together-dedicated-model-inference
Pass
Audited by Gen Agent Trust Hub on Jul 23, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill directs the user to install the official 'together' Python package and CLI tool using standard package managers (uv, pip). These are recognized vendor resources for interacting with the Together AI API.\n- [EXTERNAL_DOWNLOADS]: The skill facilitates the import of model weights and adapters from external repositories including Hugging Face and Amazon S3 (via presigned URLs) as part of its core model management functionality.\n- [PROMPT_INJECTION]: The skill acts as a proxy for user prompts, routing them to inference endpoints, which introduces a surface for indirect prompt injection.\n
- Ingestion points: Untrusted user data enters the agent context through the 'infer' function in 'scripts/deploy_model.py' and through examples in 'references/traffic-routing.md'.\n
- Boundary markers: The provided scripts do not demonstrate the use of explicit boundary markers or delimiters to isolate user-supplied prompt data from the underlying instruction set.\n
- Capability inventory: Across scripts such as 'scripts/deploy_model.py' and 'scripts/upload_custom_model.py', the skill possesses capabilities for network operations, resource deployment, configuration updates, and model deletion.\n
- Sanitization: There is no evidence of content sanitization or validation performed on the user prompts before they are transmitted to the inference API.
Audit Metadata