[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7982PVGZNP25RMHNN68094Z":3,"skill-same:01M7982PVGZNP25RMHNN68094Z":49},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":23,"commitSha":24,"skillDirectory":25,"forksCount":13,"repositoryHtmlUrl":26,"repositoryHomepageUrl":6,"ownerHtmlUrl":27,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":28,"install":29,"resources":42},"01M7982PVGZNP25RMHNN68094Z","ascend-npu-report",null,"Generate a Chinese optimization report from an existing optimize workspace. Use when an optimize session needs a human-readable summary that synthesizes environment info, round history, performance data, and pattern analysis into a structured report.","Piagari\u002Fhelix","Piagari","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F129916910?v=4","01M7980QB9Y8WK5TSSRJDC7H1X","01M7980QCA2FWSBBSSM695Y10B",0,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FPiagari\u002Fhelix\u002Fblob\u002Fmain\u002Fskills\u002Fcommon\u002Fascend-npu-report\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: ascend-npu-report\ndescription: Generate a Chinese optimization report from an existing optimize workspace. Use when an optimize session needs a human-readable summary that synthesizes environment info, round history, performance data, and pattern analysis into a structured report.\n---\n\n# ascend-npu-report\n\n## Purpose\n\nGenerate a human-readable Chinese optimization report (`report.md`) by reading an\nexisting optimize workspace. This skill is **read-only** — it reads artifacts that\nwere already produced by the `\u003CLanguage>-npu-optimize` workflow (where `\u003CLanguage>` is `triton` or `tilelang`) and synthesizes them\ninto a structured report.\n\n## Workflow\n\n### Step 1: Read hardware environment info\n\nThe invocation prompt includes a `Hardware environment information` section with:\n\n- `target_chip` — target chip series (A3 or A5)\n- `chip_name` — full chip model name\n- `cann_version` — CANN toolkit version\n- `driver_version` — Ascend driver version\n\nIf the hardware information section is absent from the prompt, note \"未记录\" for\nthe verification info section.\n\n### Step 2: Read optimization note\n\nRead `opt-note.md` at the root of the workspace. Parse:\n\n- **Round history** — each `## Round N` section gives the parent, theme, result,\n  best-status, and links to round-level artifacts.\n- **Overall Summary** — contains `Final best round`, `Geomean speedup`,\n  `Validated branches`, `Outcome`, `Next step`, and `Key optimization points`.\n\n### Step 3: Read each round summary\n\nFor each round `N` listed in `opt-note.md`, read `opt-round-N\u002Fsummary.md` for\ndetailed per-round optimization content:\n\n- Optimization points with code\u002FIR context\n- Hypothesis behind the change\n- Correctness and benchmark results\n- Assessment of the outcome\n\n### Step 4: Read best-round operator source and annotate\n\nLocate the best round from `opt-note.md` Overall Summary. Read the corresponding\n`opt-round-N\u002Fopt_\u003Coperator>.py` file for the final optimized source code.\n\nThen, for each round that contributed to the final result, read that round's\n`summary.md` or `attempts.md` to understand what changes were made and why.\nCompare the operator source across rounds to identify the exact code regions\nthat were modified.\n\nIn the final source code shown in the report, annotate each key optimization\nsite with `# -- \u003Ccomment> --` style inline comments that explain:\n\n- Which round introduced the change\n- The optimization theme or technique applied\n- Why the change was made (performance bottleneck, correctness issue, etc.)\n- What effect the change had (latency reduction, memory savings, etc.)\n\nPlace these annotation blocks directly above or adjacent to the modified code\nregions. Each annotation block should reference the relevant round number so\nthe reader can cross-reference with the per-round detail sections.\n\n### Step 5: Collect performance decomposition\n\nRead the `Key optimization points` list from `opt-note.md` Overall Summary.\nEach entry is in the form `\u003Coptimization point>: \u003Cimprovement> (round N)`.\n\nFor latency data, read each round's `round-state.json` or `*_perf.txt` to\nextract the geomean (average) latency.\n\n### Step 6: Render report.md\n\nGenerate `report.md` following the format defined in `references\u002Freport-format.md`.\nWrite the completed report to `report.md` in the workspace root.\n\n## Rules\n\n- Write the report in **Chinese**.\n- Use the exact format from `references\u002Freport-format.md` as the structure template.\n- If the `Key optimization points` list is missing or incomplete from the Overall\n  Summary of `opt-note.md`, display \"未记录\" (not recorded) in the performance\n  decomposition section.\n- If the hardware environment information is not provided in the prompt, note \"未记录\" for the verification info section.\n- Do **not** modify any existing artifacts. Only write `report.md`.\n- In the final code structure section, annotate the source code with `# -- \u003Ccomment> --` inline comments that explain each optimization change, its origin round, reason, and effect. Do not just dump raw uncommented code.\n- Keep the report professional and suitable for external review.\n- Extract latency values from per-round benchmark output or `summary.md`; prefer\n  geomean (average) values when available.\n- Use `round-state.json` `correctness_status` field to determine whether each\n  round's correctness validation passed.\n","# ascend-npu-report\n\n## Purpose\n\nGenerate a human-readable Chinese optimization report (`report.md`) by reading an\nexisting optimize workspace. This skill is **read-only** — it reads artifacts that\nwere already produced by the `\u003CLanguage>-npu-optimize` workflow (where `\u003CLanguage>` is `triton` or `tilelang`) and synthesizes them\ninto a structured report.\n\n## Workflow\n\n### Step 1: Read hardware environment info\n\nThe invocation prompt includes a `Hardware environment information` section with:\n\n- `target_chip` — target chip series (A3 or A5)\n- `chip_name` — full chip model name\n- `cann_version` — CANN toolkit version\n- `driver_version` — Ascend driver version\n\nIf the hardware information section is absent from the prompt, note \"未记录\" for\nthe verification info section.\n\n### Step 2: Read optimization note\n\nRead `opt-note.md` at the root of the workspace. Parse:\n\n- **Round history** — each `## Round N` section gives the parent, theme, result,\n  best-status, and links to round-level artifacts.\n- **Overall Summary** — contains `Final best round`, `Geomean speedup`,\n  `Validated branches`, `Outcome`, `Next step`, and `Key optimization points`.\n\n### Step 3: Read each round summary\n\nFor each round `N` listed in `opt-note.md`, read `opt-round-N\u002Fsummary.md` for\ndetailed per-round optimization content:\n\n- Optimization points with code\u002FIR context\n- Hypothesis behind the change\n- Correctness and benchmark results\n- Assessment of the outcome\n\n### Step 4: Read best-round operator source and annotate\n\nLocate the best round from `opt-note.md` Overall Summary. Read the corresponding\n`opt-round-N\u002Fopt_\u003Coperator>.py` file for the final optimized source code.\n\nThen, for each round that contributed to the final result, read that round's\n`summary.md` or `attempts.md` to understand what changes were made and why.\nCompare the operator source across rounds to identify the exact code regions\nthat were modified.\n\nIn the final source code shown in the report, annotate each key optimization\nsite with `# -- \u003Ccomment> --` style inline comments that explain:\n\n- Which round introduced the change\n- The optimization theme or technique applied\n- Why the change was made (performance bottleneck, correctness issue, etc.)\n- What effect the change had (latency reduction, memory savings, etc.)\n\nPlace these annotation blocks directly above or adjacent to the modified code\nregions. Each annotation block should reference the relevant round number so\nthe reader can cross-reference with the per-round detail sections.\n\n### Step 5: Collect performance decomposition\n\nRead the `Key optimization points` list from `opt-note.md` Overall Summary.\nEach entry is in the form `\u003Coptimization point>: \u003Cimprovement> (round N)`.\n\nFor latency data, read each round's `round-state.json` or `*_perf.txt` to\nextract the geomean (average) latency.\n\n### Step 6: Render report.md\n\nGenerate `report.md` following the format defined in `references\u002Freport-format.md`.\nWrite the completed report to `report.md` in the workspace root.\n\n## Rules\n\n- Write the report in **Chinese**.\n- Use the exact format from `references\u002Freport-format.md` as the structure template.\n- If the `Key optimization points` list is missing or incomplete from the Overall\n  Summary of `opt-note.md`, display \"未记录\" (not recorded) in the performance\n  decomposition section.\n- If the hardware environment information is not provided in the prompt, note \"未记录\" for the verification info section.\n- Do **not** modify any existing artifacts. Only write `report.md`.\n- In the final code structure section, annotate the source code with `# -- \u003Ccomment> --` inline comments that explain each optimization change, its origin round, reason, and effect. Do not just dump raw uncommented code.\n- Keep the report professional and suitable for external review.\n- Extract latency values from per-round benchmark output or `summary.md`; prefer\n  geomean (average) values when available.\n- Use `round-state.json` `correctness_status` field to determine whether each\n  round's correctness validation passed.\n",{"name":5,"description":7},"main","5d6a3de8bf1eb43267adffc8e27a87d85fd81a4b","skills\u002Fcommon\u002Fascend-npu-report","https:\u002F\u002Fgithub.com\u002FPiagari\u002Fhelix","https:\u002F\u002Fgithub.com\u002FPiagari","2026-07-16 20:07:51",{"source":8,"skillName":5,"commands":30,"prompt":40,"zipAvailable":41},[31,34,37],{"tool":32,"command":33},"npx","npx skills add Piagari\u002Fhelix --skill ascend-npu-report",{"tool":35,"command":36},"bunx","bunx skills add Piagari\u002Fhelix --skill ascend-npu-report",{"tool":38,"command":39},"pnpm","pnpm dlx skills add Piagari\u002Fhelix --skill ascend-npu-report","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FPiagari\u002Fhelix\u002Fblob\u002Fmain\u002Fskills\u002Fcommon\u002Fascend-npu-report\u002FSKILL.md",true,[43],{"path":44,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":47,"sourceHtmlUrl":48},"references\u002Freport-format.md","REFERENCE","参考文档",1781,"https:\u002F\u002Fgithub.com\u002FPiagari\u002Fhelix\u002Fblob\u002Fmain\u002Fskills\u002Fcommon\u002Fascend-npu-report\u002Freferences\u002Freport-format.md",[50,57,64,71,78,83],{"skillNo":51,"name":52,"nameZh":6,"description":53,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":54,"categories":55,"keywords":56,"publishedAt":18},"01M7985HE7EHEXX41M79MSZCMJ","triton-npu-repair-guide","Heuristic fixes for Ascend Triton compile\u002FJIT\u002Fkernel errors and numerical precision mismatches when editing or converting the operator. Does not replace ascend-npu-gen-test, ascend-npu-gen-bench, or harness specs.","https:\u002F\u002Fgithub.com\u002FPiagari\u002Fhelix\u002Fblob\u002Fmain\u002Fskills\u002Ftriton\u002Ftriton-npu-repair-guide\u002FSKILL.md",[],[],{"skillNo":58,"name":59,"nameZh":6,"description":60,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":61,"categories":62,"keywords":63,"publishedAt":18},"01M7985DFYH5RRCDFTWAFYDFGA","triton-npu-optimize","Iteratively optimize a Triton Ascend NPU operator with correctness and performance gates. Use for operator optimization tasks that need repeated correctness validation, benchmark validation, multi-round experiment tracking, reusable optimization notes, and profiler-backed performance analysis when benchmark results need deeper explanation.","https:\u002F\u002Fgithub.com\u002FPiagari\u002Fhelix\u002Fblob\u002Fmain\u002Fskills\u002Ftriton\u002Ftriton-npu-optimize\u002FSKILL.md",[],[],{"skillNo":65,"name":66,"nameZh":6,"description":67,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":68,"categories":69,"keywords":70,"publishedAt":18},"01M79858YFWE96X7ZA1BJRVDEX","triton-npu-optimize-knowledge","Generic reference-only optimize knowledge for pattern triage and evidence-backed symptom routing. This skill does not define optimize workflow or own round artifacts.","https:\u002F\u002Fgithub.com\u002FPiagari\u002Fhelix\u002Fblob\u002Fmain\u002Fskills\u002Ftriton\u002Ftriton-npu-optimize-knowledge\u002FSKILL.md",[],[],{"skillNo":72,"name":73,"nameZh":6,"description":74,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":75,"categories":76,"keywords":77,"publishedAt":18},"01M79854BZPJ0XY5PTPQWTCY7Q","triton-npu-optimize-knowledge-v3","Generic reference-only optimize knowledge (v3 working copy) for pattern triage and evidence-backed symptom routing. This skill does not define optimize workflow or own round artifacts; prefer this tree for ongoing knowledge updates.","https:\u002F\u002Fgithub.com\u002FPiagari\u002Fhelix\u002Fblob\u002Fmain\u002Fskills\u002Ftriton\u002Ftriton-npu-optimize-knowledge-v3\u002FSKILL.md",[],[],{"skillNo":79,"name":66,"nameZh":6,"description":67,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"publishedAt":18},"01M79850QM6QZP8VP6ZRQBQHCT","https:\u002F\u002Fgithub.com\u002FPiagari\u002Fhelix\u002Fblob\u002Fmain\u002Fskills\u002Ftriton\u002Ftriton-npu-optimize-knowledge-v2\u002FSKILL.md",[],[],{"skillNo":84,"name":85,"nameZh":6,"description":86,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":87,"categories":88,"keywords":89,"publishedAt":18},"01M7984W6SW3BZM5EXGRJ1VBXY","triton-npu-convert-pytorch-operator","Convert one PyTorch operator into a Triton NPU-backed PyTorch operator, preserve the trailing input-helper block, and validate the converted output through standalone or differential testing.","https:\u002F\u002Fgithub.com\u002FPiagari\u002Fhelix\u002Fblob\u002Fmain\u002Fskills\u002Ftriton\u002Ftriton-npu-convert-pytorch-operator\u002FSKILL.md",[],[]]