[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M79027XWVAT4G8EP9DW5N2AC":3,"skill-same:01M79027XWVAT4G8EP9DW5N2AC":71},{"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},"01M79027XWVAT4G8EP9DW5N2AC","evaluation-and-export",null,"Evaluate MedMNIST predictions and export 2D or 3D datasets safely, including metric calculation, standard result filenames, montages, CSV\u002Fimage output, and the Fire CLI.","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\u002Fmedmnist\u002Fsub-skills\u002Fevaluation-and-export\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: evaluation-and-export\ndescription: \"Evaluate MedMNIST predictions and export 2D or 3D datasets safely,\n  including metric calculation, standard result filenames, montages, CSV\u002Fimage\n  output, and the Fire CLI.\"\ndisable-model-invocation: true\nmetadata:\n  disco-role: operating\nlicense: Apache 2.0\n---\n\n# MedMNIST evaluation and export\n\nUse this skill when the task concerns `Evaluator`, `getAUC`, `getACC`, score\nshapes, metric interpretation, standard result CSV names, `parse_and_evaluate`,\n`MedMNIST2D\u002F3D.save`, `montage`, PNG\u002FGIF\u002FCSV export, or the public `save` and\n`evaluate` CLI commands. This skill is for evaluation and output handling, not\nfor loading arbitrary datasets or training models.\n\n## Route and safety boundary\n\n1. **First identify the flag, split, task, score shape, and root.** The split\n   must be `train`, `val`, or `test`; `y_score.shape[0]` must equal the label\n   count. Use `INFO[flag][\"task\"]` to choose binary, multilabel,\n   multiclass, or ordinal handling. For registry and NPZ-loading questions,\n   route to the sibling `dataset-loading` skill instead.\n2. Use an explicit, already-created `root` for programmatic evaluation. The\n   constructor never creates a missing custom root and never downloads unless\n   the dataset class is constructed with `download=True`.\n3. Keep result files and exports in a new, isolated output directory. The\n   evaluator does not create `save_folder`; create it first. Do not use bulk\n   download, default-root cleanup, broad development `test()`, or network\n   downloads as part of a smoke check.\n4. Scores are probabilities\u002Fscores, not class labels. Binary and multilabel\n   scores are thresholded only for ACC; multiclass and ordinal predictions use\n   `argmax`. Preserve row order and do not add a CSV header to an evaluator\n   input file.\n\nFor a compact API contract, see [references\u002Fapi-reference.md](references\u002Fapi-reference.md).\nFor CLI syntax and network\u002Fdestructive boundaries, see\n[references\u002Fcli-reference.md](references\u002Fcli-reference.md).\nFor reproducible evaluation and export procedures, see\n[references\u002Fexport-recipes.md](references\u002Fexport-recipes.md).\nFor failures and recovery, see [references\u002Ftroubleshooting.md](references\u002Ftroubleshooting.md).\nRun the safe local check with\n[scripts\u002Fmedmnist_smoke.py](scripts\u002Fmedmnist_smoke.py).\n\n## Evaluation procedure\n\n### Direct `Evaluator` use\n\nUse the installed API, with the dataset NPZ already present in `root`:\n\n```python\nimport numpy as np\nfrom medmnist import Evaluator\n\nevaluator = Evaluator(\"pneumoniamnist\", \"test\", size=64, root=\".\u002Fdata\")\ny_score = np.asarray(...)                 # N or N x 2 for binary\nassert y_score.shape[0] == evaluator.labels.shape[0]\nmetrics = evaluator.evaluate(y_score, save_folder=\".\u002Fresults\", run=\"run1\")\nprint(metrics.AUC, metrics.ACC)\n```\n\n`Evaluator(flag, split, size=None, root=\"~\u002F.medmnist\")` maps `None` and `28`\nto the unsuffixed NPZ. A non-28 size is written as `flag_size.npz`; the\navailable 2D sizes are 28\u002F64\u002F128\u002F224 and 3D sizes are 28\u002F64. The root must\nexist and contain the exact file. `evaluate` computes a `Metrics(AUC, ACC)`\nnamedtuple and, when `save_folder` is supplied, writes the score matrix with\npandas using no header and an index column. Pass a stable `run` value when the\nfilename must be reproducible.\n\n### Shape and metric decision table\n\n| `INFO[flag][\"task\"]` | Accepted score shape | AUC | ACC |\n|---|---|---|---|\n| `binary-class` | `(N,)` or `(N, 2)` | ROC one-vs-rest on the positive score; for 2-D input the last column is used | threshold the selected score at `0.5` |\n| `multi-label, binary-class` | `(N, L)` | compute ROC AUC independently for each label and average | threshold each label at `0.5`, compute per-label accuracy, then average |\n| `multi-class` | `(N, C)` | one-vs-rest AUC for each class column and average | `argmax` over columns |\n| `ordinal-regression` | `(N, C)` | same one-vs-rest loop as multiclass | `argmax` over columns |\n\nLabels should be class indices `0..C-1` for multiclass\u002Fordinal and binary\n0\u002F1 for binary. Every one-vs-rest target must contain both positive and\nnegative examples; otherwise scikit-learn cannot define ROC AUC. For\nmultilabel, retain the 2-D `(N, L)` shape even when inspecting one label.\n\n`getAUC(y_true, y_score, task)` and `getACC(y_true, y_score, task,\nthreshold=0.5)` are the low-level functions. They call `squeeze()`, so an\nunexpected singleton axis can change dimensionality; assert shapes before\ncalling them. Do not interpret ACC as a probability metric, and do not pass\nalready-argmaxed class labels in place of scores.\n\n`get_dummy_prediction()` is useful for discovering a rough output shape, but\nit is random and is not a reproducible test fixture. In the inspected version,\nits multiclass branch derives the column count from `labels.max()`; construct\nexplicit `(N, C)` scores when all classes must be represented.\n\n### Standard result filenames and round trips\n\nWith `metrics = Metrics(AUC, ACC)`, the standard name is:\n\n```text\n{flag}{size_flag}_{split}_[AUC]{auc:.3f}_[ACC]{acc:.3f}@{run}.csv\n```\n\n`size_flag` is empty for 28\u002Fdefault and `_64`, `_128`, or `_224` for the\ncorresponding 2-D size (or `_64` for a 3-D size). For example:\n`pneumoniamnist_64_test_[AUC]1.000_[ACC]1.000@run1.csv`.\n`run=None` uses the current time, so use a string or number for stable output.\n\n`Evaluator.parse_and_evaluate(path, run=None)` parses the flag, optional size,\nand split from the underscore-separated filename, reads the CSV with\n`index_col=0, header=None`, sorts by that index, evaluates the resulting score\nmatrix, prints `Metrics(...)`, and writes a standardized evaluated CSV beside\nit. If `run` is omitted, the filename must contain `@`. The parser constructs\nan evaluator with the default root; direct evaluation with an explicit root is\nsafer for isolated fixtures. Put parsed inputs in an isolated output directory\nbecause a matching standard name can be rewritten.\n\n## Export and montage procedure\n\n### 2-D\n\n```python\nfrom medmnist import PathMNIST\n\ndataset = PathMNIST(split=\"test\", root=\".\u002Fdata\", size=28)\ndataset.save(\".\u002Fexport\", postfix=\"png\", write_csv=True)\nmontage = dataset.montage(length=2, replace=True, save_folder=\".\u002Fexport\")\n```\n\n`MedMNIST2D.save` calls `save2d`, creates `export\u002F{flag}{size_flag}\u002F`, and\nwrites one image per sample plus `{flag}{size_flag}.csv` when `write_csv=True`.\nThe default postfix is `png`; the implementation can pass another image\nextension accepted by Pillow. Image names are `{SPLIT}{index}_{labels}.{postfix}`\nwhere `SPLIT` is `TRAIN`, `VALIDATION`, or `TEST`. The CSV is append-mode,\nhas no generated header, and contains split, filename, and label values;\nre-running against the same CSV can append duplicates.\n\n`montage(length=20, replace=False, save_folder=None)` randomly selects\n`length * length` samples. For a tiny fixture, either use `replace=True` or\nensure the dataset has at least that many samples. A saved 2-D montage is\n`{flag}{size_flag}_{split}_montage.jpg`; the method returns a PIL image.\nRandom selection is not deterministic unless the caller seeds NumPy before\ncalling it.\n\n### 3-D\n\n```python\nfrom medmnist import OrganMNIST3D\n\ndataset = OrganMNIST3D(split=\"test\", root=\".\u002Fdata\", size=28)\ndataset.save(\".\u002Fexport\", postfix=\"gif\", write_csv=True)\nframes = dataset.montage(length=2, replace=True, save_folder=\".\u002Fexport\")\n```\n\n`MedMNIST3D.save` requires `postfix=\"gif\"`; it emits one GIF per volume and a\nCSV under the same naming layout. A 3-D montage returns a list of frames and,\nwhen saved, writes `{flag}{size_flag}_{split}_montage.gif`. The current 3-D\nmontage path requires a single channel. GIF output requires a nonempty volume\nand compatible PIL frames.\n\n## CLI procedure\n\nThe package exposes Fire commands. With data already local, use:\n\n```bash\npython -m medmnist available\npython -m medmnist info --flag=pneumoniamnist\npython -m medmnist save --flag=pneumoniamnist --folder=.\u002Fexport --postfix=png --download=False --size=28 --root=.\u002Fdata\npython -m medmnist save --flag=organmnist3d --folder=.\u002Fexport --postfix=gif --download=False --size=28 --root=.\u002Fdata\npython -m medmnist evaluate --path=.\u002Fresults\u002Fpneumoniamnist_test_[AUC]1.000_[ACC]1.000@run1.csv\n```\n\n`save` loops over all three splits, so all three corresponding NPZ arrays are\nrequired. `--download=True` is a network operation and is intentionally not\npart of a safe smoke test. `evaluate` has only `--path`; its parser resolves\nthe dataset through the default root rather than exposing a custom-root flag.\nFor a custom root, use the direct `Evaluator` API (or arrange the data in the\nexpected default root only with explicit user approval). `available` and\n`info` are read-only metadata commands. Do not use `clean` in this skill: it\ndeletes downloaded `*mnist*.npz` files.\n\n## Verification and return handoff\n\nRun the bundled helper before relying on a new environment:\n\n```bash\npython scripts\u002Fmedmnist_smoke.py --help\npython scripts\u002Fmedmnist_smoke.py\n```\n\nA successful run reports deterministic binary, multilabel, and multiclass\n`AUC=1.000 ACC=1.000` checks, a `_64\u002F@run` CSV parse round trip in a temporary\nroot, and 2-D\u002F3-D export checks without network access. For a concise handoff,\nreport: **files** used or emitted; **evidence** (task, split, root, score\nshapes); **checks** and metric values; **difficult cases** (singleton axes,\nmissing ROC classes, `_64\u002F@run`, tiny montage sampling, 3-D GIF); **gaps** such\nas default-root-only parsing or untested optional backends; and **intentional\nomissions** (training, bulk downloads, default-root clean, and broad\n`test()`).\n","# MedMNIST evaluation and export\n\nUse this skill when the task concerns `Evaluator`, `getAUC`, `getACC`, score\nshapes, metric interpretation, standard result CSV names, `parse_and_evaluate`,\n`MedMNIST2D\u002F3D.save`, `montage`, PNG\u002FGIF\u002FCSV export, or the public `save` and\n`evaluate` CLI commands. This skill is for evaluation and output handling, not\nfor loading arbitrary datasets or training models.\n\n## Route and safety boundary\n\n1. **First identify the flag, split, task, score shape, and root.** The split\n   must be `train`, `val`, or `test`; `y_score.shape[0]` must equal the label\n   count. Use `INFO[flag][\"task\"]` to choose binary, multilabel,\n   multiclass, or ordinal handling. For registry and NPZ-loading questions,\n   route to the sibling `dataset-loading` skill instead.\n2. Use an explicit, already-created `root` for programmatic evaluation. The\n   constructor never creates a missing custom root and never downloads unless\n   the dataset class is constructed with `download=True`.\n3. Keep result files and exports in a new, isolated output directory. The\n   evaluator does not create `save_folder`; create it first. Do not use bulk\n   download, default-root cleanup, broad development `test()`, or network\n   downloads as part of a smoke check.\n4. Scores are probabilities\u002Fscores, not class labels. Binary and multilabel\n   scores are thresholded only for ACC; multiclass and ordinal predictions use\n   `argmax`. Preserve row order and do not add a CSV header to an evaluator\n   input file.\n\nFor a compact API contract, see [references\u002Fapi-reference.md](references\u002Fapi-reference.md).\nFor CLI syntax and network\u002Fdestructive boundaries, see\n[references\u002Fcli-reference.md](references\u002Fcli-reference.md).\nFor reproducible evaluation and export procedures, see\n[references\u002Fexport-recipes.md](references\u002Fexport-recipes.md).\nFor failures and recovery, see [references\u002Ftroubleshooting.md](references\u002Ftroubleshooting.md).\nRun the safe local check with\n[scripts\u002Fmedmnist_smoke.py](scripts\u002Fmedmnist_smoke.py).\n\n## Evaluation procedure\n\n### Direct `Evaluator` use\n\nUse the installed API, with the dataset NPZ already present in `root`:\n\n```python\nimport numpy as np\nfrom medmnist import Evaluator\n\nevaluator = Evaluator(\"pneumoniamnist\", \"test\", size=64, root=\".\u002Fdata\")\ny_score = np.asarray(...)                 # N or N x 2 for binary\nassert y_score.shape[0] == evaluator.labels.shape[0]\nmetrics = evaluator.evaluate(y_score, save_folder=\".\u002Fresults\", run=\"run1\")\nprint(metrics.AUC, metrics.ACC)\n```\n\n`Evaluator(flag, split, size=None, root=\"~\u002F.medmnist\")` maps `None` and `28`\nto the unsuffixed NPZ. A non-28 size is written as `flag_size.npz`; the\navailable 2D sizes are 28\u002F64\u002F128\u002F224 and 3D sizes are 28\u002F64. The root must\nexist and contain the exact file. `evaluate` computes a `Metrics(AUC, ACC)`\nnamedtuple and, when `save_folder` is supplied, writes the score matrix with\npandas using no header and an index column. Pass a stable `run` value when the\nfilename must be reproducible.\n\n### Shape and metric decision table\n\n| `INFO[flag][\"task\"]` | Accepted score shape | AUC | ACC |\n|---|---|---|---|\n| `binary-class` | `(N,)` or `(N, 2)` | ROC one-vs-rest on the positive score; for 2-D input the last column is used | threshold the selected score at `0.5` |\n| `multi-label, binary-class` | `(N, L)` | compute ROC AUC independently for each label and average | threshold each label at `0.5`, compute per-label accuracy, then average |\n| `multi-class` | `(N, C)` | one-vs-rest AUC for each class column and average | `argmax` over columns |\n| `ordinal-regression` | `(N, C)` | same one-vs-rest loop as multiclass | `argmax` over columns |\n\nLabels should be class indices `0..C-1` for multiclass\u002Fordinal and binary\n0\u002F1 for binary. Every one-vs-rest target must contain both positive and\nnegative examples; otherwise scikit-learn cannot define ROC AUC. For\nmultilabel, retain the 2-D `(N, L)` shape even when inspecting one label.\n\n`getAUC(y_true, y_score, task)` and `getACC(y_true, y_score, task,\nthreshold=0.5)` are the low-level functions. They call `squeeze()`, so an\nunexpected singleton axis can change dimensionality; assert shapes before\ncalling them. Do not interpret ACC as a probability metric, and do not pass\nalready-argmaxed class labels in place of scores.\n\n`get_dummy_prediction()` is useful for discovering a rough output shape, but\nit is random and is not a reproducible test fixture. In the inspected version,\nits multiclass branch derives the column count from `labels.max()`; construct\nexplicit `(N, C)` scores when all classes must be represented.\n\n### Standard result filenames and round trips\n\nWith `metrics = Metrics(AUC, ACC)`, the standard name is:\n\n```text\n{flag}{size_flag}_{split}_[AUC]{auc:.3f}_[ACC]{acc:.3f}@{run}.csv\n```\n\n`size_flag` is empty for 28\u002Fdefault and `_64`, `_128`, or `_224` for the\ncorresponding 2-D size (or `_64` for a 3-D size). For example:\n`pneumoniamnist_64_test_[AUC]1.000_[ACC]1.000@run1.csv`.\n`run=None` uses the current time, so use a string or number for stable output.\n\n`Evaluator.parse_and_evaluate(path, run=None)` parses the flag, optional size,\nand split from the underscore-separated filename, reads the CSV with\n`index_col=0, header=None`, sorts by that index, evaluates the resulting score\nmatrix, prints `Metrics(...)`, and writes a standardized evaluated CSV beside\nit. If `run` is omitted, the filename must contain `@`. The parser constructs\nan evaluator with the default root; direct evaluation with an explicit root is\nsafer for isolated fixtures. Put parsed inputs in an isolated output directory\nbecause a matching standard name can be rewritten.\n\n## Export and montage procedure\n\n### 2-D\n\n```python\nfrom medmnist import PathMNIST\n\ndataset = PathMNIST(split=\"test\", root=\".\u002Fdata\", size=28)\ndataset.save(\".\u002Fexport\", postfix=\"png\", write_csv=True)\nmontage = dataset.montage(length=2, replace=True, save_folder=\".\u002Fexport\")\n```\n\n`MedMNIST2D.save` calls `save2d`, creates `export\u002F{flag}{size_flag}\u002F`, and\nwrites one image per sample plus `{flag}{size_flag}.csv` when `write_csv=True`.\nThe default postfix is `png`; the implementation can pass another image\nextension accepted by Pillow. Image names are `{SPLIT}{index}_{labels}.{postfix}`\nwhere `SPLIT` is `TRAIN`, `VALIDATION`, or `TEST`. The CSV is append-mode,\nhas no generated header, and contains split, filename, and label values;\nre-running against the same CSV can append duplicates.\n\n`montage(length=20, replace=False, save_folder=None)` randomly selects\n`length * length` samples. For a tiny fixture, either use `replace=True` or\nensure the dataset has at least that many samples. A saved 2-D montage is\n`{flag}{size_flag}_{split}_montage.jpg`; the method returns a PIL image.\nRandom selection is not deterministic unless the caller seeds NumPy before\ncalling it.\n\n### 3-D\n\n```python\nfrom medmnist import OrganMNIST3D\n\ndataset = OrganMNIST3D(split=\"test\", root=\".\u002Fdata\", size=28)\ndataset.save(\".\u002Fexport\", postfix=\"gif\", write_csv=True)\nframes = dataset.montage(length=2, replace=True, save_folder=\".\u002Fexport\")\n```\n\n`MedMNIST3D.save` requires `postfix=\"gif\"`; it emits one GIF per volume and a\nCSV under the same naming layout. A 3-D montage returns a list of frames and,\nwhen saved, writes `{flag}{size_flag}_{split}_montage.gif`. The current 3-D\nmontage path requires a single channel. GIF output requires a nonempty volume\nand compatible PIL frames.\n\n## CLI procedure\n\nThe package exposes Fire commands. With data already local, use:\n\n```bash\npython -m medmnist available\npython -m medmnist info --flag=pneumoniamnist\npython -m medmnist save --flag=pneumoniamnist --folder=.\u002Fexport --postfix=png --download=False --size=28 --root=.\u002Fdata\npython -m medmnist save --flag=organmnist3d --folder=.\u002Fexport --postfix=gif --download=False --size=28 --root=.\u002Fdata\npython -m medmnist evaluate --path=.\u002Fresults\u002Fpneumoniamnist_test_[AUC]1.000_[ACC]1.000@run1.csv\n```\n\n`save` loops over all three splits, so all three corresponding NPZ arrays are\nrequired. `--download=True` is a network operation and is intentionally not\npart of a safe smoke test. `evaluate` has only `--path`; its parser resolves\nthe dataset through the default root rather than exposing a custom-root flag.\nFor a custom root, use the direct `Evaluator` API (or arrange the data in the\nexpected default root only with explicit user approval). `available` and\n`info` are read-only metadata commands. Do not use `clean` in this skill: it\ndeletes downloaded `*mnist*.npz` files.\n\n## Verification and return handoff\n\nRun the bundled helper before relying on a new environment:\n\n```bash\npython scripts\u002Fmedmnist_smoke.py --help\npython scripts\u002Fmedmnist_smoke.py\n```\n\nA successful run reports deterministic binary, multilabel, and multiclass\n`AUC=1.000 ACC=1.000` checks, a `_64\u002F@run` CSV parse round trip in a temporary\nroot, and 2-D\u002F3-D export checks without network access. For a concise handoff,\nreport: **files** used or emitted; **evidence** (task, split, root, score\nshapes); **checks** and metric values; **difficult cases** (singleton axes,\nmissing ROC classes, `_64\u002F@run`, tiny montage sampling, 3-D GIF); **gaps** such\nas default-root-only parsing or untested optional backends; and **intentional\nomissions** (training, bulk downloads, default-root clean, and broad\n`test()`).\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\u002Fmedmnist\u002Fsub-skills\u002Fevaluation-and-export",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 evaluation-and-export",{"tool":40,"command":41},"bunx","bunx skills add VectorSpaceLab\u002FAREX-Skill --skill evaluation-and-export",{"tool":43,"command":44},"pnpm","pnpm dlx skills add VectorSpaceLab\u002FAREX-Skill --skill evaluation-and-export","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fmedmnist\u002Fsub-skills\u002Fevaluation-and-export\u002FSKILL.md",[47,53,57,61,65],{"path":48,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":51,"sourceHtmlUrl":52},"references\u002Fapi-reference.md","REFERENCE","参考文档",4372,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fmedmnist\u002Fsub-skills\u002Fevaluation-and-export\u002Freferences\u002Fapi-reference.md",{"path":54,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":55,"sourceHtmlUrl":56},"references\u002Fcli-reference.md",3230,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fmedmnist\u002Fsub-skills\u002Fevaluation-and-export\u002Freferences\u002Fcli-reference.md",{"path":58,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":59,"sourceHtmlUrl":60},"references\u002Fexport-recipes.md",5104,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fmedmnist\u002Fsub-skills\u002Fevaluation-and-export\u002Freferences\u002Fexport-recipes.md",{"path":62,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":63,"sourceHtmlUrl":64},"references\u002Ftroubleshooting.md",6293,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fmedmnist\u002Fsub-skills\u002Fevaluation-and-export\u002Freferences\u002Ftroubleshooting.md",{"path":66,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":69,"sourceHtmlUrl":70},"scripts\u002Fmedmnist_smoke.py","SCRIPT","脚本",10619,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fmedmnist\u002Fsub-skills\u002Fevaluation-and-export\u002Fscripts\u002Fmedmnist_smoke.py",[72,79,86,93,100,107],{"skillNo":73,"name":74,"nameZh":6,"description":75,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":76,"categories":77,"keywords":78,"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":80,"name":81,"nameZh":6,"description":82,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":83,"categories":84,"keywords":85,"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":87,"name":88,"nameZh":6,"description":89,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":90,"categories":91,"keywords":92,"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":94,"name":95,"nameZh":6,"description":96,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":97,"categories":98,"keywords":99,"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":101,"name":102,"nameZh":6,"description":103,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":104,"categories":105,"keywords":106,"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":108,"name":109,"nameZh":6,"description":110,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":111,"categories":112,"keywords":113,"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",[],[]]