ingest-training-datalake

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

• Third-party content exposure detected (high risk: 0.90). The skill reads and acts on open/public URLs from candidate manifest files and then invokes the external fetcher (see datalake_planning._read_candidate_urls and datalake_execution._run_fetcher_manifest which runs the fetcher in FETCHER_DIR), and it subsequently parses fetcher outputs/consumer_summary.json in datalake_coverage to drive selection and further fetch decisions—so untrusted third‑party content from arbitrary URLs is ingested and can materially influence next actions.