nemo-speech-asr-finetune

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

• Third-party content exposure detected (high risk: 0.75). The skill’s runtime workflow ingests outsider-authored free text via the user-provided ASR JSONL manifests (e.g., model.train_ds.manifest_filepath=/data/train.json / model.validation_ds.manifest_filepath=/data/val.json), where the text field contains transcripts written by parties other than the operating user; this text is then read by NeMo/Lhotse dataloaders and passed into evaluation/training code paths that feed it into the LLM context for any text-processing/decoding steps that use it as readable prose.