The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides educational examples for implementing the stdio transport layer using Python's subprocess module. These examples are documented as best practices for local process communication and include instructions for running servers with least-privilege isolation.
[DATA_EXFILTRATION]: Documentation describes remote transport options like SSE and WebSockets. The skill includes specific security references that teach how to implement authentication (API keys, OAuth 2.0, mTLS) and rate limiting to protect against unauthorized data access or service abuse.
[PROMPT_INJECTION]: The skill contains various prompt templates for agent-tool interaction. It explicitly addresses indirect prompt injection risks in its 'Security Patterns' reference, advocating for Pydantic-based schema validation and input sanitization to prevent malicious user input from influencing agent behavior.
[SAFE]: The skill acts as a security-centric knowledge base. It identifies various anti-patterns (e.g., monolithic servers, description-less tools, secret leakage) and provides corrective 'Production Ready' patterns to ensure safe deployment of MCP-based systems.

ai-mcp-patterns