[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M78VMJQ8Z2E32RW3GG14E2HT":3,"skill-same:01M78VMJQ8Z2E32RW3GG14E2HT":74},{"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},"01M78VMJQ8Z2E32RW3GG14E2HT","model-training",null,"Configure, preflight, run, resume, and troubleshoot DeepDanbooru 1.0.0 model training with a CPU-first TensorFlow workflow.","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\u002Fdeep-danbooru\u002Fsub-skills\u002Fmodel-training\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: model-training\ndescription: \"Configure, preflight, run, resume, and troubleshoot DeepDanbooru\n  1.0.0 model training with a CPU-first TensorFlow workflow.\"\ndisable-model-invocation: true\nmetadata:\n  disco-role: operating\nlicense: MIT\n---\n\n# Model training\n\nUse this sub-skill after a DeepDanbooru project and Danbooru-style dataset\nexist. It covers configuration, a read-only preflight, training dispatch,\ncheckpoint resume, and saved-model artifacts for DeepDanbooru 1.0.0.\n\nCPU TensorFlow is the required verified backend. GPU execution is optional and\nunverified; never infer GPU readiness from device discovery or a CPU success.\nFull training is intentionally **skip-expensive** during skill verification:\nthe ResNet variants, image pipeline, multiple epochs, checkpoint I\u002FO, and model\nexports can consume substantial compute, memory, and storage. Verify structure\nand a bounded data sample first, then launch training only with an explicit\nruntime and resource decision.\n\n## Route the request\n\n| Need | Read or run |\n|---|---|\n| Create `project.json`, `tags.txt`, or the SQLite\u002Fimage layout | [project-data-setup](..\u002Fproject-data-setup\u002FSKILL.md) |\n| Validate configuration values without training | [`validate_project_config.py`](scripts\u002Fvalidate_project_config.py) |\n| Validate config, tags, SQLite rows, image paths, checkpoints, and packages | [`training_preflight.py`](scripts\u002Ftraining_preflight.py) |\n| Start, resume, or initialize from a saved model | [Training workflows](references\u002Ftraining-workflows.md) |\n| Understand row selection, image paths, labels, and augmentation | [Training data pipeline](references\u002Ftraining-data-pipeline.md) |\n| Choose a model, optimizer, loss, schedule, or precision policy | [Model and loss reference](references\u002Fmodel-and-loss-reference.md) |\n| Diagnose empty runs, missing images, unsupported values, or resume surprises | [Troubleshooting](references\u002Ftroubleshooting.md) |\n| Evaluate the resulting Keras model | [inference-evaluation](..\u002Finference-evaluation\u002FSKILL.md) |\n| Convert to TFLite or create Grad-CAM output | [post-training-tools](..\u002Fpost-training-tools\u002FSKILL.md) |\n\n## Supported dispatch values\n\nUse these exact lowercase strings in `project.json`:\n\n- Models: `resnet_152`, `resnet_custom_v1`, `resnet_custom_v2`,\n  `resnet_custom_v3`, `resnet_custom_v4`.\n- Optimizers: `adam`, `sgd`, `rmsprop`.\n- Losses: `binary_crossentropy`, `focal_loss`.\n\nAny other value fails before training. The default project uses\n`resnet_custom_v2`, `adam`, `binary_crossentropy`, and an image shape of\n`299` by `299` with three channels. Shape order in the model is\n`(image_height, image_width, 3)` even though the JSON fields list width and\nheight separately.\n\n## Safe preflight\n\nFrom this sub-skill directory, run:\n\n```console\npython scripts\u002Fvalidate_project_config.py PROJECT --check-paths\npython scripts\u002Ftraining_preflight.py PROJECT --check-packages\n```\n\nBoth helpers are read-only, deterministic, offline, and use the standard\nlibrary for configuration, filesystem, tag, and SQLite checks. They do not\nlaunch training. Package and TensorFlow import probes are opt-in so basic\nvalidation still works when optional runtime packages are absent.\n\nThe preflight checks a bounded, ID-ordered image sample by default. Use\n`--max-image-checks 0` only when a complete filesystem scan is intended and its\nI\u002FO cost is acceptable. Use `--probe-imports --probe-tensorflow` to test the\ninstalled modules and confirm that TensorFlow exposes a CPU; this still does\nnot train or verify a GPU.\n\nDo not launch training unless all of these hold:\n\n1. `project.json` parses and all dispatch values and numeric ranges are valid.\n2. `tags.txt` is non-empty and is the intended fixed output vocabulary.\n3. The SQLite `posts` table has the required columns and at least one eligible\n   row at `minimum_tag_count`.\n4. Derived image paths exist for the checked rows, and the image format is\n   actually decodable by the training environment.\n5. Checkpoint state is understood: an existing latest checkpoint causes an\n   automatic resume.\n6. The CPU TensorFlow environment can import DeepDanbooru, TensorFlow, and\n   TensorFlow I\u002FO.\n\n## Launch boundary\n\nThe side-effecting command is:\n\n```console\ndeepdanbooru train-project PROJECT_PATH\n```\n\nTo load initial weights\u002Fmodel structure from an existing saved model:\n\n```console\ndeepdanbooru train-project PROJECT_PATH --source-model MODEL.keras\n```\n\n`--source-model` is not a no-resume flag. After the model is loaded, any latest\ncheckpoint under `PROJECT_PATH\u002Fcheckpoints\u002F` is restored and can supersede its\nstate. Preserve or deliberately relocate checkpoints before changing training\nlineage. Also confirm source-model input\u002Foutput compatibility with the image\nshape and number of tags; the training function does not perform a friendly\ncompatibility check.\n\n## Runtime behavior to preserve\n\nTraining compiles with precision and recall metrics, shuffles records once per\nepoch from a checkpointed seed, applies an optional epoch learning-rate\nschedule, trains in checkpoint-sized slices, and saves after each slice.\nCheckpoint state includes optimizer\u002Fmodel variables plus `used_epoch`,\n`used_minibatch`, `used_sample`, `offset`, and `random_seed`; only the latest\nthree checkpoints are retained.\n\nPeriodic exports are named:\n\n```text\nmodel-MODEL.eEPOCH.keras\n```\n\nThe final export is:\n\n```text\nmodel-MODEL.keras\n```\n\nExports omit optimizer state. Resume therefore depends on `checkpoints\u002F`, not\non a `.keras` export. `export_model_per_epoch: 0` means export every epoch;\notherwise an epoch export occurs when the completed epoch is divisible by that\npositive interval.\n\nWith `mixed_precision: true`, training uses a `mixed_float16` model and a\nloss-scaled optimizer, then attempts separate float32 exports ending in\n`.float32.keras`. The 1.0.0 implementation reconstructs a float32 model and\nrestores checkpointed model variables into it. Treat those exports as needing\nexplicit load\u002Finference validation. In particular, combining\n`--source-model` with mixed precision leaves the float32 export model\nunconstructed and can fail during export; the bundled preflight rejects that\ncombination.\n\n## Failure semantics\n\nA missing image is not reliably fatal at the dataset boundary because\n`Dataset.ignore_errors()` can drop load\u002Fdecode failures. A run with many bad\npaths can therefore appear to progress while using fewer samples; a run with\nno usable records can produce checkpoints or model files without meaningful\nupdates. Treat missing files, empty eligible selections, empty tag records, and\nzero usable batches as failures, not warnings to ignore.\n\nUnsupported model, optimizer, or loss strings fail during dispatch. A\n`checkpoint_frequency_mb` of zero can prevent offset progress; a\n`console_logging_frequency_mb` of zero causes modulo-by-zero; non-positive\nbatch\u002Fimage\u002Fepoch values are invalid for an intended training run. Use the\nvalidators rather than discovering these conditions after allocating a long\njob.\n\n## Handoff\n\nAfter training, retain `project.json`, the exact ordered `tags.txt`, the final\n`.keras` model, and any checkpoints needed for further resume. Validate the\nsaved model through [inference-evaluation](..\u002Finference-evaluation\u002FSKILL.md)\nbefore conversion or Grad-CAM work in\n[post-training-tools](..\u002Fpost-training-tools\u002FSKILL.md). If project\u002Fdata checks\nfail, return to [project-data-setup](..\u002Fproject-data-setup\u002FSKILL.md) instead of\nchanging model hyperparameters to mask a data problem.\n","# Model training\n\nUse this sub-skill after a DeepDanbooru project and Danbooru-style dataset\nexist. It covers configuration, a read-only preflight, training dispatch,\ncheckpoint resume, and saved-model artifacts for DeepDanbooru 1.0.0.\n\nCPU TensorFlow is the required verified backend. GPU execution is optional and\nunverified; never infer GPU readiness from device discovery or a CPU success.\nFull training is intentionally **skip-expensive** during skill verification:\nthe ResNet variants, image pipeline, multiple epochs, checkpoint I\u002FO, and model\nexports can consume substantial compute, memory, and storage. Verify structure\nand a bounded data sample first, then launch training only with an explicit\nruntime and resource decision.\n\n## Route the request\n\n| Need | Read or run |\n|---|---|\n| Create `project.json`, `tags.txt`, or the SQLite\u002Fimage layout | [project-data-setup](..\u002Fproject-data-setup\u002FSKILL.md) |\n| Validate configuration values without training | [`validate_project_config.py`](scripts\u002Fvalidate_project_config.py) |\n| Validate config, tags, SQLite rows, image paths, checkpoints, and packages | [`training_preflight.py`](scripts\u002Ftraining_preflight.py) |\n| Start, resume, or initialize from a saved model | [Training workflows](references\u002Ftraining-workflows.md) |\n| Understand row selection, image paths, labels, and augmentation | [Training data pipeline](references\u002Ftraining-data-pipeline.md) |\n| Choose a model, optimizer, loss, schedule, or precision policy | [Model and loss reference](references\u002Fmodel-and-loss-reference.md) |\n| Diagnose empty runs, missing images, unsupported values, or resume surprises | [Troubleshooting](references\u002Ftroubleshooting.md) |\n| Evaluate the resulting Keras model | [inference-evaluation](..\u002Finference-evaluation\u002FSKILL.md) |\n| Convert to TFLite or create Grad-CAM output | [post-training-tools](..\u002Fpost-training-tools\u002FSKILL.md) |\n\n## Supported dispatch values\n\nUse these exact lowercase strings in `project.json`:\n\n- Models: `resnet_152`, `resnet_custom_v1`, `resnet_custom_v2`,\n  `resnet_custom_v3`, `resnet_custom_v4`.\n- Optimizers: `adam`, `sgd`, `rmsprop`.\n- Losses: `binary_crossentropy`, `focal_loss`.\n\nAny other value fails before training. The default project uses\n`resnet_custom_v2`, `adam`, `binary_crossentropy`, and an image shape of\n`299` by `299` with three channels. Shape order in the model is\n`(image_height, image_width, 3)` even though the JSON fields list width and\nheight separately.\n\n## Safe preflight\n\nFrom this sub-skill directory, run:\n\n```console\npython scripts\u002Fvalidate_project_config.py PROJECT --check-paths\npython scripts\u002Ftraining_preflight.py PROJECT --check-packages\n```\n\nBoth helpers are read-only, deterministic, offline, and use the standard\nlibrary for configuration, filesystem, tag, and SQLite checks. They do not\nlaunch training. Package and TensorFlow import probes are opt-in so basic\nvalidation still works when optional runtime packages are absent.\n\nThe preflight checks a bounded, ID-ordered image sample by default. Use\n`--max-image-checks 0` only when a complete filesystem scan is intended and its\nI\u002FO cost is acceptable. Use `--probe-imports --probe-tensorflow` to test the\ninstalled modules and confirm that TensorFlow exposes a CPU; this still does\nnot train or verify a GPU.\n\nDo not launch training unless all of these hold:\n\n1. `project.json` parses and all dispatch values and numeric ranges are valid.\n2. `tags.txt` is non-empty and is the intended fixed output vocabulary.\n3. The SQLite `posts` table has the required columns and at least one eligible\n   row at `minimum_tag_count`.\n4. Derived image paths exist for the checked rows, and the image format is\n   actually decodable by the training environment.\n5. Checkpoint state is understood: an existing latest checkpoint causes an\n   automatic resume.\n6. The CPU TensorFlow environment can import DeepDanbooru, TensorFlow, and\n   TensorFlow I\u002FO.\n\n## Launch boundary\n\nThe side-effecting command is:\n\n```console\ndeepdanbooru train-project PROJECT_PATH\n```\n\nTo load initial weights\u002Fmodel structure from an existing saved model:\n\n```console\ndeepdanbooru train-project PROJECT_PATH --source-model MODEL.keras\n```\n\n`--source-model` is not a no-resume flag. After the model is loaded, any latest\ncheckpoint under `PROJECT_PATH\u002Fcheckpoints\u002F` is restored and can supersede its\nstate. Preserve or deliberately relocate checkpoints before changing training\nlineage. Also confirm source-model input\u002Foutput compatibility with the image\nshape and number of tags; the training function does not perform a friendly\ncompatibility check.\n\n## Runtime behavior to preserve\n\nTraining compiles with precision and recall metrics, shuffles records once per\nepoch from a checkpointed seed, applies an optional epoch learning-rate\nschedule, trains in checkpoint-sized slices, and saves after each slice.\nCheckpoint state includes optimizer\u002Fmodel variables plus `used_epoch`,\n`used_minibatch`, `used_sample`, `offset`, and `random_seed`; only the latest\nthree checkpoints are retained.\n\nPeriodic exports are named:\n\n```text\nmodel-MODEL.eEPOCH.keras\n```\n\nThe final export is:\n\n```text\nmodel-MODEL.keras\n```\n\nExports omit optimizer state. Resume therefore depends on `checkpoints\u002F`, not\non a `.keras` export. `export_model_per_epoch: 0` means export every epoch;\notherwise an epoch export occurs when the completed epoch is divisible by that\npositive interval.\n\nWith `mixed_precision: true`, training uses a `mixed_float16` model and a\nloss-scaled optimizer, then attempts separate float32 exports ending in\n`.float32.keras`. The 1.0.0 implementation reconstructs a float32 model and\nrestores checkpointed model variables into it. Treat those exports as needing\nexplicit load\u002Finference validation. In particular, combining\n`--source-model` with mixed precision leaves the float32 export model\nunconstructed and can fail during export; the bundled preflight rejects that\ncombination.\n\n## Failure semantics\n\nA missing image is not reliably fatal at the dataset boundary because\n`Dataset.ignore_errors()` can drop load\u002Fdecode failures. A run with many bad\npaths can therefore appear to progress while using fewer samples; a run with\nno usable records can produce checkpoints or model files without meaningful\nupdates. Treat missing files, empty eligible selections, empty tag records, and\nzero usable batches as failures, not warnings to ignore.\n\nUnsupported model, optimizer, or loss strings fail during dispatch. A\n`checkpoint_frequency_mb` of zero can prevent offset progress; a\n`console_logging_frequency_mb` of zero causes modulo-by-zero; non-positive\nbatch\u002Fimage\u002Fepoch values are invalid for an intended training run. Use the\nvalidators rather than discovering these conditions after allocating a long\njob.\n\n## Handoff\n\nAfter training, retain `project.json`, the exact ordered `tags.txt`, the final\n`.keras` model, and any checkpoints needed for further resume. Validate the\nsaved model through [inference-evaluation](..\u002Finference-evaluation\u002FSKILL.md)\nbefore conversion or Grad-CAM work in\n[post-training-tools](..\u002Fpost-training-tools\u002FSKILL.md). If project\u002Fdata checks\nfail, return to [project-data-setup](..\u002Fproject-data-setup\u002FSKILL.md) instead of\nchanging model hyperparameters to mask a data problem.\n",{"name":5,"license":14,"metadata":23,"description":7,"disable-model-invocation":25},{"disco-role":24},"operating",true,"main","ac3fe1afa80fb9a09775ecfb2b6cc3ba850a2db6","skills\u002Frepositories\u002Frepo-skills\u002Fdeep-danbooru\u002Fsub-skills\u002Fmodel-training",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 model-training",{"tool":39,"command":40},"bunx","bunx skills add VectorSpaceLab\u002FAREX-Skill --skill model-training",{"tool":42,"command":43},"pnpm","pnpm dlx skills add VectorSpaceLab\u002FAREX-Skill --skill model-training","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdeep-danbooru\u002Fsub-skills\u002Fmodel-training\u002FSKILL.md",[46,52,56,60,64,70],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"references\u002Fmodel-and-loss-reference.md","REFERENCE","参考文档",5657,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdeep-danbooru\u002Fsub-skills\u002Fmodel-training\u002Freferences\u002Fmodel-and-loss-reference.md",{"path":53,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":54,"sourceHtmlUrl":55},"references\u002Ftraining-data-pipeline.md",5508,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdeep-danbooru\u002Fsub-skills\u002Fmodel-training\u002Freferences\u002Ftraining-data-pipeline.md",{"path":57,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":58,"sourceHtmlUrl":59},"references\u002Ftraining-workflows.md",5991,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdeep-danbooru\u002Fsub-skills\u002Fmodel-training\u002Freferences\u002Ftraining-workflows.md",{"path":61,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":62,"sourceHtmlUrl":63},"references\u002Ftroubleshooting.md",7247,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdeep-danbooru\u002Fsub-skills\u002Fmodel-training\u002Freferences\u002Ftroubleshooting.md",{"path":65,"resourceType":66,"resourceTypeDesc":67,"sizeBytes":68,"sourceHtmlUrl":69},"scripts\u002Ftraining_preflight.py","SCRIPT","脚本",18302,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdeep-danbooru\u002Fsub-skills\u002Fmodel-training\u002Fscripts\u002Ftraining_preflight.py",{"path":71,"resourceType":66,"resourceTypeDesc":67,"sizeBytes":72,"sourceHtmlUrl":73},"scripts\u002Fvalidate_project_config.py",12022,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdeep-danbooru\u002Fsub-skills\u002Fmodel-training\u002Fscripts\u002Fvalidate_project_config.py",[75,83,90,97,104,111],{"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":79,"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.","Apache-2.0","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":79,"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":79,"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":79,"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":79,"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":79,"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",[],[]]