Loading ToolBench
Server has 13 tools with consistent naming (verb_noun pattern) and descriptions present for all. However, parameter descriptions are minimal (often single phrases), output schemas are not documented, and error handling guidance is absent. Tool names are clear (search, ingest, delete_memory, etc.) but descriptions lack context on WHEN to use each tool vs. alternatives. Parameters like 'timezone', 'sources', 'project' appear across multiple tools but lack format/constraint documentation. No evidence of pagination guidance, result limits, or chaining IDs in responses. Security considerations (destructive operations) are noted in risk labels but not reflected in tool descriptions or confirmation patterns.
Create a wiki document with title, content, and optional project routing
Delete a memory episode by UUID
Delete a wiki document by ID
Retrieve user profile settings from Membase
Retrieve user profile memory from Membase
Retrieve a wiki document by ID
Ingest content into memory with optional display summary and project routing
Output schemas not documented. Tools return data but LLMs cannot infer response structure, field names, or chaining IDs needed for downstream calls. E.g., search() returns results but response format is undocumented.
Parameter descriptions lack format/constraint details. 'timezone' appears in 3 tools but no format specified (IANA? UTC offset?). 'sources' is array but no item type or valid values documented. 'limit' has no min/max bounds.
Destructive tools (delete_memory, delete_wiki_document) lack confirmation or dry-run patterns. No guidance on recovery or undo. Descriptions do not warn of irreversibility.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | <=2025-11-25 | v2 |
Record agent usage statistics with Membase (best-effort, non-blocking)
Register agent connection with Membase (best-effort, non-blocking)
Search memory with optional filtering by date range, timezone, sources, and project
Search memory bundles with optional filtering by date range, timezone, sources, and project
Search wiki documents with optional filtering by collection and project
Update an existing wiki document by ID with new content and optional project routing
register_connection and record_agent_usage have minimal descriptions ('best-effort, non-blocking') and empty input schemas. LLMs cannot determine what data these tools expect or why to call them.
No pagination or result-limit guidance. search() and search_bundles() accept 'limit' and 'offset' but no documentation of max results, default behavior, or total count in response.