The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a Node.js script (share.mjs) to start a local Python-based HTTP server for hosting artifacts. The script interacts with the platform's CLI (amp) and implements robust shell escaping for directory paths to prevent command injection.
[INDIRECT_PROMPT_INJECTION]: The skill processes local project data to generate HTML reports, which creates a potential surface for indirect injection. The risk is mitigated by explicit instructions for the agent to escape HTML entities and by a verification script that warns against loading external resources. Ingestion points include local project documentation and code; capabilities include filesystem writes and starting a local network service; sanitization involves manual escaping guidelines and automated linting for remote asset references.
[DATA_EXFILTRATION]: The skill provides a mechanism to share project information via a restricted portal. It includes safety checks in share.mjs to prevent loading remote assets, and instructions for the agent emphasize that artifact visibility must remain under the user's manual control.
[EXTERNAL_DOWNLOADS]: The skill relies on standard environment tools like Node.js and Python. It does not perform any unverifiable package installations or remote script executions.

creating-artifacts