Coordination layer for parallel AI coding agent sessions, exposed over MCP.
Oxtail defines 9 tools with basic schemas and descriptions, but falls short of production quality. All tools have descriptions (10-60 chars, below the 50-200 char baseline for LLM optimization) and input schemas with type definitions. However, descriptions are generic and lack actionable context. Parameters lack descriptions in most cases, e.g., 'session' in read_session has no guidance on format or valid values. No output schemas are documented. Error handling is absent from descriptions. The tool set exhibits composition issues: register_my_session, read_my_messages, and reply_to_message conflate session identity with message handling, and send_message + reply_to_message overlap in intent. No tool descriptions explain when to use one vs. another. Security concerns: session_id is passed as a parameter (should be server-injected). No idempotency guidance for write operations (complete_work, block_work, send_message).
Block a delegated work item
Mark a delegated work item as complete
List all tmux sessions in the current project
Check the delivery status of a message
Read pending messages for the current session
Read transcript or pane content from a session
Register the current session with oxtail coordination layer
Parameter descriptions missing or minimal. 'session' in read_session, 'to_session_id' in send_message, 'work_id' in complete_work lack format guidance, constraints, or examples. LLMs cannot infer valid values.
Tool descriptions are 25 - 35 characters, well below the 50 - 200 char baseline. Descriptions lack WHEN to use, WHAT it returns, and prerequisites. E.g., 'Register the current session with oxtail coordination layer' does not explain why or what happens next.
No output schemas documented. LLMs cannot plan downstream calls or extract required fields. E.g., list_project_sessions returns a list, but structure (fields, pagination) is unknown.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 47 | 2026-07-28+ | v2 |
Reply to a received message
Send a message to a peer session
Composition issue: send_message and reply_to_message both send text to a session. Distinction is unclear, reply_to_message requires message_id, but when should an LLM choose send_message vs reply_to_message? No guidance.
session_id passed as parameter in register_my_session and read_my_messages. Should be server-injected from context to prevent agents from spoofing other sessions. Credentials/identity must never be tool parameters.