The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of local processes to run and test MCP servers via the stdio transport protocol.
Evidence: scripts/connections.py uses the official mcp library's stdio_client to spawn subprocesses based on user-provided commands and arguments.
Evidence: scripts/evaluation.py provides a command-line interface allowing users to specify commands for server execution during evaluation tasks.
[EXTERNAL_DOWNLOADS]: The skill references documentation and SDK resources from trusted and well-known sources associated with the Model Context Protocol.
Evidence: SKILL.md contains instructions to fetch protocol specifications from the official modelcontextprotocol.io website.
Evidence: SKILL.md directs the agent to fetch SDK documentation (README files) from the official modelcontextprotocol organization repositories on GitHub.
[PROMPT_INJECTION]: The skill includes an evaluation harness that processes external task data, presenting a potential surface for indirect instructions.
Ingestion points: scripts/evaluation.py reads task data from XML files specified by the user via the eval_file argument.
Boundary markers: The EVALUATION_PROMPT in scripts/evaluation.py uses XML-style tags (<summary>, <feedback>, <response>) to delimit agent output components.
Capability inventory: The scripts/connections.py and scripts/evaluation.py utilities have the capability to execute local commands and perform network requests as part of the MCP client operations.
Sanitization: No explicit sanitization or filtering of the input XML question text is performed before it is interpolated into the agent's message context.

mcp-builder