ga-taxonomy

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

• Third-party content exposure detected (medium risk: 0.30). In SKILL.md → Scenario A/steps 1-2, the workflow ingests outsider-authored free text by reading a user-supplied Google Sheets URL (or local xlsx/csv path) and parsing all cell text via parse_taxonomy.py (and sheets_reader.py for URLs) into the LLM at runtime.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill fetches a Google Sheets URL at runtime (e.g. "https://docs.google.com/spreadsheets/d//edit") via sheets_reader.fetch_sheets(args.source) (called from scripts/parse_taxonomy.py), and the fetched spreadsheet contents directly control the generated taxonomy and user-facing instructions.