[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-doca-bare-metal-deployment":3,"mdc-uix3bb-key":37,"related-repo-nvidia-doca-bare-metal-deployment":994,"related-org-nvidia-doca-bare-metal-deployment":1098},{"slug":4,"name":4,"fn":5,"description":6,"org":7,"tags":11,"stars":26,"repoUrl":27,"updatedAt":28,"license":29,"forks":30,"topics":31,"repo":32,"sourceUrl":35,"mdContent":36},"doca-bare-metal-deployment","deploy and debug BlueField DPU platforms","Use this skill for launching, supervising, debugging, OR platform lifecycle on a BlueField — BFB install, RShim\u002FTMFIFO, host PF rebind, post-BFB recovery — taking a DOCA-linked binary to a healthy run directly on hardware (host x86 + BlueField NIC over PCIe, or BlueField Arm bare-metal). No container, no kubelet. Covers launch mode (direct, tmux, systemd), PCI\u002FNUMA\u002F CPU\u002FIRQ binding, co-tenant isolation (cgroup-v2\u002Fnetns\u002Fnumactl), a seven-layer error taxonomy, and a six-state BlueField lifecycle classifier. Trigger even when user does not say \"bare-metal\" — implicit phrasings include \"binary exits 1 right after launch\", \"systemd keeps restarting it\", \"no matching device on the BF\", \"bfb-install exited 0 but DPU is dead\", \"ping 192.168.100.2 works but ssh fails\", \"host PFs aren't showing netdevs\". Destructive firmware burn \u002F mlxconfig set requires explicit confirmation via doca-hardware-safety; containers, library APIs, env prep, and build use other skills.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},"nvidia","NVIDIA","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fnvidia.png",[12,16,19,20,23],{"name":13,"slug":14,"type":15},"Hardware","hardware","tag",{"name":17,"slug":18,"type":15},"Deployment","deployment",{"name":9,"slug":8,"type":15},{"name":21,"slug":22,"type":15},"Engineering","engineering",{"name":24,"slug":25,"type":15},"Debugging","debugging",2473,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills","2026-07-30T05:28:30.448801","Apache-2.0",281,[],{"repoUrl":27,"stars":26,"forks":30,"topics":33,"description":34},[],"AI agent skills published by NVIDIA","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills\u002Ftree\u002FHEAD\u002Fskills\u002Fdoca-bare-metal-deployment","---\nlicense: Apache-2.0\nname: doca-bare-metal-deployment\ndescription: >\n  Use this skill for launching, supervising, debugging, OR\n  platform lifecycle on a BlueField — BFB install, RShim\u002FTMFIFO,\n  host PF rebind, post-BFB recovery — taking a DOCA-linked binary\n  to a healthy run directly on hardware (host x86 + BlueField NIC\n  over PCIe, or BlueField Arm bare-metal). No container, no\n  kubelet. Covers launch mode (direct, tmux, systemd), PCI\u002FNUMA\u002F\n  CPU\u002FIRQ binding, co-tenant isolation (cgroup-v2\u002Fnetns\u002Fnumactl),\n  a seven-layer error taxonomy, and a six-state BlueField\n  lifecycle classifier. Trigger even when user does not say\n  \"bare-metal\" — implicit phrasings include \"binary exits 1 right\n  after launch\", \"systemd keeps restarting it\", \"no matching\n  device on the BF\", \"bfb-install exited 0 but DPU is dead\",\n  \"ping 192.168.100.2 works but ssh fails\", \"host PFs aren't\n  showing netdevs\". Destructive firmware burn \u002F mlxconfig set\n  requires explicit confirmation via doca-hardware-safety;\n  containers, library APIs, env prep, and build use other skills.\nmetadata:\n  kind: library\ncompatibility: >\n  No DOCA install required to read this skill (it is an overlay\n  loaded against any DOCA artifact skill); the validation steps\n  within this skill require a live DOCA install at \u002Fopt\u002Fmellanox\u002Fdoca on\n  a host or BlueField with a built DOCA-linked binary.\n---\n\n# DOCA bare-metal deployment\n\n**Where to start:** This skill is the bundle's home for *operating*\na DOCA-linked application binary **directly on hardware** — no\ncontainer, no kubelet, no static-pod manifest. It is the parallel\nof [`doca-container-deployment`](..\u002Fdoca-container-deployment\u002FSKILL.md)\nfor the non-container path. If the user has a DOCA-linked binary\nthey built (per the canonical workflow in\n[`doca-programming-guide`](..\u002Fdoca-programming-guide\u002FSKILL.md))\nand they want to know *how to actually run it on the host or on\nthe BlueField Arm cores correctly*, open\n[`TASKS.md`](TASKS.md) and start at\n[`## configure`](TASKS.md#configure). If the question is *what\nshape does the bare-metal runtime even have and what is the\ndeployment contract*, start at [`CAPABILITIES.md`](CAPABILITIES.md).\nIf the user is not yet sure whether their target system shape is\nthe container path or the bare-metal path, route the recognition\nstep to [`doca-setup`](..\u002Fdoca-setup\u002FSKILL.md) first; only return\nhere once *bare-metal* is the confirmed shape.\n\n## Audience\n\nThis skill serves **external DOCA developers and operators who\nhave a DOCA-linked application binary they built and want to run\nit directly on hardware** — i.e., people who already have:\n\n- a DOCA-linked application binary they built per\n  [`doca-programming-guide ## build`](..\u002Fdoca-programming-guide\u002FTASKS.md#build),\n- a real BlueField NIC and a host that talks to it (the **host\n  x86** path — DOCA host install on the host talks to the\n  BlueField NIC over PCIe), OR a BlueField with a console or SSH\n  to the Arm side (the **BlueField Arm bare-metal** path — DOCA\n  installed on the DPU Arm cores; the binary runs there\n  directly), and\n- a desire to RUN that binary directly on the hardware, not\n  inside a kubelet-standalone-managed container.\n\nIt is **not** for:\n\n- kernel-driver developers contributing to `mlx5_*` or the\n  BlueField OS,\n- DOCA library contributors (those changes go to the internal\n  DOCA tree, not to a bare-metal deployment),\n- full-Kubernetes-cluster operators managing a fleet of\n  BlueFields (the bundle covers\n  [`doca-container-deployment`](..\u002Fdoca-container-deployment\u002FSKILL.md)\n  for the single-host kubelet-standalone shape; **fleet\u002Fproduction-scale\n  deployment is fleet-orchestration scope** — route to the orchestration\n  entry-point in\n  [`doca-public-knowledge-map ## Deploying DOCA services at scale`](..\u002Fdoca-public-knowledge-map\u002Freferences\u002Fmap.md#deploying-doca-services-at-scale--orchestration-entry-point-personascale-routing)\n  (DPF \u002F Network Operator \u002F Launch Kit), not hand-rolled static-pod loops),\n- fresh-laptop-no-hardware users with no DOCA install yet — those\n  belong on\n  [`doca-setup ## no-install`](..\u002Fdoca-setup\u002FTASKS.md#no-install).\n\nThe skill teaches the agent the bare-metal-deployment *procedure*\nand the rules for quoting documented commands from the public DOCA\nProgramming Guide and the public BlueField \u002F DPU User Manual via\n[`doca-public-knowledge-map`](..\u002Fdoca-public-knowledge-map\u002FSKILL.md);\nit does not invent flag names, PCI BDFs, NUMA numbers, devlink\npaths, representor strings, or systemd `Restart=` mode names from\nmemory.\n\n## When to load this skill\n\nLoad this skill when the user is doing **hands-on bare-metal\ndeployment of a DOCA-linked application binary** on either of the\ntwo supported host modes (host x86 or BlueField Arm), or asking a\ncross-cutting bare-metal question that is not specific to one\nlibrary's API. Concretely:\n\n- Launching a DOCA-linked binary for the first time on a host\n  with a BlueField NIC in a PCIe slot, with DOCA installed on the\n  host.\n- Launching a DOCA-linked binary on the BlueField Arm cores\n  directly (BlueField Arm bare-metal mode), with DOCA installed\n  on the Arm side per the BlueField OS image.\n- Deciding which launch mode to use (direct foreground for\n  interactive debug; tmux\u002Fscreen for long-running with manual\n  reattach; systemd-supervised for restart-after-reboot,\n  journald-integrated logs, and Restart= policy).\n- Binding the DOCA process to the right PCIe function, the right\n  representor, the right NUMA node, and the right CPU set — and\n  pinning IRQs to match — without inventing the addresses or the\n  flag names.\n- Setting up per-tenant isolation (cgroup-v2 cpu \u002F memory \u002F io\n  controllers, network namespaces for multi-tenant deployments,\n  `numactl` \u002F `taskset` for CPU + NUMA binding) so multiple DOCA\n  processes co-tenant on the same BlueField without crushing each\n  other.\n- Diagnosing a bare-metal launch that is misbehaving — won't\n  start, starts and exits immediately, runs but can't find the\n  device, attaches to the device but the workload errors, OOMs or\n  is signal-killed, is in a restart loop under a supervisor, or\n  is being interfered with by a co-tenant.\n- Cross-cutting questions: *\"should I run this in tmux or as a\n  systemd unit\"*, *\"what is the smoke-before-bulk loop for a\n  binary on bare metal\"*, *\"my binary works in a container on the\n  BlueField but not when I run it directly on the Arm — what\n  changed\"*.\n\nDo **not** load this skill for the container-path equivalent\n(those questions go to\n[`doca-container-deployment`](..\u002Fdoca-container-deployment\u002FSKILL.md));\nfor full-Kubernetes-cluster operations (out of scope per the\nbundle's non-goals); for library-API questions (route to the\nmatching `libs\u002F\u003Clibrary>` skill); for env-preparation questions\nincluding hugepages, IOMMU, pkg-config, and devlink mode flips\n(use [`doca-setup`](..\u002Fdoca-setup\u002FSKILL.md)); for any\nhardware-state-changing operation including `mlxconfig` writes\nand BFB reflashes (route to\n[`doca-hardware-safety`](..\u002Fdoca-hardware-safety\u002FSKILL.md) for the\ncross-cutting meta-policy); or for cross-library programming\nquestions (use\n[`doca-programming-guide`](..\u002Fdoca-programming-guide\u002FSKILL.md)).\n\n## What this skill provides\n\nThis is a **thin loader**. Substantive material lives in two\ncompanion files:\n\n- `CAPABILITIES.md` — the bare-metal deployment runtime contract\n  for a DOCA-linked binary: the two host modes (host x86 vs\n  BlueField Arm bare-metal), the three launch modes (direct,\n  tmux\u002Fscreen, systemd-supervised), the hardware-resource-binding\n  surface (PF \u002F VF \u002F representor enumeration; NUMA topology\n  discovery; CPU pinning rationale; IRQ affinity rules), the\n  per-tenant isolation surface (cgroup-v2 cpu \u002F memory \u002F io,\n  network namespaces, `numactl` \u002F `taskset`), the restart and\n  recovery semantics (documented `systemd` `Restart=` modes vs\n  crash-and-investigate vs supervisor-driven restart), the\n  bare-metal-specific version overlay on the four-way version\n  match owned by\n  [`doca-version`](..\u002Fdoca-version\u002FSKILL.md), the cross-cutting\n  error taxonomy (seven layers, walked in order), the observability\n  surface (stdout\u002Fstderr discipline by launch mode; device-state\n  introspection via `devlink` \u002F `sysfs` \u002F `mlxconfig` *query*;\n  per-tenant resource visibility), and the safety policy (overlay\n  on\n  [`doca-hardware-safety`](..\u002Fdoca-hardware-safety\u002FSKILL.md):\n  smoke-before-bulk for binaries; failed bare-metal process is\n  HIGH-STAKES; do not invent PCI addresses, NUMA numbers,\n  representor names, devlink paths, or systemd `Restart=` mode\n  names; confirm tenant-isolation primitives BEFORE the workload\n  starts).\n- `TASKS.md` — step-by-step workflows for the in-scope bare-metal\n  verbs: `configure`, `build`, `modify`, `run` (with an explicit\n  `### isolation` sub-anchor covering cgroup-v2 \u002F namespaces \u002F\n  numactl per-tenant primitives), `test`, `debug`,\n  `bluefield-lifecycle` (the BFB-install → RShim\u002FTMFIFO →\n  post-BFB-recovery operational sequencing ladder, with the\n  six-state `bluefield-state-classifier` sub-anchor), the\n  `Command appendix` (documented commands the agent may quote,\n  each cross-linked to its public-doc source — no invented\n  commands), and the `Deferred task verbs` block routing\n  container-path \u002F cluster \u002F library-API \u002F env-prep \u002F\n  hardware-state-change \u002F cross-library questions out to their\n  owning skills. (The change-application discipline for any\n  mutating burn invoked from `## bluefield-lifecycle` is still\n  meta-policy owned by\n  [`doca-hardware-safety`](..\u002Fdoca-hardware-safety\u002FSKILL.md),\n  loaded alongside.)\n\nThe skill assumes a host or BlueField target where:\n\n- DOCA is already installed and healthy (per\n  [`doca-setup ## test`](..\u002Fdoca-setup\u002FTASKS.md#test)),\n- the user has a DOCA-linked application binary they built (per\n  [`doca-programming-guide ## build`](..\u002Fdoca-programming-guide\u002FTASKS.md#build)),\n- the user has the host-OS permissions to enumerate devices,\n  reserve hugepages, write systemd units (if they choose that\n  launch mode), and bind processes to NUMA nodes.\n\nIt does not cover installing DOCA — that path goes through\n[`doca-setup`](..\u002Fdoca-setup\u002FSKILL.md) — and it does not cover\nbuilding the binary — that path goes through\n[`doca-programming-guide`](..\u002Fdoca-programming-guide\u002FSKILL.md).\n\n## Loading order\n\n1. Read this `SKILL.md` first to confirm the user's question is\n   in scope (bare-metal launch of a DOCA-linked binary on host\n   x86 or BlueField Arm; NOT the container path, NOT a full\n   cluster, NOT a library-API question).\n2. **For the runtime contract (two host modes, three launch\n   modes, hardware-binding surface, per-tenant isolation, version\n   overlay, seven-layer error taxonomy, observability surface,\n   bare-metal safety overlay), see\n   [CAPABILITIES.md](CAPABILITIES.md).**\n3. **For step-by-step workflows — `configure`, `build` (routing\n   stub), `modify` (routing stub), `run` (with `### isolation`\n   sub-anchor), `test`, `debug`, `bluefield-lifecycle` (BFB\n   install + RShim\u002FTMFIFO + post-BFB recovery + the six-state\n   `bluefield-state-classifier`), plus the `Command appendix` and\n   the `Deferred task verbs` block — see [TASKS.md](TASKS.md).**\n\n## Example questions this skill answers well\n\nSee [`references\u002Fdetails.md`](references\u002Fdetails.md#example-questions-this-skill-answers-well).\n## What this skill deliberately does not ship\n\nSee [`references\u002Fdetails.md`](references\u002Fdetails.md#what-this-skill-deliberately-does-not-ship).\n## Related skills\n\nSee [`references\u002Fdetails.md`](references\u002Fdetails.md#related-skills).\n",{"data":38,"body":42},{"license":29,"name":4,"description":6,"metadata":39,"compatibility":41},{"kind":40},"library","No DOCA install required to read this skill (it is an overlay loaded against any DOCA artifact skill); the validation steps within this skill require a live DOCA install at \u002Fopt\u002Fmellanox\u002Fdoca on a host or BlueField with a built DOCA-linked binary.\n",{"type":43,"children":44},"root",[45,53,173,180,192,238,250,322,354,360,372,445,514,520,532,741,746,785,809,815,932,938,954,960,974,980],{"type":46,"tag":47,"props":48,"children":49},"element","h1",{"id":4},[50],{"type":51,"value":52},"text","DOCA bare-metal deployment",{"type":46,"tag":54,"props":55,"children":56},"p",{},[57,63,65,71,73,78,80,92,94,104,106,111,113,122,124,134,136,141,143,152,154,164,166,171],{"type":46,"tag":58,"props":59,"children":60},"strong",{},[61],{"type":51,"value":62},"Where to start:",{"type":51,"value":64}," This skill is the bundle's home for ",{"type":46,"tag":66,"props":67,"children":68},"em",{},[69],{"type":51,"value":70},"operating",{"type":51,"value":72},"\na DOCA-linked application binary ",{"type":46,"tag":58,"props":74,"children":75},{},[76],{"type":51,"value":77},"directly on hardware",{"type":51,"value":79}," — no\ncontainer, no kubelet, no static-pod manifest. It is the parallel\nof ",{"type":46,"tag":81,"props":82,"children":84},"a",{"href":83},"..\u002Fdoca-container-deployment\u002FSKILL.md",[85],{"type":46,"tag":86,"props":87,"children":89},"code",{"className":88},[],[90],{"type":51,"value":91},"doca-container-deployment",{"type":51,"value":93},"\nfor the non-container path. If the user has a DOCA-linked binary\nthey built (per the canonical workflow in\n",{"type":46,"tag":81,"props":95,"children":97},{"href":96},"..\u002Fdoca-programming-guide\u002FSKILL.md",[98],{"type":46,"tag":86,"props":99,"children":101},{"className":100},[],[102],{"type":51,"value":103},"doca-programming-guide",{"type":51,"value":105},")\nand they want to know ",{"type":46,"tag":66,"props":107,"children":108},{},[109],{"type":51,"value":110},"how to actually run it on the host or on\nthe BlueField Arm cores correctly",{"type":51,"value":112},", open\n",{"type":46,"tag":81,"props":114,"children":116},{"href":115},"TASKS.md",[117],{"type":46,"tag":86,"props":118,"children":120},{"className":119},[],[121],{"type":51,"value":115},{"type":51,"value":123}," and start at\n",{"type":46,"tag":81,"props":125,"children":127},{"href":126},"TASKS.md#configure",[128],{"type":46,"tag":86,"props":129,"children":131},{"className":130},[],[132],{"type":51,"value":133},"## configure",{"type":51,"value":135},". If the question is ",{"type":46,"tag":66,"props":137,"children":138},{},[139],{"type":51,"value":140},"what\nshape does the bare-metal runtime even have and what is the\ndeployment contract",{"type":51,"value":142},", start at ",{"type":46,"tag":81,"props":144,"children":146},{"href":145},"CAPABILITIES.md",[147],{"type":46,"tag":86,"props":148,"children":150},{"className":149},[],[151],{"type":51,"value":145},{"type":51,"value":153},".\nIf the user is not yet sure whether their target system shape is\nthe container path or the bare-metal path, route the recognition\nstep to ",{"type":46,"tag":81,"props":155,"children":157},{"href":156},"..\u002Fdoca-setup\u002FSKILL.md",[158],{"type":46,"tag":86,"props":159,"children":161},{"className":160},[],[162],{"type":51,"value":163},"doca-setup",{"type":51,"value":165}," first; only return\nhere once ",{"type":46,"tag":66,"props":167,"children":168},{},[169],{"type":51,"value":170},"bare-metal",{"type":51,"value":172}," is the confirmed shape.",{"type":46,"tag":174,"props":175,"children":177},"h2",{"id":176},"audience",[178],{"type":51,"value":179},"Audience",{"type":46,"tag":54,"props":181,"children":182},{},[183,185,190],{"type":51,"value":184},"This skill serves ",{"type":46,"tag":58,"props":186,"children":187},{},[188],{"type":51,"value":189},"external DOCA developers and operators who\nhave a DOCA-linked application binary they built and want to run\nit directly on hardware",{"type":51,"value":191}," — i.e., people who already have:",{"type":46,"tag":193,"props":194,"children":195},"ul",{},[196,214,233],{"type":46,"tag":197,"props":198,"children":199},"li",{},[200,202,212],{"type":51,"value":201},"a DOCA-linked application binary they built per\n",{"type":46,"tag":81,"props":203,"children":205},{"href":204},"..\u002Fdoca-programming-guide\u002FTASKS.md#build",[206],{"type":46,"tag":86,"props":207,"children":209},{"className":208},[],[210],{"type":51,"value":211},"doca-programming-guide ## build",{"type":51,"value":213},",",{"type":46,"tag":197,"props":215,"children":216},{},[217,219,224,226,231],{"type":51,"value":218},"a real BlueField NIC and a host that talks to it (the ",{"type":46,"tag":58,"props":220,"children":221},{},[222],{"type":51,"value":223},"host\nx86",{"type":51,"value":225}," path — DOCA host install on the host talks to the\nBlueField NIC over PCIe), OR a BlueField with a console or SSH\nto the Arm side (the ",{"type":46,"tag":58,"props":227,"children":228},{},[229],{"type":51,"value":230},"BlueField Arm bare-metal",{"type":51,"value":232}," path — DOCA\ninstalled on the DPU Arm cores; the binary runs there\ndirectly), and",{"type":46,"tag":197,"props":234,"children":235},{},[236],{"type":51,"value":237},"a desire to RUN that binary directly on the hardware, not\ninside a kubelet-standalone-managed container.",{"type":46,"tag":54,"props":239,"children":240},{},[241,243,248],{"type":51,"value":242},"It is ",{"type":46,"tag":58,"props":244,"children":245},{},[246],{"type":51,"value":247},"not",{"type":51,"value":249}," for:",{"type":46,"tag":193,"props":251,"children":252},{},[253,266,271,305],{"type":46,"tag":197,"props":254,"children":255},{},[256,258,264],{"type":51,"value":257},"kernel-driver developers contributing to ",{"type":46,"tag":86,"props":259,"children":261},{"className":260},[],[262],{"type":51,"value":263},"mlx5_*",{"type":51,"value":265}," or the\nBlueField OS,",{"type":46,"tag":197,"props":267,"children":268},{},[269],{"type":51,"value":270},"DOCA library contributors (those changes go to the internal\nDOCA tree, not to a bare-metal deployment),",{"type":46,"tag":197,"props":272,"children":273},{},[274,276,284,286,291,293,303],{"type":51,"value":275},"full-Kubernetes-cluster operators managing a fleet of\nBlueFields (the bundle covers\n",{"type":46,"tag":81,"props":277,"children":278},{"href":83},[279],{"type":46,"tag":86,"props":280,"children":282},{"className":281},[],[283],{"type":51,"value":91},{"type":51,"value":285},"\nfor the single-host kubelet-standalone shape; ",{"type":46,"tag":58,"props":287,"children":288},{},[289],{"type":51,"value":290},"fleet\u002Fproduction-scale\ndeployment is fleet-orchestration scope",{"type":51,"value":292}," — route to the orchestration\nentry-point in\n",{"type":46,"tag":81,"props":294,"children":296},{"href":295},"..\u002Fdoca-public-knowledge-map\u002Freferences\u002Fmap.md#deploying-doca-services-at-scale--orchestration-entry-point-personascale-routing",[297],{"type":46,"tag":86,"props":298,"children":300},{"className":299},[],[301],{"type":51,"value":302},"doca-public-knowledge-map ## Deploying DOCA services at scale",{"type":51,"value":304},"\n(DPF \u002F Network Operator \u002F Launch Kit), not hand-rolled static-pod loops),",{"type":46,"tag":197,"props":306,"children":307},{},[308,310,320],{"type":51,"value":309},"fresh-laptop-no-hardware users with no DOCA install yet — those\nbelong on\n",{"type":46,"tag":81,"props":311,"children":313},{"href":312},"..\u002Fdoca-setup\u002FTASKS.md#no-install",[314],{"type":46,"tag":86,"props":315,"children":317},{"className":316},[],[318],{"type":51,"value":319},"doca-setup ## no-install",{"type":51,"value":321},".",{"type":46,"tag":54,"props":323,"children":324},{},[325,327,332,334,344,346,352],{"type":51,"value":326},"The skill teaches the agent the bare-metal-deployment ",{"type":46,"tag":66,"props":328,"children":329},{},[330],{"type":51,"value":331},"procedure",{"type":51,"value":333},"\nand the rules for quoting documented commands from the public DOCA\nProgramming Guide and the public BlueField \u002F DPU User Manual via\n",{"type":46,"tag":81,"props":335,"children":337},{"href":336},"..\u002Fdoca-public-knowledge-map\u002FSKILL.md",[338],{"type":46,"tag":86,"props":339,"children":341},{"className":340},[],[342],{"type":51,"value":343},"doca-public-knowledge-map",{"type":51,"value":345},";\nit does not invent flag names, PCI BDFs, NUMA numbers, devlink\npaths, representor strings, or systemd ",{"type":46,"tag":86,"props":347,"children":349},{"className":348},[],[350],{"type":51,"value":351},"Restart=",{"type":51,"value":353}," mode names from\nmemory.",{"type":46,"tag":174,"props":355,"children":357},{"id":356},"when-to-load-this-skill",[358],{"type":51,"value":359},"When to load this skill",{"type":46,"tag":54,"props":361,"children":362},{},[363,365,370],{"type":51,"value":364},"Load this skill when the user is doing ",{"type":46,"tag":58,"props":366,"children":367},{},[368],{"type":51,"value":369},"hands-on bare-metal\ndeployment of a DOCA-linked application binary",{"type":51,"value":371}," on either of the\ntwo supported host modes (host x86 or BlueField Arm), or asking a\ncross-cutting bare-metal question that is not specific to one\nlibrary's API. Concretely:",{"type":46,"tag":193,"props":373,"children":374},{},[375,380,385,390,395,416,421],{"type":46,"tag":197,"props":376,"children":377},{},[378],{"type":51,"value":379},"Launching a DOCA-linked binary for the first time on a host\nwith a BlueField NIC in a PCIe slot, with DOCA installed on the\nhost.",{"type":46,"tag":197,"props":381,"children":382},{},[383],{"type":51,"value":384},"Launching a DOCA-linked binary on the BlueField Arm cores\ndirectly (BlueField Arm bare-metal mode), with DOCA installed\non the Arm side per the BlueField OS image.",{"type":46,"tag":197,"props":386,"children":387},{},[388],{"type":51,"value":389},"Deciding which launch mode to use (direct foreground for\ninteractive debug; tmux\u002Fscreen for long-running with manual\nreattach; systemd-supervised for restart-after-reboot,\njournald-integrated logs, and Restart= policy).",{"type":46,"tag":197,"props":391,"children":392},{},[393],{"type":51,"value":394},"Binding the DOCA process to the right PCIe function, the right\nrepresentor, the right NUMA node, and the right CPU set — and\npinning IRQs to match — without inventing the addresses or the\nflag names.",{"type":46,"tag":197,"props":396,"children":397},{},[398,400,406,408,414],{"type":51,"value":399},"Setting up per-tenant isolation (cgroup-v2 cpu \u002F memory \u002F io\ncontrollers, network namespaces for multi-tenant deployments,\n",{"type":46,"tag":86,"props":401,"children":403},{"className":402},[],[404],{"type":51,"value":405},"numactl",{"type":51,"value":407}," \u002F ",{"type":46,"tag":86,"props":409,"children":411},{"className":410},[],[412],{"type":51,"value":413},"taskset",{"type":51,"value":415}," for CPU + NUMA binding) so multiple DOCA\nprocesses co-tenant on the same BlueField without crushing each\nother.",{"type":46,"tag":197,"props":417,"children":418},{},[419],{"type":51,"value":420},"Diagnosing a bare-metal launch that is misbehaving — won't\nstart, starts and exits immediately, runs but can't find the\ndevice, attaches to the device but the workload errors, OOMs or\nis signal-killed, is in a restart loop under a supervisor, or\nis being interfered with by a co-tenant.",{"type":46,"tag":197,"props":422,"children":423},{},[424,426,431,433,438,439,444],{"type":51,"value":425},"Cross-cutting questions: ",{"type":46,"tag":66,"props":427,"children":428},{},[429],{"type":51,"value":430},"\"should I run this in tmux or as a\nsystemd unit\"",{"type":51,"value":432},", ",{"type":46,"tag":66,"props":434,"children":435},{},[436],{"type":51,"value":437},"\"what is the smoke-before-bulk loop for a\nbinary on bare metal\"",{"type":51,"value":432},{"type":46,"tag":66,"props":440,"children":441},{},[442],{"type":51,"value":443},"\"my binary works in a container on the\nBlueField but not when I run it directly on the Arm — what\nchanged\"",{"type":51,"value":321},{"type":46,"tag":54,"props":446,"children":447},{},[448,450,454,456,464,466,472,474,482,484,490,492,502,504,512],{"type":51,"value":449},"Do ",{"type":46,"tag":58,"props":451,"children":452},{},[453],{"type":51,"value":247},{"type":51,"value":455}," load this skill for the container-path equivalent\n(those questions go to\n",{"type":46,"tag":81,"props":457,"children":458},{"href":83},[459],{"type":46,"tag":86,"props":460,"children":462},{"className":461},[],[463],{"type":51,"value":91},{"type":51,"value":465},");\nfor full-Kubernetes-cluster operations (out of scope per the\nbundle's non-goals); for library-API questions (route to the\nmatching ",{"type":46,"tag":86,"props":467,"children":469},{"className":468},[],[470],{"type":51,"value":471},"libs\u002F\u003Clibrary>",{"type":51,"value":473}," skill); for env-preparation questions\nincluding hugepages, IOMMU, pkg-config, and devlink mode flips\n(use ",{"type":46,"tag":81,"props":475,"children":476},{"href":156},[477],{"type":46,"tag":86,"props":478,"children":480},{"className":479},[],[481],{"type":51,"value":163},{"type":51,"value":483},"); for any\nhardware-state-changing operation including ",{"type":46,"tag":86,"props":485,"children":487},{"className":486},[],[488],{"type":51,"value":489},"mlxconfig",{"type":51,"value":491}," writes\nand BFB reflashes (route to\n",{"type":46,"tag":81,"props":493,"children":495},{"href":494},"..\u002Fdoca-hardware-safety\u002FSKILL.md",[496],{"type":46,"tag":86,"props":497,"children":499},{"className":498},[],[500],{"type":51,"value":501},"doca-hardware-safety",{"type":51,"value":503}," for the\ncross-cutting meta-policy); or for cross-library programming\nquestions (use\n",{"type":46,"tag":81,"props":505,"children":506},{"href":96},[507],{"type":46,"tag":86,"props":508,"children":510},{"className":509},[],[511],{"type":51,"value":103},{"type":51,"value":513},").",{"type":46,"tag":174,"props":515,"children":517},{"id":516},"what-this-skill-provides",[518],{"type":51,"value":519},"What this skill provides",{"type":46,"tag":54,"props":521,"children":522},{},[523,525,530],{"type":51,"value":524},"This is a ",{"type":46,"tag":58,"props":526,"children":527},{},[528],{"type":51,"value":529},"thin loader",{"type":51,"value":531},". Substantive material lives in two\ncompanion files:",{"type":46,"tag":193,"props":533,"children":534},{},[535,629],{"type":46,"tag":197,"props":536,"children":537},{},[538,543,545,550,551,556,558,564,566,571,573,583,585,591,592,598,599,604,605,610,612,620,622,627],{"type":46,"tag":86,"props":539,"children":541},{"className":540},[],[542],{"type":51,"value":145},{"type":51,"value":544}," — the bare-metal deployment runtime contract\nfor a DOCA-linked binary: the two host modes (host x86 vs\nBlueField Arm bare-metal), the three launch modes (direct,\ntmux\u002Fscreen, systemd-supervised), the hardware-resource-binding\nsurface (PF \u002F VF \u002F representor enumeration; NUMA topology\ndiscovery; CPU pinning rationale; IRQ affinity rules), the\nper-tenant isolation surface (cgroup-v2 cpu \u002F memory \u002F io,\nnetwork namespaces, ",{"type":46,"tag":86,"props":546,"children":548},{"className":547},[],[549],{"type":51,"value":405},{"type":51,"value":407},{"type":46,"tag":86,"props":552,"children":554},{"className":553},[],[555],{"type":51,"value":413},{"type":51,"value":557},"), the restart and\nrecovery semantics (documented ",{"type":46,"tag":86,"props":559,"children":561},{"className":560},[],[562],{"type":51,"value":563},"systemd",{"type":51,"value":565}," ",{"type":46,"tag":86,"props":567,"children":569},{"className":568},[],[570],{"type":51,"value":351},{"type":51,"value":572}," modes vs\ncrash-and-investigate vs supervisor-driven restart), the\nbare-metal-specific version overlay on the four-way version\nmatch owned by\n",{"type":46,"tag":81,"props":574,"children":576},{"href":575},"..\u002Fdoca-version\u002FSKILL.md",[577],{"type":46,"tag":86,"props":578,"children":580},{"className":579},[],[581],{"type":51,"value":582},"doca-version",{"type":51,"value":584},", the cross-cutting\nerror taxonomy (seven layers, walked in order), the observability\nsurface (stdout\u002Fstderr discipline by launch mode; device-state\nintrospection via ",{"type":46,"tag":86,"props":586,"children":588},{"className":587},[],[589],{"type":51,"value":590},"devlink",{"type":51,"value":407},{"type":46,"tag":86,"props":593,"children":595},{"className":594},[],[596],{"type":51,"value":597},"sysfs",{"type":51,"value":407},{"type":46,"tag":86,"props":600,"children":602},{"className":601},[],[603],{"type":51,"value":489},{"type":51,"value":565},{"type":46,"tag":66,"props":606,"children":607},{},[608],{"type":51,"value":609},"query",{"type":51,"value":611},";\nper-tenant resource visibility), and the safety policy (overlay\non\n",{"type":46,"tag":81,"props":613,"children":614},{"href":494},[615],{"type":46,"tag":86,"props":616,"children":618},{"className":617},[],[619],{"type":51,"value":501},{"type":51,"value":621},":\nsmoke-before-bulk for binaries; failed bare-metal process is\nHIGH-STAKES; do not invent PCI addresses, NUMA numbers,\nrepresentor names, devlink paths, or systemd ",{"type":46,"tag":86,"props":623,"children":625},{"className":624},[],[626],{"type":51,"value":351},{"type":51,"value":628}," mode\nnames; confirm tenant-isolation primitives BEFORE the workload\nstarts).",{"type":46,"tag":197,"props":630,"children":631},{},[632,637,639,645,646,652,653,659,660,666,668,674,676,682,683,689,691,697,699,705,707,713,715,721,723,729,731,739],{"type":46,"tag":86,"props":633,"children":635},{"className":634},[],[636],{"type":51,"value":115},{"type":51,"value":638}," — step-by-step workflows for the in-scope bare-metal\nverbs: ",{"type":46,"tag":86,"props":640,"children":642},{"className":641},[],[643],{"type":51,"value":644},"configure",{"type":51,"value":432},{"type":46,"tag":86,"props":647,"children":649},{"className":648},[],[650],{"type":51,"value":651},"build",{"type":51,"value":432},{"type":46,"tag":86,"props":654,"children":656},{"className":655},[],[657],{"type":51,"value":658},"modify",{"type":51,"value":432},{"type":46,"tag":86,"props":661,"children":663},{"className":662},[],[664],{"type":51,"value":665},"run",{"type":51,"value":667}," (with an explicit\n",{"type":46,"tag":86,"props":669,"children":671},{"className":670},[],[672],{"type":51,"value":673},"### isolation",{"type":51,"value":675}," sub-anchor covering cgroup-v2 \u002F namespaces \u002F\nnumactl per-tenant primitives), ",{"type":46,"tag":86,"props":677,"children":679},{"className":678},[],[680],{"type":51,"value":681},"test",{"type":51,"value":432},{"type":46,"tag":86,"props":684,"children":686},{"className":685},[],[687],{"type":51,"value":688},"debug",{"type":51,"value":690},",\n",{"type":46,"tag":86,"props":692,"children":694},{"className":693},[],[695],{"type":51,"value":696},"bluefield-lifecycle",{"type":51,"value":698}," (the BFB-install → RShim\u002FTMFIFO →\npost-BFB-recovery operational sequencing ladder, with the\nsix-state ",{"type":46,"tag":86,"props":700,"children":702},{"className":701},[],[703],{"type":51,"value":704},"bluefield-state-classifier",{"type":51,"value":706}," sub-anchor), the\n",{"type":46,"tag":86,"props":708,"children":710},{"className":709},[],[711],{"type":51,"value":712},"Command appendix",{"type":51,"value":714}," (documented commands the agent may quote,\neach cross-linked to its public-doc source — no invented\ncommands), and the ",{"type":46,"tag":86,"props":716,"children":718},{"className":717},[],[719],{"type":51,"value":720},"Deferred task verbs",{"type":51,"value":722}," block routing\ncontainer-path \u002F cluster \u002F library-API \u002F env-prep \u002F\nhardware-state-change \u002F cross-library questions out to their\nowning skills. (The change-application discipline for any\nmutating burn invoked from ",{"type":46,"tag":86,"props":724,"children":726},{"className":725},[],[727],{"type":51,"value":728},"## bluefield-lifecycle",{"type":51,"value":730}," is still\nmeta-policy owned by\n",{"type":46,"tag":81,"props":732,"children":733},{"href":494},[734],{"type":46,"tag":86,"props":735,"children":737},{"className":736},[],[738],{"type":51,"value":501},{"type":51,"value":740},",\nloaded alongside.)",{"type":46,"tag":54,"props":742,"children":743},{},[744],{"type":51,"value":745},"The skill assumes a host or BlueField target where:",{"type":46,"tag":193,"props":747,"children":748},{},[749,766,780],{"type":46,"tag":197,"props":750,"children":751},{},[752,754,764],{"type":51,"value":753},"DOCA is already installed and healthy (per\n",{"type":46,"tag":81,"props":755,"children":757},{"href":756},"..\u002Fdoca-setup\u002FTASKS.md#test",[758],{"type":46,"tag":86,"props":759,"children":761},{"className":760},[],[762],{"type":51,"value":763},"doca-setup ## test",{"type":51,"value":765},"),",{"type":46,"tag":197,"props":767,"children":768},{},[769,771,779],{"type":51,"value":770},"the user has a DOCA-linked application binary they built (per\n",{"type":46,"tag":81,"props":772,"children":773},{"href":204},[774],{"type":46,"tag":86,"props":775,"children":777},{"className":776},[],[778],{"type":51,"value":211},{"type":51,"value":765},{"type":46,"tag":197,"props":781,"children":782},{},[783],{"type":51,"value":784},"the user has the host-OS permissions to enumerate devices,\nreserve hugepages, write systemd units (if they choose that\nlaunch mode), and bind processes to NUMA nodes.",{"type":46,"tag":54,"props":786,"children":787},{},[788,790,798,800,808],{"type":51,"value":789},"It does not cover installing DOCA — that path goes through\n",{"type":46,"tag":81,"props":791,"children":792},{"href":156},[793],{"type":46,"tag":86,"props":794,"children":796},{"className":795},[],[797],{"type":51,"value":163},{"type":51,"value":799}," — and it does not cover\nbuilding the binary — that path goes through\n",{"type":46,"tag":81,"props":801,"children":802},{"href":96},[803],{"type":46,"tag":86,"props":804,"children":806},{"className":805},[],[807],{"type":51,"value":103},{"type":51,"value":321},{"type":46,"tag":174,"props":810,"children":812},{"id":811},"loading-order",[813],{"type":51,"value":814},"Loading order",{"type":46,"tag":816,"props":817,"children":818},"ol",{},[819,832,845],{"type":46,"tag":197,"props":820,"children":821},{},[822,824,830],{"type":51,"value":823},"Read this ",{"type":46,"tag":86,"props":825,"children":827},{"className":826},[],[828],{"type":51,"value":829},"SKILL.md",{"type":51,"value":831}," first to confirm the user's question is\nin scope (bare-metal launch of a DOCA-linked binary on host\nx86 or BlueField Arm; NOT the container path, NOT a full\ncluster, NOT a library-API question).",{"type":46,"tag":197,"props":833,"children":834},{},[835],{"type":46,"tag":58,"props":836,"children":837},{},[838,840,844],{"type":51,"value":839},"For the runtime contract (two host modes, three launch\nmodes, hardware-binding surface, per-tenant isolation, version\noverlay, seven-layer error taxonomy, observability surface,\nbare-metal safety overlay), see\n",{"type":46,"tag":81,"props":841,"children":842},{"href":145},[843],{"type":51,"value":145},{"type":51,"value":321},{"type":46,"tag":197,"props":846,"children":847},{},[848],{"type":46,"tag":58,"props":849,"children":850},{},[851,853,858,859,864,866,871,873,878,880,885,887,892,893,898,899,904,906,911,913,918,920,925,927,931],{"type":51,"value":852},"For step-by-step workflows — ",{"type":46,"tag":86,"props":854,"children":856},{"className":855},[],[857],{"type":51,"value":644},{"type":51,"value":432},{"type":46,"tag":86,"props":860,"children":862},{"className":861},[],[863],{"type":51,"value":651},{"type":51,"value":865}," (routing\nstub), ",{"type":46,"tag":86,"props":867,"children":869},{"className":868},[],[870],{"type":51,"value":658},{"type":51,"value":872}," (routing stub), ",{"type":46,"tag":86,"props":874,"children":876},{"className":875},[],[877],{"type":51,"value":665},{"type":51,"value":879}," (with ",{"type":46,"tag":86,"props":881,"children":883},{"className":882},[],[884],{"type":51,"value":673},{"type":51,"value":886},"\nsub-anchor), ",{"type":46,"tag":86,"props":888,"children":890},{"className":889},[],[891],{"type":51,"value":681},{"type":51,"value":432},{"type":46,"tag":86,"props":894,"children":896},{"className":895},[],[897],{"type":51,"value":688},{"type":51,"value":432},{"type":46,"tag":86,"props":900,"children":902},{"className":901},[],[903],{"type":51,"value":696},{"type":51,"value":905}," (BFB\ninstall + RShim\u002FTMFIFO + post-BFB recovery + the six-state\n",{"type":46,"tag":86,"props":907,"children":909},{"className":908},[],[910],{"type":51,"value":704},{"type":51,"value":912},"), plus the ",{"type":46,"tag":86,"props":914,"children":916},{"className":915},[],[917],{"type":51,"value":712},{"type":51,"value":919}," and\nthe ",{"type":46,"tag":86,"props":921,"children":923},{"className":922},[],[924],{"type":51,"value":720},{"type":51,"value":926}," block — see ",{"type":46,"tag":81,"props":928,"children":929},{"href":115},[930],{"type":51,"value":115},{"type":51,"value":321},{"type":46,"tag":174,"props":933,"children":935},{"id":934},"example-questions-this-skill-answers-well",[936],{"type":51,"value":937},"Example questions this skill answers well",{"type":46,"tag":54,"props":939,"children":940},{},[941,943,953],{"type":51,"value":942},"See ",{"type":46,"tag":81,"props":944,"children":946},{"href":945},"references\u002Fdetails.md#example-questions-this-skill-answers-well",[947],{"type":46,"tag":86,"props":948,"children":950},{"className":949},[],[951],{"type":51,"value":952},"references\u002Fdetails.md",{"type":51,"value":321},{"type":46,"tag":174,"props":955,"children":957},{"id":956},"what-this-skill-deliberately-does-not-ship",[958],{"type":51,"value":959},"What this skill deliberately does not ship",{"type":46,"tag":54,"props":961,"children":962},{},[963,964,973],{"type":51,"value":942},{"type":46,"tag":81,"props":965,"children":967},{"href":966},"references\u002Fdetails.md#what-this-skill-deliberately-does-not-ship",[968],{"type":46,"tag":86,"props":969,"children":971},{"className":970},[],[972],{"type":51,"value":952},{"type":51,"value":321},{"type":46,"tag":174,"props":975,"children":977},{"id":976},"related-skills",[978],{"type":51,"value":979},"Related skills",{"type":46,"tag":54,"props":981,"children":982},{},[983,984,993],{"type":51,"value":942},{"type":46,"tag":81,"props":985,"children":987},{"href":986},"references\u002Fdetails.md#related-skills",[988],{"type":46,"tag":86,"props":989,"children":991},{"className":990},[],[992],{"type":51,"value":952},{"type":51,"value":321},{"items":995,"total":1097},[996,1013,1025,1039,1051,1068,1083],{"slug":997,"name":997,"fn":998,"description":999,"org":1000,"tags":1001,"stars":26,"repoUrl":27,"updatedAt":1012},"accelerated-computing-cudf","accelerate data processing with cuDF","Official NVIDIA-authored guidance for NVIDIA cuDF GPU DataFrames, pandas acceleration, dask-cuDF, ETL, joins, groupby, CSV\u002FParquet I\u002FO, nullable semantics, and multi-GPU DataFrame workloads.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[1002,1005,1008,1009],{"name":1003,"slug":1004,"type":15},"Data Analysis","data-analysis",{"name":1006,"slug":1007,"type":15},"Data Engineering","data-engineering",{"name":9,"slug":8,"type":15},{"name":1010,"slug":1011,"type":15},"Performance","performance","2026-07-14T05:28:43.176466",{"slug":1014,"name":1014,"fn":1015,"description":1016,"org":1017,"tags":1018,"stars":26,"repoUrl":27,"updatedAt":1024},"aiq-deploy","deploy and manage NVIDIA AI-Q infrastructure","Use when asked to install, deploy, run, validate, troubleshoot, or stop NVIDIA AI-Q Blueprint infrastructure.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[1019,1020,1023],{"name":17,"slug":18,"type":15},{"name":1021,"slug":1022,"type":15},"Infrastructure","infrastructure",{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":1026,"name":1026,"fn":1027,"description":1028,"org":1029,"tags":1030,"stars":26,"repoUrl":27,"updatedAt":1038},"aiq-research","conduct deep research with AI-Q","Use when asked to run deep research or AI-Q research through a reachable NVIDIA AI-Q Blueprint backend.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[1031,1034,1035],{"name":1032,"slug":1033,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":1036,"slug":1037,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":1040,"name":1040,"fn":1041,"description":1042,"org":1043,"tags":1044,"stars":26,"repoUrl":27,"updatedAt":1050},"amc-run-sample-calibration","run AMC sample dataset calibration","Run end-to-end calibration on the shipped sample dataset (sdg_08_2_sample_data_010926.zip) against a running AMC microservice. Use when user says 'test sample dataset', 'run sample calibration', 'verify AMC install', or 'launch and test'.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[1045,1046,1047],{"name":1003,"slug":1004,"type":15},{"name":9,"slug":8,"type":15},{"name":1048,"slug":1049,"type":15},"Testing","testing","2026-07-17T05:29:03.913266",{"slug":1052,"name":1052,"fn":1053,"description":1054,"org":1055,"tags":1056,"stars":26,"repoUrl":27,"updatedAt":1067},"amc-run-video-calibration","calibrate video datasets with AutoMagicCalib","Calibrate a new dataset from pre-recorded video files via the AutoMagicCalib REST API. Use when user has local MP4s and says 'calibrate my videos', 'run AMC on these videos', or similar. For RTSP\u002Flive streams, use amc-run-rtsp-calibration instead.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[1057,1060,1063,1064],{"name":1058,"slug":1059,"type":15},"Automation","automation",{"name":1061,"slug":1062,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":1065,"slug":1066,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":1069,"name":1069,"fn":1070,"description":1071,"org":1072,"tags":1073,"stars":26,"repoUrl":27,"updatedAt":1082},"amc-setup-calibration-stack","deploy AutoMagicCalib microservice with Docker","Launch AutoMagicCalib microservice and web UI from NGC release images via Docker Compose. Use when user says 'deploy auto calibration', 'launch auto calibration', 'launch AMC', 'start MS+UI', or 'set up auto-magic-calib'. Requires NGC API key.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[1074,1075,1078,1079],{"name":17,"slug":18,"type":15},{"name":1076,"slug":1077,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":1080,"slug":1081,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":1084,"name":1084,"fn":1085,"description":1086,"org":1087,"tags":1088,"stars":26,"repoUrl":27,"updatedAt":1096},"cudaq-guide","develop quantum applications with CUDA-Q","CUDA-Q onboarding guide for installation, test programs, GPU simulation, QPU hardware, and quantum applications.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[1089,1090,1093],{"name":9,"slug":8,"type":15},{"name":1091,"slug":1092,"type":15},"Quantum Computing","quantum-computing",{"name":1094,"slug":1095,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305,{"items":1099,"total":1248},[1100,1118,1134,1145,1157,1169,1182,1196,1207,1216,1230,1239],{"slug":1101,"name":1101,"fn":1102,"description":1103,"org":1104,"tags":1105,"stars":1115,"repoUrl":1116,"updatedAt":1117},"nemoclaw-user-guide","retrieve NemoClaw documentation and configuration","Guides human users' AI agents to the NemoClaw docs MCP server and canonical Fern documentation in Markdown form. Use when users ask how to install, configure, operate, troubleshoot, secure, or learn NemoClaw with an AI coding assistant. Trigger keywords - nemoclaw docs, use nemoclaw with ai agent, nemoclaw mcp docs, nemoclaw install help, nemoclaw quickstart, nemoclaw markdown docs, llms.txt, agent skills.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[1106,1109,1112],{"name":1107,"slug":1108,"type":15},"Documentation","documentation",{"name":1110,"slug":1111,"type":15},"MCP","mcp",{"name":1113,"slug":1114,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":1119,"name":1119,"fn":1120,"description":1121,"org":1122,"tags":1123,"stars":1131,"repoUrl":1132,"updatedAt":1133},"mcore-build-and-dependency","manage Megatron-LM development environments","Container-based dev environment setup and dependency management for Megatron-LM. Covers acquiring and launching the CI container, uv package management, and updating uv.lock.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[1124,1127,1128],{"name":1125,"slug":1126,"type":15},"Containers","containers",{"name":17,"slug":18,"type":15},{"name":1129,"slug":1130,"type":15},"Python","python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":1135,"name":1135,"fn":1136,"description":1137,"org":1138,"tags":1139,"stars":1131,"repoUrl":1132,"updatedAt":1144},"mcore-bump-base-image","update NVIDIA PyTorch base images","Bump the NVIDIA PyTorch base image (`nvcr.io\u002Fnvidia\u002Fpytorch:YY.MM-py3`) used by Megatron-LM CI. Covers the two pin sites (GitHub CI in `docker\u002F.ngc_version.dev` and GitLab CI in `.gitlab\u002Fstages\u002F01.build.yml`), the post-bump CI loop (re-run functional tests, refresh golden values, mark broken tests), and the gotchas that bit PRs",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[1140,1143],{"name":1141,"slug":1142,"type":15},"CI\u002FCD","ci-cd",{"name":17,"slug":18,"type":15},"2026-07-14T05:25:59.97109",{"slug":1146,"name":1146,"fn":1147,"description":1148,"org":1149,"tags":1150,"stars":1131,"repoUrl":1132,"updatedAt":1156},"mcore-cicd","manage CI\u002FCD pipelines for Megatron-LM","CI\u002FCD reference for Megatron-LM. Covers CI pipeline structure, PR scope labels, triggering internal GitLab CI (which force-pushes the current branch to a pull-request\u002FBRANCH ref — always dry-run and verify the destination first; never run against shared or protected branches), and CI failure investigation.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[1151,1152,1153],{"name":1141,"slug":1142,"type":15},{"name":17,"slug":18,"type":15},{"name":1154,"slug":1155,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":1158,"name":1158,"fn":1159,"description":1160,"org":1161,"tags":1162,"stars":1131,"repoUrl":1132,"updatedAt":1168},"mcore-create-issue","investigate CI failures and create issues","Investigate a failing GitHub Actions run or job and create a GitHub issue for the failure.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[1163,1164,1165],{"name":24,"slug":25,"type":15},{"name":1154,"slug":1155,"type":15},{"name":1166,"slug":1167,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":1170,"name":1170,"fn":1171,"description":1172,"org":1173,"tags":1174,"stars":1131,"repoUrl":1132,"updatedAt":1181},"mcore-linting-and-formatting","lint and format Megatron-LM code","Linting and formatting for Megatron-LM. Covers running autoformat.sh, tools (ruff, black, isort, pylint, mypy), and code style rules.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[1175,1178],{"name":1176,"slug":1177,"type":15},"Best Practices","best-practices",{"name":1179,"slug":1180,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":1183,"name":1183,"fn":1184,"description":1185,"org":1186,"tags":1187,"stars":1131,"repoUrl":1132,"updatedAt":1195},"mcore-migrate-gpt-to-hybrid","migrate Megatron-LM models to HybridModel","Migration guide for moving Megatron Core GPTModel checkpoints, model providers, training commands, and layer mappings to HybridModel.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[1188,1191,1194],{"name":1189,"slug":1190,"type":15},"Machine Learning","machine-learning",{"name":1192,"slug":1193,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":1197,"name":1197,"fn":1198,"description":1199,"org":1200,"tags":1201,"stars":1131,"repoUrl":1132,"updatedAt":1206},"mcore-onboard-gb200-1node-tests","onboard functional tests for GB200","Onboard 1-node GitHub MR functional tests for GB200 from existing mr-scoped 2-node tests.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[1202,1205],{"name":1203,"slug":1204,"type":15},"QA","qa",{"name":1048,"slug":1049,"type":15},"2026-07-14T05:25:53.673039",{"slug":1208,"name":1208,"fn":1209,"description":1210,"org":1211,"tags":1212,"stars":1131,"repoUrl":1132,"updatedAt":1215},"mcore-run-on-slurm","launch distributed training jobs on SLURM","How to launch distributed Megatron-LM training jobs on a SLURM cluster. Covers a minimal sbatch skeleton, environment-variable setup for torch.distributed.run, CUDA_DEVICE_MAX_CONNECTIONS rules across hardware and parallelism modes, container conventions, monitoring, and per-rank failure diagnosis.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[1213,1214],{"name":17,"slug":18,"type":15},{"name":1021,"slug":1022,"type":15},"2026-07-14T05:25:49.362534",{"slug":1217,"name":1217,"fn":1218,"description":1219,"org":1220,"tags":1221,"stars":1131,"repoUrl":1132,"updatedAt":1229},"mcore-split-pr","split pull requests to reduce review load","Split a PR into multiple PRs to reduce the number of required CODEOWNERS reviewer groups.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[1222,1225,1226],{"name":1223,"slug":1224,"type":15},"Code Review","code-review",{"name":1154,"slug":1155,"type":15},{"name":1227,"slug":1228,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":1231,"name":1231,"fn":1232,"description":1233,"org":1234,"tags":1235,"stars":1131,"repoUrl":1132,"updatedAt":1238},"mcore-testing","run and manage Megatron-LM tests","Test system for Megatron-LM. Covers test layout, recipe YAML structure, adding and running unit and functional tests, golden values, marker filters, and CI parity.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[1236,1237],{"name":1203,"slug":1204,"type":15},{"name":1048,"slug":1049,"type":15},"2026-07-14T05:25:54.928983",{"slug":1240,"name":1240,"fn":1241,"description":1242,"org":1243,"tags":1244,"stars":1131,"repoUrl":1132,"updatedAt":1247},"nightly-sync","manage nightly main-to-dev sync workflows","Domain knowledge for the nightly main-to-dev sync workflow. Covers merge strategy, CI architecture, failure investigation, and known issues.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[1245,1246],{"name":1058,"slug":1059,"type":15},{"name":1141,"slug":1142,"type":15},"2026-07-30T05:29:03.275638",496]