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
Risk Level
SAFE
Analyzed
Apr 7, 2026, 01:31 AM