The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run in scripts/__init__.py, scripts/setup_environment.py, and scripts/run.py to automate environment setup, install Python dependencies, and wrap the execution of its own scripts. These calls are used for local management and do not use shell=True.
[EXTERNAL_DOWNLOADS]: The skill downloads Python packages from PyPI and installs Google Chrome via the patchright library during its first-run setup to enable browser automation. These are standard operations for this type of tool.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data directly from Google NotebookLM's user interface, which could contain instructions from uploaded documents or the underlying Gemini model.
Ingestion points: scripts/ask_question.py and scripts/browser_session.py retrieve text content from the NotebookLM response area.
Boundary markers: The skill uses a follow-up reminder but does not provide explicit delimiters or instructions to ignore commands within the retrieved text.
Capability inventory: The agent has access to the local filesystem, network operations, and can execute the skill's Python scripts via a provided wrapper.
Sanitization: The ingested text is extracted from the DOM without explicit sanitization or filtering.
[DATA_EXPOSURE]: Authentication cookies and browser profile data are stored locally in the data/ directory to allow persistent sessions. The skill documentation explicitly warns that this data is sensitive and should not be shared.

notebooklm