The Agent Skills Directory

[SAFE]: The skill acts as a comprehensive reference guide for the CrewAI framework, providing documentation and code examples across 13 reference files.
[EXTERNAL_DOWNLOADS]: The documentation includes instructions for installing the crewai library and its tools from PyPI, as well as examples of utilizing Model Context Protocol (MCP) servers from trusted organizations like Anthropic and the MCP project via npx. A utility script (scripts/check-updates.py) is included to query version information from the official PyPI API, which is a trusted service.
[COMMAND_EXECUTION]: Documentation provides examples for using the CrewAI CLI for project scaffolding, training, and testing. It also details the CodeInterpreterTool for executing Python code, which is a standard feature of the framework designed for agent-based computation.
[INDIRECT_PROMPT_INJECTION]: The skill describes a framework architecture that processes external data, establishing an indirect prompt injection attack surface. 1. Ingestion points: Agents ingest data via TextFileKnowledgeSource and ScrapeWebsiteTool (references/07-memory-knowledge.md). 2. Boundary markers: The framework utilizes defined roles and backstories to scope agent behavior (references/01-agents.md). 3. Capability inventory: Agents can utilize tools for file operations (FileWriterTool), code execution (CodeInterpreterTool), and network access (SerperDevTool) as described in references/05-tools.md. 4. Sanitization: The framework supports output validation via Pydantic models and custom guardrail functions (references/02-tasks.md, references/10-structured-output.md).

crewai