The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements a behavioral instruction injection in scripts/ask_question.py. The FOLLOW_UP_REMINDER constant appends a directive to tool outputs using high-priority language ("EXTREMELY IMPORTANT") designed to override the agent's default completion logic and force a recursive research loop.
[INDIRECT_PROMPT_INJECTION]: The skill possess an indirect prompt injection attack surface by processing data from external sources.
Ingestion points: Retrieval of synthesized answers and document excerpts from the Google NotebookLM web interface in scripts/ask_question.py.
Boundary markers: The skill does not wrap external content in delimiters or provide "ignore embedded instructions" warnings before passing the data to the agent.
Capability inventory: The skill can execute subprocesses (scripts/run.py), perform local file writes for library management (scripts/notebook_manager.py), and conduct network operations via browser automation.
Sanitization: Content retrieved from NotebookLM is returned to the agent without filtering or escaping of potential instructions embedded in the processed documents.
[EXTERNAL_DOWNLOADS]: During initialization, the skill automatically fetches and installs external resources. scripts/setup_environment.py and scripts/__init__.py utilize pip to install Python dependencies and the patchright library to download Google Chrome or Chromium binaries into a local virtual environment.
[COMMAND_EXECUTION]: The skill relies on the subprocess module to manage its local environment. The scripts/run.py script dynamically constructs and executes shell commands to launch other skill scripts within the isolated Python virtual environment.

notebooklm