Loading ToolBench
MCP server for the Hobbes API, generated from their published OpenAPI spec, with a client-side policy gate on mutating operations.
Single tool 'hobbes_apply_playbook_update_preview' has critical definition gaps. Description is generic and under 100 chars (fails baseline of 194 chars avg). Input schema present but severely underdocumented: 'preview_id' lacks description; 'body' is vague ('shaped by the Hobbes OpenAPI schema') with no field-level documentation. No output schema documented. Parameters lack type constraints, enums, or validation guidance. Tool name is verbose and unclear about what 'apply' means in context. No error handling guidance. Generated from OpenAPI but abstraction layer lost critical LLM-facing metadata.
Tool generated from Hobbes OpenAPI spec for applying a playbook update preview
Tool description is 89 chars, well below 194-char baseline. Generic phrasing ('Tool generated from Hobbes OpenAPI spec') provides no LLM guidance on WHEN to use this tool or WHAT it does beyond the name.
'preview_id' parameter has no description. LLM cannot infer whether this is a UUID, numeric ID, or opaque token. 'body' parameter description is circular ('shaped by the Hobbes OpenAPI schema'), does not explain what fields are required, what they control, or constraints.
No output schema documented. LLM cannot plan downstream actions or extract relevant fields from the response. Baseline: 100% of A+ tools have documented return types.
Tool name 'hobbes_apply_playbook_update_preview' is 40 chars (p90=27 baseline). Verbose and unclear: does 'apply' mean execute, confirm, or preview? Lacks action verb clarity. Should be shorter and more specific (e.g., 'confirm_playbook_update').
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 35 | 2026-07-28+ | v2 |
No error handling guidance. If preview_id is invalid or expired, LLM receives raw HTTP error with no recovery path. Baseline: error responses must tell LLM what to do next.