minimax-h3-director
Pass
Audited by Gen Agent Trust Hub on Aug 24, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill provides a highly structured environment for cinematic preproduction and prompt engineering. Technical analysis of the instructions and referenced files revealed no malicious patterns, backdoors, or security vulnerabilities.
- [DATA_EXFILTRATION]: No hardcoded credentials or sensitive file path accesses were found. Example scripts provided in the reference documentation use placeholders for authentication tokens and target official vendor API endpoints (api.minimax.io, api.minimaxi.com) and CDNs (cdn.hailuoai.com). All network interactions are transparent and legitimate.
- [REMOTE_CODE_EXECUTION]: No instances of remote code execution or unverifiable dependency installations were detected. The skill instructions focus on text transformation and guidance, and included shell scripts are provided solely for user reference regarding API interaction.
- [PROMPT_INJECTION]: The skill implements defensive measures against prompt injection by using specific delimiters for user-supplied dialogue and enforcing strict structural compliance for final output. It explicitly instructs the agent to strip planning metadata and commentary before final assembly, reducing the impact of any instructions embedded in input data.
Audit Metadata