kunpengcompute

OmniOperator 列式算子调试专家。Use when Codex needs to 做 Omni 与 Spark 结果对比、关闭或开启列式算子定位问题、分析物理计划、处理 Coredump、排查数据不一致、或阅读 Omni 调试工作流。

Full-repo scan of Velox / Spark SQL / OmniOperator expression support status, generating an ecosystem benchmark report. Triggers: expression coverage research, competitive analysis, Omni missing expression inventory, type matrix analysis.

Full-repo scan of Spark SQL / Gluten / OmniOperator / Velox physical operator support status, generating an ecosystem benchmark report. Triggers: operator coverage research, Omni missing operator inventory, Spark physical operator pushdown status, Gluten…

Execute pytest test scripts via SSH with connection pool. Runs S1 (Native vs Omni comparison) and E2 (execution plan verification) patterns. Sets up core modules, SSH pool, logger, and handles timeout calculation. Use this skill when user mentions 'run…

Generate pytest test scripts for Spark OmniOperator testing. Reads test_cases.json and creates tests/test_*.py files with 1:1 mapping for pytest-xdist parallel execution. Trigger when user mentions '生成测试脚本', 'generate test scripts', 'create test files' in the…

Analyze async-profiler flame graph HTML files to identify Java, Velox C++, and kernel bottlenecks, map hotspots to Velox modules, provide ARM/aarch64 optimization insights, and generate an interactive HTML report with links to original flame graphs.

Use when modifying velox or gluten-velox C++ files containing x86 SIMD intrinsics (_mm_*, _mm256_*, __m128i, __m256i, immintrin headers) to port them to ARM SVE with fixed 256-bit vector length (AWS Graviton3 or similar). Performs closed-loop verification via…

用于分析离在线混部后的在线业务性能劣化、混部干扰、微架构计数器、PMU/topdown 指标、缓存压力、内存带宽、IPC 下降、纯在线基线与混部阶段对比等场景；单纯在线性能优化、无混部对比、只做通用 perf 调优时不要触发。

Use for OmniOperator expression development. Trigger when implementing new vectorized expressions, functions, or operators in OmniOperator framework. This includes scalar functions, aggregate functions, window functions, variadic functions, bitwise…

Full-repo scan of Spark SQL / Gluten / OmniOperator / Velox physical operator support status, generating an ecosystem benchmark report. Triggers: operator coverage research, Omni missing operator inventory, Spark physical operator pushdown status, Gluten…

为Spark表达式生成测试SQL用例。读取包含Spark函数信息的CSV文件，为每个函数生成19种数据类型的测试SQL并输出到新CSV。使用场景：用户需要为Spark函数生成测试SQL、处理CSV中的表达式列表、批量生成SQL测试用例、或涉及OmniOperator/gluten-velox表达式兼容性测试。触发关键词包括但不限于：生成测试SQL、表达式测试、SQL测试用例、Spark函数测试、批量生成SQL、数据类型测试、Spark表达式CSV、OmniOperator测试、gluten-velox测试。即使用户…

Full-repo scan of Velox / Spark SQL / OmniOperator expression support status, generating an ecosystem benchmark report. Triggers: expression coverage research, competitive analysis, Omni missing expression inventory, type matrix analysis.

Install and configure the OpenViking long-term memory plugin for OpenClaw via natural conversation. Once installed, the plugin automatically captures important facts from chats and recalls relevant context before each reply (auto-capture + auto-recall,…

This skill adds data(like resources) to OpenViking Context Database (aka. ov). Use when an agent needs to add files, data from URLs, or external knowledge during interactions. Trigger this tool when 1. is explicitly requested adding files or knowledge; 2.…

Use when the user explicitly types `ov dream` or `ov recall <query>` and the request should be routed to the OpenViking sync/recall CLI instead of handled as normal chat.

Use when benchmarking denoise latency or profiling a diffusion bottleneck in SGLang.

Use when quantizing a diffusion DiT with NVIDIA ModelOpt and making the resulting FP8 or NVFP4 checkpoint loadable, verifiable, and benchmarkable in SGLang Diffusion.

Use when choosing the fastest SGLang Diffusion flags for a model, GPU, and VRAM budget.

内核补丁迁移与冲突解决工作流。用于导出 Git 提交为 patch、检查补丁是否已存在、分析 .rej 冲突、按顺序批量合入 patch_sets，并在每个补丁合入后执行确定性 post-apply 门禁校验与自动复检循环。适用于 Linux 内核补丁迁移、config 文件映射同步、顺序合入多个补丁集、补丁边界漂移检查和补丁合入质量校验。

校验 git format-patch 格式的补丁文件与目标 git 仓库的合入情况。支持两种模式：(1) 应用校验，逐个分析每个补丁的每个 hunk 是否可以应用到目标分支；(2) 已合入差异校验，将本地补丁与目标仓库中已合入提交的 diff 做逐行比较，识别缺失行、额外行和删除侧不一致，并生成详细的 Markdown 格式报告。当用户需要验证补丁是否可以应用到目标仓库、检查本地补丁与上游已合入补丁是否完全一致、或生成补丁状态报告时使用此技能。

Build, test, format, and develop CubeSandbox components. Use when asked to build, compile, run tests, format code, or do development work on this project.