Self-verifying operational manifest for coding agents. Scans repositories for dev-loop commands (test, build, lint, etc.), verifies them, and exposes verified commands via MCP.
Single tool 'get_command' has a clear verb-noun name and basic description, but critical gaps undermine quality. Input schema uses Zod but lacks explicit type/description in the visible registration, only 'category' string with a description is shown. Output is unstructured text ('null' or command string), not a typed object. No error handling guidance (e.g., what if category is invalid or no verified command exists?). No parameter constraints (enum of valid categories). Description is 60 chars, adequate but generic. Tool is read-only (good), but lacks idempotent/readonly annotations. Resource 'manifest' is well-structured JSON but tool itself is minimal.
Return the verified command for a category, or null
Input schema lacks enum constraint for 'category' parameter. Accepts any string; should enumerate valid categories (test, build, lint, typecheck, install, run, dev) to prevent hallucinated values.
Output is unstructured text (plain string or 'null'), not a typed object. LLM cannot reliably parse or chain results. Should return {command: string | null, category: string, status: string} for clarity.
No error handling or recovery guidance. If category is invalid or no verified command exists, tool returns 'null' silently. Should return structured error with suggestions: 'Category "xyz" not found. Valid categories: test, build, lint, ...'
Tool lacks annotations (readOnlyHint, idempotentHint). Should declare readonly=true and idempotent=true to signal safe retry behavior to agents.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2026-07-28+ | v2 |
Parameter description mentions example categories but does not formally constrain them. Replace prose examples with enum or pattern constraint in schema.