The Agent Skills Directory

COMMAND_EXECUTION (MEDIUM): The scripts/connections.py file implements a stdio transport that allows launching arbitrary subprocesses via the stdio_client. This capability, while intended for testing local MCP servers, can be abused to execute malicious shell commands if the agent is manipulated into providing a harmful command string.- EXTERNAL_DOWNLOADS (MEDIUM): SKILL.md instructs the agent to fetch and process markdown documentation from modelcontextprotocol.io and raw.githubusercontent.com. Since these domains are not on the explicit trusted list, processing this untrusted content creates a vector for indirect prompt injection.- DATA_EXFILTRATION (LOW): The connection helpers in scripts/connections.py allow the agent to establish outbound network connections via SSE and Streamable HTTP. While no exfiltration logic is hardcoded, the capability to send data to arbitrary URLs is present.

mcp-builder