replicate
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes the OpenFilm platform's command-line tools, specifically
npx openfilm lookfor video frame comparison andnpx openfilm getfor asset management. These are integrated vendor tools required for the skill's primary functionality. - [EXTERNAL_DOWNLOADS]: The instructions involve downloading media assets (images, video, music, and voice) via the
npx openfilm getcommand. The skill also allows for the ingestion of reference videos from user-provided files or external links. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of reference videos, which represents an indirect prompt injection surface.
- Ingestion points: Reference video files or URLs provided by the user in the
references/directory or as links. - Boundary markers: The skill does not explicitly define markers to ignore instructions that might be theoretically embedded in video metadata.
- Capability inventory: The agent can execute shell commands via
npx, fetch network assets, and perform file-system operations withinassets/andreferences/directories. - Sanitization: No explicit sanitization or validation logic for the content of reference videos is described.
Audit Metadata