Loading ToolBench
A Claude plugin for web search and content fetching with optional integrations for Exa, Jina Reader, and Playwright-based browser control
Scout provides 6 browser-control and fetch tools with basic descriptions and partial schemas. Tool names follow verb_noun convention (list-pages, snapshot, screenshot, evaluate, navigate, fetch-page), which is good. However, descriptions are minimal (10-60 chars), parameter descriptions are sparse or missing, and output schemas are not documented. No error handling guidance, no parameter validation rules, and no indication of which operations are idempotent or destructive. The code shows implementation details (Playwright CDP, Chrome requirements) but the tool definitions lack LLM-optimized structure. Most parameters lack type constraints (e.g., 'selector' accepts string but no enum or format guidance). No pagination support despite fetch-page potentially returning large HTML. Risk annotations (READ_ONLY, WRITE) are present in metadata but not reflected in tool definitions.
Run JavaScript and print result
Fetch a web page via Playwright headless Chrome → stdout as HTML/text. Privacy-safe: all processing is local (no external API calls).
List open tabs (title + URL)
Navigate a tab to URL (default: first tab)
Save page screenshot to file
Get page text content (default: first tab)
Descriptions are too brief (10-60 chars) and lack LLM-optimized context. 'Get page text content' does not explain when to use snapshot vs fetch-page, what format is returned, or prerequisites. Baseline for A-grade tools is 50-200 chars with WHAT/WHEN/HOW guidance.
Parameter descriptions are missing or minimal. 'selector' appears in 4 tools but is described identically as 'Tab index or title/URL substring (optional)', no guidance on format, precedence, or error cases. 'wait_for' lacks explanation of timeout behavior or what happens if selector never appears.
No output schemas documented. Tools return text, JSON, or file paths but LLMs cannot plan downstream calls without knowing structure. fetch-page returns 'HTML/text' but no field names, size limits, or pagination info. Baseline: 100% of A+ tools document return types.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2026-07-28+ | v2 |
Error handling lacks recovery guidance. Code exits with codes 1 or 2 but tool definitions do not explain what these mean or what the LLM should do next. 'Error: no open tabs found' is raw; should be 'No tabs open. Start Chrome with --remote-debugging-port=9222 or call list-pages to verify connection.'
fetch-page lacks pagination and result-size limits. Fetching a large HTML page could return megabytes of content, exhausting context window. No mention of max_size, truncation, or how to handle timeouts. Baseline: tools returning lists should cap results at 20-50 items and document the limit.