Single tool with basic schema and description. Tool name 'scan_security_patterns' is action-oriented but generic. Description (95 chars) is present but minimal, lacks WHEN to use, prerequisites, or output structure guidance. Input schema is properly typed with required field, but output schema is completely undocumented, LLM cannot predict response structure. No parameter descriptions beyond the code parameter. No error handling guidance, no pagination, no result limits documented. Tool performs static analysis (read-only, safe to retry) but this is not stated. Missing 40+ patterns from the Arcade baseline.

• Document output schema: return type is object with 'status' (string: 'success'|'error'), 'message' (string), 'findings' (array of {rule_id, cwe, severity, line, message, remediation}). Include example response.
• Expand tool description to 150-200 chars: 'Scans Python source code using AST analysis to detect security vulnerabilities (SQL injection, command injection, deserialization, hardcoded secrets, code injection). Returns findings with CWE tags, line numbers, and remediation guidance. Safe to retry. Supports code up to 1MB.'
• Add parameter description for 'code': 'Python source code to audit (string, max 1MB, UTF-8 encoded). Supports any valid or invalid Python syntax.'
• Add error handling guidance: on SyntaxError, return 'Syntax error at line X: {details}. Ensure code is valid Python or contains the snippet you want to audit.' On timeout, return 'Code too large or analysis timeout. Try scanning smaller files or specific functions.'
• Document result limits: 'Returns up to 100 findings per scan. If more exist, findings are sorted by severity (critical first). Use multiple scans on different code sections if needed.'