skill-opt-lite

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

• Third-party content exposure detected (medium risk: 0.30). The training loop ingests outsider-authored free text via the user-supplied tasks/train.json and tasks/valid.json fields (especially instruction/expected_outcome/test_command) at runtime in both ROLLOUT and GATE validation, before any specific item selection in the external text source is enforced.