The Agent Skills Directory

[SAFE]: The skill instructions and reference documents focus purely on the technical task of generating Triton kernel code. The code snippets provided in the templates follow best practices for high-performance computing on NPU hardware, including memory alignment, precision management, and tiling strategies.
[SAFE]: Hardware information is accessed using standard torch.npu and triton.runtime.driver APIs to optimize kernel execution. This is a common and necessary practice for specialized hardware like NPUs to determine core counts for grid calculation.
[SAFE]: No external network requests, suspicious dependencies, or obfuscation techniques were found in the provided files. The skill correctly instructs the agent to avoid high-level PyTorch abstractions in favor of direct Triton kernel implementations for performance reasons.

triton-operator-code-gen