The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to fetch official protocol documentation and SDK README files from modelcontextprotocol.io and raw.githubusercontent.com/modelcontextprotocol. These are recognized as well-known, official sources for the technology described in the skill.
[COMMAND_EXECUTION]: The evaluation harness in scripts/evaluation.py and the connection utility in scripts/connections.py utilize the mcp SDK to launch local processes via stdio transport. This behavior allows the execution of local commands (e.g., python server.py) to facilitate testing, which is the intended primary purpose of the developer tool.
[DATA_EXFILTRATION]: No malicious data exfiltration was detected. The scripts communicate with the Anthropic API and specified MCP endpoints solely to perform the requested evaluations.
[REMOTE_CODE_EXECUTION]: The skill does not perform remote code execution from untrusted sources. It relies on standard, versioned packages from official registries (PyPI).
[PROMPT_INJECTION]: The EVALUATION_PROMPT in scripts/evaluation.py uses structured XML delimiters (<summary>, <feedback>, <response>) and clear system instructions to maintain control over the agent's behavior during tool-use tasks, reducing the risk of accidental or indirect instruction overrides.

mcp-builder