llamaguard
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted user-generated content for moderation purposes.
- Ingestion points: The moderate, check_input, and moderate_vllm functions in SKILL.md accept arbitrary user text.
- Boundary markers: The skill utilizes tokenizer.apply_chat_template, which provides standard structural delimiters for chat conversations.
- Capability inventory: Operations are limited to LLM inference (model.generate) and basic string parsing; no file-system writes or shell execution capabilities are exposed to the input.
- Sanitization: The skill relies on LlamaGuard's internal classification logic to identify and flag malicious content.
Audit Metadata