mamba-architecture

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.80). The skill's runtime code explicitly loads pretrained models and tokenizers from the public HuggingFace model hub (e.g., MambaLMHeadModel.from_pretrained("state-spaces/mamba-2.8b") and AutoTokenizer.from_pretrained("EleutherAI/gpt-neox-20b")), which are open/public third‑party artifacts that the agent consumes and could carry untrusted/user-provided content.