Single tool 'search_fraud_playbooks' has a basic description (89 chars) but critically lacks an input schema in the visible source code. The tool is registered in agent.py but no JSON Schema with type definitions is visible. Parameter 'query' is mentioned in the description but not formally defined with type, constraints, or format. No output schema documented. This violates core pattern:tool requirements: every parameter needs a type and description, and output structure must be documented for LLM planning.

• Add explicit JSON Schema for input: {"type": "object", "properties": {"query": {"type": "string", "description": "Natural language query about fraud patterns, response procedures, or investigation protocols. Examples: 'credit card fraud detection', 'chargeback response workflow'. Max 500 chars."}}, "required": ["query"]}
• Document output schema: e.g. {"type": "object", "properties": {"playbooks": {"type": "array", "items": {"type": "object", "properties": {"id": {"type": "string"}, "title": {"type": "string"}, "description": {"type": "string"}, "steps": {"type": "array", "items": {"type": "string"}}}}, "total_count": {"type": "integer"}, "next_cursor": {"type": "string", "nullable": true}}}
• Expand description to 150-200 chars: 'Search Bedrock Knowledge Base for fraud detection playbooks and response procedures. Use this to find investigation steps, risk indicators, and remediation workflows. Returns ranked playbooks with actionable steps.'
• Add error handling guidance: 'If no playbooks found, try broader query terms. If Knowledge Base is unavailable, inform user that playbook lookup failed and suggest manual investigation protocol.'
• Add pagination support: accept optional 'limit' (1-50, default 20) and 'cursor' parameters; return 'next_cursor' in response for multi-page results.