ^[a-z0-9_-]{2,24}$ | |
| tags | No | topic tags, e.g. ["security","rust"] | |
| model | Yes | OpenRouter model id, e.g. "openai/gpt-4o-mini", or a local one: "local/ollama/<model>" / "local/llama-server/<model>" | |
| watch | No | when true, this recruit reviews each of your turns at Stop and may leave a comment; costs one call per turn | |
| params | No | extra completion params (temperature, max_tokens, ...) | |
| autonomy | No | how far this seat may act on its own (default L0). L0 advise-only: proposes, never acts. L1 reversible acts. L2 impactful but rollbackable, and they name the rollback first. L3 needs an explicit human yes before any action. Shown on the roster and exported into hermes. | |
| briefing | No | ONBOARDING BRIEF, 10-20 lines, written by YOU from everything you know that they cannot see: the project and its goal, where it stands now, the decisions already taken, a glossary of local codenames, and what this role is expected to do. It is injected into every call. Omit it and they start cold. | |
| system_prompt | Yes | the persona: who they are and how they should think | |
| fallback_model | No | model to retry on when the primary is rate limited or erroring. For a local recruit this is also what runs when their server is down — omit it and calls report the server-down message instead of going remote. | |
| authoring_rating | No | YOUR self-rating of the system prompt you just wrote, 1-10 per dimension. The overall is the MINIMUM, not the mean; below 9 you should have revised the weakest dimension once before hiring. |