ai-infrastructure-together-ai

Pass

Audited by Gen Agent Trust Hub on Apr 7, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill includes patterns for processing untrusted user input (e.g., userMessage in examples/chat.md and transcript in examples/structured-output.md) within prompts.
  • Ingestion points: User-supplied variables passed to client.chat.completions.create in examples/chat.md and examples/structured-output.md.
  • Boundary markers: The skill encourages using response_format: { type: 'json_schema' } and including the schema in the system prompt to constrain output.
  • Capability inventory: The patterns demonstrate multi-step tool calling and file uploads for model fine-tuning.
  • Sanitization: No explicit input sanitization or delimiter-based isolation of user content is demonstrated in the patterns.
  • [EXTERNAL_DOWNLOADS]: The skill utilizes the together-ai and openai npm packages to interact with well-known AI inference services.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 7, 2026, 01:31 AM