Critical failures across naming, descriptions, and schemas. 28 tools registered but many lack proper definitions or are duplicated (query_llm, web_scrape, take_screenshot appear 2-3 times). Descriptions are present but generic (avg ~80 chars, well below 194-char baseline). Parameter descriptions often missing or trivial. No output schemas documented. Tool names lack clear verb-noun structure (e.g., 'query_llm' is acceptable, but 'setup', 'learn_lesson', 'save_to_memory' are vague). Many tools expose dangerous operations (db_query, batch_process, lint_code with --fix) without dry-run or confirmation patterns. No error handling guidance visible in schemas.

• Deduplicate tools: merge query_llm variants into one with clear provider/model selection; consolidate web_scrape and take_screenshot into single definitions with distinct parameters.
• Add dry-run parameters to all WRITE tools (db_query, lint_code, batch_process, smart_refactor). Default dry_run=true for safety; require explicit dry_run=false to execute.
• Document output schemas for all tools. E.g., code_search returns [{file_path, line_number, match_text, context}]; web_crawler returns {urls: [], content: {}, metadata: {}}. Include in tool description or separate schema field.
• Rename vague tools: 'search' → 'web_search' (already exists, remove duplicate); 'setup' → 'install_mcp_server' or split into install_mcp, configure_mcp, verify_mcp; 'learn_lesson' → 'record_lesson' or 'store_lesson'.
• Add parameter constraints and format hints: db_path should note 'SQLite file path, e.g. /path/to/db.sqlite'; pattern should note 'glob pattern, e.g. *.js or src/**/*.ts'; query should note 'SQL query string, parameterized with ? placeholders'.