[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6DBJSJ9GRN4A5DSPYF63PG4":3,"skill-same:01M6DBJSJ9GRN4A5DSPYF63PG4":202},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":23,"publishedAt":43,"linkOnly":44,"markdownContent":45,"markdownBody":46,"frontMatter":47,"refName":51,"commitSha":52,"skillDirectory":53,"forksCount":14,"repositoryHtmlUrl":54,"repositoryHomepageUrl":6,"ownerHtmlUrl":55,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":56,"install":57,"resources":70},"01M6DBJSJ9GRN4A5DSPYF63PG4","finetuning",null,"Fine-tune models on Azure AI Foundry using SFT (supervised), DPO (preference), or RFT (reinforcement with graders). Covers dataset preparation, training job submission, deployment, and evaluation. USE FOR: fine-tune, SFT, DPO, RFT, training data, grader, distillation, fine-tuned model, training job, large file upload, calibrate grader, deploy fine-tuned model, evaluate fine-tuned model. DO NOT USE FOR: general model deployment without fine-tuning (use deploy-model), agent creation (use agents), prompt optimization without training (use prompt-optimizer).","Prekzursil\u002Fagent-skills-toolchain","Prekzursil","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F54636077?v=4","01M6DACZ1PCQHD99SCR1YXG28B","Andrei Mihai Visalon","01M6DACZ670WZF535V3WYYSS4Z",0,"MIT","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002FSKILL.md",[18],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[5,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42],"creation","placeholder","optimizer","upload","do","general","use","prompt","model","DPO","Fine-Tuning","Optimization","calibrate","file","azure ai foundry","deployment","training","model deployment","azure","2026-11-10 10:12:30",false,"---\nname: finetuning\ndescription: \"Fine-tune models on Azure AI Foundry using SFT (supervised), DPO (preference), or RFT (reinforcement with graders). Covers dataset preparation, training job submission, deployment, and evaluation. USE FOR: fine-tune, SFT, DPO, RFT, training data, grader, distillation, fine-tuned model, training job, large file upload, calibrate grader, deploy fine-tuned model, evaluate fine-tuned model. DO NOT USE FOR: general model deployment without fine-tuning (use deploy-model), agent creation (use agents), prompt optimization without training (use prompt-optimizer).\"\nlicense: MIT\nmetadata:\n  author: Microsoft\n  version: \"0.0.0-placeholder\"\n---\n\n# Fine-Tuning on Azure AI Foundry\n\nFine-tune models using SFT (supervised), DPO (preference), or RFT (reinforcement with graders). Covers dataset prep, training, deployment, and evaluation.\n\n## When to Use\n\nUse this sub-skill when the user asks about:\n- Fine-tuning a model (SFT, DPO, or RFT)\n- Preparing, validating, or formatting training data\n- Submitting, monitoring, or diagnosing training jobs\n- Calibrating graders or pass thresholds for RFT\n- Deploying or evaluating a fine-tuned model\n- Choosing between training types (SFT vs DPO vs RFT)\n- Distillation, synthetic data generation, or dataset quality scoring\n- Large file uploads for training data\n- Cleaning up fine-tuning resources (files, deployments)\n\n**Do NOT use for:** General model deployment without fine-tuning (use deploy-model), agent creation (use agents), prompt optimization without training (use prompt-optimizer).\n\n## Workflows\n\n| Stage | Guide |\n|-------|-------|\n| **Quick start** | [workflows\u002Fquickstart.md](workflows\u002Fquickstart.md) |\n| **Full pipeline** | [workflows\u002Ffull-pipeline.md](workflows\u002Ffull-pipeline.md) |\n| **Create data** | [workflows\u002Fdataset-creation.md](workflows\u002Fdataset-creation.md) |\n| **Iterate** | [workflows\u002Fiterative-training.md](workflows\u002Fiterative-training.md) |\n| **Diagnose** | [workflows\u002Fdiagnose-poor-results.md](workflows\u002Fdiagnose-poor-results.md) |\n\n## References\n\n| Topic | File |\n|-------|------|\n| SFT vs DPO vs RFT | [references\u002Ftraining-types.md](references\u002Ftraining-types.md) |\n| Hyperparameters | [references\u002Fhyperparameters.md](references\u002Fhyperparameters.md) |\n| Data formats | [references\u002Fdataset-formats.md](references\u002Fdataset-formats.md) |\n| Grader design (RFT) | [references\u002Fgrader-design.md](references\u002Fgrader-design.md) |\n| Reward hacking | [references\u002Freward-hacking.md](references\u002Freward-hacking.md) |\n| Agentic RFT (tools) | [references\u002Fagentic-rft.md](references\u002Fagentic-rft.md) |\n| Deployment | [references\u002Fdeployment.md](references\u002Fdeployment.md) |\n| Training curves | [references\u002Ftraining-curves.md](references\u002Ftraining-curves.md) |\n| Evaluation | [references\u002Fevaluation.md](references\u002Fevaluation.md) |\n| Vision fine-tuning | [references\u002Fvision-fine-tuning.md](references\u002Fvision-fine-tuning.md) |\n| Large file uploads | [references\u002Flarge-file-uploads.md](references\u002Flarge-file-uploads.md) |\n| Platform gotchas | [references\u002Fplatform-gotchas.md](references\u002Fplatform-gotchas.md) |\n\n## Scripts\n\n| Script | Purpose |\n|--------|---------|\n| `scripts\u002Fsubmit_training.py` | Submit SFT\u002FDPO\u002FRFT jobs |\n| `scripts\u002Fmonitor_training.py` | Poll job until completion |\n| `scripts\u002Fcalibrate_grader.py` | Find optimal RFT pass_threshold |\n| `scripts\u002Fcheck_training.py` | Analyze curves, list checkpoints |\n| `scripts\u002Fdeploy_model.py` | Deploy via ARM REST API |\n| `scripts\u002Fevaluate_model.py` | LLM judge evaluation |\n| `scripts\u002Fconvert_dataset.py` | Convert between SFT\u002FDPO\u002FRFT formats |\n| `scripts\u002Fgenerate_distillation_data.py` | Generate synthetic training data |\n| `scripts\u002Fscore_dataset.py` | Quality scoring on training data |\n| `scripts\u002Fcleanup.py` | Delete old files and deployments |\n| `scripts\u002Fvalidate\u002F` | Data validators (SFT, DPO, RFT) + stats |\n\n## Rules\n\n1. **Always baseline first** — evaluate the base model before fine-tuning\n2. **Validate data** before submitting — run `scripts\u002Fvalidate\u002Fvalidate_sft.py`\n3. **Calibrate RFT graders** — target 25-50% failure rate on the base model\n4. **Evaluate checkpoints** — don't blindly deploy the final one\n5. **Measure token cost** alongside accuracy when comparing models\n\n## Quick Reference\n\n| Task | Command |\n|------|---------|\n| Validate SFT data | `python scripts\u002Fvalidate\u002Fvalidate_sft.py data.jsonl` |\n| Submit SFT job | `python scripts\u002Fsubmit_training.py --model gpt-4.1-mini --training-file train.jsonl --validation-file val.jsonl --type sft` |\n| Monitor job | `python scripts\u002Fmonitor_training.py --job-id ftjob-xxx` |\n| Analyze curves | `python scripts\u002Fcheck_training.py --job-id ftjob-xxx` |\n| Deploy model | `python scripts\u002Fdeploy_model.py --model-id ft:gpt-4.1-mini:... --name my-eval` |\n| Evaluate model | `python scripts\u002Fevaluate_model.py --deployment-name my-eval --test-file test.jsonl` |\n\n## Error Handling\n\n| Error | Cause | Fix |\n|-------|-------|-----|\n| \"API version not supported\" | Older `openai` SDK on `\u002Fv1\u002F` endpoint | Upgrade to `openai>=1.0` |\n| \"does not support fine-tuning with Standard TrainingType\" | OSS model needs `globalStandard` | Use `--use-rest` flag or script auto-falls back |\n| Job stuck in post-training eval | Under-provisioned tool endpoint (RFT) | Scale to S2+, enable Always On |\n| \"DeploymentNotReady\" after ARM succeeds | ARM\u002Fdata-plane race condition | Delete and recreate deployment, wait 5 min |\n| Content safety block at deployment | PII-dense training data | Remove problematic document types |\n","# Fine-Tuning on Azure AI Foundry\n\nFine-tune models using SFT (supervised), DPO (preference), or RFT (reinforcement with graders). Covers dataset prep, training, deployment, and evaluation.\n\n## When to Use\n\nUse this sub-skill when the user asks about:\n- Fine-tuning a model (SFT, DPO, or RFT)\n- Preparing, validating, or formatting training data\n- Submitting, monitoring, or diagnosing training jobs\n- Calibrating graders or pass thresholds for RFT\n- Deploying or evaluating a fine-tuned model\n- Choosing between training types (SFT vs DPO vs RFT)\n- Distillation, synthetic data generation, or dataset quality scoring\n- Large file uploads for training data\n- Cleaning up fine-tuning resources (files, deployments)\n\n**Do NOT use for:** General model deployment without fine-tuning (use deploy-model), agent creation (use agents), prompt optimization without training (use prompt-optimizer).\n\n## Workflows\n\n| Stage | Guide |\n|-------|-------|\n| **Quick start** | [workflows\u002Fquickstart.md](workflows\u002Fquickstart.md) |\n| **Full pipeline** | [workflows\u002Ffull-pipeline.md](workflows\u002Ffull-pipeline.md) |\n| **Create data** | [workflows\u002Fdataset-creation.md](workflows\u002Fdataset-creation.md) |\n| **Iterate** | [workflows\u002Fiterative-training.md](workflows\u002Fiterative-training.md) |\n| **Diagnose** | [workflows\u002Fdiagnose-poor-results.md](workflows\u002Fdiagnose-poor-results.md) |\n\n## References\n\n| Topic | File |\n|-------|------|\n| SFT vs DPO vs RFT | [references\u002Ftraining-types.md](references\u002Ftraining-types.md) |\n| Hyperparameters | [references\u002Fhyperparameters.md](references\u002Fhyperparameters.md) |\n| Data formats | [references\u002Fdataset-formats.md](references\u002Fdataset-formats.md) |\n| Grader design (RFT) | [references\u002Fgrader-design.md](references\u002Fgrader-design.md) |\n| Reward hacking | [references\u002Freward-hacking.md](references\u002Freward-hacking.md) |\n| Agentic RFT (tools) | [references\u002Fagentic-rft.md](references\u002Fagentic-rft.md) |\n| Deployment | [references\u002Fdeployment.md](references\u002Fdeployment.md) |\n| Training curves | [references\u002Ftraining-curves.md](references\u002Ftraining-curves.md) |\n| Evaluation | [references\u002Fevaluation.md](references\u002Fevaluation.md) |\n| Vision fine-tuning | [references\u002Fvision-fine-tuning.md](references\u002Fvision-fine-tuning.md) |\n| Large file uploads | [references\u002Flarge-file-uploads.md](references\u002Flarge-file-uploads.md) |\n| Platform gotchas | [references\u002Fplatform-gotchas.md](references\u002Fplatform-gotchas.md) |\n\n## Scripts\n\n| Script | Purpose |\n|--------|---------|\n| `scripts\u002Fsubmit_training.py` | Submit SFT\u002FDPO\u002FRFT jobs |\n| `scripts\u002Fmonitor_training.py` | Poll job until completion |\n| `scripts\u002Fcalibrate_grader.py` | Find optimal RFT pass_threshold |\n| `scripts\u002Fcheck_training.py` | Analyze curves, list checkpoints |\n| `scripts\u002Fdeploy_model.py` | Deploy via ARM REST API |\n| `scripts\u002Fevaluate_model.py` | LLM judge evaluation |\n| `scripts\u002Fconvert_dataset.py` | Convert between SFT\u002FDPO\u002FRFT formats |\n| `scripts\u002Fgenerate_distillation_data.py` | Generate synthetic training data |\n| `scripts\u002Fscore_dataset.py` | Quality scoring on training data |\n| `scripts\u002Fcleanup.py` | Delete old files and deployments |\n| `scripts\u002Fvalidate\u002F` | Data validators (SFT, DPO, RFT) + stats |\n\n## Rules\n\n1. **Always baseline first** — evaluate the base model before fine-tuning\n2. **Validate data** before submitting — run `scripts\u002Fvalidate\u002Fvalidate_sft.py`\n3. **Calibrate RFT graders** — target 25-50% failure rate on the base model\n4. **Evaluate checkpoints** — don't blindly deploy the final one\n5. **Measure token cost** alongside accuracy when comparing models\n\n## Quick Reference\n\n| Task | Command |\n|------|---------|\n| Validate SFT data | `python scripts\u002Fvalidate\u002Fvalidate_sft.py data.jsonl` |\n| Submit SFT job | `python scripts\u002Fsubmit_training.py --model gpt-4.1-mini --training-file train.jsonl --validation-file val.jsonl --type sft` |\n| Monitor job | `python scripts\u002Fmonitor_training.py --job-id ftjob-xxx` |\n| Analyze curves | `python scripts\u002Fcheck_training.py --job-id ftjob-xxx` |\n| Deploy model | `python scripts\u002Fdeploy_model.py --model-id ft:gpt-4.1-mini:... --name my-eval` |\n| Evaluate model | `python scripts\u002Fevaluate_model.py --deployment-name my-eval --test-file test.jsonl` |\n\n## Error Handling\n\n| Error | Cause | Fix |\n|-------|-------|-----|\n| \"API version not supported\" | Older `openai` SDK on `\u002Fv1\u002F` endpoint | Upgrade to `openai>=1.0` |\n| \"does not support fine-tuning with Standard TrainingType\" | OSS model needs `globalStandard` | Use `--use-rest` flag or script auto-falls back |\n| Job stuck in post-training eval | Under-provisioned tool endpoint (RFT) | Scale to S2+, enable Always On |\n| \"DeploymentNotReady\" after ARM succeeds | ARM\u002Fdata-plane race condition | Delete and recreate deployment, wait 5 min |\n| Content safety block at deployment | PII-dense training data | Remove problematic document types |\n",{"name":5,"license":15,"metadata":48,"description":7},{"author":49,"version":50},"Microsoft","0.0.0-placeholder","main","a298c929b6cbd662e1fce54302ce520d5ceea1a5",".agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain","https:\u002F\u002Fgithub.com\u002FPrekzursil","2026-11-02 14:30:57",{"source":8,"skillName":5,"commands":58,"prompt":68,"zipAvailable":69},[59,62,65],{"tool":60,"command":61},"npx","npx skills add Prekzursil\u002Fagent-skills-toolchain --skill finetuning",{"tool":63,"command":64},"bunx","bunx skills add Prekzursil\u002Fagent-skills-toolchain --skill finetuning",{"tool":66,"command":67},"pnpm","pnpm dlx skills add Prekzursil\u002Fagent-skills-toolchain --skill finetuning","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002FSKILL.md",true,[71,77,81,85,89,93,96,100,104,108,112,116,120,126,130,134,138,142,146,150,154,158,162,166,170,174,178,182,186,190,194,198],{"path":72,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":75,"sourceHtmlUrl":76},"references\u002Fagentic-rft.md","REFERENCE","参考文档",3233,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Freferences\u002Fagentic-rft.md",{"path":78,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":79,"sourceHtmlUrl":80},"references\u002Fdataset-formats.md",3740,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Freferences\u002Fdataset-formats.md",{"path":82,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":83,"sourceHtmlUrl":84},"references\u002Fdeployment.md",3503,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Freferences\u002Fdeployment.md",{"path":86,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":87,"sourceHtmlUrl":88},"references\u002Fevaluation.md",5924,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Freferences\u002Fevaluation.md",{"path":90,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":91,"sourceHtmlUrl":92},"references\u002Fgrader-design.md",3437,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Freferences\u002Fgrader-design.md",{"path":94,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":75,"sourceHtmlUrl":95},"references\u002Fhyperparameters.md","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Freferences\u002Fhyperparameters.md",{"path":97,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":98,"sourceHtmlUrl":99},"references\u002Flarge-file-uploads.md",927,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Freferences\u002Flarge-file-uploads.md",{"path":101,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":102,"sourceHtmlUrl":103},"references\u002Fplatform-gotchas.md",2069,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Freferences\u002Fplatform-gotchas.md",{"path":105,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":106,"sourceHtmlUrl":107},"references\u002Freward-hacking.md",2945,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Freferences\u002Freward-hacking.md",{"path":109,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":110,"sourceHtmlUrl":111},"references\u002Ftraining-curves.md",4183,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Freferences\u002Ftraining-curves.md",{"path":113,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":114,"sourceHtmlUrl":115},"references\u002Ftraining-types.md",3158,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Freferences\u002Ftraining-types.md",{"path":117,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":118,"sourceHtmlUrl":119},"references\u002Fvision-fine-tuning.md",4205,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Freferences\u002Fvision-fine-tuning.md",{"path":121,"resourceType":122,"resourceTypeDesc":123,"sizeBytes":124,"sourceHtmlUrl":125},"scripts\u002Fcalibrate_grader.py","SCRIPT","脚本",9907,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Fscripts\u002Fcalibrate_grader.py",{"path":127,"resourceType":122,"resourceTypeDesc":123,"sizeBytes":128,"sourceHtmlUrl":129},"scripts\u002Fcheck_training.py",7983,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Fscripts\u002Fcheck_training.py",{"path":131,"resourceType":122,"resourceTypeDesc":123,"sizeBytes":132,"sourceHtmlUrl":133},"scripts\u002Fcleanup.py",9687,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Fscripts\u002Fcleanup.py",{"path":135,"resourceType":122,"resourceTypeDesc":123,"sizeBytes":136,"sourceHtmlUrl":137},"scripts\u002Fcommon.py",7799,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Fscripts\u002Fcommon.py",{"path":139,"resourceType":122,"resourceTypeDesc":123,"sizeBytes":140,"sourceHtmlUrl":141},"scripts\u002Fconvert_dataset.py",10936,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Fscripts\u002Fconvert_dataset.py",{"path":143,"resourceType":122,"resourceTypeDesc":123,"sizeBytes":144,"sourceHtmlUrl":145},"scripts\u002Fdeploy_model.py",9616,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Fscripts\u002Fdeploy_model.py",{"path":147,"resourceType":122,"resourceTypeDesc":123,"sizeBytes":148,"sourceHtmlUrl":149},"scripts\u002Fevaluate_model.py",11690,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Fscripts\u002Fevaluate_model.py",{"path":151,"resourceType":122,"resourceTypeDesc":123,"sizeBytes":152,"sourceHtmlUrl":153},"scripts\u002Fgenerate_distillation_data.py",10297,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Fscripts\u002Fgenerate_distillation_data.py",{"path":155,"resourceType":122,"resourceTypeDesc":123,"sizeBytes":156,"sourceHtmlUrl":157},"scripts\u002Fmonitor_training.py",5863,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Fscripts\u002Fmonitor_training.py",{"path":159,"resourceType":122,"resourceTypeDesc":123,"sizeBytes":160,"sourceHtmlUrl":161},"scripts\u002Fscore_dataset.py",7872,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Fscripts\u002Fscore_dataset.py",{"path":163,"resourceType":122,"resourceTypeDesc":123,"sizeBytes":164,"sourceHtmlUrl":165},"scripts\u002Fsubmit_training.py",11067,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Fscripts\u002Fsubmit_training.py",{"path":167,"resourceType":122,"resourceTypeDesc":123,"sizeBytes":168,"sourceHtmlUrl":169},"scripts\u002Fvalidate\u002Fdata_stats.py",6323,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Fscripts\u002Fvalidate\u002Fdata_stats.py",{"path":171,"resourceType":122,"resourceTypeDesc":123,"sizeBytes":172,"sourceHtmlUrl":173},"scripts\u002Fvalidate\u002Fvalidate_dpo.py",3802,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Fscripts\u002Fvalidate\u002Fvalidate_dpo.py",{"path":175,"resourceType":122,"resourceTypeDesc":123,"sizeBytes":176,"sourceHtmlUrl":177},"scripts\u002Fvalidate\u002Fvalidate_rft.py",8789,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Fscripts\u002Fvalidate\u002Fvalidate_rft.py",{"path":179,"resourceType":122,"resourceTypeDesc":123,"sizeBytes":180,"sourceHtmlUrl":181},"scripts\u002Fvalidate\u002Fvalidate_sft.py",4568,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Fscripts\u002Fvalidate\u002Fvalidate_sft.py",{"path":183,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":184,"sourceHtmlUrl":185},"workflows\u002Fdataset-creation.md",3161,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Fworkflows\u002Fdataset-creation.md",{"path":187,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":188,"sourceHtmlUrl":189},"workflows\u002Fdiagnose-poor-results.md",2401,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Fworkflows\u002Fdiagnose-poor-results.md",{"path":191,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":192,"sourceHtmlUrl":193},"workflows\u002Ffull-pipeline.md",3165,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Fworkflows\u002Ffull-pipeline.md",{"path":195,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":196,"sourceHtmlUrl":197},"workflows\u002Fiterative-training.md",3178,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Fworkflows\u002Fiterative-training.md",{"path":199,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":200,"sourceHtmlUrl":201},"workflows\u002Fquickstart.md",4598,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fmicrosoft-foundry\u002Ffinetuning\u002Fworkflows\u002Fquickstart.md",[203,223,249,276,290,306],{"skillNo":204,"name":205,"nameZh":6,"description":206,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":207,"categories":208,"keywords":209,"publishedAt":43},"01M6DF7D7478R55P46REYH4QAC","clarify-requirements","Resolve genuine pre-execution ambiguity by scanning repo context, generating 1-6 grounded options, and asking 1-3 focused questions via AskUserQuestion. NOT for mid-task use (injection risk); NOT when the goal is already unambiguous; use to disambiguate vague prompts before any file edits.","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Fworkspace-hooks\u002Fskills\u002Fclarify-requirements\u002FSKILL.md",[],[210,211,212,213,214,215,216,217,218,30,219,220,221,222,37],"clarify requirements","clarify","requirements","risk","options","task","prompts","goal","scanning","context","repo","injection","execution",{"skillNo":224,"name":225,"nameZh":6,"description":226,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":227,"categories":228,"keywords":230,"publishedAt":43},"01M6DF79CA4BHJS66J8EZDBPWJ","openmm-md","Run FREE, open-source, Windows-native GPU molecular dynamics via the openmm-md MCP (OpenMM + CUDA on the local NVIDIA GPU). Use for protein \u002F protein-ligand MD when you want a free stack or are on Windows without a Schrodinger license. Covers the full workflow (prepare, solvate, minimize, equilibrate, production, analyze, mmgbsa) AND an honest decision guide vs Schrodinger Desmond, including how to run Desmond legitimately on Linux or a cloud GPU with a real license. This is a deliberately LESSER free alternative and does NOT replace the schrodinger MCP tools. Do NOT use it to pick a free tool when a validated FEP+, WaterMap, IFD-MD, or Prime MM-GBSA result is actually required.","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Fmcp\u002Fopenmm-md-mcp\u002Fskill\u002Fopenmm-md\u002FSKILL.md",[229],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},[231,232,233,234,235,236,237,238,239,240,241,242,243,244,30,245,246,247,28,248],"openmm md","how","decision","pick","linux","analyze","tool","stack","open","NVIDIA","gpu","license","workflow","windows","free","molecular dynamics","local","MCP",{"skillNo":250,"name":251,"nameZh":6,"description":252,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":253,"categories":254,"keywords":261,"publishedAt":43},"01M6DF76C1MG85VEWP8VZCBNMN","visual-review","Take screenshots of web pages and UI using Playwright for visual review and iteration","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Feditor\u002Fmetaswarm\u002Fskills\u002Fvisual-review\u002FSKILL.md",[255,258],{"categoryNo":256,"categoryName":257,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000003","演示文稿",{"categoryNo":259,"categoryName":260,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000008","测试与质量",[262,263,264,265,266,267,268,269,270,271,232,272,273,274,275],"review","visual","visual review","pages","web","false","iteration","take","check","slides","ui","screenshot","playwright","screenshots",{"skillNo":277,"name":278,"nameZh":6,"description":279,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":280,"categories":281,"keywords":282,"publishedAt":43},"01M6DF73AQVCBZKEN0YXEX9FXP","status","Diagnostic status report — shows metaswarm installation state, project setup, and potential issues","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Feditor\u002Fmetaswarm\u002Fskills\u002Fstatus\u002FSKILL.md",[],[278,283,284,285,286,287,288,289],"setup","report","project","installation","project setup","issues","state",{"skillNo":291,"name":292,"nameZh":6,"description":293,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":294,"categories":295,"keywords":297,"publishedAt":43},"01M6DF6X6C2XB0P6GX76P7TTYV","start","Use when starting work on any task, when the user mentions metaswarm, or when the user wants to begin tracked development work","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Feditor\u002Fmetaswarm\u002Fskills\u002Fstart\u002FSKILL.md",[296],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},[292,298,215,299,300,301,30,302,303,304,305],"start issue","issue","ready","work on issue","user","work","work on","development",{"skillNo":307,"name":283,"nameZh":6,"description":308,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":309,"categories":310,"keywords":311,"publishedAt":43},"01M6DF6QMJ9AJKFKHHKRT005PQ","Interactive project setup — detects your project, configures metaswarm, writes project-local files","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Feditor\u002Fmetaswarm\u002Fskills\u002Fsetup\u002FSKILL.md",[],[283,285,247,312,287,313],"files","interactive"]