huashu-seedance
Pass
Audited by Gen Agent Trust Hub on Oct 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill uses Node.js, Python (OpenCV), and ffmpeg to process video files and generate configuration. These operations are restricted to the skill's local directory and standard temporary directories for rendering purposes.
- [EXTERNAL_DOWNLOADS]: The skill utilizes
npx hyperframesto render Three.js scenes into video. This is an official tool used for deterministic video rendering and falls within standard developer tool usage. - [REMOTE_CODE_EXECUTION]: While it executes a rendering tool via
npx, the execution is scoped to rendering thescene.jsonfile provided by the user or the skill's own templates. No arbitrary remote script piping (e.g., curl|bash) is present. - [INDIRECT_PROMPT_INJECTION]: The skill processes
scene.jsonfiles and user-provided photos. It implements strict validation logic inengine/probe.mjsandengine/make_prompt.mjsto ensure data corresponds to valid geometric parameters, mitigating the risk of instructions embedded in data influencing the agent.
Audit Metadata