Loading ToolBench
MCP server that exposes a product build plan as a work queue for agents. Provides three tools to retrieve tasks, mark them complete, and check plan status.
Three well-named tools with clear, action-oriented verbs (preflight_next_task, preflight_complete_task, preflight_plan_status). Descriptions are concise (50 - 100 chars) and explain WHAT and WHEN. Input schemas present with type definitions and descriptions. However, output schemas are not formally documented in the tool definitions, only inferred from code. Parameter descriptions are minimal but adequate. No error handling guidance in descriptions (e.g., what to do if plan not found). Tool composition is sound: each does one thing, and outputs chain naturally (task IDs flow to complete_task). Missing: explicit output schema documentation, error recovery hints, and parameter constraints (e.g., planId format).
Marks one task as done and reports how many remain. Call this only after the work is actually finished and verified, not when starting it.
Returns the next unfinished task from the user's Preflight plan, including a self-contained instruction to carry it out. Call this to start work, and again after each completed task. Tasks come back in build order, so the one returned is the one to do next.
Summarises the plan: its title, how many tasks are done, and progress per feature. Use it to report progress or to find the plan id when the user has several.
Output schemas not formally documented in tool definitions. LLMs cannot predict response structure (e.g., task object fields, feature array format) without explicit schema documentation.
Error responses lack recovery guidance. When plan not found, description says 'Omit to use the most recently updated plan' but does not explain what happens if no plan exists or how to recover.
planId parameter lacks format/constraint documentation. No indication of expected format (UUID, alphanumeric, length limits), forcing LLMs to guess or trial-and-error.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 65 | 2026-07-28+ | v2 |
Tool descriptions do not explicitly state idempotency or side effects. preflight_complete_task modifies state but description does not warn against retries or confirm-before-execute pattern.
No pagination or result limits documented. If a plan has hundreds of tasks, preflight_next_task returns one; but preflight_plan_status returns all features and tasks without limit guidance.