The Agent Skills Directory

[SAFE]: No malicious patterns were detected. The skill provides legitimate documentation, architectural best practices, and code snippets for the CrewAI framework. It encourages secure practices such as using environment variables for API keys.
[INDIRECT_PROMPT_INJECTION]: The skill documentation illustrates how to build agents that ingest untrusted data from external sources (e.g., using SerperDevTool for web search or ScrapeWebsiteTool for site content) and perform file-system operations (e.g., FileWriterTool). This configuration establishes a surface for indirect prompt injection, where an agent could potentially be manipulated by instructions embedded in external content. This is identified as an inherent capability of the framework rather than a malicious implementation.
[DYNAMIC_EXECUTION]: The skill mentions the CodeInterpreterTool, which enables agents to execute Python code within a Docker sandbox. This is a standard, isolated execution feature designed for automated coding and data analysis tasks within the framework.
[COMMAND_EXECUTION]: The skill provides instructions for using framework-specific CLI tools for project setup and execution, such as crewai create flow, crewai install, and crewai run. These are legitimate operations for developers using the platform.
[EXTERNAL_DOWNLOADS]: The documentation references official Model Context Protocol (MCP) servers and third-party APIs (e.g., GitHub, Exa, Serper) for extending agent capabilities. These downloads and connections target well-known services and are standard for the AI agent ecosystem.

getting-started