Skill Claude CodeCodex
Generic workflow and guardrails for developing, verifying, benchmarking, and profiling CUTLASS-based CUDA kernels across any operator family, including GEMM, convolution, attention, fused epilogues, and related experiments. Use when creating or modifying CUTLASS experiments, writing .bat-driven build/verify/bench…
Skill Claude CodeCodex
Profile CUDA kernels with Nsight Compute on SM89 (RTX 4090/4070). Use when the user asks to profile a kernel, analyze its performance, diagnose bottlenecks, read an ncu report, or write an optimization plan — including variants like "profile 一下", "为什么慢", "ncu 报告", "性能分析".
Instructions file CodexOpenCode
AGENTS.md instructions for yy6768/tiny-cutlass, covering tiny-cutlass agent 指南, 子工作区索引 and 全局规则.