kermt-add-cmim-pretrain

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

• Third-party content exposure detected (medium risk: 0.30). Source-independent runtime ingests outsider-authored free text via the user-supplied pretraining corpus CSV (--csv <path>), which the workflow reads during prepare_data.py --mode pretrain and then uses to build smiles_vocab and manifests for the subsequent upgrade and training.