Single tool with mixed quality. Tool name 'interactive_feedback' lacks action verb (should be 'request_feedback' or 'get_user_feedback'). Description is lengthy (500+ chars) and mixes Chinese usage guidelines with English, reducing clarity for LLM parsing. Input schema present with types and descriptions, but parameter descriptions are minimal (10-20 chars each). No output schema documented, tool returns MCP content blocks but this is not declared. Error handling is silent (returns empty result on timeout/failure) with no guidance to LLM on recovery. Tool is READ_ONLY per risk classification, which is correct, but not annotated in schema.

• Rename tool to 'request_user_feedback' to start with action verb and clarify intent.
• Rewrite description in English only, 80-150 chars: 'Request user input via interactive UI. Supports text response and image attachments. Returns user's text answer and any uploaded images.'
• Document output schema: 'Returns array of content blocks: [{type: "text", text: string}, {type: "image", data: base64, mimeType: string}]. Empty array if user cancelled.'
• Add error handling: return structured error on timeout (60s default) with message 'User did not respond within timeout. Retry or ask user to try again.' Include retryable: true flag.
• Enhance parameter descriptions: 'message_images: Array of file paths (PNG, JPG, GIF, WebP) to display in UI. Max 5 images, each <10MB. Optional.' 'predefined_options: Array of 1-10 choice strings (1-100 chars each). If provided, user selects one; if empty, user enters free text.'