---
name: "ascend-triton-kernel-optimization"
description: "Profile and iteratively optimize a correctness-passed Ascend Triton kernel with explicit NPU baselines, per-shape measurements, UB live-set and physical-core reasoning, MTE/Vector/Scalar bottleneck attribution, one-hypothesis rounds, noise-aware KEEP/DISCARD decisions, and Run Manifest evidence. Use for single-kernel latency or throughput improvement after all planned correctness cases pass. Do not use to create or migrate the first correct kernel, bypass failed validation, assess whole-model serving regressions, attribute model HBM, or diagnose a non-Triton operator."
---

<!-- Generated from .agents/skills/ascend-triton-kernel-optimization/SKILL.md. Do not edit. -->

# ascend-triton-kernel-optimization

Read `.agents/skills/ascend-triton-kernel-optimization/SKILL.md` and only the references needed
for the current task. The canonical skill owns the workflow.