The Agent Skills Directory

[COMMAND_EXECUTION]: The skill spawns a local Chrome/Chromium browser process (scripts/auth.ts) to allow the user to log in to their Google account. It uses the Chrome DevTools Protocol to monitor the session and capture authentication cookies.
[EXTERNAL_DOWNLOADS]: The skill downloads generated media and documents from Google's infrastructure (scripts/rpc-client.ts). These operations are restricted to trusted domains associated with Google Services.
[INDIRECT_PROMPT_INJECTION]: The skill provides an interface to import external content from URLs, YouTube, and web research into a notebook (scripts/source-manager.ts, scripts/research-manager.ts). While this data is processed by Google's AI, it represents a surface for indirect prompt injection attacks if malicious content is imported.
Ingestion points: sources add-url, sources add-youtube, research fast/deep commands in scripts/main.ts.
Boundary markers: None explicitly implemented in the skill; behavior depends on NotebookLM's internal parsing.
Capability inventory: File system writes (scripts/artifact-generator.ts), network requests to Google APIs (scripts/rpc-client.ts), and browser spawning (scripts/auth.ts).
Sanitization: Content is forwarded to Google's backend without local sanitization or filtering.

notebooklm