The Agent Skills Directory

[COMMAND_EXECUTION]: The skill employs a wrapper script, scripts/run.py, which uses subprocess.run to execute internal Python scripts. This architecture ensures all skill operations run within an isolated virtual environment. Additional subprocess calls in scripts/setup_environment.py and scripts/__init__.py manage the automated installation of dependencies and browser binaries.
[EXTERNAL_DOWNLOADS]: Upon first execution, the skill's setup scripts automatically download and install required Python packages including patchright (a Playwright fork for anti-detection) and the Google Chrome/Chromium browser engines to support its automation functionality.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content retrieved from Google NotebookLM via browser automation. The agent synthesizes this data to provide answers. While no explicit boundary markers or content sanitization are implemented in the scripts, the risk is inherent to the skill's primary purpose of document-based research and is mitigated by the agent's contextual synthesis.

notebooklm