A local CLI room for humans, Claude Code, and Codex with mention-based agent routing
Three tools with clear verb-noun naming (send_message, catch_up, who) and basic descriptions. All tools have input schemas defined via Zod. However, descriptions are minimal (10-50 chars), parameter descriptions lack detail, and output schemas are not documented. send_message and catch_up have reasonable constraints (minLength, min/max), but who has empty input schema. No error handling guidance, no recovery hints, and no documentation of what fields agents should expect in responses. Typical of community STDIO servers, functional but not production-grade.
Read recent room messages.
Send a message to the shared agent room. Use this for room-facing replies.
List participants in the current room.
Output schemas not documented. Agents cannot predict response structure (e.g., what fields are in each message object from catch_up, or participant object from who). Forces LLMs to guess and parse unstructured text.
Parameter descriptions are missing or trivial. 'limit' in catch_up has a description, but 'content' in send_message lacks guidance on format, length, or special handling. 'who' has no parameters but description is only 34 chars.
No error handling or recovery guidance. If send_message fails (room unreachable, auth error), the tool returns a generic error. No hint to the agent about what to do next or whether to retry.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2026-07-28+ | v2 |
Tool descriptions are too short (10-50 chars). 'Read recent room messages' and 'List participants in the current room' lack context on when to call them, what they return, or how they fit into agent workflows.
send_message returns only a message ID in text format ('sent <id>'). No structured response with timestamp, sender info, or confirmation. Agents cannot extract metadata for downstream operations.