Two tools with minimal documentation. execute_python_code and fetch_web_content have descriptions (10-20 chars each), but lack parameter descriptions entirely. Input schemas are visible but parameters have no descriptions, violating the critical rule that every parameter needs one. No output schemas documented. Error handling is generic ('Error: {str(e)}') with no recovery guidance. No tool annotations (readOnlyHint/destructiveHint). execute_python_code is IRREVERSIBLE but has no confirmation pattern or dry-run support. fetch_web_content lacks max_length parameter description. Naming is acceptable (verb_noun), but definitions fall far short of production baseline.

• Add parameter descriptions: 'code' → 'Python code to execute (string, max 10000 chars). Executed in isolated subprocess with 15s timeout.' 'url' → 'HTTP(S) URL to fetch (string, must start with http:// or https://). max_length' → 'Maximum characters to return (integer, 1-50000, default 5000).'
• Document output schemas: execute_python_code returns {type: string, description: 'stdout or stderr from subprocess execution'}. fetch_web_content returns {type: string, description: 'HTML/text content truncated to max_length'}.
• Add tool annotations: execute_python_code with destructiveHint=true (can modify system state). fetch_web_content with readOnlyHint=true (no side effects).
• Implement error classification: distinguish retryable (timeout, network) from user-fixable (invalid URL format) from fatal (permission denied). Return actionable messages: 'URL must start with http:// or https://' instead of 'Error: invalid URL'.
• Add confirmation pattern for execute_python_code: require explicit approval before running code, or implement a dry-run mode that parses/validates without executing.