The Agent Skills Directory

COMMAND_EXECUTION (HIGH): The MCPConnectionStdio class in scripts/connections.py utilizes mcp.client.stdio.stdio_client to spawn subprocesses. This allows for arbitrary command execution on the host machine if the command or args parameters are not strictly validated.
REMOTE_CODE_EXECUTION (HIGH): The skill's primary function is to run and interact with MCP servers. Executing unverified local binaries or connecting to untrusted remote servers provides a direct path for remote code execution.
EXTERNAL_DOWNLOADS (MEDIUM): Through MCPConnectionSSE and MCPConnectionHTTP, the skill can connect to arbitrary network endpoints. This facilitates Server-Side Request Forgery (SSRF) if the target URL is provided by an untrusted source.
DATA_EXFILTRATION (MEDIUM): The network transport classes (SSE and HTTP) can be abused to transmit sensitive information from the agent's environment to external servers controlled by an attacker.
INDIRECT_PROMPT_INJECTION (HIGH):
Ingestion points: scripts/connections.py in the call_tool method, which ingests raw data from external tools/servers.
Boundary markers: Absent. The skill does not implement delimiters or warnings to treat external tool output as untrusted.
Capability inventory: Includes subprocess execution (stdio_client) and network requests (sse_client, streamablehttp_client).
Sanitization: Absent. Data returned from MCP tools is passed back to the agent without any filtering or validation.

mcp-builder