The Agent Skills Directory

[COMMAND_EXECUTION]: The utility scripts/connections.py and the evaluation harness scripts/evaluation.py enable the execution of local shell commands to interact with MCP servers via the stdio transport. This functionality is the intended mechanism for testing locally developed servers within the framework.
[INDIRECT_PROMPT_INJECTION]: The test harness scripts/evaluation.py creates a potential surface for indirect prompt injection by parsing external XML files and integrating tool outputs from MCP servers into prompts sent to the LLM.
Ingestion points: The script scripts/evaluation.py parses evaluation questions and answers from a user-specified XML file.
Boundary markers: The system prompt in scripts/evaluation.py uses XML-style tags (<summary>, <feedback>, <response>) to separate model reasoning from the final answer.
Capability inventory: The toolkit includes subprocess execution (scripts/connections.py), network access to the Anthropic API (scripts/evaluation.py), and the ability to connect to remote HTTP/SSE servers (scripts/connections.py).
Sanitization: Data from the evaluation XML and tool results are interpolated into prompts without explicit escaping or sanitization.
[EXTERNAL_DOWNLOADS]: The skill fetches protocol documentation and SDK README files from the official Model Context Protocol website and GitHub repositories. These are well-known and established sources for developers using the protocol.

mcp-builder