Loading ToolBench
Single tool with incomplete definition. Tool name is reasonable but descriptions are sparse. Input schema is visible and properly structured with types, but parameter descriptions are minimal (1-3 words each, below the 10-char baseline). No output schema documented. No error handling guidance. The server appears functional but lacks LLM-optimized documentation critical for agent tool selection and planning.
query chat messages with given parameters
Tool description is extremely brief ('query chat messages with given parameters') and lacks context on WHEN to use it, WHAT it returns, or HOW results should be interpreted. LLMs cannot determine appropriate tool selection with such minimal guidance.
Parameter descriptions are 1-3 words each ('chat room names', 'talker names', 'chat messages limit'), below the 10-character minimum. These trivial descriptions add no actionable context for LLM parameter selection. For example, 'chat room names' does not explain format (array of strings), whether partial names are accepted, or case sensitivity.
No output schema documented. The implementation returns JSON.stringify(messages) but provides no schema specification of the ChatMessage structure. LLMs cannot plan downstream operations without knowing what fields are returned (e.g., is there a room_id, talker_id, timestamp?).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 51 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 33 | - | v1 |
No error handling or recovery guidance. The CallToolRequestSchema handler throws 'Unknown tool' but provides no guidance to LLMs on retryability, user-fixable errors, or next steps when a query returns empty results or fails.
All parameters are optional (required: []) but the tool is not usable with zero arguments (empty call returns all 1M+ chat messages, context explosion). No default pagination, no limit validation, no guidance on typical usage patterns.
No pagination strategy. Large result sets have no next_cursor or offset/limit mechanism in the response definition. The limit parameter caps results but the response does not indicate whether more results exist or how to fetch them.