[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-microsoft-msresearch-dayhoff":3,"mdc--kilcdo-key":34,"related-org-microsoft-msresearch-dayhoff":1173,"related-repo-microsoft-msresearch-dayhoff":1370},{"slug":4,"name":4,"fn":5,"description":6,"org":7,"tags":11,"stars":23,"repoUrl":24,"updatedAt":25,"license":26,"forks":27,"topics":28,"repo":29,"sourceUrl":32,"mdContent":33},"msresearch-dayhoff","generate and score protein sequences with Dayhoff","Microsoft Research's Dayhoff — generates novel protein sequences and scores variants using hybrid Mamba+Transformer models trained on the 3.34B-sequence Dayhoff Atlas. Use when users mention Dayhoff by name, want to generate protein sequences from scratch or from a prompt, score mutations \u002F variants \u002F fitness without experimental data, design sequences in a protein family from homologs, scaffold sequences around a motif, or are setting up \u002F running \u002F troubleshooting the Dayhoff prototype locally.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},"microsoft","Microsoft","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fmicrosoft.png",[12,16,19,22],{"name":13,"slug":14,"type":15},"Proteomics","proteomics","tag",{"name":17,"slug":18,"type":15},"Deep Learning","deep-learning",{"name":20,"slug":21,"type":15},"Bioinformatics","bioinformatics",{"name":9,"slug":8,"type":15},34,"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fvibe-kit","2026-07-07T06:53:47.296995","MIT",7,[],{"repoUrl":24,"stars":23,"forks":27,"topics":30,"description":31},[],"Microsoft Research Vibe Kit is a customized vibe coding environment to enable faster prototyping and integration of Microsoft Research innovations using GitHub Copilot inside Visual Studio Code. ","https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fvibe-kit\u002Ftree\u002FHEAD\u002Fskills\u002Fmsresearch-dayhoff","---\nname: msresearch-dayhoff\ndescription: Microsoft Research's Dayhoff — generates novel protein sequences and scores variants using hybrid Mamba+Transformer models trained on the 3.34B-sequence Dayhoff Atlas. Use when users mention Dayhoff by name, want to generate protein sequences from scratch or from a prompt, score mutations \u002F variants \u002F fitness without experimental data, design sequences in a protein family from homologs, scaffold sequences around a motif, or are setting up \u002F running \u002F troubleshooting the Dayhoff prototype locally.\nlicense: MIT\n---\n\n## Scope\n\n- Run the Dayhoff reference app locally to generate novel protein sequences, score variants, scaffold motifs, and (for some variants) condition on homologous sequences.\n- Three ways to run the reference app, in increasing order of capability and setup cost:\n  1. **Path A — Cached demo:** frontend only (`npm run dev`), no GPU, no backend. Returns precomputed real Dayhoff outputs for four bundled prompts (Cas9, insulin, DNA polymerase, SARS-CoV-2 spike) across all four model variants. Best for evaluating the UX or demoing on a laptop with no GPU access.\n  2. **Path B — Proxy mode:** Flask proxy on `:8000` + frontend on `:5173`, calling a `dayhoff-multi` Azure ML endpoint. Two terminals, no local model load. Live arbitrary-prompt generation.\n  3. **Path C — Fully local:** adds a self-hosted score server on `:5001` (and optionally a fold server) running on a CUDA-capable NVIDIA GPU. Three to four terminals.\n- Hosted models are also published to Hugging Face and Azure AI Foundry — use the proxy in `backend\u002Fapp.py` to call a `dayhoff-multi` AML endpoint, or use the `transformers` library directly against [the Dayhoff model collection](https:\u002F\u002Fhuggingface.co\u002Fcollections\u002Fmicrosoft\u002Fdayhoff-atlas-6866d679465a2685b06ee969).\n- Four model variants (`3b-GR-HM-c`, `3b-GR-HM`, `3b-UR90`, `170m-UR50-BRn`) with different strengths — see `docs\u002Fabout-dayhoff.md` for the picker.\n- Bundled biosecurity guardrail: `backend\u002Fsequence_screening.py` runs a 3-layer Select Agent toxin screen on inputs and outputs. This is a guardrail, not a guarantee — see `docs\u002Fresponsible-use.md`.\n\n## Prerequisites\n\n- **Windows users:** Dayhoff prototype is developed and tested on Linux \u002F WSL2. On Windows 11, run everything inside a WSL2 distro (Ubuntu recommended) with NVIDIA drivers installed on the **Windows host** (not inside WSL). Native Windows Python \u002F PowerShell is not supported. See [Microsoft's WSL install guide](https:\u002F\u002Flearn.microsoft.com\u002Fwindows\u002Fwsl\u002Finstall) and [NVIDIA's CUDA-on-WSL guide](https:\u002F\u002Fdocs.nvidia.com\u002Fcuda\u002Fwsl-user-guide\u002Findex.html). (Path A — cached demo — is the one exception: if a Windows user just wants to see Dayhoff outputs in the UI, `npm run dev` from the frontend on native Windows works fine because it never invokes the Python backend or the GPU.)\n- Linux (or WSL2) with a CUDA-capable NVIDIA GPU for Path C. Path B needs only network access to the AML endpoint. Path A needs only Node.js. CPU works only for the 170M variant on toy sequences and is unusably slow for the 3B variants.\n- Python 3.10+ for the backend (Path B and Path C); Node.js 18+ for the frontend (all paths).\n- Disk space:\n  - Path A (cached demo): ~200 MB for `node_modules`.\n  - Path B (proxy-only, calling a remote AML endpoint): ~1 GB for Python deps.\n  - Path C with the 170M variant: ~2 GB additional for weights.\n  - Path C with all four variants loaded: ~20 GB GPU memory and ~25 GB disk for weights.\n\n## Workflow\n\n1. Load `docs\u002Fabout-dayhoff.md` when users ask what Dayhoff is, how it works, which variant to use, or for benchmark numbers.\n2. Follow `docs\u002Fquick-start.md` to get the reference app running locally — proxy first, then frontend, then optional score \u002F fold servers.\n3. For code examples (generation, fitness scoring, motif preservation, batch sampling, FASTA \u002F CSV \u002F JSON export), route to `docs\u002Fapplication-patterns.md`.\n4. For loading the Dayhoff Atlas datasets (GigaRef, BackboneRef, DayhoffRef), route to `docs\u002Fdata-integration.md`.\n5. For extending the prototype (new endpoints, custom scoring filters, batch jobs, database integration), route to `docs\u002Fprototype-expansion.md`.\n6. For GPU sizing, throughput tuning, and scaling, route to `docs\u002Fperformance-guide.md`.\n7. For errors, route to `docs\u002Ftroubleshooting.md`.\n8. **Before any sequence is exported, shared, or sent to a wet lab**, route to `docs\u002Fresponsible-use.md`.\n\n### Operating rules\n\n- **Windows users:** Dayhoff is Linux\u002FWSL2-only for Path B and Path C. On Windows, all backend \u002F score-server commands must run inside a WSL2 distro with NVIDIA drivers installed on the Windows host. Native Windows Python and PowerShell are not supported. Point Windows users at [Microsoft's WSL install guide](https:\u002F\u002Flearn.microsoft.com\u002Fwindows\u002Fwsl\u002Finstall) and [NVIDIA's CUDA-on-WSL guide](https:\u002F\u002Fdocs.nvidia.com\u002Fcuda\u002Fwsl-user-guide\u002Findex.html) before running any Path B\u002FC step. (Path A — cached demo — is the one exception: pure-frontend `npm run dev` works on native Windows because it never invokes Python or the GPU.)\n- **No-GPU demo path exists:** If the user has no GPU, no AML endpoint, or just wants a quick look at what Dayhoff produces, route to **Path A** in `docs\u002Fquick-start.md` — frontend only with cached real Dayhoff outputs for four bundled prompts. Don't gatekeep on hardware. The cache lives in `frontend\u002Fsrc\u002FdemoCache.ts` and is what the UI shows when the backend badge says \"Backend offline · cached examples only.\"\n- **GPU probe first (Path C only):** Before suggesting Path C (local model loading), run `nvidia-smi` and confirm a CUDA-capable GPU is visible. If none is present, default to Path A (cached demo) or Path B (remote AML endpoint) — do not try to load the 3B models on CPU.\n- **First-run weight download (Path C):** First call to the local score server downloads model weights from Hugging Face. The 170M variant is ~1 GB; loading all four variants is ~25 GB. Tell the user this will be slow and must not be interrupted. Subsequent runs reuse the cache. Recommend `huggingface-cli login` and `HUGGINGFACE_HUB_ENABLE_HF_TRANSFER=1` to avoid throttling.\n- **Multi-terminal discipline:** Path A needs one terminal (frontend on `:5173`). Path B needs two (proxy backend on `:8000` + frontend on `:5173`). Path C needs three (add local score server on `:5001`), and up to four with a local fold server. Each must stay running in its own terminal — never run other commands in a terminal hosting a live server.\n- **Local credentials handling:** Copy `.env.example` to `.env` and fill in `DAYHOFF_ENDPOINT` \u002F `DAYHOFF_API_KEY` for the AML proxy mode, or leave them pointing at `http:\u002F\u002Flocalhost:5001\u002Fscore` for fully-local mode. Never ask the user to paste credentials into chat — always reference the `.env` file. Keep the `DAYHOFF_*` and `ESMFOLD_*` variable names exactly as written; the proxy reads those exact names.\n- **Honest scope:** The reference app is a *research prototype*. Generated sequences are model predictions of plausibility, not validated designs. The bundled `sequence_screening.py` toxin filter catches a small Select Agent reference set — it is a guardrail, not a guarantee. Do not claim safety, function, or foldability for any generated sequence.\n- **Biosecurity gate:** Before assisting with any sequence export, sharing, or wet-lab handoff, surface `docs\u002Fresponsible-use.md` and confirm the user has reviewed the pre-export checklist. Never assist with sequences targeting pathogens, toxins, or dual-use applications. Treat every generated sequence as a research draft requiring qualified human review.\n- **Variant selection matters:** When a user asks for sequence generation, pick the right variant (see `docs\u002Fabout-dayhoff.md` matrix). Reject homolog inputs for `3b-UR90` and `170m-UR50-BRn` early — the score server will return 400 if a model that doesn't support homolog conditioning receives a `homologs` payload.\n- **Execute, don't display:** When terminal execution is available, run quick-start commands directly rather than printing bash blocks for the user to copy.\n- **Always offer the next step:** After loading any explainer-style doc (e.g. `docs\u002Fabout-dayhoff.md`), end your response with a concrete offer to advance the user. Default phrasing: *\"Want to try Dayhoff? Fastest path is just `npm run dev` in the bundled frontend — no GPU, no backend, four cached prompts (Cas9, insulin, DNA polymerase, SARS-CoV-2 spike) return real Dayhoff outputs instantly. From there we can stand up proxy mode against an Azure ML endpoint or fully local mode on your GPU when you want custom prompts.\"* Adapt wording to context.\n\n## Routing\n\n| Doc | When to load |\n|---|---|\n| `docs\u002Fabout-dayhoff.md` | User asks what Dayhoff is, how it works, which variant to use, benchmarks, or scientific background |\n| `docs\u002Fquick-start.md` | User wants to run the reference app locally |\n| `docs\u002Fapplication-patterns.md` | User wants code examples for generation, fitness scoring, motif preservation, batch sampling, or export |\n| `docs\u002Fdata-integration.md` | User wants to load Dayhoff Atlas datasets (GigaRef, BackboneRef, DayhoffRef) |\n| `docs\u002Fprototype-expansion.md` | User wants to extend the reference app — new endpoints, custom scoring, batch processing, database, validation |\n| `docs\u002Fperformance-guide.md` | User asks about GPU sizing, throughput, batching, or scaling |\n| `docs\u002Ftroubleshooting.md` | User hits an error: missing GPU, Mamba kernel error, weight download stall, port conflict, AML 401, HF rate limit |\n| `docs\u002Fresponsible-use.md` | **Before** any sequence export, sharing, or wet-lab handoff; or when user asks about safety, biosecurity, or responsible use |\n\n## Learning Path\n\n1. `docs\u002Fabout-dayhoff.md` — Understand what Dayhoff is, the four model variants, and which one fits your task\n2. `docs\u002Fquick-start.md` — Run the reference app locally and generate your first sequences\n3. `docs\u002Fresponsible-use.md` — Read the biosecurity guardrails and pre-export checklist before exporting anything\n4. `docs\u002Fapplication-patterns.md` — Build your own generation, scoring, and export workflows\n5. `docs\u002Fdata-integration.md` and `docs\u002Fprototype-expansion.md` — Scale up with Atlas datasets and custom prototype features\n\n## Reference Links\n\n- Research paper (preprint): https:\u002F\u002Faka.ms\u002Fdayhoff\u002Fpreprint\n- GitHub: https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fdayhoff\n- Hugging Face model collection: https:\u002F\u002Fhuggingface.co\u002Fcollections\u002Fmicrosoft\u002Fdayhoff-atlas-6866d679465a2685b06ee969\n- Azure AI Foundry model card: https:\u002F\u002Fai.azure.com\u002Fcatalog\u002Fmodels\u002FDayhoff-170m-GR\n\n## Assets\n\n- `assets\u002Fdayhoff-prototype\u002F` — Self-contained reference app: Flask proxy backend (`backend\u002Fapp.py`, port `:8000`), local score server (`backend\u002Fscore\u002F`, port `:5001`), optional ESMFold structure server (`backend\u002Ffold\u002F`), Vite + React + TypeScript frontend (`frontend\u002F`, port `:5173`), Azure ML deployment YAMLs (`infra\u002F`), and `Dockerfile` for App Service deployment.\n- `assets\u002Fpaper\u002Fdayhoff-biorxiv-paper.pdf` — Bundled offline copy of the bioRxiv preprint.\n",{"data":35,"body":36},{"name":4,"description":6,"license":26},{"type":37,"children":38},"root",[39,48,241,247,337,343,447,454,792,798,959,965,1024,1030,1076,1082],{"type":40,"tag":41,"props":42,"children":44},"element","h2",{"id":43},"scope",[45],{"type":46,"value":47},"text","Scope",{"type":40,"tag":49,"props":50,"children":51},"ul",{},[52,58,139,178,221],{"type":40,"tag":53,"props":54,"children":55},"li",{},[56],{"type":46,"value":57},"Run the Dayhoff reference app locally to generate novel protein sequences, score variants, scaffold motifs, and (for some variants) condition on homologous sequences.",{"type":40,"tag":53,"props":59,"children":60},{},[61,63],{"type":46,"value":62},"Three ways to run the reference app, in increasing order of capability and setup cost:\n",{"type":40,"tag":64,"props":65,"children":66},"ol",{},[67,87,121],{"type":40,"tag":53,"props":68,"children":69},{},[70,76,78,85],{"type":40,"tag":71,"props":72,"children":73},"strong",{},[74],{"type":46,"value":75},"Path A — Cached demo:",{"type":46,"value":77}," frontend only (",{"type":40,"tag":79,"props":80,"children":82},"code",{"className":81},[],[83],{"type":46,"value":84},"npm run dev",{"type":46,"value":86},"), no GPU, no backend. Returns precomputed real Dayhoff outputs for four bundled prompts (Cas9, insulin, DNA polymerase, SARS-CoV-2 spike) across all four model variants. Best for evaluating the UX or demoing on a laptop with no GPU access.",{"type":40,"tag":53,"props":88,"children":89},{},[90,95,97,103,105,111,113,119],{"type":40,"tag":71,"props":91,"children":92},{},[93],{"type":46,"value":94},"Path B — Proxy mode:",{"type":46,"value":96}," Flask proxy on ",{"type":40,"tag":79,"props":98,"children":100},{"className":99},[],[101],{"type":46,"value":102},":8000",{"type":46,"value":104}," + frontend on ",{"type":40,"tag":79,"props":106,"children":108},{"className":107},[],[109],{"type":46,"value":110},":5173",{"type":46,"value":112},", calling a ",{"type":40,"tag":79,"props":114,"children":116},{"className":115},[],[117],{"type":46,"value":118},"dayhoff-multi",{"type":46,"value":120}," Azure ML endpoint. Two terminals, no local model load. Live arbitrary-prompt generation.",{"type":40,"tag":53,"props":122,"children":123},{},[124,129,131,137],{"type":40,"tag":71,"props":125,"children":126},{},[127],{"type":46,"value":128},"Path C — Fully local:",{"type":46,"value":130}," adds a self-hosted score server on ",{"type":40,"tag":79,"props":132,"children":134},{"className":133},[],[135],{"type":46,"value":136},":5001",{"type":46,"value":138}," (and optionally a fold server) running on a CUDA-capable NVIDIA GPU. Three to four terminals.",{"type":40,"tag":53,"props":140,"children":141},{},[142,144,150,152,157,159,165,167,176],{"type":46,"value":143},"Hosted models are also published to Hugging Face and Azure AI Foundry — use the proxy in ",{"type":40,"tag":79,"props":145,"children":147},{"className":146},[],[148],{"type":46,"value":149},"backend\u002Fapp.py",{"type":46,"value":151}," to call a ",{"type":40,"tag":79,"props":153,"children":155},{"className":154},[],[156],{"type":46,"value":118},{"type":46,"value":158}," AML endpoint, or use the ",{"type":40,"tag":79,"props":160,"children":162},{"className":161},[],[163],{"type":46,"value":164},"transformers",{"type":46,"value":166}," library directly against ",{"type":40,"tag":168,"props":169,"children":173},"a",{"href":170,"rel":171},"https:\u002F\u002Fhuggingface.co\u002Fcollections\u002Fmicrosoft\u002Fdayhoff-atlas-6866d679465a2685b06ee969",[172],"nofollow",[174],{"type":46,"value":175},"the Dayhoff model collection",{"type":46,"value":177},".",{"type":40,"tag":53,"props":179,"children":180},{},[181,183,189,191,197,198,204,205,211,213,219],{"type":46,"value":182},"Four model variants (",{"type":40,"tag":79,"props":184,"children":186},{"className":185},[],[187],{"type":46,"value":188},"3b-GR-HM-c",{"type":46,"value":190},", ",{"type":40,"tag":79,"props":192,"children":194},{"className":193},[],[195],{"type":46,"value":196},"3b-GR-HM",{"type":46,"value":190},{"type":40,"tag":79,"props":199,"children":201},{"className":200},[],[202],{"type":46,"value":203},"3b-UR90",{"type":46,"value":190},{"type":40,"tag":79,"props":206,"children":208},{"className":207},[],[209],{"type":46,"value":210},"170m-UR50-BRn",{"type":46,"value":212},") with different strengths — see ",{"type":40,"tag":79,"props":214,"children":216},{"className":215},[],[217],{"type":46,"value":218},"docs\u002Fabout-dayhoff.md",{"type":46,"value":220}," for the picker.",{"type":40,"tag":53,"props":222,"children":223},{},[224,226,232,234,240],{"type":46,"value":225},"Bundled biosecurity guardrail: ",{"type":40,"tag":79,"props":227,"children":229},{"className":228},[],[230],{"type":46,"value":231},"backend\u002Fsequence_screening.py",{"type":46,"value":233}," runs a 3-layer Select Agent toxin screen on inputs and outputs. This is a guardrail, not a guarantee — see ",{"type":40,"tag":79,"props":235,"children":237},{"className":236},[],[238],{"type":46,"value":239},"docs\u002Fresponsible-use.md",{"type":46,"value":177},{"type":40,"tag":41,"props":242,"children":244},{"id":243},"prerequisites",[245],{"type":46,"value":246},"Prerequisites",{"type":40,"tag":49,"props":248,"children":249},{},[250,292,297,302],{"type":40,"tag":53,"props":251,"children":252},{},[253,258,260,265,267,274,276,283,285,290],{"type":40,"tag":71,"props":254,"children":255},{},[256],{"type":46,"value":257},"Windows users:",{"type":46,"value":259}," Dayhoff prototype is developed and tested on Linux \u002F WSL2. On Windows 11, run everything inside a WSL2 distro (Ubuntu recommended) with NVIDIA drivers installed on the ",{"type":40,"tag":71,"props":261,"children":262},{},[263],{"type":46,"value":264},"Windows host",{"type":46,"value":266}," (not inside WSL). Native Windows Python \u002F PowerShell is not supported. See ",{"type":40,"tag":168,"props":268,"children":271},{"href":269,"rel":270},"https:\u002F\u002Flearn.microsoft.com\u002Fwindows\u002Fwsl\u002Finstall",[172],[272],{"type":46,"value":273},"Microsoft's WSL install guide",{"type":46,"value":275}," and ",{"type":40,"tag":168,"props":277,"children":280},{"href":278,"rel":279},"https:\u002F\u002Fdocs.nvidia.com\u002Fcuda\u002Fwsl-user-guide\u002Findex.html",[172],[281],{"type":46,"value":282},"NVIDIA's CUDA-on-WSL guide",{"type":46,"value":284},". (Path A — cached demo — is the one exception: if a Windows user just wants to see Dayhoff outputs in the UI, ",{"type":40,"tag":79,"props":286,"children":288},{"className":287},[],[289],{"type":46,"value":84},{"type":46,"value":291}," from the frontend on native Windows works fine because it never invokes the Python backend or the GPU.)",{"type":40,"tag":53,"props":293,"children":294},{},[295],{"type":46,"value":296},"Linux (or WSL2) with a CUDA-capable NVIDIA GPU for Path C. Path B needs only network access to the AML endpoint. Path A needs only Node.js. CPU works only for the 170M variant on toy sequences and is unusably slow for the 3B variants.",{"type":40,"tag":53,"props":298,"children":299},{},[300],{"type":46,"value":301},"Python 3.10+ for the backend (Path B and Path C); Node.js 18+ for the frontend (all paths).",{"type":40,"tag":53,"props":303,"children":304},{},[305,307],{"type":46,"value":306},"Disk space:\n",{"type":40,"tag":49,"props":308,"children":309},{},[310,322,327,332],{"type":40,"tag":53,"props":311,"children":312},{},[313,315,321],{"type":46,"value":314},"Path A (cached demo): ~200 MB for ",{"type":40,"tag":79,"props":316,"children":318},{"className":317},[],[319],{"type":46,"value":320},"node_modules",{"type":46,"value":177},{"type":40,"tag":53,"props":323,"children":324},{},[325],{"type":46,"value":326},"Path B (proxy-only, calling a remote AML endpoint): ~1 GB for Python deps.",{"type":40,"tag":53,"props":328,"children":329},{},[330],{"type":46,"value":331},"Path C with the 170M variant: ~2 GB additional for weights.",{"type":40,"tag":53,"props":333,"children":334},{},[335],{"type":46,"value":336},"Path C with all four variants loaded: ~20 GB GPU memory and ~25 GB disk for weights.",{"type":40,"tag":41,"props":338,"children":340},{"id":339},"workflow",[341],{"type":46,"value":342},"Workflow",{"type":40,"tag":64,"props":344,"children":345},{},[346,358,371,383,395,407,419,431],{"type":40,"tag":53,"props":347,"children":348},{},[349,351,356],{"type":46,"value":350},"Load ",{"type":40,"tag":79,"props":352,"children":354},{"className":353},[],[355],{"type":46,"value":218},{"type":46,"value":357}," when users ask what Dayhoff is, how it works, which variant to use, or for benchmark numbers.",{"type":40,"tag":53,"props":359,"children":360},{},[361,363,369],{"type":46,"value":362},"Follow ",{"type":40,"tag":79,"props":364,"children":366},{"className":365},[],[367],{"type":46,"value":368},"docs\u002Fquick-start.md",{"type":46,"value":370}," to get the reference app running locally — proxy first, then frontend, then optional score \u002F fold servers.",{"type":40,"tag":53,"props":372,"children":373},{},[374,376,382],{"type":46,"value":375},"For code examples (generation, fitness scoring, motif preservation, batch sampling, FASTA \u002F CSV \u002F JSON export), route to ",{"type":40,"tag":79,"props":377,"children":379},{"className":378},[],[380],{"type":46,"value":381},"docs\u002Fapplication-patterns.md",{"type":46,"value":177},{"type":40,"tag":53,"props":384,"children":385},{},[386,388,394],{"type":46,"value":387},"For loading the Dayhoff Atlas datasets (GigaRef, BackboneRef, DayhoffRef), route to ",{"type":40,"tag":79,"props":389,"children":391},{"className":390},[],[392],{"type":46,"value":393},"docs\u002Fdata-integration.md",{"type":46,"value":177},{"type":40,"tag":53,"props":396,"children":397},{},[398,400,406],{"type":46,"value":399},"For extending the prototype (new endpoints, custom scoring filters, batch jobs, database integration), route to ",{"type":40,"tag":79,"props":401,"children":403},{"className":402},[],[404],{"type":46,"value":405},"docs\u002Fprototype-expansion.md",{"type":46,"value":177},{"type":40,"tag":53,"props":408,"children":409},{},[410,412,418],{"type":46,"value":411},"For GPU sizing, throughput tuning, and scaling, route to ",{"type":40,"tag":79,"props":413,"children":415},{"className":414},[],[416],{"type":46,"value":417},"docs\u002Fperformance-guide.md",{"type":46,"value":177},{"type":40,"tag":53,"props":420,"children":421},{},[422,424,430],{"type":46,"value":423},"For errors, route to ",{"type":40,"tag":79,"props":425,"children":427},{"className":426},[],[428],{"type":46,"value":429},"docs\u002Ftroubleshooting.md",{"type":46,"value":177},{"type":40,"tag":53,"props":432,"children":433},{},[434,439,441,446],{"type":40,"tag":71,"props":435,"children":436},{},[437],{"type":46,"value":438},"Before any sequence is exported, shared, or sent to a wet lab",{"type":46,"value":440},", route to ",{"type":40,"tag":79,"props":442,"children":444},{"className":443},[],[445],{"type":46,"value":239},{"type":46,"value":177},{"type":40,"tag":448,"props":449,"children":451},"h3",{"id":450},"operating-rules",[452],{"type":46,"value":453},"Operating rules",{"type":40,"tag":49,"props":455,"children":456},{},[457,486,518,536,561,598,670,696,713,751,761],{"type":40,"tag":53,"props":458,"children":459},{},[460,464,466,471,472,477,479,484],{"type":40,"tag":71,"props":461,"children":462},{},[463],{"type":46,"value":257},{"type":46,"value":465}," Dayhoff is Linux\u002FWSL2-only for Path B and Path C. On Windows, all backend \u002F score-server commands must run inside a WSL2 distro with NVIDIA drivers installed on the Windows host. Native Windows Python and PowerShell are not supported. Point Windows users at ",{"type":40,"tag":168,"props":467,"children":469},{"href":269,"rel":468},[172],[470],{"type":46,"value":273},{"type":46,"value":275},{"type":40,"tag":168,"props":473,"children":475},{"href":278,"rel":474},[172],[476],{"type":46,"value":282},{"type":46,"value":478}," before running any Path B\u002FC step. (Path A — cached demo — is the one exception: pure-frontend ",{"type":40,"tag":79,"props":480,"children":482},{"className":481},[],[483],{"type":46,"value":84},{"type":46,"value":485}," works on native Windows because it never invokes Python or the GPU.)",{"type":40,"tag":53,"props":487,"children":488},{},[489,494,496,501,503,508,510,516],{"type":40,"tag":71,"props":490,"children":491},{},[492],{"type":46,"value":493},"No-GPU demo path exists:",{"type":46,"value":495}," If the user has no GPU, no AML endpoint, or just wants a quick look at what Dayhoff produces, route to ",{"type":40,"tag":71,"props":497,"children":498},{},[499],{"type":46,"value":500},"Path A",{"type":46,"value":502}," in ",{"type":40,"tag":79,"props":504,"children":506},{"className":505},[],[507],{"type":46,"value":368},{"type":46,"value":509}," — frontend only with cached real Dayhoff outputs for four bundled prompts. Don't gatekeep on hardware. The cache lives in ",{"type":40,"tag":79,"props":511,"children":513},{"className":512},[],[514],{"type":46,"value":515},"frontend\u002Fsrc\u002FdemoCache.ts",{"type":46,"value":517}," and is what the UI shows when the backend badge says \"Backend offline · cached examples only.\"",{"type":40,"tag":53,"props":519,"children":520},{},[521,526,528,534],{"type":40,"tag":71,"props":522,"children":523},{},[524],{"type":46,"value":525},"GPU probe first (Path C only):",{"type":46,"value":527}," Before suggesting Path C (local model loading), run ",{"type":40,"tag":79,"props":529,"children":531},{"className":530},[],[532],{"type":46,"value":533},"nvidia-smi",{"type":46,"value":535}," and confirm a CUDA-capable GPU is visible. If none is present, default to Path A (cached demo) or Path B (remote AML endpoint) — do not try to load the 3B models on CPU.",{"type":40,"tag":53,"props":537,"children":538},{},[539,544,546,552,553,559],{"type":40,"tag":71,"props":540,"children":541},{},[542],{"type":46,"value":543},"First-run weight download (Path C):",{"type":46,"value":545}," First call to the local score server downloads model weights from Hugging Face. The 170M variant is ~1 GB; loading all four variants is ~25 GB. Tell the user this will be slow and must not be interrupted. Subsequent runs reuse the cache. Recommend ",{"type":40,"tag":79,"props":547,"children":549},{"className":548},[],[550],{"type":46,"value":551},"huggingface-cli login",{"type":46,"value":275},{"type":40,"tag":79,"props":554,"children":556},{"className":555},[],[557],{"type":46,"value":558},"HUGGINGFACE_HUB_ENABLE_HF_TRANSFER=1",{"type":46,"value":560}," to avoid throttling.",{"type":40,"tag":53,"props":562,"children":563},{},[564,569,571,576,578,583,584,589,591,596],{"type":40,"tag":71,"props":565,"children":566},{},[567],{"type":46,"value":568},"Multi-terminal discipline:",{"type":46,"value":570}," Path A needs one terminal (frontend on ",{"type":40,"tag":79,"props":572,"children":574},{"className":573},[],[575],{"type":46,"value":110},{"type":46,"value":577},"). Path B needs two (proxy backend on ",{"type":40,"tag":79,"props":579,"children":581},{"className":580},[],[582],{"type":46,"value":102},{"type":46,"value":104},{"type":40,"tag":79,"props":585,"children":587},{"className":586},[],[588],{"type":46,"value":110},{"type":46,"value":590},"). Path C needs three (add local score server on ",{"type":40,"tag":79,"props":592,"children":594},{"className":593},[],[595],{"type":46,"value":136},{"type":46,"value":597},"), and up to four with a local fold server. Each must stay running in its own terminal — never run other commands in a terminal hosting a live server.",{"type":40,"tag":53,"props":599,"children":600},{},[601,606,608,614,616,622,624,630,632,638,640,646,648,653,655,661,662,668],{"type":40,"tag":71,"props":602,"children":603},{},[604],{"type":46,"value":605},"Local credentials handling:",{"type":46,"value":607}," Copy ",{"type":40,"tag":79,"props":609,"children":611},{"className":610},[],[612],{"type":46,"value":613},".env.example",{"type":46,"value":615}," to ",{"type":40,"tag":79,"props":617,"children":619},{"className":618},[],[620],{"type":46,"value":621},".env",{"type":46,"value":623}," and fill in ",{"type":40,"tag":79,"props":625,"children":627},{"className":626},[],[628],{"type":46,"value":629},"DAYHOFF_ENDPOINT",{"type":46,"value":631}," \u002F ",{"type":40,"tag":79,"props":633,"children":635},{"className":634},[],[636],{"type":46,"value":637},"DAYHOFF_API_KEY",{"type":46,"value":639}," for the AML proxy mode, or leave them pointing at ",{"type":40,"tag":79,"props":641,"children":643},{"className":642},[],[644],{"type":46,"value":645},"http:\u002F\u002Flocalhost:5001\u002Fscore",{"type":46,"value":647}," for fully-local mode. Never ask the user to paste credentials into chat — always reference the ",{"type":40,"tag":79,"props":649,"children":651},{"className":650},[],[652],{"type":46,"value":621},{"type":46,"value":654}," file. Keep the ",{"type":40,"tag":79,"props":656,"children":658},{"className":657},[],[659],{"type":46,"value":660},"DAYHOFF_*",{"type":46,"value":275},{"type":40,"tag":79,"props":663,"children":665},{"className":664},[],[666],{"type":46,"value":667},"ESMFOLD_*",{"type":46,"value":669}," variable names exactly as written; the proxy reads those exact names.",{"type":40,"tag":53,"props":671,"children":672},{},[673,678,680,686,688,694],{"type":40,"tag":71,"props":674,"children":675},{},[676],{"type":46,"value":677},"Honest scope:",{"type":46,"value":679}," The reference app is a ",{"type":40,"tag":681,"props":682,"children":683},"em",{},[684],{"type":46,"value":685},"research prototype",{"type":46,"value":687},". Generated sequences are model predictions of plausibility, not validated designs. The bundled ",{"type":40,"tag":79,"props":689,"children":691},{"className":690},[],[692],{"type":46,"value":693},"sequence_screening.py",{"type":46,"value":695}," toxin filter catches a small Select Agent reference set — it is a guardrail, not a guarantee. Do not claim safety, function, or foldability for any generated sequence.",{"type":40,"tag":53,"props":697,"children":698},{},[699,704,706,711],{"type":40,"tag":71,"props":700,"children":701},{},[702],{"type":46,"value":703},"Biosecurity gate:",{"type":46,"value":705}," Before assisting with any sequence export, sharing, or wet-lab handoff, surface ",{"type":40,"tag":79,"props":707,"children":709},{"className":708},[],[710],{"type":46,"value":239},{"type":46,"value":712}," and confirm the user has reviewed the pre-export checklist. Never assist with sequences targeting pathogens, toxins, or dual-use applications. Treat every generated sequence as a research draft requiring qualified human review.",{"type":40,"tag":53,"props":714,"children":715},{},[716,721,723,728,730,735,736,741,743,749],{"type":40,"tag":71,"props":717,"children":718},{},[719],{"type":46,"value":720},"Variant selection matters:",{"type":46,"value":722}," When a user asks for sequence generation, pick the right variant (see ",{"type":40,"tag":79,"props":724,"children":726},{"className":725},[],[727],{"type":46,"value":218},{"type":46,"value":729}," matrix). Reject homolog inputs for ",{"type":40,"tag":79,"props":731,"children":733},{"className":732},[],[734],{"type":46,"value":203},{"type":46,"value":275},{"type":40,"tag":79,"props":737,"children":739},{"className":738},[],[740],{"type":46,"value":210},{"type":46,"value":742}," early — the score server will return 400 if a model that doesn't support homolog conditioning receives a ",{"type":40,"tag":79,"props":744,"children":746},{"className":745},[],[747],{"type":46,"value":748},"homologs",{"type":46,"value":750}," payload.",{"type":40,"tag":53,"props":752,"children":753},{},[754,759],{"type":40,"tag":71,"props":755,"children":756},{},[757],{"type":46,"value":758},"Execute, don't display:",{"type":46,"value":760}," When terminal execution is available, run quick-start commands directly rather than printing bash blocks for the user to copy.",{"type":40,"tag":53,"props":762,"children":763},{},[764,769,771,776,778,790],{"type":40,"tag":71,"props":765,"children":766},{},[767],{"type":46,"value":768},"Always offer the next step:",{"type":46,"value":770}," After loading any explainer-style doc (e.g. ",{"type":40,"tag":79,"props":772,"children":774},{"className":773},[],[775],{"type":46,"value":218},{"type":46,"value":777},"), end your response with a concrete offer to advance the user. Default phrasing: ",{"type":40,"tag":681,"props":779,"children":780},{},[781,783,788],{"type":46,"value":782},"\"Want to try Dayhoff? Fastest path is just ",{"type":40,"tag":79,"props":784,"children":786},{"className":785},[],[787],{"type":46,"value":84},{"type":46,"value":789}," in the bundled frontend — no GPU, no backend, four cached prompts (Cas9, insulin, DNA polymerase, SARS-CoV-2 spike) return real Dayhoff outputs instantly. From there we can stand up proxy mode against an Azure ML endpoint or fully local mode on your GPU when you want custom prompts.\"",{"type":46,"value":791}," Adapt wording to context.",{"type":40,"tag":41,"props":793,"children":795},{"id":794},"routing",[796],{"type":46,"value":797},"Routing",{"type":40,"tag":799,"props":800,"children":801},"table",{},[802,821],{"type":40,"tag":803,"props":804,"children":805},"thead",{},[806],{"type":40,"tag":807,"props":808,"children":809},"tr",{},[810,816],{"type":40,"tag":811,"props":812,"children":813},"th",{},[814],{"type":46,"value":815},"Doc",{"type":40,"tag":811,"props":817,"children":818},{},[819],{"type":46,"value":820},"When to load",{"type":40,"tag":822,"props":823,"children":824},"tbody",{},[825,842,858,874,890,906,922,938],{"type":40,"tag":807,"props":826,"children":827},{},[828,837],{"type":40,"tag":829,"props":830,"children":831},"td",{},[832],{"type":40,"tag":79,"props":833,"children":835},{"className":834},[],[836],{"type":46,"value":218},{"type":40,"tag":829,"props":838,"children":839},{},[840],{"type":46,"value":841},"User asks what Dayhoff is, how it works, which variant to use, benchmarks, or scientific background",{"type":40,"tag":807,"props":843,"children":844},{},[845,853],{"type":40,"tag":829,"props":846,"children":847},{},[848],{"type":40,"tag":79,"props":849,"children":851},{"className":850},[],[852],{"type":46,"value":368},{"type":40,"tag":829,"props":854,"children":855},{},[856],{"type":46,"value":857},"User wants to run the reference app locally",{"type":40,"tag":807,"props":859,"children":860},{},[861,869],{"type":40,"tag":829,"props":862,"children":863},{},[864],{"type":40,"tag":79,"props":865,"children":867},{"className":866},[],[868],{"type":46,"value":381},{"type":40,"tag":829,"props":870,"children":871},{},[872],{"type":46,"value":873},"User wants code examples for generation, fitness scoring, motif preservation, batch sampling, or export",{"type":40,"tag":807,"props":875,"children":876},{},[877,885],{"type":40,"tag":829,"props":878,"children":879},{},[880],{"type":40,"tag":79,"props":881,"children":883},{"className":882},[],[884],{"type":46,"value":393},{"type":40,"tag":829,"props":886,"children":887},{},[888],{"type":46,"value":889},"User wants to load Dayhoff Atlas datasets (GigaRef, BackboneRef, DayhoffRef)",{"type":40,"tag":807,"props":891,"children":892},{},[893,901],{"type":40,"tag":829,"props":894,"children":895},{},[896],{"type":40,"tag":79,"props":897,"children":899},{"className":898},[],[900],{"type":46,"value":405},{"type":40,"tag":829,"props":902,"children":903},{},[904],{"type":46,"value":905},"User wants to extend the reference app — new endpoints, custom scoring, batch processing, database, validation",{"type":40,"tag":807,"props":907,"children":908},{},[909,917],{"type":40,"tag":829,"props":910,"children":911},{},[912],{"type":40,"tag":79,"props":913,"children":915},{"className":914},[],[916],{"type":46,"value":417},{"type":40,"tag":829,"props":918,"children":919},{},[920],{"type":46,"value":921},"User asks about GPU sizing, throughput, batching, or scaling",{"type":40,"tag":807,"props":923,"children":924},{},[925,933],{"type":40,"tag":829,"props":926,"children":927},{},[928],{"type":40,"tag":79,"props":929,"children":931},{"className":930},[],[932],{"type":46,"value":429},{"type":40,"tag":829,"props":934,"children":935},{},[936],{"type":46,"value":937},"User hits an error: missing GPU, Mamba kernel error, weight download stall, port conflict, AML 401, HF rate limit",{"type":40,"tag":807,"props":939,"children":940},{},[941,949],{"type":40,"tag":829,"props":942,"children":943},{},[944],{"type":40,"tag":79,"props":945,"children":947},{"className":946},[],[948],{"type":46,"value":239},{"type":40,"tag":829,"props":950,"children":951},{},[952,957],{"type":40,"tag":71,"props":953,"children":954},{},[955],{"type":46,"value":956},"Before",{"type":46,"value":958}," any sequence export, sharing, or wet-lab handoff; or when user asks about safety, biosecurity, or responsible use",{"type":40,"tag":41,"props":960,"children":962},{"id":961},"learning-path",[963],{"type":46,"value":964},"Learning Path",{"type":40,"tag":64,"props":966,"children":967},{},[968,978,988,998,1008],{"type":40,"tag":53,"props":969,"children":970},{},[971,976],{"type":40,"tag":79,"props":972,"children":974},{"className":973},[],[975],{"type":46,"value":218},{"type":46,"value":977}," — Understand what Dayhoff is, the four model variants, and which one fits your task",{"type":40,"tag":53,"props":979,"children":980},{},[981,986],{"type":40,"tag":79,"props":982,"children":984},{"className":983},[],[985],{"type":46,"value":368},{"type":46,"value":987}," — Run the reference app locally and generate your first sequences",{"type":40,"tag":53,"props":989,"children":990},{},[991,996],{"type":40,"tag":79,"props":992,"children":994},{"className":993},[],[995],{"type":46,"value":239},{"type":46,"value":997}," — Read the biosecurity guardrails and pre-export checklist before exporting anything",{"type":40,"tag":53,"props":999,"children":1000},{},[1001,1006],{"type":40,"tag":79,"props":1002,"children":1004},{"className":1003},[],[1005],{"type":46,"value":381},{"type":46,"value":1007}," — Build your own generation, scoring, and export workflows",{"type":40,"tag":53,"props":1009,"children":1010},{},[1011,1016,1017,1022],{"type":40,"tag":79,"props":1012,"children":1014},{"className":1013},[],[1015],{"type":46,"value":393},{"type":46,"value":275},{"type":40,"tag":79,"props":1018,"children":1020},{"className":1019},[],[1021],{"type":46,"value":405},{"type":46,"value":1023}," — Scale up with Atlas datasets and custom prototype features",{"type":40,"tag":41,"props":1025,"children":1027},{"id":1026},"reference-links",[1028],{"type":46,"value":1029},"Reference Links",{"type":40,"tag":49,"props":1031,"children":1032},{},[1033,1044,1055,1065],{"type":40,"tag":53,"props":1034,"children":1035},{},[1036,1038],{"type":46,"value":1037},"Research paper (preprint): ",{"type":40,"tag":168,"props":1039,"children":1042},{"href":1040,"rel":1041},"https:\u002F\u002Faka.ms\u002Fdayhoff\u002Fpreprint",[172],[1043],{"type":46,"value":1040},{"type":40,"tag":53,"props":1045,"children":1046},{},[1047,1049],{"type":46,"value":1048},"GitHub: ",{"type":40,"tag":168,"props":1050,"children":1053},{"href":1051,"rel":1052},"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fdayhoff",[172],[1054],{"type":46,"value":1051},{"type":40,"tag":53,"props":1056,"children":1057},{},[1058,1060],{"type":46,"value":1059},"Hugging Face model collection: ",{"type":40,"tag":168,"props":1061,"children":1063},{"href":170,"rel":1062},[172],[1064],{"type":46,"value":170},{"type":40,"tag":53,"props":1066,"children":1067},{},[1068,1070],{"type":46,"value":1069},"Azure AI Foundry model card: ",{"type":40,"tag":168,"props":1071,"children":1074},{"href":1072,"rel":1073},"https:\u002F\u002Fai.azure.com\u002Fcatalog\u002Fmodels\u002FDayhoff-170m-GR",[172],[1075],{"type":46,"value":1072},{"type":40,"tag":41,"props":1077,"children":1079},{"id":1078},"assets",[1080],{"type":46,"value":1081},"Assets",{"type":40,"tag":49,"props":1083,"children":1084},{},[1085,1162],{"type":40,"tag":53,"props":1086,"children":1087},{},[1088,1094,1096,1101,1103,1108,1110,1116,1117,1122,1124,1130,1132,1138,1139,1144,1146,1152,1154,1160],{"type":40,"tag":79,"props":1089,"children":1091},{"className":1090},[],[1092],{"type":46,"value":1093},"assets\u002Fdayhoff-prototype\u002F",{"type":46,"value":1095}," — Self-contained reference app: Flask proxy backend (",{"type":40,"tag":79,"props":1097,"children":1099},{"className":1098},[],[1100],{"type":46,"value":149},{"type":46,"value":1102},", port ",{"type":40,"tag":79,"props":1104,"children":1106},{"className":1105},[],[1107],{"type":46,"value":102},{"type":46,"value":1109},"), local score server (",{"type":40,"tag":79,"props":1111,"children":1113},{"className":1112},[],[1114],{"type":46,"value":1115},"backend\u002Fscore\u002F",{"type":46,"value":1102},{"type":40,"tag":79,"props":1118,"children":1120},{"className":1119},[],[1121],{"type":46,"value":136},{"type":46,"value":1123},"), optional ESMFold structure server (",{"type":40,"tag":79,"props":1125,"children":1127},{"className":1126},[],[1128],{"type":46,"value":1129},"backend\u002Ffold\u002F",{"type":46,"value":1131},"), Vite + React + TypeScript frontend (",{"type":40,"tag":79,"props":1133,"children":1135},{"className":1134},[],[1136],{"type":46,"value":1137},"frontend\u002F",{"type":46,"value":1102},{"type":40,"tag":79,"props":1140,"children":1142},{"className":1141},[],[1143],{"type":46,"value":110},{"type":46,"value":1145},"), Azure ML deployment YAMLs (",{"type":40,"tag":79,"props":1147,"children":1149},{"className":1148},[],[1150],{"type":46,"value":1151},"infra\u002F",{"type":46,"value":1153},"), and ",{"type":40,"tag":79,"props":1155,"children":1157},{"className":1156},[],[1158],{"type":46,"value":1159},"Dockerfile",{"type":46,"value":1161}," for App Service deployment.",{"type":40,"tag":53,"props":1163,"children":1164},{},[1165,1171],{"type":40,"tag":79,"props":1166,"children":1168},{"className":1167},[],[1169],{"type":46,"value":1170},"assets\u002Fpaper\u002Fdayhoff-biorxiv-paper.pdf",{"type":46,"value":1172}," — Bundled offline copy of the bioRxiv preprint.",{"items":1174,"total":1369},[1175,1197,1218,1239,1254,1271,1282,1295,1310,1325,1344,1357],{"slug":1176,"name":1176,"fn":1177,"description":1178,"org":1179,"tags":1180,"stars":1194,"repoUrl":1195,"updatedAt":1196},"rushstack-best-practices","manage Rush monorepos with best practices","Provides best practices and guidance for working with Rush monorepos. Use when the user is working in a Rush-based repository, asks about Rush commands (install, update, build, rebuild), needs help with project selection, dependency management, build caching, subspace configuration, or troubleshooting Rush-specific issues.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1181,1184,1187,1188,1191],{"name":1182,"slug":1183,"type":15},"Engineering","engineering",{"name":1185,"slug":1186,"type":15},"Local Development","local-development",{"name":9,"slug":8,"type":15},{"name":1189,"slug":1190,"type":15},"Project Management","project-management",{"name":1192,"slug":1193,"type":15},"Rush","rush",6484,"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Frushstack","2026-04-06T18:34:44.965032",{"slug":1198,"name":1198,"fn":1199,"description":1200,"org":1201,"tags":1202,"stars":1215,"repoUrl":1216,"updatedAt":1217},"azure-ai-agents-persistent-dotnet","build AI agents with Azure .NET SDK","Azure AI Agents Persistent SDK for .NET. Low-level SDK for creating and managing AI agents with threads, messages, runs, and tools. Use for agent CRUD, conversation threads, streaming responses, function calling, file search, and code interpreter. Triggers: \"PersistentAgentsClient\", \"persistent agents\", \"agent threads\", \"agent runs\", \"streaming agents\", \"function calling agents .NET\".\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1203,1206,1209,1212],{"name":1204,"slug":1205,"type":15},".NET","net",{"name":1207,"slug":1208,"type":15},"Agents","agents",{"name":1210,"slug":1211,"type":15},"Azure","azure",{"name":1213,"slug":1214,"type":15},"LLM","llm",2804,"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fskills","2026-07-03T16:32:10.297433",{"slug":1219,"name":1219,"fn":1220,"description":1221,"org":1222,"tags":1223,"stars":1215,"repoUrl":1216,"updatedAt":1238},"azure-ai-anomalydetector-java","build anomaly detection applications with Java","Build anomaly detection applications with Azure AI Anomaly Detector SDK for Java. Use when implementing univariate\u002Fmultivariate anomaly detection, time-series analysis, or AI-powered monitoring.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1224,1227,1228,1231,1234,1235],{"name":1225,"slug":1226,"type":15},"Analytics","analytics",{"name":1210,"slug":1211,"type":15},{"name":1229,"slug":1230,"type":15},"Data Analysis","data-analysis",{"name":1232,"slug":1233,"type":15},"Java","java",{"name":9,"slug":8,"type":15},{"name":1236,"slug":1237,"type":15},"Monitoring","monitoring","2026-05-13T06:14:16.261754",{"slug":1240,"name":1240,"fn":1241,"description":1242,"org":1243,"tags":1244,"stars":1215,"repoUrl":1216,"updatedAt":1253},"azure-ai-contentsafety-java","build content moderation applications with Azure AI","Build content moderation applications with Azure AI Content Safety SDK for Java. Use when implementing text\u002Fimage analysis, blocklist management, or harm detection for hate, violence, sexual content, and self-harm.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1245,1248,1249,1250],{"name":1246,"slug":1247,"type":15},"AI Infrastructure","ai-infrastructure",{"name":1210,"slug":1211,"type":15},{"name":1232,"slug":1233,"type":15},{"name":1251,"slug":1252,"type":15},"Security","security","2026-07-07T06:53:31.293235",{"slug":1255,"name":1255,"fn":1256,"description":1257,"org":1258,"tags":1259,"stars":1215,"repoUrl":1216,"updatedAt":1270},"azure-ai-contentsafety-py","detect harmful content with Azure AI Content Safety","Azure AI Content Safety SDK for Python. Use for detecting harmful content in text and images with multi-severity classification.\nTriggers: \"azure-ai-contentsafety\", \"ContentSafetyClient\", \"content moderation\", \"harmful content\", \"text analysis\", \"image analysis\".\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1260,1261,1264,1265,1266,1269],{"name":1210,"slug":1211,"type":15},{"name":1262,"slug":1263,"type":15},"Compliance","compliance",{"name":1213,"slug":1214,"type":15},{"name":9,"slug":8,"type":15},{"name":1267,"slug":1268,"type":15},"Python","python",{"name":1251,"slug":1252,"type":15},"2026-07-18T05:14:23.017504",{"slug":1272,"name":1272,"fn":1273,"description":1274,"org":1275,"tags":1276,"stars":1215,"repoUrl":1216,"updatedAt":1281},"azure-ai-language-conversations-py","implement conversational language understanding with Python","Implement Conversational Language Understanding (CLU) using the azure-ai-language-conversations Python SDK. Use when working with ConversationAnalysisClient to analyze conversation intent and entities, building NLP features, or integrating language understanding into applications.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1277,1278,1279,1280],{"name":1225,"slug":1226,"type":15},{"name":1210,"slug":1211,"type":15},{"name":1213,"slug":1214,"type":15},{"name":1267,"slug":1268,"type":15},"2026-07-31T05:54:29.068751",{"slug":1283,"name":1283,"fn":1284,"description":1285,"org":1286,"tags":1287,"stars":1215,"repoUrl":1216,"updatedAt":1294},"azure-ai-translation-text-py","translate text using Azure AI services","Azure AI Text Translation SDK for real-time text translation, transliteration, language detection, and dictionary lookup. Use for translating text content in applications.\nTriggers: \"text translation\", \"translator\", \"translate text\", \"transliterate\", \"TextTranslationClient\".\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1288,1291,1292,1293],{"name":1289,"slug":1290,"type":15},"API Development","api-development",{"name":1210,"slug":1211,"type":15},{"name":9,"slug":8,"type":15},{"name":1267,"slug":1268,"type":15},"2026-07-18T05:14:16.988376",{"slug":1296,"name":1296,"fn":1297,"description":1298,"org":1299,"tags":1300,"stars":1215,"repoUrl":1216,"updatedAt":1309},"azure-ai-vision-imageanalysis-py","analyze images with Azure AI Vision","Azure AI Vision Image Analysis SDK for captions, tags, objects, OCR, people detection, and smart cropping. Use for computer vision and image understanding tasks.\nTriggers: \"image analysis\", \"computer vision\", \"OCR\", \"object detection\", \"ImageAnalysisClient\", \"image caption\".\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1301,1302,1305,1308],{"name":1210,"slug":1211,"type":15},{"name":1303,"slug":1304,"type":15},"Computer Vision","computer-vision",{"name":1306,"slug":1307,"type":15},"Images","images",{"name":1267,"slug":1268,"type":15},"2026-07-18T05:14:18.007737",{"slug":1311,"name":1311,"fn":1312,"description":1313,"org":1314,"tags":1315,"stars":1215,"repoUrl":1216,"updatedAt":1324},"azure-appconfiguration-java","manage configuration with Azure App Configuration","Azure App Configuration SDK for Java. Centralized application configuration management with key-value settings, feature flags, and snapshots.\nTriggers: \"ConfigurationClient java\", \"app configuration java\", \"feature flag java\", \"configuration setting java\", \"azure config java\".\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1316,1317,1320,1323],{"name":1210,"slug":1211,"type":15},{"name":1318,"slug":1319,"type":15},"Configuration","configuration",{"name":1321,"slug":1322,"type":15},"Feature Flags","feature-flags",{"name":1232,"slug":1233,"type":15},"2026-07-03T16:32:01.278468",{"slug":1326,"name":1326,"fn":1327,"description":1328,"org":1329,"tags":1330,"stars":1215,"repoUrl":1216,"updatedAt":1343},"azure-cosmos-rust","build applications with Azure Cosmos DB","Azure Cosmos DB library for Rust (NoSQL API). Document CRUD, containers, and globally distributed data.\nTriggers: \"cosmos db rust\", \"CosmosClient rust\", \"document crud rust\", \"NoSQL rust\", \"partition key rust\".\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1331,1334,1337,1340],{"name":1332,"slug":1333,"type":15},"Cosmos DB","cosmos-db",{"name":1335,"slug":1336,"type":15},"Database","database",{"name":1338,"slug":1339,"type":15},"NoSQL","nosql",{"name":1341,"slug":1342,"type":15},"Rust","rust","2026-07-31T05:54:27.021432",{"slug":1345,"name":1345,"fn":1327,"description":1346,"org":1347,"tags":1348,"stars":1215,"repoUrl":1216,"updatedAt":1356},"azure-cosmos-ts","Azure Cosmos DB JavaScript\u002FTypeScript SDK (@azure\u002Fcosmos) for data plane operations. Use for CRUD operations on documents, queries, bulk operations, and container management. Triggers: \"Cosmos DB\", \"@azure\u002Fcosmos\", \"CosmosClient\", \"document CRUD\", \"NoSQL queries\", \"bulk operations\", \"partition key\", \"container.items\".\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1349,1350,1351,1352,1353],{"name":1332,"slug":1333,"type":15},{"name":1335,"slug":1336,"type":15},{"name":9,"slug":8,"type":15},{"name":1338,"slug":1339,"type":15},{"name":1354,"slug":1355,"type":15},"TypeScript","typescript","2026-07-03T16:31:19.368382",{"slug":1358,"name":1358,"fn":1359,"description":1360,"org":1361,"tags":1362,"stars":1215,"repoUrl":1216,"updatedAt":1368},"azure-data-tables-java","build table storage applications with Java","Build table storage applications with Azure Tables SDK for Java. Use when working with Azure Table Storage or Cosmos DB Table API for NoSQL key-value data, schemaless storage, or structured data at scale.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1363,1364,1365,1366,1367],{"name":1210,"slug":1211,"type":15},{"name":1332,"slug":1333,"type":15},{"name":1335,"slug":1336,"type":15},{"name":1232,"slug":1233,"type":15},{"name":1338,"slug":1339,"type":15},"2026-05-13T06:14:17.582229",267,{"items":1371,"total":1407},[1372,1387,1400],{"slug":1373,"name":1373,"fn":1374,"description":1375,"org":1376,"tags":1377,"stars":23,"repoUrl":24,"updatedAt":1386},"msresearch-aurora","forecast weather with Aurora foundation model","Aurora — Microsoft's foundation model for the Earth system. Use this skill for both inference (regional weather\u002Fclimate forecasting, Norway example, ERA5\u002FCDS data, wind\u002Ftemperature\u002Fpressure mapping) and fine-tuning Aurora on new variables, regions, or domains (training, evaluation, checkpoint extension, gradient\u002Floss debugging). Triggers on Aurora, Microsoft Aurora, foundation models for Earth systems, ERA5\u002FCDS\u002FNetCDF weather data, weather forecast validation, and adapting Aurora to new use cases.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1378,1379,1380,1383],{"name":1213,"slug":1214,"type":15},{"name":9,"slug":8,"type":15},{"name":1381,"slug":1382,"type":15},"Physics","physics",{"name":1384,"slug":1385,"type":15},"Research","research","2026-07-03T16:32:28.677219",{"slug":1388,"name":1388,"fn":1389,"description":1390,"org":1391,"tags":1392,"stars":23,"repoUrl":24,"updatedAt":1399},"msresearch-bioemu","generate protein conformational ensembles with BioEmu","Microsoft Research's BioEmu — generates protein conformational ensembles from amino acid sequence on a local GPU. Use when users mention BioEmu by name, want to predict protein dynamics or flexibility from sequence, or are setting up \u002F running \u002F troubleshooting BioEmu locally.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1393,1394,1395,1398],{"name":20,"slug":21,"type":15},{"name":17,"slug":18,"type":15},{"name":1396,"slug":1397,"type":15},"Life Sciences","life-sciences",{"name":9,"slug":8,"type":15},"2026-07-07T06:53:46.003518",{"slug":4,"name":4,"fn":5,"description":6,"org":1401,"tags":1402,"stars":23,"repoUrl":24,"updatedAt":25},{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1403,1404,1405,1406],{"name":20,"slug":21,"type":15},{"name":17,"slug":18,"type":15},{"name":9,"slug":8,"type":15},{"name":13,"slug":14,"type":15},3]