The Agent Skills Directory

Command Execution Surface: The scripts/connections.py and scripts/evaluation.py scripts are designed to launch and connect to MCP servers. When using the stdio transport, the system executes arbitrary commands and arguments provided via the command line. While this is a core requirement for testing local server implementations, users should ensure they only execute trusted code.
Indirect Prompt Injection Surface: The evaluation harness (scripts/evaluation.py) ingests data from external XML files to form test questions.
Ingestion points: The parse_evaluation_file function in scripts/evaluation.py reads questions from a user-supplied XML file.
Boundary markers: The questions are interpolated directly into the message history without specific boundary markers or "ignore embedded instructions" warnings.
Capability inventory: The environment has the capability to execute subprocesses (via connections.py) and perform network operations to the Anthropic API.
Sanitization: There is no explicit sanitization or filtering of the question text before it is sent to the LLM.
Context: This is a potential surface for indirect instructions if a user processes an untrusted evaluation file.
External Resource References: The skill documentation (SKILL.md) references official resources and documentation from the Model Context Protocol project, including their website and GitHub repositories. These are reputable sources providing necessary technical context for the skill's primary purpose.

mcp-builder