minimax-h3-adviser
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted user data, including draft prompts and media assets, to generate instructions for specialized generation tasks.
- Ingestion points: Processes user-supplied media ideas, draft prompts, and failure reports through the main SKILL.md logic.
- Boundary markers: Implements a human-in-the-loop confirmation step, instructing the agent to summarize the brief and wait for user approval before final prompt production.
- Capability inventory: Possesses the ability to route requests to other local skills (audio, text-to-video, video-editor) and trigger media execution via a ComfyUI integration.
- Sanitization: Employs basic input validation for execution flags (e.g., interpreting boolean values) to prevent malformed control signals.
- [EXTERNAL_DOWNLOADS]: The skill references a modular ecosystem of related tools, such as the h3style plugin and relative skill paths (e.g., ../minimax-h3-audio/SKILL.md). These references point to local components within the same development environment rather than untrusted remote sources.
Audit Metadata