ai-patterns-tool-use-patterns
Pass
Audited by Gen Agent Trust Hub on Apr 7, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is documentation-focused, providing provider-agnostic patterns for LLM function calling. It contains no executable malicious scripts or instructions.
- [SAFE]: The content explicitly mandates security controls, including guarding tool loops with maximum step limits to prevent resource exhaustion and validating all tool input arguments before execution.
- [SAFE]: It provides specific 'Defense-in-Depth' patterns for tool execution, such as schema validation using Zod, domain allowlists to prevent data exfiltration, and timeouts to prevent hanging processes.
- [SAFE]: The skill includes patterns for 'Human-in-the-loop' approvals for dangerous operations like fund transfers or database deletions, reducing the risk of autonomous agent accidents.
Audit Metadata