blender-dreamina-export
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill guides the installation of a Blender add-on from the official Jimeng website (
jimeng.jianying.com), which is a well-known service in the creative software ecosystem. The instructions explicitly warn against using third-party mirrors or unverified files. - [COMMAND_EXECUTION]: The agent is instructed to perform environment changes such as installing software and running render commands within the Blender environment. The workflow mandates obtaining explicit user authorization before performing any such tasks.
- [DATA_EXFILTRATION]: The skill's primary purpose involves uploading video data to the Jimeng platform. It correctly identifies this as a data transfer operation and provides defensive guidance, such as advising the agent not to log sensitive temporary tokens generated during the link creation process.
- [INDIRECT_PROMPT_INJECTION]: The skill has an attack surface as it interacts with external web content and local media files generated or selected by the user.
- Ingestion points: Local video files (MP4/MOV) and the Jimeng web application interface where the agent verifies the status of uploaded references.
- Boundary markers: The skill utilizes a structured verification reporting system (PASS/FAIL/UNVERIFIED) to track the state and integrity of the workflow steps.
- Capability inventory: File system access (reading and rendering videos), plugin management within Blender, and network communication via the browser or plugin interface.
- Sanitization: The instructions focus on manual verification of file integrity (e.g., checking if the video is playable and non-empty) and page status rather than automated sanitization of data content.
Audit Metadata