The Agent Skills Directory

[COMMAND_EXECUTION]: The application's local server (serve.mjs) executes system commands to support a 'Reveal File' feature. It uses the platform's default file manager (explorer.exe, open, or xdg-open) via Node.js spawn. This is implemented safely using argument arrays without a shell, which mitigates command injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied Markdown and text files to build the knowledge graph. This is the primary ingestion surface for indirect prompt injection. The implementation mitigates risks by using DOMPurify to sanitize HTML output in the frontend before rendering note content.
[SAFE]: The local server implementation in serve.mjs is security-conscious, binding strictly to 127.0.0.1, validating Host and Origin headers to prevent cross-site requests, and restricting access to an allowlist of static files and specific API endpoints.

3d-brain