[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7900AT6FFCC04WPGM24X6DA":3,"skill-same:01M7900AT6FFCC04WPGM24X6DA":75},{"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":27,"commitSha":28,"skillDirectory":29,"forksCount":30,"repositoryHtmlUrl":31,"repositoryHomepageUrl":6,"ownerHtmlUrl":32,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":33,"install":34,"resources":46},"01M7900AT6FFCC04WPGM24X6DA","data-and-preprocessing",null,"Prepare, inspect, and safely adapt MedicalDetectionToolkit array datasets, labels, preprocessing, patching, and batchgenerators pipelines for toy, LIDC, and PET-CT-style experiments.","VectorSpaceLab\u002FAREX-Skill","VectorSpaceLab","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F180637102?v=4","01M7538C5WPSJGWF7KHR4EFZ9F","01M7538C6R2GFP53NPXTCZNBRT",351,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fmedical-detection-toolkit\u002Fsub-skills\u002Fdata-and-preprocessing\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: data-and-preprocessing\ndescription: \"Prepare, inspect, and safely adapt MedicalDetectionToolkit array\n  datasets, labels, preprocessing, patching, and batchgenerators pipelines for\n  toy, LIDC, and PET-CT-style experiments.\"\nmetadata:\n  disco-role: operating\ndisable-model-invocation: true\nlicense: Apache 2.0\n---\n\n# Data and preprocessing\n\nUse this sub-skill when a Researcher must turn a preprocessed medical-image case\ninto the dictionaries and arrays expected by MedicalDetectionToolkit (MDT), or\nwhen adapting one of the example loaders without silently changing axes, label\nsemantics, or patch coordinates. This is an operating reference, not a data\nacquisition recipe. The repository is legacy and explicitly no longer\nmaintained; reproduce the contracts below and verify the installed environment\nbefore changing them.\n\n## Operating contract\n\nGiven an experiment config `cf`, a logger, a manifest\u002Fdata directory, and\npreprocessed image\u002Fsegmentation arrays, produce one of these observable\ncontracts:\n\n- `load_dataset` returns an `OrderedDict` keyed by `pid`. Each value contains\n  paths under `data` and `seg`, the `pid`, and a `class_target` list. LIDC and\n  PET-CT additionally carry `fg_slices`; toy uses one array path for a packed\n  two-plane `[image, segmentation]` file and does not carry foreground-slice\n  metadata.\n- A training batch contains `data`, `seg`, `pid`, and `class_target`. `data` is\n  batched channels-first and `seg` has a singleton segmentation-channel axis.\n  The spatial axes are `(x, y)` for 2D or `(x, y, z)` for 3D at the point the\n  network sees them; the exact loader transposes are documented in\n  [data-formats](references\u002Fdata-formats.md).\n- After `ConvertSegToBoundingBoxCoordinates`, a batch also contains\n  `bb_target` and `roi_labels`. Patient iterators preserve patient-level\n  targets as `patient_bb_target` and `patient_roi_labels`, plus\n  `original_img_shape`; tiled batches additionally carry `patch_crop_coords`.\n\nBefore loading a real case, validate a portable image\u002Fsegmentation pair with\n[`validate_preprocessed_case.py`](scripts\u002Fvalidate_preprocessed_case.py). The\nvalidator accepts only `.npy`\u002F`.npz` arrays and optional JSON metadata, has hard\nsize\u002Fvoxel bounds, performs no repository imports, and never downloads or\nmodifies data. Its JSON metadata format is described in\n[troubleshooting](references\u002Ftroubleshooting.md).\n\n## Required workflow\n\n1. **Select the evidence-backed family.** Use the toy loader only for small 2D\n   onboarding data; use LIDC for per-patient CT volumes with instance-valued\n   ROIs and optional 2D\u002F3D operation; use PET-CT for the two-channel CT\u002FPET\n   classification-style example with a binary foreground segmentation. Read\n   [preprocessing-workflows](references\u002Fpreprocessing-workflows.md) before\n   changing normalization, resampling, or label creation.\n2. **Freeze the manifest contract.** The configured `input_df_name` is normally\n   `info_df.pickle`. Its rows must agree with the filenames derived from `pid`.\n   Do not infer a different axis order from a filename. Check `class_target`,\n   `fg_slices` (where present), and spacing\u002Fshape metadata before training.\n3. **Validate arrays and metadata.** Use the bundled validator on a copied or\n   synthetic case. Never pass arbitrary pickle files to a new utility. The\n   source loaders read pandas pickle manifests, so treat them as trusted local\n   inputs only and keep their paths outside portable skills.\n4. **Construct the generator.** Instantiate the appropriate `BatchGenerator`,\n   then use `create_data_gen_pipeline`. Training composes optional mirror and\n   `SpatialTransform`; validation\u002Ftesting uses `CenterCropTransform`; both then\n   call `ConvertSegToBoundingBoxCoordinates`. Exact flags and version-sensitive\n   signatures are in [data-loader-contracts](references\u002Fdata-loader-contracts.md).\n5. **Check labels after every geometric transform.** Image and segmentation\n   must be transformed together. Use interpolation order 0 \u002F background 0 for\n   segmentation. Confirm the resulting `bb_target` coordinates fit the\n   post-transform patch and that class labels still correspond to ROI ids.\n6. **Record reversible customization.** Keep a small manifest schema note and\n   a before\u002Fafter shape example for every custom loader. If a change affects\n   axes, label mode, patch size, or channel count, update config and tests\n   together rather than patching a downstream model.\n\n## Label and geometry rules\n\n- Background is label `0`; positive integer ROI ids are foreground. In LIDC,\n  each nonzero ROI id is an individual lesion and `class_target` is aligned to\n  those ids after the preprocessing script's `rix` assignment. Do not relabel\n  ROIs without updating the aligned class vector.\n- A **pixelwise instance map** has increasing ROI ids and needs\n  `get_rois_from_seg_flag=False`. A **binary map** has foreground `1` and does\n  not identify individual lesions; use `get_rois_from_seg_flag=True` so the\n  batchgenerators converter performs connected-component labeling. The PET-CT\n  pipeline uses this binary mode. A box-only source must first be represented\n  as a pseudo-mask if it is to go through the same spatial augmentation path.\n- `class_specific_seg_flag` controls whether converted segmentation targets\n  preserve class-specific channels\u002Flabels. It is not a substitute for making\n  a binary mask into instance ids. Check `cf.head_classes` and\n  `cf.num_seg_classes` when changing it.\n- The loader may return no foreground in a sampled patch even when the patient\n  has lesions. Do not treat an empty patch as a corrupt case; distinguish it\n  from an invalid all-negative patient or an out-of-range `fg_slices` entry.\n- Coordinates are local to the current patch. Patient-level targets must be\n  retained separately when tiling so predictions can be mapped\u002Fconsolidated.\n\n## Dimensions, channels, and patching\n\n- Toy is 2D only: its packed case stores image at index `0` and segmentation at\n  index `1`; `BatchGenerator` adds a batch axis and a singleton data\u002Fseg channel\n  axis. It uses `320 x 320` pre-crop and patch settings in the supplied config.\n- LIDC preprocessing writes `(z, y, x)` image and ROI arrays. The loader\n  transposes them to `(y, x, z)` before sampling; 2D mode selects a z slice,\n  while 3D mode retains the volume. The supplied config uses `300 x 300`\n  pre-crops \u002F `288 x 288` 2D patches, or `156 x 156 x 96` pre-crops \u002F\n  `128 x 128 x 64` 3D patches, and enables 2D-to-3D prediction merging.\n- PET-CT preprocessing writes a two-channel image `(c, z, y, x)` and a binary\n  `(z, y, x)` mask. The loader transposes the image to `(c, y, x, z)` and\n  samples 2D slices or 3D volumes. The supplied config is 3D with\n  `280 x 280 x 48` pre-crops and `192 x 192 x 32` patches, two channels, and\n  2D-to-3D merging enabled if 2D is selected.\n- If an input is smaller than the configured pre-crop or patient patch, the\n  loaders call `pad_nd_image`; padding is centered and defaults to constant\n  zero in the shown paths. If larger, training samples a pre-crop (with\n  foreground-biased sampling) and patient inference creates a coordinate grid\n  using `get_patch_crop_coords`. A patch size of `1` in z means one patch per\n  slice, not a three-dimensional slab.\n- `n_3D_context` adds neighboring slices as channels in 2D mode. It changes\n  `cf.n_channels` and requires matching model input configuration; padding is\n  applied at the volume boundary. Do not enable it without checking channel\n  count and the context coordinate adjustment.\n\n## Safety boundaries\n\n- The example raw data roots, mounted paths, `rsync` copies, and cloud\u002Fserver\n  branches are environment-specific. LIDC-IDRI, PET\u002FCT\u002FTNM, and the toy data\n  directories are external datasets; the PET\u002FCT paths and clinical labels may\n  be private. Do not claim access, redistribute them, or run acquisition from\n  this skill. The supplied examples are evidence of schema and behavior only.\n- `experiments\u002Flidc_exp\u002Fpack_dataset.py` and\n  `utils\u002Fdataloader_utils.py::unpack_dataset`\u002F`delete_npy` are storage helpers,\n  not safe validation steps. Packing\u002Funpacking can create many files and\n  `delete_npy` is destructive. Treat pack, delete, `rsync`, multiprocessing\n  preprocessing, and raw-data conversion as **non-runnable** unless a user\n  explicitly reviews paths, backups, and a bounded dry run outside this skill.\n- Do not run native final cases, download data, or import this operating graph\n  during construction. Use signatures\u002Fimport checks and bounded synthetic\n  validator fixtures only.\n\n## References and bundled script\n\n- [Data formats and array axes](references\u002Fdata-formats.md)\n- [Data-loader and batchgenerators contracts](references\u002Fdata-loader-contracts.md)\n- [Preprocessing workflows](references\u002Fpreprocessing-workflows.md)\n- [Data utilities: padding, tiling, balancing, packing boundaries](references\u002Fdata-utilities.md)\n- [Troubleshooting and safe customization](references\u002Ftroubleshooting.md)\n- [Standalone preprocessed-case validator](scripts\u002Fvalidate_preprocessed_case.py)\n\nAll references link back here and to the other references so the graph remains\nusable when a Researcher opens a leaf directly. Source evidence is the checked-\nout `README.md`, the three experiment `configs.py`\u002F`data_loader.py` pairs,\nLIDC `preprocessing.py`\u002F`pack_dataset.py`, PET-CT `preprocessing.py`, and\n`utils\u002Fdataloader_utils.py`.\n\n## Verification gate\n\nAccept a customized data path only when: (a) the three example loaders import\nwith the documented loader dependency\u002FAPI variant; (b) the validator passes a\nbounded synthetic valid case and rejects mismatched shapes, non-integral labels,\nor oversized input;\n(c) the manifest-to-file naming and `class_target`\u002FROI semantics are recorded;\nand (d) no native dataset case or destructive helper was run. Unresolved\nbackend, private-data, axis, or label ambiguities stay explicit in the handoff.\n","# Data and preprocessing\n\nUse this sub-skill when a Researcher must turn a preprocessed medical-image case\ninto the dictionaries and arrays expected by MedicalDetectionToolkit (MDT), or\nwhen adapting one of the example loaders without silently changing axes, label\nsemantics, or patch coordinates. This is an operating reference, not a data\nacquisition recipe. The repository is legacy and explicitly no longer\nmaintained; reproduce the contracts below and verify the installed environment\nbefore changing them.\n\n## Operating contract\n\nGiven an experiment config `cf`, a logger, a manifest\u002Fdata directory, and\npreprocessed image\u002Fsegmentation arrays, produce one of these observable\ncontracts:\n\n- `load_dataset` returns an `OrderedDict` keyed by `pid`. Each value contains\n  paths under `data` and `seg`, the `pid`, and a `class_target` list. LIDC and\n  PET-CT additionally carry `fg_slices`; toy uses one array path for a packed\n  two-plane `[image, segmentation]` file and does not carry foreground-slice\n  metadata.\n- A training batch contains `data`, `seg`, `pid`, and `class_target`. `data` is\n  batched channels-first and `seg` has a singleton segmentation-channel axis.\n  The spatial axes are `(x, y)` for 2D or `(x, y, z)` for 3D at the point the\n  network sees them; the exact loader transposes are documented in\n  [data-formats](references\u002Fdata-formats.md).\n- After `ConvertSegToBoundingBoxCoordinates`, a batch also contains\n  `bb_target` and `roi_labels`. Patient iterators preserve patient-level\n  targets as `patient_bb_target` and `patient_roi_labels`, plus\n  `original_img_shape`; tiled batches additionally carry `patch_crop_coords`.\n\nBefore loading a real case, validate a portable image\u002Fsegmentation pair with\n[`validate_preprocessed_case.py`](scripts\u002Fvalidate_preprocessed_case.py). The\nvalidator accepts only `.npy`\u002F`.npz` arrays and optional JSON metadata, has hard\nsize\u002Fvoxel bounds, performs no repository imports, and never downloads or\nmodifies data. Its JSON metadata format is described in\n[troubleshooting](references\u002Ftroubleshooting.md).\n\n## Required workflow\n\n1. **Select the evidence-backed family.** Use the toy loader only for small 2D\n   onboarding data; use LIDC for per-patient CT volumes with instance-valued\n   ROIs and optional 2D\u002F3D operation; use PET-CT for the two-channel CT\u002FPET\n   classification-style example with a binary foreground segmentation. Read\n   [preprocessing-workflows](references\u002Fpreprocessing-workflows.md) before\n   changing normalization, resampling, or label creation.\n2. **Freeze the manifest contract.** The configured `input_df_name` is normally\n   `info_df.pickle`. Its rows must agree with the filenames derived from `pid`.\n   Do not infer a different axis order from a filename. Check `class_target`,\n   `fg_slices` (where present), and spacing\u002Fshape metadata before training.\n3. **Validate arrays and metadata.** Use the bundled validator on a copied or\n   synthetic case. Never pass arbitrary pickle files to a new utility. The\n   source loaders read pandas pickle manifests, so treat them as trusted local\n   inputs only and keep their paths outside portable skills.\n4. **Construct the generator.** Instantiate the appropriate `BatchGenerator`,\n   then use `create_data_gen_pipeline`. Training composes optional mirror and\n   `SpatialTransform`; validation\u002Ftesting uses `CenterCropTransform`; both then\n   call `ConvertSegToBoundingBoxCoordinates`. Exact flags and version-sensitive\n   signatures are in [data-loader-contracts](references\u002Fdata-loader-contracts.md).\n5. **Check labels after every geometric transform.** Image and segmentation\n   must be transformed together. Use interpolation order 0 \u002F background 0 for\n   segmentation. Confirm the resulting `bb_target` coordinates fit the\n   post-transform patch and that class labels still correspond to ROI ids.\n6. **Record reversible customization.** Keep a small manifest schema note and\n   a before\u002Fafter shape example for every custom loader. If a change affects\n   axes, label mode, patch size, or channel count, update config and tests\n   together rather than patching a downstream model.\n\n## Label and geometry rules\n\n- Background is label `0`; positive integer ROI ids are foreground. In LIDC,\n  each nonzero ROI id is an individual lesion and `class_target` is aligned to\n  those ids after the preprocessing script's `rix` assignment. Do not relabel\n  ROIs without updating the aligned class vector.\n- A **pixelwise instance map** has increasing ROI ids and needs\n  `get_rois_from_seg_flag=False`. A **binary map** has foreground `1` and does\n  not identify individual lesions; use `get_rois_from_seg_flag=True` so the\n  batchgenerators converter performs connected-component labeling. The PET-CT\n  pipeline uses this binary mode. A box-only source must first be represented\n  as a pseudo-mask if it is to go through the same spatial augmentation path.\n- `class_specific_seg_flag` controls whether converted segmentation targets\n  preserve class-specific channels\u002Flabels. It is not a substitute for making\n  a binary mask into instance ids. Check `cf.head_classes` and\n  `cf.num_seg_classes` when changing it.\n- The loader may return no foreground in a sampled patch even when the patient\n  has lesions. Do not treat an empty patch as a corrupt case; distinguish it\n  from an invalid all-negative patient or an out-of-range `fg_slices` entry.\n- Coordinates are local to the current patch. Patient-level targets must be\n  retained separately when tiling so predictions can be mapped\u002Fconsolidated.\n\n## Dimensions, channels, and patching\n\n- Toy is 2D only: its packed case stores image at index `0` and segmentation at\n  index `1`; `BatchGenerator` adds a batch axis and a singleton data\u002Fseg channel\n  axis. It uses `320 x 320` pre-crop and patch settings in the supplied config.\n- LIDC preprocessing writes `(z, y, x)` image and ROI arrays. The loader\n  transposes them to `(y, x, z)` before sampling; 2D mode selects a z slice,\n  while 3D mode retains the volume. The supplied config uses `300 x 300`\n  pre-crops \u002F `288 x 288` 2D patches, or `156 x 156 x 96` pre-crops \u002F\n  `128 x 128 x 64` 3D patches, and enables 2D-to-3D prediction merging.\n- PET-CT preprocessing writes a two-channel image `(c, z, y, x)` and a binary\n  `(z, y, x)` mask. The loader transposes the image to `(c, y, x, z)` and\n  samples 2D slices or 3D volumes. The supplied config is 3D with\n  `280 x 280 x 48` pre-crops and `192 x 192 x 32` patches, two channels, and\n  2D-to-3D merging enabled if 2D is selected.\n- If an input is smaller than the configured pre-crop or patient patch, the\n  loaders call `pad_nd_image`; padding is centered and defaults to constant\n  zero in the shown paths. If larger, training samples a pre-crop (with\n  foreground-biased sampling) and patient inference creates a coordinate grid\n  using `get_patch_crop_coords`. A patch size of `1` in z means one patch per\n  slice, not a three-dimensional slab.\n- `n_3D_context` adds neighboring slices as channels in 2D mode. It changes\n  `cf.n_channels` and requires matching model input configuration; padding is\n  applied at the volume boundary. Do not enable it without checking channel\n  count and the context coordinate adjustment.\n\n## Safety boundaries\n\n- The example raw data roots, mounted paths, `rsync` copies, and cloud\u002Fserver\n  branches are environment-specific. LIDC-IDRI, PET\u002FCT\u002FTNM, and the toy data\n  directories are external datasets; the PET\u002FCT paths and clinical labels may\n  be private. Do not claim access, redistribute them, or run acquisition from\n  this skill. The supplied examples are evidence of schema and behavior only.\n- `experiments\u002Flidc_exp\u002Fpack_dataset.py` and\n  `utils\u002Fdataloader_utils.py::unpack_dataset`\u002F`delete_npy` are storage helpers,\n  not safe validation steps. Packing\u002Funpacking can create many files and\n  `delete_npy` is destructive. Treat pack, delete, `rsync`, multiprocessing\n  preprocessing, and raw-data conversion as **non-runnable** unless a user\n  explicitly reviews paths, backups, and a bounded dry run outside this skill.\n- Do not run native final cases, download data, or import this operating graph\n  during construction. Use signatures\u002Fimport checks and bounded synthetic\n  validator fixtures only.\n\n## References and bundled script\n\n- [Data formats and array axes](references\u002Fdata-formats.md)\n- [Data-loader and batchgenerators contracts](references\u002Fdata-loader-contracts.md)\n- [Preprocessing workflows](references\u002Fpreprocessing-workflows.md)\n- [Data utilities: padding, tiling, balancing, packing boundaries](references\u002Fdata-utilities.md)\n- [Troubleshooting and safe customization](references\u002Ftroubleshooting.md)\n- [Standalone preprocessed-case validator](scripts\u002Fvalidate_preprocessed_case.py)\n\nAll references link back here and to the other references so the graph remains\nusable when a Researcher opens a leaf directly. Source evidence is the checked-\nout `README.md`, the three experiment `configs.py`\u002F`data_loader.py` pairs,\nLIDC `preprocessing.py`\u002F`pack_dataset.py`, PET-CT `preprocessing.py`, and\n`utils\u002Fdataloader_utils.py`.\n\n## Verification gate\n\nAccept a customized data path only when: (a) the three example loaders import\nwith the documented loader dependency\u002FAPI variant; (b) the validator passes a\nbounded synthetic valid case and rejects mismatched shapes, non-integral labels,\nor oversized input;\n(c) the manifest-to-file naming and `class_target`\u002FROI semantics are recorded;\nand (d) no native dataset case or destructive helper was run. Unresolved\nbackend, private-data, axis, or label ambiguities stay explicit in the handoff.\n",{"name":5,"license":23,"metadata":24,"description":7,"disable-model-invocation":26},"Apache 2.0",{"disco-role":25},"operating",true,"main","ac3fe1afa80fb9a09775ecfb2b6cc3ba850a2db6","skills\u002Frepositories\u002Frepo-skills\u002Fmedical-detection-toolkit\u002Fsub-skills\u002Fdata-and-preprocessing",27,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab","2026-09-03 10:25:21",{"source":8,"skillName":5,"commands":35,"prompt":45,"zipAvailable":26},[36,39,42],{"tool":37,"command":38},"npx","npx skills add VectorSpaceLab\u002FAREX-Skill --skill data-and-preprocessing",{"tool":40,"command":41},"bunx","bunx skills add VectorSpaceLab\u002FAREX-Skill --skill data-and-preprocessing",{"tool":43,"command":44},"pnpm","pnpm dlx skills add VectorSpaceLab\u002FAREX-Skill --skill data-and-preprocessing","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fmedical-detection-toolkit\u002Fsub-skills\u002Fdata-and-preprocessing\u002FSKILL.md",[47,53,57,61,65,69],{"path":48,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":51,"sourceHtmlUrl":52},"references\u002Fdata-formats.md","REFERENCE","参考文档",6087,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fmedical-detection-toolkit\u002Fsub-skills\u002Fdata-and-preprocessing\u002Freferences\u002Fdata-formats.md",{"path":54,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":55,"sourceHtmlUrl":56},"references\u002Fdata-loader-contracts.md",6185,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fmedical-detection-toolkit\u002Fsub-skills\u002Fdata-and-preprocessing\u002Freferences\u002Fdata-loader-contracts.md",{"path":58,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":59,"sourceHtmlUrl":60},"references\u002Fdata-utilities.md",5318,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fmedical-detection-toolkit\u002Fsub-skills\u002Fdata-and-preprocessing\u002Freferences\u002Fdata-utilities.md",{"path":62,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":63,"sourceHtmlUrl":64},"references\u002Fpreprocessing-workflows.md",5205,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fmedical-detection-toolkit\u002Fsub-skills\u002Fdata-and-preprocessing\u002Freferences\u002Fpreprocessing-workflows.md",{"path":66,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":67,"sourceHtmlUrl":68},"references\u002Ftroubleshooting.md",5109,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fmedical-detection-toolkit\u002Fsub-skills\u002Fdata-and-preprocessing\u002Freferences\u002Ftroubleshooting.md",{"path":70,"resourceType":71,"resourceTypeDesc":72,"sizeBytes":73,"sourceHtmlUrl":74},"scripts\u002Fvalidate_preprocessed_case.py","SCRIPT","脚本",3600,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fmedical-detection-toolkit\u002Fsub-skills\u002Fdata-and-preprocessing\u002Fscripts\u002Fvalidate_preprocessed_case.py",[76,83,90,97,104,111],{"skillNo":77,"name":78,"nameZh":6,"description":79,"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},"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.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-triton-opt\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},"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":91,"name":92,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":94,"categories":95,"keywords":96,"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":98,"name":99,"nameZh":6,"description":100,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":101,"categories":102,"keywords":103,"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":105,"name":106,"nameZh":6,"description":107,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":108,"categories":109,"keywords":110,"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":112,"name":113,"nameZh":6,"description":114,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":115,"categories":116,"keywords":117,"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",[],[]]