The Agent Skills Directory

[COMMAND_EXECUTION]: The evaluation harness in scripts/evaluation.py and connection utility in scripts/connections.py allow for arbitrary command execution on the host. The MCPConnectionStdio class enables the execution of subprocesses using a command and arguments provided through the script's command-line interface.\n- [REMOTE_CODE_EXECUTION]: The skill supports interacting with remote MCP servers via SSE and HTTP transports defined in scripts/connections.py. Executing tools provided by untrusted or compromised remote servers can lead to unintended code execution.\n- [PROMPT_INJECTION]: The scripts/evaluation.py script contains a surface for indirect prompt injection. It ingests untrusted data from an XML evaluation file (ingestion point) and interpolates the question content directly into the agent's prompt context without boundary markers or sanitization. The agent has significant capabilities including command execution and remote tool calling, which can be exploited if the XML content is malicious.\n- [EXTERNAL_DOWNLOADS]: The SKILL.md file contains instructions for fetching protocol specifications and SDK documentation from modelcontextprotocol.io and official GitHub repositories, which are well-known and trusted sources for this technology.

mcp-builder