rwkv-architecture
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is a legitimate technical resource for machine learning practitioners. It provides comprehensive documentation and code snippets for the RWKV architecture, following standard development practices.
- [EXTERNAL_DOWNLOADS]: Dependencies are retrieved from official registries (PyPI), and technical references point to trusted domains such as pytorch.org, arxiv.org, and established open-source repositories on GitHub (github.com/BlinkDL).
- [COMMAND_EXECUTION]: The provided installation steps involve standard 'pip install' commands for well-known and trusted packages including 'torch', 'deepspeed', 'wandb', and 'ninja'.
- [DYNAMIC_EXECUTION]: The code snippets utilize environment variables to enable JIT (Just-In-Time) compilation and CUDA kernel optimization via the 'ninja' build system. This is standard procedure for high-performance neural network architectures.
- [INDIRECT_PROMPT_INJECTION]: The skill documentation describes how text and vision data are processed by the model, which represents a standard input surface for AI agents. As a technical guide, it does not implement specific safety logic, which is expected to be handled at the application level.
- Ingestion points: Text prompts and image tokens are ingested through the model's 'forward' and 'pipeline' methods in SKILL.md and references/rwkv7.md.
- Boundary markers: None explicitly defined in the architectural templates.
- Capability inventory: Local model inference, token generation, and fine-tuning capabilities.
- Sanitization: Not applicable for architectural documentation.
Audit Metadata