video-dashboard

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

• Third-party content exposure detected (high risk: 0.75). SKILL.md requires loading and fetching outsider-authored free text into LLM context via “transcripts in transcripts/{platform}/{id}.txt … aggregate transcripts and frame analysis data into structured analysis JSONs” and later generating topic/sentiment/cross-platform analysis from those transcripts; since transcripts/metadata are untrusted input content (outsider-authored), the workflow implies those strings are ingested for analysis.