The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes functionality to execute local commands and scripts. Specifically, scripts/connections.py and scripts/evaluation.py use the stdio_client from the MCP SDK to spawn subprocesses representing the servers being tested. This is a primary function of the test harness.
[PROMPT_INJECTION]: The scripts/evaluation.py script is susceptible to indirect prompt injection. It parses external XML files (evaluation.xml) and interpolates the <question> content directly into the user message for the LLM without sanitization or protective boundary markers.
Ingestion points: The parse_evaluation_file function in scripts/evaluation.py reads data from an XML file.
Boundary markers: Absent; the question text is passed directly into the agent context.
Capability inventory: The script has capabilities to execute local commands (via MCP connections) and perform network operations (via the Anthropic client and remote MCP transports).
Sanitization: Absent; the raw text from the external file is utilized directly.
[EXTERNAL_DOWNLOADS]: The documentation and guides within the skill (SKILL.md, reference/mcp_best_practices.md) instruct the agent to fetch the official MCP protocol specification and SDK documentation from the official Model Context Protocol website and GitHub repositories. These references target well-known and trusted sources.

mcp-builder