Loading ToolBench
Agent framework with tool execution, session management, and plugin support. Provides tool call handling, middleware system, and multi-client server capabilities.
Three tools with partial schemas and inconsistent documentation. agentxx_share_store and agentxx_subagent have descriptions (94-108 chars, within baseline 34-392 range) but lack critical parameter details. tool_skill_search has minimal description (95 chars). All three tools show input schemas with type definitions, but parameter descriptions are sparse or missing entirely. No output schemas documented. Error handling guidance absent. Tool names lack clear action verbs (share_store, subagent are vague). No evidence of idempotency, permission gates, or recovery guidance.
Session shared storage tool for get/insert/set/delete operations on shared data within a session
Sub-agent delegation tool for single and batch task execution through sub-agents
Search available tools and skills for loading based on user requirements. Uses filesystem tools to search SKILL.md files and determine which tools/skills to load.
Tool names lack action verbs. 'agentxx_share_store' and 'agentxx_subagent' are vague; should be 'get_shared_storage', 'insert_shared_storage', 'delegate_to_subagent', etc. LLMs cannot infer intent from generic names.
Parameter descriptions missing or incomplete. 'opt' parameter in agentxx_share_store lacks guidance on valid values (insert/get/set/delete); 'line_offset' and 'line_limit' lack context. LLMs cannot determine when to use these parameters.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields. agentxx_share_store should document return structure for get/insert/set/delete operations.
No error handling guidance. Tools lack recovery hints (e.g., 'If storage ID not found, try listing available IDs'). Agents cannot self-correct on failures.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 38 | 2026-07-28+ | v2 |
agentxx_subagent accepts both 'message' (single) and 'messages' (array) but lacks documentation of mutual exclusivity or behavior when both provided. Parameter relationships undocumented.