The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a testing harness in scripts/evaluation.py and scripts/connections.py that allows for the execution of local MCP servers. The stdio transport mechanism launches server processes using user-provided commands and arguments, which is a necessary and intended feature for testing local development.
[EXTERNAL_DOWNLOADS]: The implementation guides and documentation reference and fetch content from official sources, including modelcontextprotocol.io and the modelcontextprotocol GitHub organization. These references are used to provide up-to-date protocol specifications and SDK documentation from trusted sources.
[PROMPT_INJECTION]: The evaluation harness (scripts/evaluation.py) processes test questions from XML files, representing a surface for indirect prompt injection. The skill mitigates this risk by using structured XML tags as boundary markers (<summary>, <feedback>, <response>) within its system prompt to maintain context separation.
[DATA_EXFILTRATION]: The evaluation script transmits tool outputs and test results to the Anthropic API to simulate agent behavior. This network activity is a core part of the skill's evaluation functionality and targets a well-known service domain.
[CREDENTIALS_UNSAFE]: The documentation provides instructions for setting environment variables for authentication (e.g., ANTHROPIC_API_KEY). The skill uses standard placeholders and follows best practices for secret management without hardcoding any sensitive data.

mcp-builder