The Agent Skills Directory

[SAFE]: All software installation instructions refer to legitimate, widely-used machine learning packages, including mamba-ssm, torch, transformers, and causal-conv1d. No suspicious or unauthorized package sources were identified.
[SAFE]: The skill references models and documentation from reputable and established organizations in the AI community, such as the state-spaces project on GitHub and HuggingFace, as well as peer-reviewed research papers on ArXiv.
[INDIRECT_PROMPT_INJECTION]: The skill includes code snippets for generating text based on user-provided prompts, which represents a standard interaction surface for language models.
Ingestion points: User-defined prompt strings passed to the model.generate function in SKILL.md.
Boundary markers: None explicitly defined in the provided code examples.
Capability inventory: Execution of Python code, model inference, and package management via pip.
Sanitization: No input sanitization is present in the illustrative code snippets.
This finding represents the intended primary function of the skill and does not indicate an adversarial pattern.

mamba-architecture