The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions in SKILL.md establish a "Follow-Up Mechanism" that directs the agent to override its standard response flow when it identifies a specific string ("EXTREMELY IMPORTANT: Is that ALL you need to know?") within the output from NotebookLM. This creates a surface for indirect prompt injection, as the agent's logic is influenced by content retrieved from an external knowledge base that may contain untrusted data.
Ingestion points: scripts/ask_question.py extracts raw text responses from the NotebookLM web interface.
Boundary markers: The skill does not implement specific delimiters or safety warnings to wrap the untrusted content before presenting it to the agent.
Capability inventory: The skill possesses the capability to execute shell commands and perform full browser automation through scripts/run.py and scripts/ask_question.py.
Sanitization: No sanitization or filtering is performed on the data fetched from the external service before interpolation into the agent's context.
[EXTERNAL_DOWNLOADS]: As part of its automatic setup routine in scripts/setup_environment.py, the skill downloads and installs Google Chrome using the patchright library. This download originates from Google's official repositories and is a required component for the skill's core functionality of automating browser-based research.
[COMMAND_EXECUTION]: The skill utilizes the subprocess.run function across several scripts, including scripts/run.py and scripts/setup_environment.py, to handle its operational environment. These commands are used to create a local virtual environment, install required Python packages, and orchestrate the execution of the skill's internal automation scripts.

notebooklm