edit-davinci-video
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as word-timed transcripts and source screenshots while utilizing high-privilege tools like 'computer use' and the DaVinci Resolve API. This creates a surface for indirect prompt injection where instructions embedded in media could influence agent behavior.
- Ingestion points: The agent reads transcripts (e.g., references/waveform-word-timing.md), video source clips, and screenshots.
- Boundary markers: There are no explicit instructions to the agent to treat external content as untrusted data or to ignore instructions embedded within those files.
- Capability inventory: The skill utilizes 'computer use' for system/GUI automation, the Resolve API for project modification, and performs file system operations in the asset directory.
- Sanitization: No sanitization, validation, or filtering of ingested data content is specified.
- [DYNAMIC_EXECUTION]: The skill provides technical templates for Fusion expressions (e.g., animation curves and wiggle functions) to be defined and applied at runtime within the video editing environment.
- Evidence: SKILL.md and references/approved-look.md provide specific math expressions (e.g., 'u=clamp((time-start)/duration,0,1)...') for the agent to implement in Fusion.
Audit Metadata