crewai-multi-agent
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents and enables several data ingestion surfaces which could theoretically be used for indirect prompt injection if processing untrusted content.
- Ingestion points: The skill utilizes tools such as ScrapeWebsiteTool, SerperDevTool, PDFSearchTool, and FileReadTool to bring external content into the agent context (as seen in references/tools.md and SKILL.md).
- Boundary markers: The framework uses structured Agent roles and Task descriptions to separate instructions from data, though explicit 'ignore embedded instructions' warnings are not present in the provided examples.
- Capability inventory: The skill identifies capabilities including CodeInterpreterTool (python execution), FileWriterTool, and MCPServerAdapter across its documentation and guides.
- Sanitization: The skill provides a specific hardening script scripts/safe_arithmetic.py that uses AST-based parsing to evaluate math expressions without using eval(), mitigating risks associated with untrusted mathematical input.
Audit Metadata