The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/notebooklm_gate.py executes the notebooklm CLI tool using subprocess.run to manage notebooks, upload sources, and query the model.
Evidence: Multiple calls to subprocess.run are used to interface with the notebooklm executable, passing parameters for source files and queries.
[EXTERNAL_DOWNLOADS]: The script scripts/extract.py fetches article content from user-provided URLs using the trafilatura library and the playwright headless browser.
Evidence: Implementation of trafilatura.fetch_url() and a Playwright-based fallback mechanism to download content from remote websites.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data from external news articles which is later interpreted by an LLM via the NotebookLM service.
Ingestion points: scripts/extract.py reads content from arbitrary external URLs.
Boundary markers: scripts/build_notebook_sources.py places content within structured text blocks, and references/prompts.md provides instructions to the AI to only use the provided sources and avoid speculation.
Capability inventory: The skill's scripts can execute shell commands (notebooklm_gate.py), write to the local filesystem (render_dashboard.py, slide.py), and delete files in specified output directories (cleanup_output.py).
Sanitization: scripts/render_dashboard.py uses HTML escaping for data interpolated into the dashboard template.

news-scrap-v2