Single tool with critical definition gaps. Tool name 'analyze_massive_document' is somewhat descriptive but lacks action verb clarity. Description is in Indonesian and very brief (82 chars), below the 194-char baseline. Input schema present but lacks per-parameter descriptions, parameters 'file_path' and 'query' have minimal context. Output is undocumented free-text string with no schema. No error categorization or recovery guidance. No input validation constraints (file size limits, format specs). The implementation shows basic error handling (file not found, empty file, exception catch) but responses are unstructured and non-actionable for LLM self-correction. Security concern: tool reads arbitrary files from disk without validation or sandboxing, potential path traversal risk.

• Rewrite tool description in English, expand to 150-250 characters explaining: (1) what it does, analyzes large documents using recursive LLM chunking, (2) when to use it, for documents too large to fit in a single context window, (3) what it returns, structured findings with line numbers or relevant excerpts
• Add explicit description to 'file_path' parameter: 'Local filesystem path to the document (txt, code, log). Max 1 GB. Will be sandboxed to /allowed/documents/ directory. Use relative paths only (e.g., reports/2024.txt).'
• Add explicit description to 'query' parameter: 'Natural language question or task to perform on the document (e.g., "Find all security vulnerabilities", "Extract API endpoints"). Max 500 chars.'
• Document output schema as: {success: bool, findings: string, metadata: {chunks_processed: int, tokens_used: int, execution_time_sec: float}}
• Implement strict input validation: (1) validate file_path against a whitelist or sandbox directory, reject ../ and absolute paths, (2) enforce max file size (e.g., 50 MB), (3) return structured error responses with actionable recovery steps
• Redesign error responses to include categorization and guidance. Example: '{error: "file_not_found", path: "/path/to/file", suggestion: "Check the path spelling. Available files in this directory: [list]", retryable: false}'
• Add timeout configuration (default 30 seconds per LLM call, 5 minutes per full analysis) and return timeout errors with: 'Query exceeded 5-minute limit on 500 MB document. Try: (1) splitting the file, (2) refining the query to be more specific, (3) increasing timeout config.'