nightly-eval-investigation

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

• Third-party content exposure detected (medium risk: 0.30). At runtime, scripts/investigate.ts syncs and reads first-class text artifacts from the external author-controlled GCS bucket (gs://guidance-evals/…/evals.json and then local context extracted from task/guide/grader files) and incorporates their contents into the generated Markdown/JSON reports.