Three tools with basic schemas and descriptions, but significant gaps in parameter documentation and output structure. All tools have descriptions (10-60 chars), but descriptions are minimal and lack context for LLM selection. Input schemas present but parameters lack detailed descriptions explaining constraints, formats, or dependencies. No output schemas documented. Error handling is absent, tools return success responses without guidance on failure modes or recovery. No tool annotations (readOnlyHint/destructiveHint). Naming is clear (verb_noun pattern), but descriptions do not explain WHEN to use each tool or what happens on failure.

• Add detailed parameter descriptions to collect_intake: 'name (string, 1-100 chars, required): Customer full name or first name. Used for scheduling confirmation.', 'callback_number (string, E.164 format, required): Customer phone number in +1-555-0123 format.', 'service (string, enum: ["plumbing", "electrical", "hvac", "general"], required): Type of service requested.', 'preferred_time (string, ISO 8601 or natural language, required): Preferred appointment window, e.g. "2024-01-15T09:00-17:00" or "Monday morning".'
• Document output schemas for all tools. E.g., search_service_knowledge should return {"results": [{"title": string, "content": string, "source": string}], "total": int, "query": string}. collect_intake should return {"status": "pending_confirmation", "name": string, "callback_number": string, "service": string, "preferred_time": string, "message": string}.
• Add error handling and recovery guidance. E.g., collect_intake: 'If callback_number is invalid, return {"status": "error", "error": "Invalid phone format. Expected E.164 format like +1-555-0123.", "recovery": "Ask the user to provide their phone number in +1-555-0123 format."}'. Escalate_to_human: 'If priority is not recognized, normalize to "normal" or "urgent" and return {"status": "queued", "priority": normalized_priority, "summary": summary[:500], "escalation_id": uuid}.'
• Add tool annotations: collect_intake and escalate_to_human should include destructiveHint=true (they modify state). search_service_knowledge should include readOnlyHint=true (read-only). This enables LLMs to reason about retry safety.