crewai-multi-agent

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill provides a "CalculatorTool" example in both SKILL.md and references/tools.md that uses the Python eval() function to process input. Using eval() on unsanitized strings is a significant security risk that can lead to arbitrary code execution if an agent or an external data source provides malicious input.
  • [INDIRECT_PROMPT_INJECTION]: The skill documents agents that ingest data from various untrusted sources, creating a potential attack surface.
  • Ingestion points: Agents are configured to utilize SerperDevTool, ScrapeWebsiteTool, PDFSearchTool, and FileReadTool to incorporate external content into the LLM context (found in SKILL.md and references/tools.md).
  • Boundary markers: The provided examples do not include explicit boundary markers or instructions to ignore embedded commands in the ingested data.
  • Capability inventory: Agents are equipped with powerful tools like CodeInterpreterTool, FileWriterTool, and custom tools (such as the eval-based calculator), which allow for significant actions on the host environment.
  • Sanitization: While one example in references/tools.md demonstrates a basic character check for the calculator, the primary implementation examples lack robust sanitization or validation of external input before it is processed by the agent or its tools.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 07:07 PM
Security Audit — agent-trust-hub — crewai-multi-agent