The Agent Skills Directory

[COMMAND_EXECUTION]: The evaluation harness in scripts/evaluation.py and scripts/connections.py allows developers to launch and test local MCP servers using standard input/output (stdio). This process is initiated by the user via command-line arguments and is the intended core functionality of the testing suite.
[EXTERNAL_DOWNLOADS]: The skill provides instructions to fetch official documentation and SDKs from well-known and trusted sources, specifically the official Model Context Protocol website and its verified GitHub repositories.
[CREDENTIALS_UNSAFE]: Security best practices are followed for API authentication. The documentation explicitly directs users to set the ANTHROPIC_API_KEY as an environment variable rather than hardcoding it into scripts or configuration files.
[DATA_EXFILTRATION]: Network activity is limited to authorized operations: sending evaluation questions to the Anthropic API and connecting to developer-specified MCP server endpoints (local or remote).
[PROMPT_INJECTION]: No malicious prompt injection patterns or attempts to bypass safety filters were detected in the instructions or scripts.
[SAFE]: No signs of obfuscation, unauthorized persistence mechanisms, or privilege escalation attempts were found. The skill is consistent with its stated purpose as a developer productivity tool.

mcp-builder