minimax-h3-video
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill requires downloading model files and workflow configurations from external repositories.\n
- Evidence: Model weights are sourced from
huggingface.co/Comfy-Org/MiniMax-H3.\n - Evidence: Workflow templates are fetched from
github.com/Comfy-Org/workflow_templates.\n - Evidence: Community-developed LoRAs are referenced from
lightx2v/Minimax-h3-TurboandKijai/MiniMax-H3_comfy.\n - Evidence: The skill mentions installing a vendor skill via
npx skills addfromgithub.com/MiniMax-AI/MiniMax-H3.\n- [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow that processes user-supplied text for video generation, which is a potential surface for indirect prompt injection.\n - Ingestion points: The prompt widget of the
MiniMaxH3ImageToVideonode in theSKILL.mdnode graph description.\n - Boundary markers: Absent; there are no instructions to delimit user input or prevent the model from following embedded instructions.\n
- Capability inventory: The skill leverages capabilities for model downloading (
download_model), workflow execution (panel_load_workflow), and file system interaction (save_workflow,upload_image).\n - Sanitization: No sanitization logic for input prompts is specified in the skill.
Audit Metadata