video-cut

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

• Third-party content exposure detected (high risk: 0.70). Outsider free text can enter the LLM context via work_dir/narration.json (authored by others) when the legacy path maps narration (scripts/cut.py loads narration.json, reads narration text, and writes narration_mapped.json/narration_mapped_lint.json), which downstream LLM steps may ingest; the skill itself does not sanitize narration text.