livestream-to-clips
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, specifically livestream transcripts and audience chat logs, creating a surface for indirect prompt injection. However, the risk is effectively mitigated by the skill's design, which mandates cross-verifying all data against the original source media.
- Ingestion points: Transcripts, interaction data (chat messages), and stream markers ingested from the livestream project files.
- Boundary markers: The skill provides explicit instructions to snap boundaries to clean word, phrase, and shot transitions, reducing the likelihood of accidental instruction obedience.
- Capability inventory: The agent utilizes project editing tools including timeline management (
manage_timelines), visual inspection (view_asset_frames,view_timeline_frames), and video rendering (submit_render_job). - Sanitization: The workflow requires multimodal verification (speech vs visual vs audio) and rigorous QA metrics to ensure that the final output preserves the original intent and factual accuracy of the source recording.
Audit Metadata