The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (MEDIUM): The install.sh script and requirements.txt specify the installation of notebooklm-py and httpx. These are third-party packages from PyPI that do not originate from the defined trusted organizations, posing a supply chain risk.
[COMMAND_EXECUTION] (MEDIUM): The scripts/upload.py file uses subprocess.run to orchestrate the notebooklm CLI. It passes dynamic strings, such as stock names and notebook titles, as command arguments. While it avoids shell=True, the safety of the CLI tool's argument parsing is unverified.
[REMOTE_CODE_EXECUTION] (MEDIUM): The installation process includes playwright install chromium, which downloads and installs external browser binaries. This execution of remote binary code is a standard but high-privilege operation.
[PROMPT_INJECTION] (LOW): Indirect prompt injection surface detected via financial reports.
Ingestion points: PDF reports downloaded from cninfo.com.cn (referenced in scripts/run.py and download.py).
Boundary markers: None; external PDF files are uploaded directly to the AI service as primary context sources.
Capability inventory: Subprocess execution of notebooklm CLI and file system cleanup (shutil.rmtree) in scripts/upload.py and scripts/run.py.
Sanitization: The skill does not sanitize the contents of the downloaded PDFs before processing.
[DATA_EXFILTRATION] (LOW): The skill is designed to transmit potentially sensitive financial data (A-share reports) to an external cloud service (Google NotebookLM). This is the intended behavior but represents data movement outside the local environment.

cninfo-to-notebooklm