The Agent Skills Directory

[DYNAMIC_EXECUTION]: The references/tool-integration.md file contains a code example for a calculation tool that uses the eval() function on a string parameter. If an agent is provided this tool, a prompt injection or malicious data source could lead to arbitrary code execution within the environment where the crew is running.
[DYNAMIC_EXECUTION]: The scripts/check-setup.py utility uses __import__ to dynamically load modules for verification purposes. While targeting standard libraries, dynamic module loading is a technique that can be used to obfuscate imports or load malicious code if package names are manipulated.
[INDIRECT_PROMPT_INJECTION]: The skill documents the use of agent context passing (e.g., in references/task-design.md and templates/research-crew.py), where the output of one agent is fed directly into the prompt of another. This creates a surface for indirect prompt injection where a compromised or misbehaving agent/tool can influence downstream agents without adequate boundary markers or sanitization described in the templates.

crewai