news-scrap-v2

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.90). Yes — the pipeline explicitly fetches and extracts content from public URLs (scripts/extract.py using trafilatura and Playwright), writes those article texts into verified_articles.json and per-article source files (scripts/build_notebook_sources.py), and then uploads/asks NotebookLM to analyze those sources (scripts/notebooklm_gate.py for Q0–Q5), so untrusted third-party webpages are ingested and directly influence agent actions.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The workflow (scripts/extract.py) fetches arbitrary article URLs listed in verified_articles.json (e.g., "https://...") at runtime and those fetched raw texts are uploaded as NotebookLM sources and injected into the model context, thereby directly controlling the agent's prompts/answers.