The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a Node.js script (scripts/herdr-preview.mjs) for managing localhost previews via a local broker socket. The command execution is scoped to managing local dev-server visibility and does not involve arbitrary shell command execution.
[DATA_EXFILTRATION]: No data exfiltration risks were identified. The script is explicitly restricted to communicating with a local Unix domain socket (socketPath) and strictly validates that only localhost or 127.0.0.1 URLs are registered for previewing.
[PROMPT_INJECTION]: The instructions include clear boundaries and warnings against fabricating links or falling back to unauthenticated tunnels. It requires explicit user intent or environment-provided IDs for workspace identification, reducing the risk of redirection or spoofing.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a minimal attack surface for indirect injection. It processes URLs and JSON responses from a local broker. The requestPreview function includes a response size limit (256KB) and uses standard JSON parsing to prevent potential memory or injection issues from the local service output.
[DYNAMIC_EXECUTION]: The script uses standard Node.js modules (node:http, node:fs, etc.) and does not employ eval(), exec(), or dynamic path loading for modules. Logic is entirely static and focused on argument parsing and HTTP-over-Unix-socket communication.

herdr-preview