skills/mukul975/anthropic-cybersecurity-skills/securing-agentic-ai-tool-invocation/Gen Agent Trust Hub
securing-agentic-ai-tool-invocation
Pass
Audited by Gen Agent Trust Hub on Aug 6, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is a defensive resource that provides templates and instructions for securing AI agent tool calls. No malicious patterns, obfuscation, or data exfiltration attempts were detected.
- [EXTERNAL_DOWNLOADS]: The skill suggests installing standard, well-known Python libraries (nemoguardrails, jsonschema, boto3) required for its security functions. These are sourced from official registries.
- [COMMAND_EXECUTION]: The implementation includes a run_shell control pattern that uses strict allowlisting via JSON schemas to permit only specific, benign commands, demonstrating a secure-by-default approach.
- [CREDENTIALS_UNSAFE]: The skill provides code for managing AWS identities via STS (Security Token Service). It correctly promotes the use of short-lived, scoped session credentials over long-lived secrets, following the principle of least privilege.
Audit Metadata