The Agent Skills Directory

COMMAND_EXECUTION (SAFE): The MCPConnectionStdio class in scripts/connections.py utilizes mcp.client.stdio.stdio_client to execute local commands. This is a standard and necessary function for connecting to MCP servers via standard input/output streams. As long as the commands and arguments are not derived from untrusted user input without validation, this is a routine capability.
EXTERNAL_DOWNLOADS (SAFE): The MCPConnectionSSE and MCPConnectionHTTP classes in scripts/connections.py facilitate network connections to external URLs using Server-Sent Events and HTTP. This allows the agent to communicate with remote MCP servers. No hardcoded or suspicious URLs were found in the codebase.
[Category 8] INDIRECT_PROMPT_INJECTION (LOW): The skill represents an attack surface for indirect prompt injection because it ingests tool definitions and execution results from external MCP servers.
Ingestion points: Data enters the agent context through list_tools and call_tool methods in scripts/connections.py.
Boundary markers: No explicit boundary markers or instruction-ignoring delimiters are implemented in this transport-layer logic.
Capability inventory: The skill can execute local subprocesses (via stdio_client) and perform network operations (via sse_client and streamablehttp_client).
Sanitization: There is no evidence of sanitization or validation of the data returned by the MCP servers before it is passed back to the agent.

mcp-builder