The Agent Skills Directory

[COMMAND_EXECUTION]: Multiple scripts including run.py, setup_environment.py, and source_filter.py utilize subprocess.run to execute shell commands for creating virtual environments, installing dependencies, and invoking external CLI tools like Gemini and Claude.
[EXTERNAL_DOWNLOADS]: The skill automatically fetches Python packages from PyPI and browser binaries from external servers during its initialization phase to set up the execution environment.
[REMOTE_CODE_EXECUTION]: The skill downloads and executes third-party browser automation libraries and drivers. The installation process for patchright and Google Chrome involves executing code and binaries from remote sources during setup.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. 1. Ingestion points: ask_question.py and source_extractor.py retrieve text from NotebookLM notebooks which can contain untrusted data from PDFs, websites, and user uploads. 2. Boundary markers: The skill does not implement delimiters or specific instructions to prevent the agent from following instructions embedded in the notebook data. 3. Capability inventory: The skill can execute shell commands via the run.py wrapper, perform file I/O for sync operations, and manage browser authentication state. 4. Sanitization: Extracted content from the browser is not sanitized or validated before being passed into the agent context.

notebooklm