[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M78ZTHMPMDE1RJ24MBYACT1K":3,"skill-same:01M78ZTHMPMDE1RJ24MBYACT1K":66},{"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":26,"commitSha":27,"skillDirectory":28,"forksCount":29,"repositoryHtmlUrl":30,"repositoryHomepageUrl":6,"ownerHtmlUrl":31,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":32,"install":33,"resources":45},"01M78ZTHMPMDE1RJ24MBYACT1K","data-preparation",null,"This skill prepares and preflights MapTR nuScenes and Argoverse2 data layouts, annotation contracts, and vector-map schemas before model execution.","VectorSpaceLab\u002FAREX-Skill","VectorSpaceLab","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F180637102?v=4","01M7538C5WPSJGWF7KHR4EFZ9F","01M7538C6R2GFP53NPXTCZNBRT",351,"MIT","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fmaptr\u002Fsub-skills\u002Fdata-preparation\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: data-preparation\ndescription: \"This skill prepares and preflights MapTR nuScenes and Argoverse2\n  data layouts, annotation contracts, and vector-map schemas before model\n  execution.\"\ndisable-model-invocation: true\nmetadata:\n  disco-role: operating\nlicense: MIT\n---\n\n# MapTR data preparation\n\nUse this route before training, evaluation, or visualization. It turns a raw\nnuScenes or Argoverse2 download into a checked layout plan and tells the next\nagent which generated files and vector schema MapTR expects. It does **not**\ndownload data, run a full converter, install packages, or modify a user data\nroot.\n\n## Fast path\n\n1. Choose the dataset and make an explicit absolute data root.\n2. Run the bundled [CPU-only layout checker](scripts\u002Fcheck_dataset_layout.py).\n   It checks directories and file names; it does not import MapTR, `mmcv`,\n   `nuscenes`, `av2`, or `torch`.\n\n   ```bash\n   python \u003Cskill-root>\u002Fscripts\u002Fcheck_dataset_layout.py \\\n     --dataset nuscenes --root \u002Fdata\u002Fmaptr\u002Fnuscenes --canbus-root \u002Fdata\u002Fmaptr \\\n     --check-annotations\n   python \u003Cskill-root>\u002Fscripts\u002Fcheck_dataset_layout.py \\\n     --dataset av2 --root \u002Fdata\u002Fmaptr\u002Fargoverse2\u002Fsensor \\\n     --check-annotations\n   ```\n\n3. Resolve every `ERROR` before starting a converter. Use `--json` when a\n   machine-readable report is needed. `--self-test` exercises valid and\n   deliberately broken tiny fixtures without accessing external data.\n4. Select a converter only after the raw layout passes. Record the exact\n   command, version, output root, and worker count; preserve converter logs.\n5. Re-run with `--check-annotations` after conversion. A passing layout check\n   is not proof that images, calibration, transforms, or map geometry are\n   numerically correct.\n\nThe checker returns exit code 0 only when all requested checks pass. It returns\n1 for a missing required path, wrong placement, malformed map archive, or\nmissing requested annotation file. It never creates a directory in the data\nroot.\n\n## Installation and safety gates\n\nThe documented dependency target is Python 3.8, PyTorch 1.9.1+cu111,\n`mmcv-full==1.4.0`, `mmdet==2.14.0`, `mmsegmentation==0.14.1`, `timm`,\n`mmdetection3d==0.17.2`, `shapely==1.8.5.post1`, and the `av2` API for AV2.\nThe project also documents installing its mmdetection3d component and the\nGeometric Kernel Attention extension. Treat these versions as a compatibility\nplan, not as a verified installation on the current machine. A legacy CUDA\ncompiler, matching `mmcv-full`, and a successful custom extension build are\nnot assumed here.\n\nInstall prerequisites in an isolated environment and prove imports before\nconversion. Do not start with a large converter just to discover an import or\nABI mismatch. See the [workflow reference](references\u002Fworkflows.md) for the\nordered probes and documented commands. Route model and extension choices to\n[model-configuration](..\u002Fmodel-configuration\u002FSKILL.md), and route execution to\n[training-evaluation](..\u002Ftraining-evaluation\u002FSKILL.md).\n\n## Dataset choices and converter contracts\n\n### nuScenes\n\nThe documented raw root is `data\u002Fnuscenes\u002F`, with `maps\u002F`, `samples\u002F`,\n`sweeps\u002F`, and version metadata such as `v1.0-trainval\u002F` and `v1.0-test\u002F`.\nThe CAN bus expansion belongs beside the dataset root: with\n`--root-path \u002Fdata\u002Fmaptr\u002Fnuscenes --canbus \u002Fdata\u002Fmaptr`, the converter looks\nfor `\u002Fdata\u002Fmaptr\u002Fcan_bus\u002F`, **not** `\u002Fdata\u002Fmaptr\u002Fnuscenes\u002Fcan_bus\u002F`.\n\nReference command for the full release:\n\n```bash\npython tools\u002Fcreate_data.py nuscenes \\\n  --root-path \u002Fdata\u002Fmaptr\u002Fnuscenes --out-dir \u002Fdata\u002Fmaptr\u002Fnuscenes \\\n  --extra-tag nuscenes --version v1.0 --canbus \u002Fdata\u002Fmaptr \\\n  --max-sweeps 10\n```\n\n`tools\u002Fcreate_data.py` dispatches `nuscenes` twice for a non-mini version: it\npasses `v1.0-trainval`, then `v1.0-test`, to the nuScenes converter. The\ntrainval call writes `nuscenes_infos_temporal_train.pkl` and\n`nuscenes_infos_temporal_val.pkl`; the test call writes\n`nuscenes_infos_temporal_test.pkl`. A `v1.0-mini` request is handled once and\nuses the mini split. The converter also performs 2D annotation export. MapTR's\nmap dataset consumes the temporal `infos` records and derives local vectors\nfrom the map expansion at runtime.\n\n### Argoverse2 Sensor\n\nThe documented root is `data\u002Fargoverse2\u002Fsensor\u002F` with exactly the split level\n`train\u002F`, `val\u002F`, and `test\u002F`. Each log directory needs the sensor data used by\nthe loader and one map archive matching `map\u002Flog_map_archive_*.json`. The\nconverter is:\n\n```bash\npython tools\u002Fdata_converter\u002Fav2_converter.py \\\n  --data-root \u002Fdata\u002Fmaptr\u002Fargoverse2\u002Fsensor --nproc 8\n```\n\nThe CLI defaults to 64 workers; choose a bounded value for a first run. Its\n`__main__` path processes `train`, `val`, and `test`, removes a fixed list of\nknown failing logs, queries lidar timestamps, chooses the closest image for\nseven ring cameras, and discards samples with unsynchronized lidar\u002Fcamera\npaths. It loads one JSON map archive per log and writes, at the **sensor root**,\n`av2_map_infos_train.pkl`, `av2_map_infos_val.pkl`, and\n`av2_map_infos_test.pkl`. The config normally consumes train and val files and\nuses `load_interval=4` for validation.\n\nFull converters are reference-only for this route: they require downloaded\nprivate data, third-party APIs, map parsing, multiprocessing, and writes to\nlarge output files. The checker is the safe preflight; do not replace it with\na full conversion run on synthetic folders.\n\n## MapTR data contract\n\nFor nuScenes configs, set `data_root` to the nuScenes directory,\n`ann_file` to the matching temporal train\u002Fval pkl, and leave `map_ann_file`\nfor the map evaluation-format JSON (for example,\n`nuscenes_map_anns_val.json`). `map_ann_file` is not a substitute for the\nraw map expansion or the temporal annotation pkl. For AV2, set `data_root` to\nthe sensor directory and use `av2_map_infos_{train,val}.pkl`; the dataset\nloader expects each pkl to contain `samples` and `id2map`, sorted by sample\ntimestamp. `map_ann_file` similarly names the map-format output used by\nvalidation\u002Ftest formatting.\n\nBoth custom datasets expose three map classes in the standard MapTR configs:\n`divider` (label 0), `ped_crossing` (label 1), and `boundary` (label 2).\n`VectorizedLocalMap` obtains nuScenes divider lines, pedestrian-crossing\npolygons, and road\u002Flane contour boundaries from the four supported map\nlocations. `VectorizedAV2LocalMap` obtains divider polylines, pedestrian\npolygons, and drivable-area boundary polygons from the converter's `id2map`.\nUnexpected vector class names are errors; do not silently remap them.\n\nThe configured fixed-point representation is 20 points per ground-truth and\npredicted line. A nonempty `LiDARInstanceLines.fixed_num_sampled_points` has\nshape `[N, 20, 2]`, in local x\u002Fy coordinates, clamped to the patch. Shifted\nrepresentations may have shape `[N, shifts, 20, 2]`; closed polygons receive\ncyclic shifts, open lines receive forward\u002Freverse variants, and padding uses\n`-10000`. Empty vector sets must remain representable; never fabricate a\nplaceholder line. The patch is derived from `pc_range`: height is\n`y_max-y_min`, width is `x_max-x_min`. For the common nuScenes range\n`[-15,-30,-2,15,30,2]`, the local patch is `(60,30)` and x\u002Fy limits are\napproximately `[-15,15]`\u002F`[-30,30]`. AV2's common range reverses those planar\nextents.\n\n## Handoff and routing\n\nReport: dataset and release, absolute roots, raw checks, converter command\nplanned or run, generated pkl names, map-archive count, map classes, fixed\npoint count, and any unresolved dependency or data-quality issue. Do not claim\nnative conversion or model execution from a filesystem-only pass.\n\nFor configuration values and `pc_range` consistency use\n[model-configuration](..\u002Fmodel-configuration\u002FSKILL.md). For train\u002Ftest and\nruntime commands use [training-evaluation](..\u002Ftraining-evaluation\u002FSKILL.md).\nFor map annotation rendering and benchmark plots use\n[visualization-benchmarking](..\u002Fvisualization-benchmarking\u002FSKILL.md).\nUse [workflows](references\u002Fworkflows.md), [data formats](references\u002Fdata-formats.md),\nand [troubleshooting](references\u002Ftroubleshooting.md) for detailed recovery.\n","# MapTR data preparation\n\nUse this route before training, evaluation, or visualization. It turns a raw\nnuScenes or Argoverse2 download into a checked layout plan and tells the next\nagent which generated files and vector schema MapTR expects. It does **not**\ndownload data, run a full converter, install packages, or modify a user data\nroot.\n\n## Fast path\n\n1. Choose the dataset and make an explicit absolute data root.\n2. Run the bundled [CPU-only layout checker](scripts\u002Fcheck_dataset_layout.py).\n   It checks directories and file names; it does not import MapTR, `mmcv`,\n   `nuscenes`, `av2`, or `torch`.\n\n   ```bash\n   python \u003Cskill-root>\u002Fscripts\u002Fcheck_dataset_layout.py \\\n     --dataset nuscenes --root \u002Fdata\u002Fmaptr\u002Fnuscenes --canbus-root \u002Fdata\u002Fmaptr \\\n     --check-annotations\n   python \u003Cskill-root>\u002Fscripts\u002Fcheck_dataset_layout.py \\\n     --dataset av2 --root \u002Fdata\u002Fmaptr\u002Fargoverse2\u002Fsensor \\\n     --check-annotations\n   ```\n\n3. Resolve every `ERROR` before starting a converter. Use `--json` when a\n   machine-readable report is needed. `--self-test` exercises valid and\n   deliberately broken tiny fixtures without accessing external data.\n4. Select a converter only after the raw layout passes. Record the exact\n   command, version, output root, and worker count; preserve converter logs.\n5. Re-run with `--check-annotations` after conversion. A passing layout check\n   is not proof that images, calibration, transforms, or map geometry are\n   numerically correct.\n\nThe checker returns exit code 0 only when all requested checks pass. It returns\n1 for a missing required path, wrong placement, malformed map archive, or\nmissing requested annotation file. It never creates a directory in the data\nroot.\n\n## Installation and safety gates\n\nThe documented dependency target is Python 3.8, PyTorch 1.9.1+cu111,\n`mmcv-full==1.4.0`, `mmdet==2.14.0`, `mmsegmentation==0.14.1`, `timm`,\n`mmdetection3d==0.17.2`, `shapely==1.8.5.post1`, and the `av2` API for AV2.\nThe project also documents installing its mmdetection3d component and the\nGeometric Kernel Attention extension. Treat these versions as a compatibility\nplan, not as a verified installation on the current machine. A legacy CUDA\ncompiler, matching `mmcv-full`, and a successful custom extension build are\nnot assumed here.\n\nInstall prerequisites in an isolated environment and prove imports before\nconversion. Do not start with a large converter just to discover an import or\nABI mismatch. See the [workflow reference](references\u002Fworkflows.md) for the\nordered probes and documented commands. Route model and extension choices to\n[model-configuration](..\u002Fmodel-configuration\u002FSKILL.md), and route execution to\n[training-evaluation](..\u002Ftraining-evaluation\u002FSKILL.md).\n\n## Dataset choices and converter contracts\n\n### nuScenes\n\nThe documented raw root is `data\u002Fnuscenes\u002F`, with `maps\u002F`, `samples\u002F`,\n`sweeps\u002F`, and version metadata such as `v1.0-trainval\u002F` and `v1.0-test\u002F`.\nThe CAN bus expansion belongs beside the dataset root: with\n`--root-path \u002Fdata\u002Fmaptr\u002Fnuscenes --canbus \u002Fdata\u002Fmaptr`, the converter looks\nfor `\u002Fdata\u002Fmaptr\u002Fcan_bus\u002F`, **not** `\u002Fdata\u002Fmaptr\u002Fnuscenes\u002Fcan_bus\u002F`.\n\nReference command for the full release:\n\n```bash\npython tools\u002Fcreate_data.py nuscenes \\\n  --root-path \u002Fdata\u002Fmaptr\u002Fnuscenes --out-dir \u002Fdata\u002Fmaptr\u002Fnuscenes \\\n  --extra-tag nuscenes --version v1.0 --canbus \u002Fdata\u002Fmaptr \\\n  --max-sweeps 10\n```\n\n`tools\u002Fcreate_data.py` dispatches `nuscenes` twice for a non-mini version: it\npasses `v1.0-trainval`, then `v1.0-test`, to the nuScenes converter. The\ntrainval call writes `nuscenes_infos_temporal_train.pkl` and\n`nuscenes_infos_temporal_val.pkl`; the test call writes\n`nuscenes_infos_temporal_test.pkl`. A `v1.0-mini` request is handled once and\nuses the mini split. The converter also performs 2D annotation export. MapTR's\nmap dataset consumes the temporal `infos` records and derives local vectors\nfrom the map expansion at runtime.\n\n### Argoverse2 Sensor\n\nThe documented root is `data\u002Fargoverse2\u002Fsensor\u002F` with exactly the split level\n`train\u002F`, `val\u002F`, and `test\u002F`. Each log directory needs the sensor data used by\nthe loader and one map archive matching `map\u002Flog_map_archive_*.json`. The\nconverter is:\n\n```bash\npython tools\u002Fdata_converter\u002Fav2_converter.py \\\n  --data-root \u002Fdata\u002Fmaptr\u002Fargoverse2\u002Fsensor --nproc 8\n```\n\nThe CLI defaults to 64 workers; choose a bounded value for a first run. Its\n`__main__` path processes `train`, `val`, and `test`, removes a fixed list of\nknown failing logs, queries lidar timestamps, chooses the closest image for\nseven ring cameras, and discards samples with unsynchronized lidar\u002Fcamera\npaths. It loads one JSON map archive per log and writes, at the **sensor root**,\n`av2_map_infos_train.pkl`, `av2_map_infos_val.pkl`, and\n`av2_map_infos_test.pkl`. The config normally consumes train and val files and\nuses `load_interval=4` for validation.\n\nFull converters are reference-only for this route: they require downloaded\nprivate data, third-party APIs, map parsing, multiprocessing, and writes to\nlarge output files. The checker is the safe preflight; do not replace it with\na full conversion run on synthetic folders.\n\n## MapTR data contract\n\nFor nuScenes configs, set `data_root` to the nuScenes directory,\n`ann_file` to the matching temporal train\u002Fval pkl, and leave `map_ann_file`\nfor the map evaluation-format JSON (for example,\n`nuscenes_map_anns_val.json`). `map_ann_file` is not a substitute for the\nraw map expansion or the temporal annotation pkl. For AV2, set `data_root` to\nthe sensor directory and use `av2_map_infos_{train,val}.pkl`; the dataset\nloader expects each pkl to contain `samples` and `id2map`, sorted by sample\ntimestamp. `map_ann_file` similarly names the map-format output used by\nvalidation\u002Ftest formatting.\n\nBoth custom datasets expose three map classes in the standard MapTR configs:\n`divider` (label 0), `ped_crossing` (label 1), and `boundary` (label 2).\n`VectorizedLocalMap` obtains nuScenes divider lines, pedestrian-crossing\npolygons, and road\u002Flane contour boundaries from the four supported map\nlocations. `VectorizedAV2LocalMap` obtains divider polylines, pedestrian\npolygons, and drivable-area boundary polygons from the converter's `id2map`.\nUnexpected vector class names are errors; do not silently remap them.\n\nThe configured fixed-point representation is 20 points per ground-truth and\npredicted line. A nonempty `LiDARInstanceLines.fixed_num_sampled_points` has\nshape `[N, 20, 2]`, in local x\u002Fy coordinates, clamped to the patch. Shifted\nrepresentations may have shape `[N, shifts, 20, 2]`; closed polygons receive\ncyclic shifts, open lines receive forward\u002Freverse variants, and padding uses\n`-10000`. Empty vector sets must remain representable; never fabricate a\nplaceholder line. The patch is derived from `pc_range`: height is\n`y_max-y_min`, width is `x_max-x_min`. For the common nuScenes range\n`[-15,-30,-2,15,30,2]`, the local patch is `(60,30)` and x\u002Fy limits are\napproximately `[-15,15]`\u002F`[-30,30]`. AV2's common range reverses those planar\nextents.\n\n## Handoff and routing\n\nReport: dataset and release, absolute roots, raw checks, converter command\nplanned or run, generated pkl names, map-archive count, map classes, fixed\npoint count, and any unresolved dependency or data-quality issue. Do not claim\nnative conversion or model execution from a filesystem-only pass.\n\nFor configuration values and `pc_range` consistency use\n[model-configuration](..\u002Fmodel-configuration\u002FSKILL.md). For train\u002Ftest and\nruntime commands use [training-evaluation](..\u002Ftraining-evaluation\u002FSKILL.md).\nFor map annotation rendering and benchmark plots use\n[visualization-benchmarking](..\u002Fvisualization-benchmarking\u002FSKILL.md).\nUse [workflows](references\u002Fworkflows.md), [data formats](references\u002Fdata-formats.md),\nand [troubleshooting](references\u002Ftroubleshooting.md) for detailed recovery.\n",{"name":5,"license":14,"metadata":23,"description":7,"disable-model-invocation":25},{"disco-role":24},"operating",true,"main","ac3fe1afa80fb9a09775ecfb2b6cc3ba850a2db6","skills\u002Frepositories\u002Frepo-skills\u002Fmaptr\u002Fsub-skills\u002Fdata-preparation",27,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab","2026-09-03 10:25:21",{"source":8,"skillName":5,"commands":34,"prompt":44,"zipAvailable":25},[35,38,41],{"tool":36,"command":37},"npx","npx skills add VectorSpaceLab\u002FAREX-Skill --skill data-preparation",{"tool":39,"command":40},"bunx","bunx skills add VectorSpaceLab\u002FAREX-Skill --skill data-preparation",{"tool":42,"command":43},"pnpm","pnpm dlx skills add VectorSpaceLab\u002FAREX-Skill --skill data-preparation","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fmaptr\u002Fsub-skills\u002Fdata-preparation\u002FSKILL.md",[46,52,56,60],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"references\u002Fdata-formats.md","REFERENCE","参考文档",6233,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fmaptr\u002Fsub-skills\u002Fdata-preparation\u002Freferences\u002Fdata-formats.md",{"path":53,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":54,"sourceHtmlUrl":55},"references\u002Ftroubleshooting.md",7196,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fmaptr\u002Fsub-skills\u002Fdata-preparation\u002Freferences\u002Ftroubleshooting.md",{"path":57,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":58,"sourceHtmlUrl":59},"references\u002Fworkflows.md",7433,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fmaptr\u002Fsub-skills\u002Fdata-preparation\u002Freferences\u002Fworkflows.md",{"path":61,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":64,"sourceHtmlUrl":65},"scripts\u002Fcheck_dataset_layout.py","SCRIPT","脚本",14042,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fmaptr\u002Fsub-skills\u002Fdata-preparation\u002Fscripts\u002Fcheck_dataset_layout.py",[67,75,82,89,96,103],{"skillNo":68,"name":69,"nameZh":6,"description":70,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":71,"licenseFileUrl":6,"sourceHtmlUrl":72,"categories":73,"keywords":74,"publishedAt":18},"01M799NYSH1R3RVAHSM2D88Y84","passnet-triton-opt","Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block\u002Fgrid\u002Fwarp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.","Apache-2.0","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-triton-opt\u002FSKILL.md",[],[],{"skillNo":76,"name":77,"nameZh":6,"description":78,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":71,"licenseFileUrl":6,"sourceHtmlUrl":79,"categories":80,"keywords":81,"publishedAt":18},"01M799NS5APEC0N1FZ7XZVW6BN","passnet-solve","END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-solve\u002FSKILL.md",[],[],{"skillNo":83,"name":84,"nameZh":6,"description":85,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":71,"licenseFileUrl":6,"sourceHtmlUrl":86,"categories":87,"keywords":88,"publishedAt":18},"01M799NNY9BKSM21V5R7QVTBJX","passnet-skill","PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under .\u002Fpass_dir\u002F, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-skill\u002FSKILL.md",[],[],{"skillNo":90,"name":91,"nameZh":6,"description":92,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":71,"licenseFileUrl":6,"sourceHtmlUrl":93,"categories":94,"keywords":95,"publishedAt":18},"01M799NHH6NP7Z1RK77KGXAN6S","passnet-pattern-fusion","Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()\u002Freplacement_args()\u002Freplacement_func() files, when a pass \"failed to match\", or when deciding how to fuse multiple kernels\u002Fops.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-pattern-fusion\u002FSKILL.md",[],[],{"skillNo":97,"name":98,"nameZh":6,"description":99,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":71,"licenseFileUrl":6,"sourceHtmlUrl":100,"categories":101,"keywords":102,"publishedAt":18},"01M799ND2J4WNC2PGAZ0A06NDF","passnet-orchestrate","Round\u002Fbatch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep\u002Frevert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-orchestrate\u002FSKILL.md",[],[],{"skillNo":104,"name":105,"nameZh":6,"description":106,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":71,"licenseFileUrl":6,"sourceHtmlUrl":107,"categories":108,"keywords":109,"publishedAt":18},"01M799N8SBTMY02ZDG2AZF5ZWE","passnet-feedback","Fast, deterministic feedback for PassNet work: pre-flight pattern\u002Fmatch verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-feedback\u002FSKILL.md",[],[]]