Single tool 'detect' has a clear, well-structured description (194 chars) and properly typed input schema with a string parameter. However, the tool lacks output schema documentation, the description mentions 'human-readable result with attached JSON summary' but does not specify the actual response structure (fields, types). No error handling guidance is provided. The tool name 'detect' is a weak verb; 'detect_prompt_injection' would be more explicit per naming patterns. Parameter description is adequate but could specify constraints (e.g., max length, non-empty validation).

• Document the complete output schema: specify all response fields (e.g., is_injection, confidence, detection_layers, summary), their types, and what they represent. Example: {"is_injection": boolean, "confidence": number 0-1, "summary": string, "details": object}.
• Rename tool to 'detect_prompt_injection' to follow verb_noun convention and make the action explicit.
• Add parameter constraints to 'text': specify max length (e.g., 10000 chars), encoding (UTF-8), and validation rules. Update description: 'The text to analyze (1-10000 UTF-8 characters, non-empty).'
• Implement error handling with recovery guidance. Return structured errors: {"error": "text_too_long", "message": "Text exceeds 10000 characters. Truncate or split into smaller chunks.", "retryable": true}.
• Add tool annotations in the MCP registration: set readOnlyHint=true and idempotentHint=true to signal that repeated calls are safe and do not modify state.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry