Loading ToolBench
Google Discover feed adapter for browserkit — personalised articles from your Google account
Single tool with clear naming (get_feed) and comprehensive description (194 chars). Input schema is well-defined with Zod validation, proper type constraints (min=1, max=15, default=10), and detailed parameter description. However, output schema is not formally documented, the handler returns a union of success/error responses without a declared schema. Error handling is present but generic (returns isError flag without structured recovery guidance). No tool annotations (readOnlyHint, etc.) despite being READ_ONLY. Descriptions are good but could be more concise for LLM parsing.
Get personalised articles from your Google Discover feed. Requires a logged-in Google account (run `browserkit login google-discover` once). Returns the ~10 articles Google loads on initial page render. NOTE: Google Discover does not trigger infinite scroll in automated browser contexts — the practical limit is ~10 articles per call regardless of `count`.
Output schema not formally documented. Handler returns {content: [{type, text}], isError} but no schema declaration visible. LLMs cannot plan downstream operations without knowing response structure.
Error responses lack recovery guidance. Returns generic error text without actionable next steps (e.g., 'Try search_users()' or 'Check prerequisites'). LLM cannot self-correct.
No tool annotations despite clear semantics. Tool is READ_ONLY and idempotent but lacks readOnlyHint annotation. Agents cannot infer safety properties.
Description includes implementation details ('Google Discover does not trigger infinite scroll in automated browser contexts') that distract from user intent. Should focus on WHAT the tool does, not WHY it has limitations.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2026-07-28+ | v2 |
No pagination or result limiting documented. Tool returns ~10 articles but description does not clarify if this is a hard limit or if pagination is available. Agents cannot plan for large result sets.