The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and processes responses from Google NotebookLM (external data) and returns them to the agent context.
Ingestion points: scripts/ask_question.py and scripts/browser_session.py read synthesized response content directly from the browser DOM after querying NotebookLM.
Boundary markers: The skill appends a follow-up reminder string (FOLLOW_UP_REMINDER) to the ingested content but does not utilize formal data delimiters or instructions to ignore embedded commands.
Capability inventory: The skill possesses significant capabilities including command execution via subprocess.run (in run.py and setup_environment.py), local file system access for managing authentication state and notebook metadata, and full network access via an automated browser context.
Sanitization: The external content is retrieved and returned as raw text without explicit sanitization or validation of its contents.
[EXTERNAL_DOWNLOADS]: The skill downloads and installs Python packages (patchright, python-dotenv) from official registries and utilizes the patchright library to download Google Chrome or Chromium binaries to facilitate browser automation.
[COMMAND_EXECUTION]: The skill uses subprocess.run within scripts/run.py and scripts/setup_environment.py to orchestrate the creation of a local virtual environment, install required dependencies, and execute internal Python scripts for task automation.

notebooklm