Single tool 'ask_gemini_web' has a basic description (67 chars) and minimal input schema. The description lacks context on when to use this tool vs. alternatives, prerequisites, or what happens on failure. The schema defines only one required parameter 'prompt' with a generic string type and minimal description (28 chars). No output schema is documented, the LLM cannot predict the response structure. No error handling guidance, no parameter constraints (e.g., max length), and no mention of timeout behavior or failure modes. The tool name 'ask_gemini_web' is reasonably clear (verb_noun pattern), but the overall definition is sparse and would not meet production standards.

• Add output schema: document that the response is {content: [{type: 'text', text: string}]} so the LLM knows what to expect.
• Expand tool description to 100-150 chars: 'Send a prompt to Google Gemini Pro via the browser extension. Returns the model response as plain text. Requires the extension to be connected; fails if unavailable.'
• Add parameter constraints: 'prompt (string, 1-4000 chars): The question or instruction to send to Gemini. Keep under 4000 characters to avoid timeouts.'
• Document error cases: 'If the extension is not connected, returns error 'Extension not connected'. If the request times out after 30s, returns error 'Request timeout, try a shorter prompt.'
• Add idempotent hint to tool definition if repeated calls with the same prompt are safe, or destructive hint if they have side effects (e.g., if Gemini maintains conversation state).