Scoring was not performed

• Add explicit JSON Schema with type and description for every parameter. For tools with empty input schemas (intel_alert_digest, intel_weekly_trends), explicitly define {} as the input schema if truly parameterless, or add required parameters.
• Document output schemas for all tools. Specify the structure of return values so LLMs can reliably extract fields and chain tools together.
• Consolidate duplicate tools: merge fetch_classify_event into classify_event and fetch_entity_extraction into extract_entities. Let LLMs call a single canonical function.
• Expand descriptions for intel_alert_digest and intel_weekly_trends to 100-200 characters, explaining what each tool monitors and when to call it (e.g., 'Call intel_alert_digest for real-time cross-domain threats; use intel_weekly_trends for historical volatility and anomaly detection').
• Add error handling guidance. Each tool should document recoverable errors (e.g., 'If country code is invalid, return available country codes') and fatal errors. Use structured error responses with recovery hints.
• Use JSON Schema enums for constrained parameters. Replace text descriptions like 'Corridor names from infrastructure.CABLE_CORRIDORS' with an explicit 'enum' field listing valid values.
• Add pagination parameters (limit, offset, next_cursor) to tools that return lists. Cap default results at 20-50 items.
• Document the default value and valid range for numeric parameters. E.g., 'similarity_threshold: 0.0-1.0 (default 0.3)' not just 'Minimum Jaccard similarity'.