crewai
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No malicious patterns or security risks were detected in the instructions or code templates provided. The skill primarily serves as an architectural guide for the CrewAI framework.
- [INDIRECT_PROMPT_INJECTION]: The skill describes an architecture for multi-agent systems which naturally ingest untrusted data from tools (e.g., search tools) and other agents. It proactively mitigates this risk by instructing the agent to treat all retrieval outputs as untrusted and to enforce boundary markers.
- [DYNAMIC_EXECUTION]: The skill provides a template for creating custom tools, such as a database query interface. It specifically warns against executing model-supplied SQL directly and provides remediation guidance, recommending a read-only query API with bound parameters and least-privilege connections.
Audit Metadata