Loading ToolBench
CLI + MCP server + web catalog for AI agent skills, workflows, and MCP configs
Three tools with basic verb-noun naming (list_, install_, search_) and present descriptions. However, descriptions are minimal (34-72 chars, below the 50-200 char LLM-optimized baseline), parameter descriptions lack depth, and output schemas are not documented. The install_resource tool lacks error handling guidance for common failure modes (resource not found, permission denied). No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles. Code is incomplete (truncated at install_resource implementation), preventing full schema validation.
Install an AI agent skill or workflow into the current project
List all available AI agent resources (skills, workflows, MCP configs)
Search for AI agent resources by keyword or tag
Descriptions are too brief (34-72 chars vs. 50-200 char baseline). 'Search for AI agent resources by keyword or tag' lacks context on when to use this vs. list_resources, what fields are searchable, or result structure.
Output schemas not documented. Callers cannot predict response structure (e.g., does list_resources return tags, URLs, version info?). LLMs must guess what fields to extract for downstream calls.
install_resource lacks error handling guidance. No description of what happens if resource not found, already installed, or permission denied. LLM has no recovery path.
No tool annotations despite clear risk profiles. install_resource is destructive (WRITE risk) but lacks destructiveHint. list_resources and search_resources are read-only but lack readOnlyHint.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | 2026-07-28+ | v2 |
Parameter 'type' in list_resources uses enum but description is generic ('Filter by resource type'). Does not explain when to filter vs. when to list all, or what each enum value returns.