Dev Boards — harness-agnostic mission kanban for AI agents. MCP server providing five agent verbs (poll, claim, heartbeat, escalate, deliver) plus mission creation.
Agent-relay defines 6 tools with basic schemas and descriptions, but falls short of production quality. Tool names lack clear verb prefixes (poll, claim, heartbeat, escalate, deliver, create are acceptable but generic). Descriptions are present but brief (10-80 chars), below the 50-200 char LLM-optimized baseline. Parameter descriptions are sparse or missing entirely, most params lack type hints or constraints. Output schemas are not documented. Error handling is minimal. The 'create' tool definition is split across files (src/mcp/tools.ts vs scripts/mcp-server.mjs), making verification incomplete. No evidence of pagination, idempotency hints, or recovery guidance.
Atomically claim a mission (Running)
File a new mission (default inbox) on an allowed board
Deliver mission to Review with summary and optional artifacts/usage
Escalate with one clear human question (Needs human)
Progress heartbeat on a claimed mission
List claimable missions from Ready (or column). Prefers skill∩tag matches when agent has skills.
Parameter descriptions missing or incomplete. Most params (column, limit, agent, project, match_agent_skills, note, question, summary, artifacts, usage, priority, tags) lack descriptions explaining their purpose, format, or constraints. LLMs cannot infer parameter meaning from names alone.
Output schemas not documented. No tool specifies what fields are returned, their types, or structure. LLMs cannot plan downstream calls or extract required data without knowing response shape.
Tool descriptions too brief (10-80 chars vs 50-200 char baseline). 'Atomically claim a mission' and 'Progress heartbeat on a claimed mission' lack context on when to use, prerequisites, or what happens next. Descriptions should answer: what, when, and what does it return?
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 44 | 2026-07-28+ | v2 |
No error handling guidance. Tools do not document what errors can occur, whether they are retryable, or what the LLM should do next. A failed claim or deliver leaves the agent without recovery direction.
No pagination or result limits documented. 'poll' returns a list but does not specify max results, pagination mechanism, or total count. Large result sets will blow context windows.