kelly-agent-eval

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

• Third-party content exposure detected (medium risk: 0.30). At runtime the app’s Busabase provider reads free-text fields (e.g., cases rows’ prompt, baseline-transcript, candidate-transcript, and reviewer decision-note) via records.list/records.get, so outsider-authored content can be ingested from the cases/settings Bases without any prior selection step.