Loading ToolBench
Single tool with basic schema and minimal description. Tool name 'query-couchbase' lacks action verb prefix (should be 'execute_couchbase_query' or 'run_couchbase_query'). Description is 62 chars, acceptable but generic, lacks WHEN to use or prerequisites. Input schema present with proper type and description for 'statement' param. No output schema documented. No error recovery guidance. No parameter constraints (e.g., max query length, timeout hints). Tool is read-only but lacks explicit idempotent annotation.
Execute a Couchbase query using a SQL++ statement
Tool name lacks action verb. 'query-couchbase' should start with a verb like 'execute_', 'run_', or 'query_' to signal intent to LLMs.
No output schema documented. LLMs cannot plan downstream operations or extract fields from query results without knowing the response structure.
Description lacks context on WHEN to use this tool and prerequisites (e.g., 'Requires Couchbase cluster connection via env vars'). Generic descriptions reduce LLM selection confidence.
No parameter constraints documented. 'statement' param accepts unbounded strings with no guidance on max length, timeout behavior, or SQL++ syntax requirements.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 43 | 2026-07-28+ | v2 |
Error handling returns raw error messages without recovery guidance. 'Error executing query: <message>' gives LLM no actionable next step (retry? check syntax? contact admin?).