The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes responses from Google NotebookLM (Gemini) and returns them to the agent.
Ingestion points: scripts/ask_question.py and scripts/browser_session.py extract text from the NotebookLM UI using CSS selectors (.to-user-container .message-text-content).
Boundary markers: Absent. The skill does not use delimiters or instructions to prevent the agent from obeying commands embedded in the retrieved text.
Capability inventory: The skill has the ability to execute shell commands via subprocess.run (found in scripts/run.py, scripts/__init__.py, and scripts/setup_environment.py).
Sanitization: Absent. Data retrieved from the browser is returned to the agent without validation or filtering.
[EXTERNAL_DOWNLOADS]: The skill automates the setup of its own environment and dependencies.
Evidence: scripts/setup_environment.py installs Python packages from requirements.txt and downloads Google Chrome using patchright install chrome.
Context: These operations are restricted to the skill's isolated .venv and are part of the documented setup process for the browser automation library used.
[COMMAND_EXECUTION]: The skill uses a wrapper script to manage execution within a virtual environment.
Evidence: scripts/run.py uses subprocess.run to execute other Python scripts in the library (ask_question.py, auth_manager.py, etc.) using the .venv interpreter.
Context: This is used for internal orchestration and environment isolation rather than executing arbitrary user-provided commands.

notebooklm