minimax-speech
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and process untrusted user data (scripts and prompts) and send it to external APIs, creating a potential vector for indirect prompt injection.\n
- Ingestion points: User-provided script text for synchronous and asynchronous text-to-audio requests, as well as natural language voice descriptions for the
voice_designprompt field, as documented inSKILL.md.\n - Boundary markers: The instructions do not recommend using delimiters (like XML tags or markdown blocks) or specific instructions to the model to ignore commands embedded within the user-provided text segments.\n
- Capability inventory: The skill facilitates network operations via HTTP and WebSocket connections to the official
api.minimax.ioendpoints for audio generation and file uploads, as detailed in the 'Capability boundary' section ofSKILL.md.\n - Sanitization: There are no guidelines provided for validating, escaping, or sanitizing user-provided text before it is interpolated into the JSON payloads for the speech synthesis APIs.
Audit Metadata