Loading ToolBench
A local LLM client with web browsing, search, and MCP server integration capabilities
Three tools with basic schemas and descriptions, but significant gaps in parameter documentation and output schema clarity. Tool names lack action verbs (e.g., 'google-search-engine' vs 'search_google'). Descriptions are present but generic (10-60 chars, below the 50-200 char baseline for LLM optimization). Parameters have type info but lack detailed constraints, ranges, or format guidance. No documented output schemas. Error handling is minimal, no recovery guidance or categorization. The tools are read-only and relatively simple, but lack the rigor expected for production agent use.
Extracts the most relevant content from a webpage as a single string, without any HTML tags.
Extracts images from a webpage. Returns a list of images found on the webpage.
Web search. Useful for when you need to answer search questions. Input should be a search query.
Tool names do not start with action verbs. 'google-search-engine' should be 'search_google' or 'search_web'. 'extract-images-from-webpage' should be 'extract_webpage_images'. LLMs rely on verb-first naming to infer intent before reading descriptions.
Descriptions are too short (10-60 chars vs. baseline 50-200 chars). 'Web search. Useful for when you need to answer search questions. Input should be a search query.' lacks context on when to use vs. similar tools, what it returns, and prerequisites (API key, search engine ID).
No output schemas documented. LLMs cannot plan downstream calls or extract fields without knowing the response structure. google-search-engine returns ChatToolResult with 'text' and 'references' array, but this is not visible in the tool definition.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 43 | 2026-07-28+ | v2 |
Parameter 'query' and 'url' lack format constraints, examples, or length limits. No guidance on valid URL schemes, max query length, or handling of special characters. LLMs will pass arbitrary strings without validation.
Error handling is minimal. No recovery guidance (e.g., 'If search fails, try a simpler query'). No categorization of errors as retryable vs. fatal. Errors like 'Executing search failed: ...' provide no actionable next step for the agent.