video
Pass
Audited by Gen Agent Trust Hub on Jul 27, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references standard installation procedures for well-known programmatic video tools, including the
hyperframeslibrary and the Remotion framework vianpx create-video. These are established, legitimate tools in the video production ecosystem. - [PROMPT_INJECTION]: The 'Reverse-Engineer a Viral Edit' feature (detailed in
references/edit-anatomy.md) introduces an indirect prompt injection surface by ingesting external data from social media platforms. - Ingestion points: Untrusted content enters the agent's context through
watch-video(extracting transcripts and frames) andsocial-fetch(retrieving post metadata) for analysis of editing patterns. - Boundary markers: While explicit delimiters are not defined in the instructions, the skill implements 'Originality guardrails' to ensure the agent only adopts editing 'grammar' rather than reproducing the source's literal content.
- Capability inventory: The skill uses the analyzed data to generate programmatic code (HTML/CSS for Hyperframes, React for Remotion) or instructions for manual editing tools.
- Sanitization: The workflow focuses on structural decomposition of the video rhythm and style, which naturally filters out non-visual/non-rhythmic instructional data embedded in the source video.
Audit Metadata