[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-doca-hardware-safety":3,"mdc-dlu9s6-key":34,"related-org-nvidia-doca-hardware-safety":1620,"related-repo-nvidia-doca-hardware-safety":1781},{"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},"doca-hardware-safety","manage NVIDIA DPU and NIC hardware state","Use this skill whenever the agent is about to recommend or apply a change that touches DPU \u002F NIC hardware state on a live system — mlxconfig firmware-parameter write, NIC firmware burn, BFB reflash, NIC ↔ DPU mode flip, SR-IOV or device-emulation slot enable, kernel boot-parameter change (IOMMU, hugepages, VFIO), PCIe rebind \u002F rescan \u002F link-state flip, or BlueField cold reboot. Wraps the change in pre-flight inventory, OOB reachability, a maintenance window, the mlxconfig cold-power-cycle rule, replica rehearsal, and rollback. Trigger even when the user does not say \"hardware safety\" — implicit phrasings: \"flip BlueField mode over SSH\", \"enable SR-IOV and reboot\", \"burned firmware but mlxconfig shows old value\", \"reflashed BFB and lost representors\", \"reflash during business hours\", \"vendor says this is one-way\". Refuse for general DOCA orientation (doca-public-knowledge-map), install or env debug (doca-setup), and program-side debug (doca-debug, doca-programming-guide) — those belong to 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",[12,16,19,22],{"name":13,"slug":14,"type":15},"Security","security","tag",{"name":17,"slug":18,"type":15},"Operations","operations",{"name":20,"slug":21,"type":15},"Hardware","hardware",{"name":9,"slug":8,"type":15},2473,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills","2026-07-30T05:28:20.416606","Apache-2.0",281,[],{"repoUrl":24,"stars":23,"forks":27,"topics":30,"description":31},[],"AI agent skills published by NVIDIA","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills\u002Ftree\u002FHEAD\u002Fskills\u002Fdoca-hardware-safety","---\nlicense: Apache-2.0\nname: doca-hardware-safety\ndescription: >\n  Use this skill whenever the agent is about to recommend or apply a\n  change that touches DPU \u002F NIC hardware state on a live system —\n  mlxconfig firmware-parameter write, NIC firmware burn, BFB reflash,\n  NIC ↔ DPU mode flip, SR-IOV or device-emulation slot enable, kernel\n  boot-parameter change (IOMMU, hugepages, VFIO), PCIe rebind \u002F\n  rescan \u002F link-state flip, or BlueField cold reboot. Wraps the\n  change in pre-flight inventory, OOB reachability, a maintenance\n  window, the mlxconfig cold-power-cycle rule, replica rehearsal, and\n  rollback. Trigger even when the user does not say \"hardware safety\"\n  — implicit phrasings: \"flip BlueField mode over SSH\", \"enable\n  SR-IOV and reboot\", \"burned firmware but mlxconfig shows old\n  value\", \"reflashed BFB and lost representors\", \"reflash during\n  business hours\", \"vendor says this is one-way\". Refuse for general\n  DOCA orientation (doca-public-knowledge-map), install or env debug\n  (doca-setup), and program-side debug (doca-debug,\n  doca-programming-guide) — those belong to 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 DO require a live DOCA install at \u002Fopt\u002Fmellanox\u002Fdoca with a\n  BlueField DPU or ConnectX NIC, plus out-of-band console\n  reachability (BMC, RShim, or operator-managed console) for any\n  link-breaking change.\n---\n\n# DOCA hardware safety\n\n**Where to start:** This skill is the bundle's single source of truth\nfor the discipline that wraps every change touching DPU \u002F NIC hardware\nstate on a live system. Open\n[`TASKS.md`](TASKS.md) when the operator is about to *apply* a\nhardware-touching change and needs the change-application discipline\n(pre-flight inventory → out-of-band path → window → apply → verify\n→ rollback). Open [`CAPABILITIES.md`](CAPABILITIES.md) when the\nquestion is *what does hardware-safety even cover* (the class of\nchanges in scope, the failure modes the policy prevents, the\nobservability surface that gates a change, and the meta-policy that\nevery per-artifact `## Safety policy` overlays).\n\nEvery per-artifact skill (services, libraries, tools) in the bundle\nthat recommends a hardware-touching action overlays this meta-policy\nwith artifact-specific safety. The per-artifact `## Safety policy`\nanchors do NOT redefine the cross-cutting discipline — they layer the\nartifact's own concerns on top of it. This skill is the layer they all\nbuild on.\n\n## Example questions this skill answers well\n\nThe CLASSES of hardware-safety questions this skill is built to\nanswer, each with one worked example. The agent should treat the\n*class* as load-bearing — the worked example is a single instance.\n\n- **\"I'm about to apply a hardware-touching change. What do I have to\n  capture *before* I touch anything?\"** — worked example: *\"the\n  per-artifact skill told me to flip a firmware-level emulation slot;\n  what do I capture first?\"*. Answered by the pre-flight inventory\n  in [`TASKS.md ## configure`](TASKS.md#configure) plus the\n  inventory taxonomy in\n  [`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes).\n- **\"This change might drop the link I'm using to manage the\n  BlueField. Is that safe?\"** — worked example: *\"I'm about to flip\n  the BlueField between NIC and DPU mode over the same management\n  link\"*. Answered by the out-of-band access rule in\n  [`CAPABILITIES.md ## Safety policy`](CAPABILITIES.md#safety-policy)\n  plus the OOB-precondition gate in\n  [`TASKS.md ## configure`](TASKS.md#configure).\n- **\"The per-artifact skill said to write an `mlxconfig` parameter,\n  then reboot. Is that the right sequence?\"** — worked example:\n  *\"the storage-emulation skill told me to enable a firmware slot\n  via `mlxconfig` and then warm-reboot to apply it\"*. Answered by\n  the `mlxconfig`-class rule in\n  [`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes)\n  plus the apply-with-cold-power-cycle workflow in\n  [`TASKS.md ## modify`](TASKS.md#modify).\n- **\"My deployment plan reflashes the BlueField BFB during business\n  hours. Is that OK?\"** — worked example: *\"I have a one-hour\n  window during the day; can I reflash now?\"*. Answered by the\n  maintenance-window discipline in\n  [`CAPABILITIES.md ## Safety policy`](CAPABILITIES.md#safety-policy)\n  plus the firmware-burn workflow in\n  [`TASKS.md ## modify`](TASKS.md#modify).\n- **\"How do I prove the change works *before* I touch production?\"**\n  — worked example: *\"the change is small; can I skip the lab\n  replica\"*. Answered by the replica-first rule in\n  [`TASKS.md ## test`](TASKS.md#test) plus the\n  pre-hardware-validation pattern in\n  [`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes).\n- **\"How do I roll back if this change goes wrong?\"** — worked\n  example: *\"I just reflashed the BFB and the host can't see the\n  representors anymore\"*. Answered by the rollback ladder in\n  [`TASKS.md ## debug`](TASKS.md#debug) plus the\n  rollback-must-be-documented rule in\n  [`CAPABILITIES.md ## Safety policy`](CAPABILITIES.md#safety-policy).\n- **\"This change doesn't have a documented rollback. Should I still\n  apply it?\"** — worked example: *\"the vendor says this firmware\n  rev is one-way\"*. Answered by the refuse-and-escalate rule in\n  [`CAPABILITIES.md ## Safety policy`](CAPABILITIES.md#safety-policy)\n  plus the escalation path in [`TASKS.md ## debug`](TASKS.md#debug).\n\n## When to load this skill\n\nLoad this skill whenever the agent is about to recommend, or is\nhelping the operator apply, a change that touches DPU \u002F NIC hardware\nstate on a live system. The decision must be made **before** the\nagent composes its first sentence — the activation checklist below\nis the same one referenced from\n[`AGENTS.md ## Cross-cutting overlay activation triggers`](..\u002F..\u002FAGENTS.md#cross-cutting-overlay-activation-triggers),\nmirrored here so a per-artifact skill that already loaded this skill\nhas the activation rule at hand.\n\n## Agent activation checklist — load this skill at the START of the answer when any cell below is true\n\n| Trigger class | Concrete prompt-side signals (any one fires the overlay) |\n| --- | --- |\n| `mlxconfig`-class change | the prompt or the agent's next recommended action mentions `mlxconfig` directly; OR toggles BlueField between NIC \u002F DPU \u002F Separated-Host mode; OR enables SR-IOV; OR enables device-emulation slot (virtio-net, NVMe-emu, snap, virtio-blk-emu); OR changes BAR window \u002F window size; OR sets any firmware-committed parameter that requires a reset to take effect |\n| Firmware \u002F BFB | NIC firmware burn (`flint`, `mft`, `mlxfwmanager`, `mlxconfig` with `-y` reset); BlueField BFB reflash (`bfb-install`, `rshim`); BlueField mode change that requires a BFB swap |\n| Host kernel state | host kernel boot parameter change (IOMMU mode `iommu=pt` \u002F `intel_iommu=on`, `default_hugepagesz`, `hugepagesz`, `nr_hugepages`, `pci=resource_alignment`, `vfio-pci.ids`); requires host reboot |\n| Hugepages | hugepage *reservation* change (`\u002Fsys\u002Fkernel\u002Fmm\u002Fhugepages\u002F*\u002Fnr_hugepages`, `sysctl vm.nr_hugepages`); hugepage *mount* change (`mount -t hugetlbfs`); this is global state shared with every DOCA \u002F DPDK process on the host |\n| PCIe state | PCIe rebind \u002F `echo > \u002Fsys\u002Fbus\u002Fpci\u002F...\u002F{bind,unbind,remove,rescan}`; representor enable\u002Fdisable; eswitch mode change (`devlink dev eswitch set ... mode {switchdev,legacy}`); link down\u002Fup on a port currently carrying traffic |\n| BlueField reboot class | BlueField cold reboot, BlueField warm reboot to apply `mlxconfig`; any change whose blast radius is *\"every hosted service on this DPU restarts\"* |\n| Per-artifact cross-link | any per-artifact skill's `## Safety policy` cross-links here for the cross-cutting rule body |\n\nWhen any cell above fires, the agent MUST load this skill **before**\ncomposing the first sentence of the answer. For production it MUST\nwalk the change-application discipline in this order:\n[`TASKS.md ## configure`](TASKS.md#configure) (plan) →\n[`## test`](TASKS.md#test) (representative replica change + rollback\nrehearsal) → [`## modify`](TASKS.md#modify) (production apply) →\n[`## run`](TASKS.md#run) (production verification) →\n[`## debug`](TASKS.md#debug) (debug \u002F rollback). It MUST cite the\nactivation explicitly in the answer (e.g. *\"because this touches\n`mlxconfig`, the answer follows the `doca-hardware-safety`\ndiscipline …\"*) so the user can audit the reasoning.\n\nThe activation is **mandatory, not advisory.** The most common failure mode this overlay prevents is *\"the agent recommended a `mlxconfig` change with no maintenance window, no out-of-band path, and no rollback statement, the user applied it, the management link dropped, and the box was unrecoverable without a physical console.\"* The cost of one unjustified activation (a few extra paragraphs in the answer) is trivial compared to the cost of one missed activation.\n\n## Refuse-and-escalate is a hard rule\n\nIf any of the following is true, the agent MUST stop and refuse to recommend the change — not soften the warning, not proceed with a *\"this is risky but here's how\"* answer, not defer the rollback question to *\"you should think about that\"*:\n\n1. The change has no documented rollback path AND the user cannot provide one. (Per [`CAPABILITIES.md ## Safety policy`](CAPABILITIES.md#safety-policy) *rollback-must-be-documented* rule.)\n2. The change is link-breaking AND the host has no out-of-band access path. (Per [`CAPABILITIES.md ## Safety policy`](CAPABILITIES.md#safety-policy) *out-of-band-precondition* rule.)\n3. The change touches hardware state AND the user has not confirmed an\n   explicit, time-boxed maintenance window. (Per\n   [`CAPABILITIES.md ## Safety policy`](CAPABILITIES.md#safety-policy)\n   *maintenance-window* rule.)\n4. Production application is contemplated before the change and its\n   rollback have passed on a representative non-prod replica. This\n   refusal is intent-based: it applies to a plan, recommendation, or\n   next action that would reach production early, not only when the\n   user explicitly asks for \"direct application.\" A replica\n   mismatched on the required hardware, firmware, kernel, module, or\n   function-topology axes does not satisfy the gate; obtain a\n   representative replica or refuse and escalate. (Per\n   [`TASKS.md ## test`](TASKS.md#test) *replica-first* rule.)\n\nIn each of these cases the correct answer shape is *\"this change requires X (here is why); the bundle refuses to recommend it without X; here is the route to obtain X\"* — not silence and not improvisation. The refuse-and-escalate rule is what makes the bundle's hardware-safety guidance trustworthy to production operators.\n\nDo **not** load this skill for general DOCA orientation (use\n[`doca-public-knowledge-map`](..\u002Fdoca-public-knowledge-map\u002FSKILL.md)),\nfor first-time install or env-class debug (use\n[`doca-setup`](..\u002Fdoca-setup\u002FSKILL.md)), or for purely program-side\ndebug that does not touch hardware state (use\n[`doca-debug`](..\u002Fdoca-debug\u002FSKILL.md) or\n[`doca-programming-guide`](..\u002Fdoca-programming-guide\u002FSKILL.md)).\n\n## What this skill provides\n\nThis is a **thin loader**. The body keeps only the orientation needed\nto pick the right next file. The substantive content lives in two\ncompanion files:\n\n- `CAPABILITIES.md` — the meta-policy surface: the class of changes\n  in scope (the pre-flight inventory taxonomy, the\n  `mlxconfig`-class \u002F firmware-burn \u002F kernel-boot-parameter\n  groupings), the cross-cutting safety policy that every per-artifact\n  `## Safety policy` overlays, the failure modes the policy prevents\n  (bricked-link, runaway-burn, silent-mode-change, missing-rollback),\n  the observability gate the operator must satisfy before any\n  workload moves, and the thin version-compatibility overlay that\n  redirects to [`doca-version`](..\u002Fdoca-version\u002FSKILL.md).\n- `TASKS.md` — the change-application workflows: `## configure` (the\n  pre-flight inventory + out-of-band + maintenance-window plan),\n  `## build` (routing stub — hardware-touching changes do not produce\n  build artifacts), `## modify` (the apply-the-change discipline,\n  including the `mlxconfig` cold-power-cycle rule and the\n  firmware-burn discipline), `## run` (the post-change verification\n  gate), `## test` (the replica-first smoke), `## debug` (the\n  rollback ladder + the refuse-and-escalate escape valve), and the\n  `## Deferred task verbs` block.\n\n## Loading order\n\n1. Read this `SKILL.md` first to confirm the user's question is in\n   scope (the agent is about to recommend a change that touches\n   hardware state on a live system).\n2. **For the class of changes in scope, the meta-safety policy, the\n   failure-mode taxonomy, the observability gate, and the\n   version-overlay redirect, see [CAPABILITIES.md](CAPABILITIES.md).**\n3. **For the apply-a-change workflow — pre-flight inventory →\n   out-of-band → maintenance window → apply → verify → rollback —\n   see [TASKS.md](TASKS.md).**\n4. The per-artifact specifics (which exact firmware slot to flip,\n   which exact kernel parameter the operator needs, which exact\n   container tag the operator must roll back to) live in the matching\n   per-artifact skill's `## Safety policy` overlay. This skill does\n   NOT name those specifics; the agent reaches them by routing back\n   to the per-artifact skill after the meta-policy is satisfied.\n\n## Related skills\n\n- [`doca-version`](..\u002Fdoca-version\u002FSKILL.md) — the four-way match\n  rule and the host ↔ BlueField BFB ↔ container-tag pairing. Every\n  hardware-touching change has a version dimension; this skill's\n  `## Version compatibility` overlay is a 3-5 line redirect to\n  `doca-version` for the body.\n- [`doca-setup`](..\u002Fdoca-setup\u002FSKILL.md) — env-class checks that\n  precondition a hardware-touching change (hugepages, IOMMU mode,\n  `pkg-config`, representor visibility). The pre-flight inventory in\n  this skill's `## configure` cross-links to `doca-setup` for the\n  env-class half of the inventory.\n- [`doca-debug`](..\u002Fdoca-debug\u002FSKILL.md) — the cross-cutting layered\n  debug ladder. When a hardware-touching change goes wrong, the\n  rollback ladder in this skill's `## debug` hands off to\n  `doca-debug` once the rollback has restored a known state and the\n  symptom now lives at a software layer.\n- [`doca-structured-tools-contract`](..\u002Fdoca-structured-tools-contract\u002FSKILL.md) —\n  the JSON schemas the agent prefers when present. The\n  `collect-host-state` \u002F `collect-dpu-state` schemas are the\n  structured form of this skill's pre-flight inventory; the agent\n  uses them as the one-shot answer when the host has the helpers\n  installed.\n- [`doca-container-deployment`](..\u002Fdoca-container-deployment\u002FSKILL.md) —\n  the canonical container-deployment recipe shared across DOCA\n  services. Several hardware-touching changes (BlueField cold\n  reboot, BFB reflash) interrupt every hosted service container on\n  the BlueField; the rollback path quotes the\n  `doca-container-deployment` re-deploy shape.\n- [`doca-programming-guide`](..\u002Fdoca-programming-guide\u002FSKILL.md) —\n  program-side preconditions (capability discovery,\n  validate-before-commit). The post-change verification gate in this\n  skill's `## run` cross-links there for the program-side\n  observability surface that must be visible before any production\n  workload moves.\n- Per-artifact `## Safety policy` anchors in each in-bundle\n  service \u002F library \u002F tool skill — e.g. the firmware-slot\n  precondition in [`doca-argus`](..\u002Fservices\u002Fdoca-argus\u002FSKILL.md),\n  [`doca-dms`](..\u002Fservices\u002Fdoca-dms\u002FSKILL.md),\n  [`doca-firefly`](..\u002Fservices\u002Fdoca-firefly\u002FSKILL.md),\n  [`doca-urom-svc`](..\u002Fservices\u002Fdoca-urom-svc\u002FSKILL.md); the\n  device-touching libraries\n  ([`doca-flow`](..\u002Flibs\u002Fdoca-flow\u002FSKILL.md),\n  [`doca-rdma`](..\u002Flibs\u002Fdoca-rdma\u002FSKILL.md),\n  [`doca-eth`](..\u002Flibs\u002Fdoca-eth\u002FSKILL.md),\n  [`doca-pcc`](..\u002Flibs\u002Fdoca-pcc\u002FSKILL.md),\n  [`doca-rmax`](..\u002Flibs\u002Fdoca-rmax\u002FSKILL.md)); and the\n  hardware-touching tools (e.g.\n  [`doca-spcx-cc`](..\u002Ftools\u002Fdoca-spcx-cc\u002FSKILL.md),\n  [`doca-pcc-counters`](..\u002Ftools\u002Fdoca-pcc-counters\u002FSKILL.md)).\n  Every in-bundle artifact skill's `## Safety policy` overlays\n  this meta-policy with artifact-specific safety. The cross-link is\n  intentionally bidirectional: per-artifact skills link here for\n  the meta-policy; this skill enumerates the in-bundle overlays in\n  [`CAPABILITIES.md ## Safety policy`](CAPABILITIES.md#safety-policy)\n  as \"skills that overlay this meta-policy\". The externally-\n  productized analogs (`doca-virtio-net`, `doca-snap`, `doca-hbn`,\n  BlueMan, DPF) are NOT in-bundle skills — their safety policies\n  live in product documentation reached through\n  [`doca-public-knowledge-map ## Externally-productized DOCA\n  software`](..\u002Fdoca-public-knowledge-map\u002FSKILL.md#externally-productized-doca-software--not-in-this-bundle-but-here-is-where-to-route).\n",{"data":35,"body":39},{"license":26,"name":4,"description":6,"metadata":36,"compatibility":38},{"kind":37},"library","No DOCA install required to read this skill (it is an overlay loaded against any DOCA artifact skill); the validation steps within DO require a live DOCA install at \u002Fopt\u002Fmellanox\u002Fdoca with a BlueField DPU or ConnectX NIC, plus out-of-band console reachability (BMC, RShim, or operator-managed console) for any link-breaking change.\n",{"type":40,"children":41},"root",[42,50,109,121,128,140,442,448,471,477,782,869,895,901,920,1005,1017,1077,1083,1095,1202,1208,1262,1268],{"type":43,"tag":44,"props":45,"children":46},"element","h1",{"id":4},[47],{"type":48,"value":49},"text","DOCA hardware safety",{"type":43,"tag":51,"props":52,"children":53},"p",{},[54,60,62,73,75,81,83,92,94,99,101,107],{"type":43,"tag":55,"props":56,"children":57},"strong",{},[58],{"type":48,"value":59},"Where to start:",{"type":48,"value":61}," This skill is the bundle's single source of truth\nfor the discipline that wraps every change touching DPU \u002F NIC hardware\nstate on a live system. Open\n",{"type":43,"tag":63,"props":64,"children":66},"a",{"href":65},"TASKS.md",[67],{"type":43,"tag":68,"props":69,"children":71},"code",{"className":70},[],[72],{"type":48,"value":65},{"type":48,"value":74}," when the operator is about to ",{"type":43,"tag":76,"props":77,"children":78},"em",{},[79],{"type":48,"value":80},"apply",{"type":48,"value":82}," a\nhardware-touching change and needs the change-application discipline\n(pre-flight inventory → out-of-band path → window → apply → verify\n→ rollback). Open ",{"type":43,"tag":63,"props":84,"children":86},{"href":85},"CAPABILITIES.md",[87],{"type":43,"tag":68,"props":88,"children":90},{"className":89},[],[91],{"type":48,"value":85},{"type":48,"value":93}," when the\nquestion is ",{"type":43,"tag":76,"props":95,"children":96},{},[97],{"type":48,"value":98},"what does hardware-safety even cover",{"type":48,"value":100}," (the class of\nchanges in scope, the failure modes the policy prevents, the\nobservability surface that gates a change, and the meta-policy that\nevery per-artifact ",{"type":43,"tag":68,"props":102,"children":104},{"className":103},[],[105],{"type":48,"value":106},"## Safety policy",{"type":48,"value":108}," overlays).",{"type":43,"tag":51,"props":110,"children":111},{},[112,114,119],{"type":48,"value":113},"Every per-artifact skill (services, libraries, tools) in the bundle\nthat recommends a hardware-touching action overlays this meta-policy\nwith artifact-specific safety. The per-artifact ",{"type":43,"tag":68,"props":115,"children":117},{"className":116},[],[118],{"type":48,"value":106},{"type":48,"value":120},"\nanchors do NOT redefine the cross-cutting discipline — they layer the\nartifact's own concerns on top of it. This skill is the layer they all\nbuild on.",{"type":43,"tag":122,"props":123,"children":125},"h2",{"id":124},"example-questions-this-skill-answers-well",[126],{"type":48,"value":127},"Example questions this skill answers well",{"type":43,"tag":51,"props":129,"children":130},{},[131,133,138],{"type":48,"value":132},"The CLASSES of hardware-safety questions this skill is built to\nanswer, each with one worked example. The agent should treat the\n",{"type":43,"tag":76,"props":134,"children":135},{},[136],{"type":48,"value":137},"class",{"type":48,"value":139}," as load-bearing — the worked example is a single instance.",{"type":43,"tag":141,"props":142,"children":143},"ul",{},[144,193,230,290,325,369,407],{"type":43,"tag":145,"props":146,"children":147},"li",{},[148,160,162,167,169,179,181,191],{"type":43,"tag":55,"props":149,"children":150},{},[151,153,158],{"type":48,"value":152},"\"I'm about to apply a hardware-touching change. What do I have to\ncapture ",{"type":43,"tag":76,"props":154,"children":155},{},[156],{"type":48,"value":157},"before",{"type":48,"value":159}," I touch anything?\"",{"type":48,"value":161}," — worked example: ",{"type":43,"tag":76,"props":163,"children":164},{},[165],{"type":48,"value":166},"\"the\nper-artifact skill told me to flip a firmware-level emulation slot;\nwhat do I capture first?\"",{"type":48,"value":168},". Answered by the pre-flight inventory\nin ",{"type":43,"tag":63,"props":170,"children":172},{"href":171},"TASKS.md#configure",[173],{"type":43,"tag":68,"props":174,"children":176},{"className":175},[],[177],{"type":48,"value":178},"TASKS.md ## configure",{"type":48,"value":180}," plus the\ninventory taxonomy in\n",{"type":43,"tag":63,"props":182,"children":184},{"href":183},"CAPABILITIES.md#capabilities-and-modes",[185],{"type":43,"tag":68,"props":186,"children":188},{"className":187},[],[189],{"type":48,"value":190},"CAPABILITIES.md ## Capabilities and modes",{"type":48,"value":192},".",{"type":43,"tag":145,"props":194,"children":195},{},[196,201,202,207,209,219,221,229],{"type":43,"tag":55,"props":197,"children":198},{},[199],{"type":48,"value":200},"\"This change might drop the link I'm using to manage the\nBlueField. Is that safe?\"",{"type":48,"value":161},{"type":43,"tag":76,"props":203,"children":204},{},[205],{"type":48,"value":206},"\"I'm about to flip\nthe BlueField between NIC and DPU mode over the same management\nlink\"",{"type":48,"value":208},". Answered by the out-of-band access rule in\n",{"type":43,"tag":63,"props":210,"children":212},{"href":211},"CAPABILITIES.md#safety-policy",[213],{"type":43,"tag":68,"props":214,"children":216},{"className":215},[],[217],{"type":48,"value":218},"CAPABILITIES.md ## Safety policy",{"type":48,"value":220},"\nplus the OOB-precondition gate in\n",{"type":43,"tag":63,"props":222,"children":223},{"href":171},[224],{"type":43,"tag":68,"props":225,"children":227},{"className":226},[],[228],{"type":48,"value":178},{"type":48,"value":192},{"type":43,"tag":145,"props":231,"children":232},{},[233,246,248,260,262,267,269,277,279,289],{"type":43,"tag":55,"props":234,"children":235},{},[236,238,244],{"type":48,"value":237},"\"The per-artifact skill said to write an ",{"type":43,"tag":68,"props":239,"children":241},{"className":240},[],[242],{"type":48,"value":243},"mlxconfig",{"type":48,"value":245}," parameter,\nthen reboot. Is that the right sequence?\"",{"type":48,"value":247}," — worked example:\n",{"type":43,"tag":76,"props":249,"children":250},{},[251,253,258],{"type":48,"value":252},"\"the storage-emulation skill told me to enable a firmware slot\nvia ",{"type":43,"tag":68,"props":254,"children":256},{"className":255},[],[257],{"type":48,"value":243},{"type":48,"value":259}," and then warm-reboot to apply it\"",{"type":48,"value":261},". Answered by\nthe ",{"type":43,"tag":68,"props":263,"children":265},{"className":264},[],[266],{"type":48,"value":243},{"type":48,"value":268},"-class rule in\n",{"type":43,"tag":63,"props":270,"children":271},{"href":183},[272],{"type":43,"tag":68,"props":273,"children":275},{"className":274},[],[276],{"type":48,"value":190},{"type":48,"value":278},"\nplus the apply-with-cold-power-cycle workflow in\n",{"type":43,"tag":63,"props":280,"children":282},{"href":281},"TASKS.md#modify",[283],{"type":43,"tag":68,"props":284,"children":286},{"className":285},[],[287],{"type":48,"value":288},"TASKS.md ## modify",{"type":48,"value":192},{"type":43,"tag":145,"props":291,"children":292},{},[293,298,299,304,306,314,316,324],{"type":43,"tag":55,"props":294,"children":295},{},[296],{"type":48,"value":297},"\"My deployment plan reflashes the BlueField BFB during business\nhours. Is that OK?\"",{"type":48,"value":161},{"type":43,"tag":76,"props":300,"children":301},{},[302],{"type":48,"value":303},"\"I have a one-hour\nwindow during the day; can I reflash now?\"",{"type":48,"value":305},". Answered by the\nmaintenance-window discipline in\n",{"type":43,"tag":63,"props":307,"children":308},{"href":211},[309],{"type":43,"tag":68,"props":310,"children":312},{"className":311},[],[313],{"type":48,"value":218},{"type":48,"value":315},"\nplus the firmware-burn workflow in\n",{"type":43,"tag":63,"props":317,"children":318},{"href":281},[319],{"type":43,"tag":68,"props":320,"children":322},{"className":321},[],[323],{"type":48,"value":288},{"type":48,"value":192},{"type":43,"tag":145,"props":326,"children":327},{},[328,339,341,346,348,358,360,368],{"type":43,"tag":55,"props":329,"children":330},{},[331,333,337],{"type":48,"value":332},"\"How do I prove the change works ",{"type":43,"tag":76,"props":334,"children":335},{},[336],{"type":48,"value":157},{"type":48,"value":338}," I touch production?\"",{"type":48,"value":340},"\n— worked example: ",{"type":43,"tag":76,"props":342,"children":343},{},[344],{"type":48,"value":345},"\"the change is small; can I skip the lab\nreplica\"",{"type":48,"value":347},". Answered by the replica-first rule in\n",{"type":43,"tag":63,"props":349,"children":351},{"href":350},"TASKS.md#test",[352],{"type":43,"tag":68,"props":353,"children":355},{"className":354},[],[356],{"type":48,"value":357},"TASKS.md ## test",{"type":48,"value":359}," plus the\npre-hardware-validation pattern in\n",{"type":43,"tag":63,"props":361,"children":362},{"href":183},[363],{"type":43,"tag":68,"props":364,"children":366},{"className":365},[],[367],{"type":48,"value":190},{"type":48,"value":192},{"type":43,"tag":145,"props":370,"children":371},{},[372,377,379,384,386,396,398,406],{"type":43,"tag":55,"props":373,"children":374},{},[375],{"type":48,"value":376},"\"How do I roll back if this change goes wrong?\"",{"type":48,"value":378}," — worked\nexample: ",{"type":43,"tag":76,"props":380,"children":381},{},[382],{"type":48,"value":383},"\"I just reflashed the BFB and the host can't see the\nrepresentors anymore\"",{"type":48,"value":385},". Answered by the rollback ladder in\n",{"type":43,"tag":63,"props":387,"children":389},{"href":388},"TASKS.md#debug",[390],{"type":43,"tag":68,"props":391,"children":393},{"className":392},[],[394],{"type":48,"value":395},"TASKS.md ## debug",{"type":48,"value":397}," plus the\nrollback-must-be-documented rule in\n",{"type":43,"tag":63,"props":399,"children":400},{"href":211},[401],{"type":43,"tag":68,"props":402,"children":404},{"className":403},[],[405],{"type":48,"value":218},{"type":48,"value":192},{"type":43,"tag":145,"props":408,"children":409},{},[410,415,416,421,423,431,433,441],{"type":43,"tag":55,"props":411,"children":412},{},[413],{"type":48,"value":414},"\"This change doesn't have a documented rollback. Should I still\napply it?\"",{"type":48,"value":161},{"type":43,"tag":76,"props":417,"children":418},{},[419],{"type":48,"value":420},"\"the vendor says this firmware\nrev is one-way\"",{"type":48,"value":422},". Answered by the refuse-and-escalate rule in\n",{"type":43,"tag":63,"props":424,"children":425},{"href":211},[426],{"type":43,"tag":68,"props":427,"children":429},{"className":428},[],[430],{"type":48,"value":218},{"type":48,"value":432},"\nplus the escalation path in ",{"type":43,"tag":63,"props":434,"children":435},{"href":388},[436],{"type":43,"tag":68,"props":437,"children":439},{"className":438},[],[440],{"type":48,"value":395},{"type":48,"value":192},{"type":43,"tag":122,"props":443,"children":445},{"id":444},"when-to-load-this-skill",[446],{"type":48,"value":447},"When to load this skill",{"type":43,"tag":51,"props":449,"children":450},{},[451,453,457,459,469],{"type":48,"value":452},"Load this skill whenever the agent is about to recommend, or is\nhelping the operator apply, a change that touches DPU \u002F NIC hardware\nstate on a live system. The decision must be made ",{"type":43,"tag":55,"props":454,"children":455},{},[456],{"type":48,"value":157},{"type":48,"value":458}," the\nagent composes its first sentence — the activation checklist below\nis the same one referenced from\n",{"type":43,"tag":63,"props":460,"children":462},{"href":461},"..\u002F..\u002FAGENTS.md#cross-cutting-overlay-activation-triggers",[463],{"type":43,"tag":68,"props":464,"children":466},{"className":465},[],[467],{"type":48,"value":468},"AGENTS.md ## Cross-cutting overlay activation triggers",{"type":48,"value":470},",\nmirrored here so a per-artifact skill that already loaded this skill\nhas the activation rule at hand.",{"type":43,"tag":122,"props":472,"children":474},{"id":473},"agent-activation-checklist-load-this-skill-at-the-start-of-the-answer-when-any-cell-below-is-true",[475],{"type":48,"value":476},"Agent activation checklist — load this skill at the START of the answer when any cell below is true",{"type":43,"tag":478,"props":479,"children":480},"table",{},[481,500],{"type":43,"tag":482,"props":483,"children":484},"thead",{},[485],{"type":43,"tag":486,"props":487,"children":488},"tr",{},[489,495],{"type":43,"tag":490,"props":491,"children":492},"th",{},[493],{"type":48,"value":494},"Trigger class",{"type":43,"tag":490,"props":496,"children":497},{},[498],{"type":48,"value":499},"Concrete prompt-side signals (any one fires the overlay)",{"type":43,"tag":501,"props":502,"children":503},"tbody",{},[504,530,595,659,708,737,762],{"type":43,"tag":486,"props":505,"children":506},{},[507,518],{"type":43,"tag":508,"props":509,"children":510},"td",{},[511,516],{"type":43,"tag":68,"props":512,"children":514},{"className":513},[],[515],{"type":48,"value":243},{"type":48,"value":517},"-class change",{"type":43,"tag":508,"props":519,"children":520},{},[521,523,528],{"type":48,"value":522},"the prompt or the agent's next recommended action mentions ",{"type":43,"tag":68,"props":524,"children":526},{"className":525},[],[527],{"type":48,"value":243},{"type":48,"value":529}," directly; OR toggles BlueField between NIC \u002F DPU \u002F Separated-Host mode; OR enables SR-IOV; OR enables device-emulation slot (virtio-net, NVMe-emu, snap, virtio-blk-emu); OR changes BAR window \u002F window size; OR sets any firmware-committed parameter that requires a reset to take effect",{"type":43,"tag":486,"props":531,"children":532},{},[533,538],{"type":43,"tag":508,"props":534,"children":535},{},[536],{"type":48,"value":537},"Firmware \u002F BFB",{"type":43,"tag":508,"props":539,"children":540},{},[541,543,549,551,557,558,564,565,570,572,578,580,586,587,593],{"type":48,"value":542},"NIC firmware burn (",{"type":43,"tag":68,"props":544,"children":546},{"className":545},[],[547],{"type":48,"value":548},"flint",{"type":48,"value":550},", ",{"type":43,"tag":68,"props":552,"children":554},{"className":553},[],[555],{"type":48,"value":556},"mft",{"type":48,"value":550},{"type":43,"tag":68,"props":559,"children":561},{"className":560},[],[562],{"type":48,"value":563},"mlxfwmanager",{"type":48,"value":550},{"type":43,"tag":68,"props":566,"children":568},{"className":567},[],[569],{"type":48,"value":243},{"type":48,"value":571}," with ",{"type":43,"tag":68,"props":573,"children":575},{"className":574},[],[576],{"type":48,"value":577},"-y",{"type":48,"value":579}," reset); BlueField BFB reflash (",{"type":43,"tag":68,"props":581,"children":583},{"className":582},[],[584],{"type":48,"value":585},"bfb-install",{"type":48,"value":550},{"type":43,"tag":68,"props":588,"children":590},{"className":589},[],[591],{"type":48,"value":592},"rshim",{"type":48,"value":594},"); BlueField mode change that requires a BFB swap",{"type":43,"tag":486,"props":596,"children":597},{},[598,603],{"type":43,"tag":508,"props":599,"children":600},{},[601],{"type":48,"value":602},"Host kernel state",{"type":43,"tag":508,"props":604,"children":605},{},[606,608,614,616,622,623,629,630,636,637,643,644,650,651,657],{"type":48,"value":607},"host kernel boot parameter change (IOMMU mode ",{"type":43,"tag":68,"props":609,"children":611},{"className":610},[],[612],{"type":48,"value":613},"iommu=pt",{"type":48,"value":615}," \u002F ",{"type":43,"tag":68,"props":617,"children":619},{"className":618},[],[620],{"type":48,"value":621},"intel_iommu=on",{"type":48,"value":550},{"type":43,"tag":68,"props":624,"children":626},{"className":625},[],[627],{"type":48,"value":628},"default_hugepagesz",{"type":48,"value":550},{"type":43,"tag":68,"props":631,"children":633},{"className":632},[],[634],{"type":48,"value":635},"hugepagesz",{"type":48,"value":550},{"type":43,"tag":68,"props":638,"children":640},{"className":639},[],[641],{"type":48,"value":642},"nr_hugepages",{"type":48,"value":550},{"type":43,"tag":68,"props":645,"children":647},{"className":646},[],[648],{"type":48,"value":649},"pci=resource_alignment",{"type":48,"value":550},{"type":43,"tag":68,"props":652,"children":654},{"className":653},[],[655],{"type":48,"value":656},"vfio-pci.ids",{"type":48,"value":658},"); requires host reboot",{"type":43,"tag":486,"props":660,"children":661},{},[662,667],{"type":43,"tag":508,"props":663,"children":664},{},[665],{"type":48,"value":666},"Hugepages",{"type":43,"tag":508,"props":668,"children":669},{},[670,672,677,679,685,686,692,694,699,700,706],{"type":48,"value":671},"hugepage ",{"type":43,"tag":76,"props":673,"children":674},{},[675],{"type":48,"value":676},"reservation",{"type":48,"value":678}," change (",{"type":43,"tag":68,"props":680,"children":682},{"className":681},[],[683],{"type":48,"value":684},"\u002Fsys\u002Fkernel\u002Fmm\u002Fhugepages\u002F*\u002Fnr_hugepages",{"type":48,"value":550},{"type":43,"tag":68,"props":687,"children":689},{"className":688},[],[690],{"type":48,"value":691},"sysctl vm.nr_hugepages",{"type":48,"value":693},"); hugepage ",{"type":43,"tag":76,"props":695,"children":696},{},[697],{"type":48,"value":698},"mount",{"type":48,"value":678},{"type":43,"tag":68,"props":701,"children":703},{"className":702},[],[704],{"type":48,"value":705},"mount -t hugetlbfs",{"type":48,"value":707},"); this is global state shared with every DOCA \u002F DPDK process on the host",{"type":43,"tag":486,"props":709,"children":710},{},[711,716],{"type":43,"tag":508,"props":712,"children":713},{},[714],{"type":48,"value":715},"PCIe state",{"type":43,"tag":508,"props":717,"children":718},{},[719,721,727,729,735],{"type":48,"value":720},"PCIe rebind \u002F ",{"type":43,"tag":68,"props":722,"children":724},{"className":723},[],[725],{"type":48,"value":726},"echo > \u002Fsys\u002Fbus\u002Fpci\u002F...\u002F{bind,unbind,remove,rescan}",{"type":48,"value":728},"; representor enable\u002Fdisable; eswitch mode change (",{"type":43,"tag":68,"props":730,"children":732},{"className":731},[],[733],{"type":48,"value":734},"devlink dev eswitch set ... mode {switchdev,legacy}",{"type":48,"value":736},"); link down\u002Fup on a port currently carrying traffic",{"type":43,"tag":486,"props":738,"children":739},{},[740,745],{"type":43,"tag":508,"props":741,"children":742},{},[743],{"type":48,"value":744},"BlueField reboot class",{"type":43,"tag":508,"props":746,"children":747},{},[748,750,755,757],{"type":48,"value":749},"BlueField cold reboot, BlueField warm reboot to apply ",{"type":43,"tag":68,"props":751,"children":753},{"className":752},[],[754],{"type":48,"value":243},{"type":48,"value":756},"; any change whose blast radius is ",{"type":43,"tag":76,"props":758,"children":759},{},[760],{"type":48,"value":761},"\"every hosted service on this DPU restarts\"",{"type":43,"tag":486,"props":763,"children":764},{},[765,770],{"type":43,"tag":508,"props":766,"children":767},{},[768],{"type":48,"value":769},"Per-artifact cross-link",{"type":43,"tag":508,"props":771,"children":772},{},[773,775,780],{"type":48,"value":774},"any per-artifact skill's ",{"type":43,"tag":68,"props":776,"children":778},{"className":777},[],[779],{"type":48,"value":106},{"type":48,"value":781}," cross-links here for the cross-cutting rule body",{"type":43,"tag":51,"props":783,"children":784},{},[785,787,791,793,801,803,812,814,823,825,835,837,846,848,867],{"type":48,"value":786},"When any cell above fires, the agent MUST load this skill ",{"type":43,"tag":55,"props":788,"children":789},{},[790],{"type":48,"value":157},{"type":48,"value":792},"\ncomposing the first sentence of the answer. For production it MUST\nwalk the change-application discipline in this order:\n",{"type":43,"tag":63,"props":794,"children":795},{"href":171},[796],{"type":43,"tag":68,"props":797,"children":799},{"className":798},[],[800],{"type":48,"value":178},{"type":48,"value":802}," (plan) →\n",{"type":43,"tag":63,"props":804,"children":805},{"href":350},[806],{"type":43,"tag":68,"props":807,"children":809},{"className":808},[],[810],{"type":48,"value":811},"## test",{"type":48,"value":813}," (representative replica change + rollback\nrehearsal) → ",{"type":43,"tag":63,"props":815,"children":816},{"href":281},[817],{"type":43,"tag":68,"props":818,"children":820},{"className":819},[],[821],{"type":48,"value":822},"## modify",{"type":48,"value":824}," (production apply) →\n",{"type":43,"tag":63,"props":826,"children":828},{"href":827},"TASKS.md#run",[829],{"type":43,"tag":68,"props":830,"children":832},{"className":831},[],[833],{"type":48,"value":834},"## run",{"type":48,"value":836}," (production verification) →\n",{"type":43,"tag":63,"props":838,"children":839},{"href":388},[840],{"type":43,"tag":68,"props":841,"children":843},{"className":842},[],[844],{"type":48,"value":845},"## debug",{"type":48,"value":847}," (debug \u002F rollback). It MUST cite the\nactivation explicitly in the answer (e.g. ",{"type":43,"tag":76,"props":849,"children":850},{},[851,853,858,860,865],{"type":48,"value":852},"\"because this touches\n",{"type":43,"tag":68,"props":854,"children":856},{"className":855},[],[857],{"type":48,"value":243},{"type":48,"value":859},", the answer follows the ",{"type":43,"tag":68,"props":861,"children":863},{"className":862},[],[864],{"type":48,"value":4},{"type":48,"value":866},"\ndiscipline …\"",{"type":48,"value":868},") so the user can audit the reasoning.",{"type":43,"tag":51,"props":870,"children":871},{},[872,874,879,881,893],{"type":48,"value":873},"The activation is ",{"type":43,"tag":55,"props":875,"children":876},{},[877],{"type":48,"value":878},"mandatory, not advisory.",{"type":48,"value":880}," The most common failure mode this overlay prevents is ",{"type":43,"tag":76,"props":882,"children":883},{},[884,886,891],{"type":48,"value":885},"\"the agent recommended a ",{"type":43,"tag":68,"props":887,"children":889},{"className":888},[],[890],{"type":48,"value":243},{"type":48,"value":892}," change with no maintenance window, no out-of-band path, and no rollback statement, the user applied it, the management link dropped, and the box was unrecoverable without a physical console.\"",{"type":48,"value":894}," The cost of one unjustified activation (a few extra paragraphs in the answer) is trivial compared to the cost of one missed activation.",{"type":43,"tag":122,"props":896,"children":898},{"id":897},"refuse-and-escalate-is-a-hard-rule",[899],{"type":48,"value":900},"Refuse-and-escalate is a hard rule",{"type":43,"tag":51,"props":902,"children":903},{},[904,906,911,913,918],{"type":48,"value":905},"If any of the following is true, the agent MUST stop and refuse to recommend the change — not soften the warning, not proceed with a ",{"type":43,"tag":76,"props":907,"children":908},{},[909],{"type":48,"value":910},"\"this is risky but here's how\"",{"type":48,"value":912}," answer, not defer the rollback question to ",{"type":43,"tag":76,"props":914,"children":915},{},[916],{"type":48,"value":917},"\"you should think about that\"",{"type":48,"value":919},":",{"type":43,"tag":921,"props":922,"children":923},"ol",{},[924,946,966,985],{"type":43,"tag":145,"props":925,"children":926},{},[927,929,937,939,944],{"type":48,"value":928},"The change has no documented rollback path AND the user cannot provide one. (Per ",{"type":43,"tag":63,"props":930,"children":931},{"href":211},[932],{"type":43,"tag":68,"props":933,"children":935},{"className":934},[],[936],{"type":48,"value":218},{"type":48,"value":938}," ",{"type":43,"tag":76,"props":940,"children":941},{},[942],{"type":48,"value":943},"rollback-must-be-documented",{"type":48,"value":945}," rule.)",{"type":43,"tag":145,"props":947,"children":948},{},[949,951,959,960,965],{"type":48,"value":950},"The change is link-breaking AND the host has no out-of-band access path. (Per ",{"type":43,"tag":63,"props":952,"children":953},{"href":211},[954],{"type":43,"tag":68,"props":955,"children":957},{"className":956},[],[958],{"type":48,"value":218},{"type":48,"value":938},{"type":43,"tag":76,"props":961,"children":962},{},[963],{"type":48,"value":964},"out-of-band-precondition",{"type":48,"value":945},{"type":43,"tag":145,"props":967,"children":968},{},[969,971,979,984],{"type":48,"value":970},"The change touches hardware state AND the user has not confirmed an\nexplicit, time-boxed maintenance window. (Per\n",{"type":43,"tag":63,"props":972,"children":973},{"href":211},[974],{"type":43,"tag":68,"props":975,"children":977},{"className":976},[],[978],{"type":48,"value":218},{"type":43,"tag":76,"props":980,"children":981},{},[982],{"type":48,"value":983},"maintenance-window",{"type":48,"value":945},{"type":43,"tag":145,"props":986,"children":987},{},[988,990,998,999,1004],{"type":48,"value":989},"Production application is contemplated before the change and its\nrollback have passed on a representative non-prod replica. This\nrefusal is intent-based: it applies to a plan, recommendation, or\nnext action that would reach production early, not only when the\nuser explicitly asks for \"direct application.\" A replica\nmismatched on the required hardware, firmware, kernel, module, or\nfunction-topology axes does not satisfy the gate; obtain a\nrepresentative replica or refuse and escalate. (Per\n",{"type":43,"tag":63,"props":991,"children":992},{"href":350},[993],{"type":43,"tag":68,"props":994,"children":996},{"className":995},[],[997],{"type":48,"value":357},{"type":48,"value":938},{"type":43,"tag":76,"props":1000,"children":1001},{},[1002],{"type":48,"value":1003},"replica-first",{"type":48,"value":945},{"type":43,"tag":51,"props":1006,"children":1007},{},[1008,1010,1015],{"type":48,"value":1009},"In each of these cases the correct answer shape is ",{"type":43,"tag":76,"props":1011,"children":1012},{},[1013],{"type":48,"value":1014},"\"this change requires X (here is why); the bundle refuses to recommend it without X; here is the route to obtain X\"",{"type":48,"value":1016}," — not silence and not improvisation. The refuse-and-escalate rule is what makes the bundle's hardware-safety guidance trustworthy to production operators.",{"type":43,"tag":51,"props":1018,"children":1019},{},[1020,1022,1027,1029,1039,1041,1051,1053,1063,1065,1075],{"type":48,"value":1021},"Do ",{"type":43,"tag":55,"props":1023,"children":1024},{},[1025],{"type":48,"value":1026},"not",{"type":48,"value":1028}," load this skill for general DOCA orientation (use\n",{"type":43,"tag":63,"props":1030,"children":1032},{"href":1031},"..\u002Fdoca-public-knowledge-map\u002FSKILL.md",[1033],{"type":43,"tag":68,"props":1034,"children":1036},{"className":1035},[],[1037],{"type":48,"value":1038},"doca-public-knowledge-map",{"type":48,"value":1040},"),\nfor first-time install or env-class debug (use\n",{"type":43,"tag":63,"props":1042,"children":1044},{"href":1043},"..\u002Fdoca-setup\u002FSKILL.md",[1045],{"type":43,"tag":68,"props":1046,"children":1048},{"className":1047},[],[1049],{"type":48,"value":1050},"doca-setup",{"type":48,"value":1052},"), or for purely program-side\ndebug that does not touch hardware state (use\n",{"type":43,"tag":63,"props":1054,"children":1056},{"href":1055},"..\u002Fdoca-debug\u002FSKILL.md",[1057],{"type":43,"tag":68,"props":1058,"children":1060},{"className":1059},[],[1061],{"type":48,"value":1062},"doca-debug",{"type":48,"value":1064}," or\n",{"type":43,"tag":63,"props":1066,"children":1068},{"href":1067},"..\u002Fdoca-programming-guide\u002FSKILL.md",[1069],{"type":43,"tag":68,"props":1070,"children":1072},{"className":1071},[],[1073],{"type":48,"value":1074},"doca-programming-guide",{"type":48,"value":1076},").",{"type":43,"tag":122,"props":1078,"children":1080},{"id":1079},"what-this-skill-provides",[1081],{"type":48,"value":1082},"What this skill provides",{"type":43,"tag":51,"props":1084,"children":1085},{},[1086,1088,1093],{"type":48,"value":1087},"This is a ",{"type":43,"tag":55,"props":1089,"children":1090},{},[1091],{"type":48,"value":1092},"thin loader",{"type":48,"value":1094},". The body keeps only the orientation needed\nto pick the right next file. The substantive content lives in two\ncompanion files:",{"type":43,"tag":141,"props":1096,"children":1097},{},[1098,1133],{"type":43,"tag":145,"props":1099,"children":1100},{},[1101,1106,1108,1113,1115,1120,1122,1132],{"type":43,"tag":68,"props":1102,"children":1104},{"className":1103},[],[1105],{"type":48,"value":85},{"type":48,"value":1107}," — the meta-policy surface: the class of changes\nin scope (the pre-flight inventory taxonomy, the\n",{"type":43,"tag":68,"props":1109,"children":1111},{"className":1110},[],[1112],{"type":48,"value":243},{"type":48,"value":1114},"-class \u002F firmware-burn \u002F kernel-boot-parameter\ngroupings), the cross-cutting safety policy that every per-artifact\n",{"type":43,"tag":68,"props":1116,"children":1118},{"className":1117},[],[1119],{"type":48,"value":106},{"type":48,"value":1121}," overlays, the failure modes the policy prevents\n(bricked-link, runaway-burn, silent-mode-change, missing-rollback),\nthe observability gate the operator must satisfy before any\nworkload moves, and the thin version-compatibility overlay that\nredirects to ",{"type":43,"tag":63,"props":1123,"children":1125},{"href":1124},"..\u002Fdoca-version\u002FSKILL.md",[1126],{"type":43,"tag":68,"props":1127,"children":1129},{"className":1128},[],[1130],{"type":48,"value":1131},"doca-version",{"type":48,"value":192},{"type":43,"tag":145,"props":1134,"children":1135},{},[1136,1141,1143,1149,1151,1157,1159,1164,1166,1171,1173,1178,1180,1185,1187,1192,1194,1200],{"type":43,"tag":68,"props":1137,"children":1139},{"className":1138},[],[1140],{"type":48,"value":65},{"type":48,"value":1142}," — the change-application workflows: ",{"type":43,"tag":68,"props":1144,"children":1146},{"className":1145},[],[1147],{"type":48,"value":1148},"## configure",{"type":48,"value":1150}," (the\npre-flight inventory + out-of-band + maintenance-window plan),\n",{"type":43,"tag":68,"props":1152,"children":1154},{"className":1153},[],[1155],{"type":48,"value":1156},"## build",{"type":48,"value":1158}," (routing stub — hardware-touching changes do not produce\nbuild artifacts), ",{"type":43,"tag":68,"props":1160,"children":1162},{"className":1161},[],[1163],{"type":48,"value":822},{"type":48,"value":1165}," (the apply-the-change discipline,\nincluding the ",{"type":43,"tag":68,"props":1167,"children":1169},{"className":1168},[],[1170],{"type":48,"value":243},{"type":48,"value":1172}," cold-power-cycle rule and the\nfirmware-burn discipline), ",{"type":43,"tag":68,"props":1174,"children":1176},{"className":1175},[],[1177],{"type":48,"value":834},{"type":48,"value":1179}," (the post-change verification\ngate), ",{"type":43,"tag":68,"props":1181,"children":1183},{"className":1182},[],[1184],{"type":48,"value":811},{"type":48,"value":1186}," (the replica-first smoke), ",{"type":43,"tag":68,"props":1188,"children":1190},{"className":1189},[],[1191],{"type":48,"value":845},{"type":48,"value":1193}," (the\nrollback ladder + the refuse-and-escalate escape valve), and the\n",{"type":43,"tag":68,"props":1195,"children":1197},{"className":1196},[],[1198],{"type":48,"value":1199},"## Deferred task verbs",{"type":48,"value":1201}," block.",{"type":43,"tag":122,"props":1203,"children":1205},{"id":1204},"loading-order",[1206],{"type":48,"value":1207},"Loading order",{"type":43,"tag":921,"props":1209,"children":1210},{},[1211,1224,1237,1250],{"type":43,"tag":145,"props":1212,"children":1213},{},[1214,1216,1222],{"type":48,"value":1215},"Read this ",{"type":43,"tag":68,"props":1217,"children":1219},{"className":1218},[],[1220],{"type":48,"value":1221},"SKILL.md",{"type":48,"value":1223}," first to confirm the user's question is in\nscope (the agent is about to recommend a change that touches\nhardware state on a live system).",{"type":43,"tag":145,"props":1225,"children":1226},{},[1227],{"type":43,"tag":55,"props":1228,"children":1229},{},[1230,1232,1236],{"type":48,"value":1231},"For the class of changes in scope, the meta-safety policy, the\nfailure-mode taxonomy, the observability gate, and the\nversion-overlay redirect, see ",{"type":43,"tag":63,"props":1233,"children":1234},{"href":85},[1235],{"type":48,"value":85},{"type":48,"value":192},{"type":43,"tag":145,"props":1238,"children":1239},{},[1240],{"type":43,"tag":55,"props":1241,"children":1242},{},[1243,1245,1249],{"type":48,"value":1244},"For the apply-a-change workflow — pre-flight inventory →\nout-of-band → maintenance window → apply → verify → rollback —\nsee ",{"type":43,"tag":63,"props":1246,"children":1247},{"href":65},[1248],{"type":48,"value":65},{"type":48,"value":192},{"type":43,"tag":145,"props":1251,"children":1252},{},[1253,1255,1260],{"type":48,"value":1254},"The per-artifact specifics (which exact firmware slot to flip,\nwhich exact kernel parameter the operator needs, which exact\ncontainer tag the operator must roll back to) live in the matching\nper-artifact skill's ",{"type":43,"tag":68,"props":1256,"children":1258},{"className":1257},[],[1259],{"type":48,"value":106},{"type":48,"value":1261}," overlay. This skill does\nNOT name those specifics; the agent reaches them by routing back\nto the per-artifact skill after the meta-policy is satisfied.",{"type":43,"tag":122,"props":1263,"children":1265},{"id":1264},"related-skills",[1266],{"type":48,"value":1267},"Related skills",{"type":43,"tag":141,"props":1269,"children":1270},{},[1271,1299,1334,1361,1391,1413,1433],{"type":43,"tag":145,"props":1272,"children":1273},{},[1274,1282,1284,1290,1292,1297],{"type":43,"tag":63,"props":1275,"children":1276},{"href":1124},[1277],{"type":43,"tag":68,"props":1278,"children":1280},{"className":1279},[],[1281],{"type":48,"value":1131},{"type":48,"value":1283}," — the four-way match\nrule and the host ↔ BlueField BFB ↔ container-tag pairing. Every\nhardware-touching change has a version dimension; this skill's\n",{"type":43,"tag":68,"props":1285,"children":1287},{"className":1286},[],[1288],{"type":48,"value":1289},"## Version compatibility",{"type":48,"value":1291}," overlay is a 3-5 line redirect to\n",{"type":43,"tag":68,"props":1293,"children":1295},{"className":1294},[],[1296],{"type":48,"value":1131},{"type":48,"value":1298}," for the body.",{"type":43,"tag":145,"props":1300,"children":1301},{},[1302,1310,1312,1318,1320,1325,1327,1332],{"type":43,"tag":63,"props":1303,"children":1304},{"href":1043},[1305],{"type":43,"tag":68,"props":1306,"children":1308},{"className":1307},[],[1309],{"type":48,"value":1050},{"type":48,"value":1311}," — env-class checks that\nprecondition a hardware-touching change (hugepages, IOMMU mode,\n",{"type":43,"tag":68,"props":1313,"children":1315},{"className":1314},[],[1316],{"type":48,"value":1317},"pkg-config",{"type":48,"value":1319},", representor visibility). The pre-flight inventory in\nthis skill's ",{"type":43,"tag":68,"props":1321,"children":1323},{"className":1322},[],[1324],{"type":48,"value":1148},{"type":48,"value":1326}," cross-links to ",{"type":43,"tag":68,"props":1328,"children":1330},{"className":1329},[],[1331],{"type":48,"value":1050},{"type":48,"value":1333}," for the\nenv-class half of the inventory.",{"type":43,"tag":145,"props":1335,"children":1336},{},[1337,1345,1347,1352,1354,1359],{"type":43,"tag":63,"props":1338,"children":1339},{"href":1055},[1340],{"type":43,"tag":68,"props":1341,"children":1343},{"className":1342},[],[1344],{"type":48,"value":1062},{"type":48,"value":1346}," — the cross-cutting layered\ndebug ladder. When a hardware-touching change goes wrong, the\nrollback ladder in this skill's ",{"type":43,"tag":68,"props":1348,"children":1350},{"className":1349},[],[1351],{"type":48,"value":845},{"type":48,"value":1353}," hands off to\n",{"type":43,"tag":68,"props":1355,"children":1357},{"className":1356},[],[1358],{"type":48,"value":1062},{"type":48,"value":1360}," once the rollback has restored a known state and the\nsymptom now lives at a software layer.",{"type":43,"tag":145,"props":1362,"children":1363},{},[1364,1374,1376,1382,1383,1389],{"type":43,"tag":63,"props":1365,"children":1367},{"href":1366},"..\u002Fdoca-structured-tools-contract\u002FSKILL.md",[1368],{"type":43,"tag":68,"props":1369,"children":1371},{"className":1370},[],[1372],{"type":48,"value":1373},"doca-structured-tools-contract",{"type":48,"value":1375}," —\nthe JSON schemas the agent prefers when present. The\n",{"type":43,"tag":68,"props":1377,"children":1379},{"className":1378},[],[1380],{"type":48,"value":1381},"collect-host-state",{"type":48,"value":615},{"type":43,"tag":68,"props":1384,"children":1386},{"className":1385},[],[1387],{"type":48,"value":1388},"collect-dpu-state",{"type":48,"value":1390}," schemas are the\nstructured form of this skill's pre-flight inventory; the agent\nuses them as the one-shot answer when the host has the helpers\ninstalled.",{"type":43,"tag":145,"props":1392,"children":1393},{},[1394,1404,1406,1411],{"type":43,"tag":63,"props":1395,"children":1397},{"href":1396},"..\u002Fdoca-container-deployment\u002FSKILL.md",[1398],{"type":43,"tag":68,"props":1399,"children":1401},{"className":1400},[],[1402],{"type":48,"value":1403},"doca-container-deployment",{"type":48,"value":1405}," —\nthe canonical container-deployment recipe shared across DOCA\nservices. Several hardware-touching changes (BlueField cold\nreboot, BFB reflash) interrupt every hosted service container on\nthe BlueField; the rollback path quotes the\n",{"type":43,"tag":68,"props":1407,"children":1409},{"className":1408},[],[1410],{"type":48,"value":1403},{"type":48,"value":1412}," re-deploy shape.",{"type":43,"tag":145,"props":1414,"children":1415},{},[1416,1424,1426,1431],{"type":43,"tag":63,"props":1417,"children":1418},{"href":1067},[1419],{"type":43,"tag":68,"props":1420,"children":1422},{"className":1421},[],[1423],{"type":48,"value":1074},{"type":48,"value":1425}," —\nprogram-side preconditions (capability discovery,\nvalidate-before-commit). The post-change verification gate in this\nskill's ",{"type":43,"tag":68,"props":1427,"children":1429},{"className":1428},[],[1430],{"type":48,"value":834},{"type":48,"value":1432}," cross-links there for the program-side\nobservability surface that must be visible before any production\nworkload moves.",{"type":43,"tag":145,"props":1434,"children":1435},{},[1436,1438,1443,1445,1455,1457,1467,1468,1478,1479,1489,1491,1501,1502,1512,1513,1523,1524,1534,1535,1545,1547,1557,1558,1568,1570,1575,1577,1585,1587,1593,1594,1600,1601,1607,1609,1619],{"type":48,"value":1437},"Per-artifact ",{"type":43,"tag":68,"props":1439,"children":1441},{"className":1440},[],[1442],{"type":48,"value":106},{"type":48,"value":1444}," anchors in each in-bundle\nservice \u002F library \u002F tool skill — e.g. the firmware-slot\nprecondition in ",{"type":43,"tag":63,"props":1446,"children":1448},{"href":1447},"..\u002Fservices\u002Fdoca-argus\u002FSKILL.md",[1449],{"type":43,"tag":68,"props":1450,"children":1452},{"className":1451},[],[1453],{"type":48,"value":1454},"doca-argus",{"type":48,"value":1456},",\n",{"type":43,"tag":63,"props":1458,"children":1460},{"href":1459},"..\u002Fservices\u002Fdoca-dms\u002FSKILL.md",[1461],{"type":43,"tag":68,"props":1462,"children":1464},{"className":1463},[],[1465],{"type":48,"value":1466},"doca-dms",{"type":48,"value":1456},{"type":43,"tag":63,"props":1469,"children":1471},{"href":1470},"..\u002Fservices\u002Fdoca-firefly\u002FSKILL.md",[1472],{"type":43,"tag":68,"props":1473,"children":1475},{"className":1474},[],[1476],{"type":48,"value":1477},"doca-firefly",{"type":48,"value":1456},{"type":43,"tag":63,"props":1480,"children":1482},{"href":1481},"..\u002Fservices\u002Fdoca-urom-svc\u002FSKILL.md",[1483],{"type":43,"tag":68,"props":1484,"children":1486},{"className":1485},[],[1487],{"type":48,"value":1488},"doca-urom-svc",{"type":48,"value":1490},"; the\ndevice-touching libraries\n(",{"type":43,"tag":63,"props":1492,"children":1494},{"href":1493},"..\u002Flibs\u002Fdoca-flow\u002FSKILL.md",[1495],{"type":43,"tag":68,"props":1496,"children":1498},{"className":1497},[],[1499],{"type":48,"value":1500},"doca-flow",{"type":48,"value":1456},{"type":43,"tag":63,"props":1503,"children":1505},{"href":1504},"..\u002Flibs\u002Fdoca-rdma\u002FSKILL.md",[1506],{"type":43,"tag":68,"props":1507,"children":1509},{"className":1508},[],[1510],{"type":48,"value":1511},"doca-rdma",{"type":48,"value":1456},{"type":43,"tag":63,"props":1514,"children":1516},{"href":1515},"..\u002Flibs\u002Fdoca-eth\u002FSKILL.md",[1517],{"type":43,"tag":68,"props":1518,"children":1520},{"className":1519},[],[1521],{"type":48,"value":1522},"doca-eth",{"type":48,"value":1456},{"type":43,"tag":63,"props":1525,"children":1527},{"href":1526},"..\u002Flibs\u002Fdoca-pcc\u002FSKILL.md",[1528],{"type":43,"tag":68,"props":1529,"children":1531},{"className":1530},[],[1532],{"type":48,"value":1533},"doca-pcc",{"type":48,"value":1456},{"type":43,"tag":63,"props":1536,"children":1538},{"href":1537},"..\u002Flibs\u002Fdoca-rmax\u002FSKILL.md",[1539],{"type":43,"tag":68,"props":1540,"children":1542},{"className":1541},[],[1543],{"type":48,"value":1544},"doca-rmax",{"type":48,"value":1546},"); and the\nhardware-touching tools (e.g.\n",{"type":43,"tag":63,"props":1548,"children":1550},{"href":1549},"..\u002Ftools\u002Fdoca-spcx-cc\u002FSKILL.md",[1551],{"type":43,"tag":68,"props":1552,"children":1554},{"className":1553},[],[1555],{"type":48,"value":1556},"doca-spcx-cc",{"type":48,"value":1456},{"type":43,"tag":63,"props":1559,"children":1561},{"href":1560},"..\u002Ftools\u002Fdoca-pcc-counters\u002FSKILL.md",[1562],{"type":43,"tag":68,"props":1563,"children":1565},{"className":1564},[],[1566],{"type":48,"value":1567},"doca-pcc-counters",{"type":48,"value":1569},").\nEvery in-bundle artifact skill's ",{"type":43,"tag":68,"props":1571,"children":1573},{"className":1572},[],[1574],{"type":48,"value":106},{"type":48,"value":1576}," overlays\nthis meta-policy with artifact-specific safety. The cross-link is\nintentionally bidirectional: per-artifact skills link here for\nthe meta-policy; this skill enumerates the in-bundle overlays in\n",{"type":43,"tag":63,"props":1578,"children":1579},{"href":211},[1580],{"type":43,"tag":68,"props":1581,"children":1583},{"className":1582},[],[1584],{"type":48,"value":218},{"type":48,"value":1586},"\nas \"skills that overlay this meta-policy\". The externally-\nproductized analogs (",{"type":43,"tag":68,"props":1588,"children":1590},{"className":1589},[],[1591],{"type":48,"value":1592},"doca-virtio-net",{"type":48,"value":550},{"type":43,"tag":68,"props":1595,"children":1597},{"className":1596},[],[1598],{"type":48,"value":1599},"doca-snap",{"type":48,"value":550},{"type":43,"tag":68,"props":1602,"children":1604},{"className":1603},[],[1605],{"type":48,"value":1606},"doca-hbn",{"type":48,"value":1608},",\nBlueMan, DPF) are NOT in-bundle skills — their safety policies\nlive in product documentation reached through\n",{"type":43,"tag":63,"props":1610,"children":1612},{"href":1611},"..\u002Fdoca-public-knowledge-map\u002FSKILL.md#externally-productized-doca-software--not-in-this-bundle-but-here-is-where-to-route",[1613],{"type":43,"tag":68,"props":1614,"children":1616},{"className":1615},[],[1617],{"type":48,"value":1618},"doca-public-knowledge-map ## Externally-productized DOCA software",{"type":48,"value":192},{"items":1621,"total":1780},[1622,1640,1658,1669,1681,1695,1708,1722,1735,1746,1760,1769],{"slug":1623,"name":1623,"fn":1624,"description":1625,"org":1626,"tags":1627,"stars":1637,"repoUrl":1638,"updatedAt":1639},"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},[1628,1631,1634],{"name":1629,"slug":1630,"type":15},"Documentation","documentation",{"name":1632,"slug":1633,"type":15},"MCP","mcp",{"name":1635,"slug":1636,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":1641,"name":1641,"fn":1642,"description":1643,"org":1644,"tags":1645,"stars":1655,"repoUrl":1656,"updatedAt":1657},"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},[1646,1649,1652],{"name":1647,"slug":1648,"type":15},"Containers","containers",{"name":1650,"slug":1651,"type":15},"Deployment","deployment",{"name":1653,"slug":1654,"type":15},"Python","python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":1659,"name":1659,"fn":1660,"description":1661,"org":1662,"tags":1663,"stars":1655,"repoUrl":1656,"updatedAt":1668},"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},[1664,1667],{"name":1665,"slug":1666,"type":15},"CI\u002FCD","ci-cd",{"name":1650,"slug":1651,"type":15},"2026-07-14T05:25:59.97109",{"slug":1670,"name":1670,"fn":1671,"description":1672,"org":1673,"tags":1674,"stars":1655,"repoUrl":1656,"updatedAt":1680},"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},[1675,1676,1677],{"name":1665,"slug":1666,"type":15},{"name":1650,"slug":1651,"type":15},{"name":1678,"slug":1679,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":1682,"name":1682,"fn":1683,"description":1684,"org":1685,"tags":1686,"stars":1655,"repoUrl":1656,"updatedAt":1694},"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},[1687,1690,1691],{"name":1688,"slug":1689,"type":15},"Debugging","debugging",{"name":1678,"slug":1679,"type":15},{"name":1692,"slug":1693,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":1696,"name":1696,"fn":1697,"description":1698,"org":1699,"tags":1700,"stars":1655,"repoUrl":1656,"updatedAt":1707},"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},[1701,1704],{"name":1702,"slug":1703,"type":15},"Best Practices","best-practices",{"name":1705,"slug":1706,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":1709,"name":1709,"fn":1710,"description":1711,"org":1712,"tags":1713,"stars":1655,"repoUrl":1656,"updatedAt":1721},"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},[1714,1717,1720],{"name":1715,"slug":1716,"type":15},"Machine Learning","machine-learning",{"name":1718,"slug":1719,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":1723,"name":1723,"fn":1724,"description":1725,"org":1726,"tags":1727,"stars":1655,"repoUrl":1656,"updatedAt":1734},"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},[1728,1731],{"name":1729,"slug":1730,"type":15},"QA","qa",{"name":1732,"slug":1733,"type":15},"Testing","testing","2026-07-14T05:25:53.673039",{"slug":1736,"name":1736,"fn":1737,"description":1738,"org":1739,"tags":1740,"stars":1655,"repoUrl":1656,"updatedAt":1745},"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},[1741,1742],{"name":1650,"slug":1651,"type":15},{"name":1743,"slug":1744,"type":15},"Infrastructure","infrastructure","2026-07-14T05:25:49.362534",{"slug":1747,"name":1747,"fn":1748,"description":1749,"org":1750,"tags":1751,"stars":1655,"repoUrl":1656,"updatedAt":1759},"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},[1752,1755,1756],{"name":1753,"slug":1754,"type":15},"Code Review","code-review",{"name":1678,"slug":1679,"type":15},{"name":1757,"slug":1758,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":1761,"name":1761,"fn":1762,"description":1763,"org":1764,"tags":1765,"stars":1655,"repoUrl":1656,"updatedAt":1768},"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},[1766,1767],{"name":1729,"slug":1730,"type":15},{"name":1732,"slug":1733,"type":15},"2026-07-14T05:25:54.928983",{"slug":1770,"name":1770,"fn":1771,"description":1772,"org":1773,"tags":1774,"stars":1655,"repoUrl":1656,"updatedAt":1779},"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},[1775,1778],{"name":1776,"slug":1777,"type":15},"Automation","automation",{"name":1665,"slug":1666,"type":15},"2026-07-30T05:29:03.275638",496,{"items":1782,"total":1876},[1783,1800,1810,1824,1834,1849,1862],{"slug":1784,"name":1784,"fn":1785,"description":1786,"org":1787,"tags":1788,"stars":23,"repoUrl":24,"updatedAt":1799},"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},[1789,1792,1795,1796],{"name":1790,"slug":1791,"type":15},"Data Analysis","data-analysis",{"name":1793,"slug":1794,"type":15},"Data Engineering","data-engineering",{"name":9,"slug":8,"type":15},{"name":1797,"slug":1798,"type":15},"Performance","performance","2026-07-14T05:28:43.176466",{"slug":1801,"name":1801,"fn":1802,"description":1803,"org":1804,"tags":1805,"stars":23,"repoUrl":24,"updatedAt":1809},"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},[1806,1807,1808],{"name":1650,"slug":1651,"type":15},{"name":1743,"slug":1744,"type":15},{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":1811,"name":1811,"fn":1812,"description":1813,"org":1814,"tags":1815,"stars":23,"repoUrl":24,"updatedAt":1823},"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},[1816,1819,1820],{"name":1817,"slug":1818,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":1821,"slug":1822,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":1825,"name":1825,"fn":1826,"description":1827,"org":1828,"tags":1829,"stars":23,"repoUrl":24,"updatedAt":1833},"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},[1830,1831,1832],{"name":1790,"slug":1791,"type":15},{"name":9,"slug":8,"type":15},{"name":1732,"slug":1733,"type":15},"2026-07-17T05:29:03.913266",{"slug":1835,"name":1835,"fn":1836,"description":1837,"org":1838,"tags":1839,"stars":23,"repoUrl":24,"updatedAt":1848},"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},[1840,1841,1844,1845],{"name":1776,"slug":1777,"type":15},{"name":1842,"slug":1843,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":1846,"slug":1847,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":1850,"name":1850,"fn":1851,"description":1852,"org":1853,"tags":1854,"stars":23,"repoUrl":24,"updatedAt":1861},"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},[1855,1856,1859,1860],{"name":1650,"slug":1651,"type":15},{"name":1857,"slug":1858,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":17,"slug":18,"type":15},"2026-07-17T05:28:56.913999",{"slug":1863,"name":1863,"fn":1864,"description":1865,"org":1866,"tags":1867,"stars":23,"repoUrl":24,"updatedAt":1875},"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},[1868,1869,1872],{"name":9,"slug":8,"type":15},{"name":1870,"slug":1871,"type":15},"Quantum Computing","quantum-computing",{"name":1873,"slug":1874,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305]