The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The evaluation harness (scripts/evaluation.py) processes input from XML files (qa_pair elements) which are interpolated into an LLM prompt. This creates a surface where instructions embedded in the evaluation data could influence the behavior of the agent being tested during the evaluation process.
[DYNAMIC_EXECUTION]: The connection utility (scripts/connections.py) and evaluation script support a stdio transport mode that executes local commands to launch MCP servers. This is a functional requirement for testing local server implementations.
[EXTERNAL_DOWNLOADS]: The instructions direct the agent to fetch documentation and SDK information from official sources including modelcontextprotocol.io and the modelcontextprotocol GitHub organization. These are recognized sources for the protocol and are used to gather implementation details.

mcp-builder