[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-doca-collectx-deployment":3,"mdc-2f81pc-key":34,"related-repo-nvidia-doca-collectx-deployment":964,"related-org-nvidia-doca-collectx-deployment":1068},{"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-collectx-deployment","deploy DOCA telemetry collectors","Use this skill to deploy and operate a CollectX (clx) based DOCA telemetry collector on a host or BlueField — wiring providers \u002F counters into the collector, running the collection daemon, and shaping its exporters (Prometheus pull, Fluent Bit push, NetFlow, file \u002F IPC) so the metrics actually leave the box. Trigger even when the user never says CollectX or clx — implicit phrasings: {collector emits nothing downstream}, {add a provider to the clx collector}, {turn on the Prometheus endpoint}, {ship counters to Fluent Bit from the DPU}, {daemon starts but no schema rows appear}. This skill owns the CollectX collection mechanism plus the operator's own doca-telemetry \u002F doca-telemetry-exporter usage; it ROUTES the productized DOCA Telemetry Service (DTS) to public docs (AGENTS.md Non-goal #7), the reader API to doca-telemetry, and the publisher API to doca-telemetry-exporter. Refuse to invent clx symbols, provider names, schema fields, flags, or config paths — describe the class and route to the live source.\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},"Observability","observability","tag",{"name":17,"slug":18,"type":15},"Monitoring","monitoring",{"name":20,"slug":21,"type":15},"Deployment","deployment",{"name":9,"slug":8,"type":15},2473,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills","2026-07-20T06:24:25.344569","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-collectx-deployment","---\nlicense: Apache-2.0\nname: doca-collectx-deployment\ndescription: >\n  Use this skill to deploy and operate a CollectX (clx) based\n  DOCA telemetry collector on a host or BlueField — wiring\n  providers \u002F counters into the collector, running the\n  collection daemon, and shaping its exporters (Prometheus\n  pull, Fluent Bit push, NetFlow, file \u002F IPC) so the metrics\n  actually leave the box. Trigger even when\n  the user never says CollectX or clx — implicit phrasings:\n  {collector emits nothing downstream}, {add a provider to the\n  clx collector}, {turn on the Prometheus endpoint}, {ship\n  counters to Fluent Bit from the DPU}, {daemon starts but no\n  schema rows appear}. This skill owns the CollectX collection\n  mechanism plus the operator's own doca-telemetry \u002F\n  doca-telemetry-exporter usage; it ROUTES the\n  productized DOCA Telemetry Service (DTS) to public docs\n  (AGENTS.md Non-goal #7), the reader API to doca-telemetry, and\n  the publisher API to doca-telemetry-exporter. Refuse to invent\n  clx symbols, provider names, schema fields, flags, or config\n  paths — describe the class and route to the live source.\nmetadata:\n  kind: library\ncompatibility: >\n  No DOCA install required to read this skill (it is a\n  deployment \u002F operation overlay over the DOCA telemetry\n  libraries and the CollectX collection mechanism). The\n  hands-on steps DO require a live DOCA install at\n  \u002Fopt\u002Fmellanox\u002Fdoca on a host or BlueField, an operator\n  account that can run the collector and reach its exporter\n  sinks, and the public DOCA Telemetry \u002F DTS guides on\n  docs.nvidia.com for any concrete provider name, schema field,\n  flag, or config path (this skill never invents those).\n---\n\n# DOCA CollectX telemetry deployment\n\n**Where to start:** This skill is the bundle's home for\n*operating a CollectX (clx) based telemetry collector* — the\ncollection framework that gathers provider counters into a\nschema and ships them out through one or more exporters. It is a\ndeployment \u002F operation skill, parallel to\n[`doca-bare-metal-deployment`](..\u002Fdoca-bare-metal-deployment\u002FSKILL.md)\nand [`doca-container-deployment`](..\u002Fdoca-container-deployment\u002FSKILL.md):\nit owns the *runtime shape* of a telemetry collector on the\noperator's host or BlueField, not the library APIs the operator's\nown program calls. If the user wants to stand up, wire, or debug\na collector and its exporters, open [`TASKS.md`](TASKS.md) and\nstart at [`## configure`](TASKS.md#configure). If the question is\n*what surfaces does the collector even have and where is the\nscope boundary*, start at [`CAPABILITIES.md`](CAPABILITIES.md).\nIf the user has not installed DOCA yet, route to\n[`doca-setup`](..\u002Fdoca-setup\u002FSKILL.md) first.\n\n## The scope boundary (read this before anything else)\n\nCollectX (clx) is NVIDIA's telemetry **collection** framework. It\nunderpins the **DOCA Telemetry Service (DTS)** — and DTS\n*as-deployed* (the productized, NGC-shipped \u002F kubelet-started\nservice container) is **out of scope** for this bundle per\n[`AGENTS.md` Non-goal #7](..\u002F..\u002FAGENTS.md#non-goals-questions-the-agent-should-recognize-and-refuse-politely).\nThis skill therefore draws a hard line and the agent MUST state\nit up front:\n\n- **In scope here:** the CollectX *collection mechanism* as a\n  class (providers \u002F counters → schema → collector daemon →\n  exporters), and the operator deploying \u002F running \u002F debugging a\n  collector that they own, plus the operator's own usage of the\n  two in-bundle telemetry **libraries** when those feed or\n  consume the collector.\n- **Routed to the DOCA telemetry libraries:** the\n  hardware-counter **reader** API is owned by\n  [`doca-telemetry`](..\u002Flibs\u002Fdoca-telemetry\u002FSKILL.md); the\n  application-side **publisher** API (emit counters \u002F events from\n  a DOCA program) is owned by\n  [`doca-telemetry-exporter`](..\u002Flibs\u002Fdoca-telemetry-exporter\u002FSKILL.md).\n  This skill does not re-document either API surface.\n- **Routed to public docs (Non-goal #7):** the productized\n  **DTS container** — its packaged config schema, its built-in\n  provider set, its kubelet manifest, its NGC image — is\n  externally productized. Route every \"operate the DTS service\"\n  question to the\n  [`doca-public-knowledge-map` externally-productized routing row](..\u002Fdoca-public-knowledge-map\u002FSKILL.md#externally-productized-doca-software--not-in-this-bundle-but-here-is-where-to-route)\n  and the public DTS guide it points at. The agent must NOT\n  synthesize DTS config file names, provider knob names, or\n  paths from memory.\n\nThe load-bearing first-touch failure this skill exists to\nprevent is **collapsing these four surfaces into \"DOCA\ntelemetry\"**: the clx collection mechanism, the `doca-telemetry`\nreader library, the `doca-telemetry-exporter` publisher library,\nand the productized DTS container are four different things with\nfour different owners. The agent surfaces the decomposition\nBEFORE any config-level guidance.\n\n## Audience\n\nThis skill serves **external operators standing up or running a\nCollectX-based telemetry collector** on a host or BlueField they\nadminister — people who already have:\n\n- a DOCA install on the side they are collecting from (host x86\n  or BlueField Arm), verified per\n  [`doca-setup ## test`](..\u002Fdoca-setup\u002FTASKS.md#test),\n- a goal of *getting counters off the box* through a collector +\n  exporter, not of writing the reader \u002F publisher library code\n  (that is the two `libs\u002F` skills above), and\n- access to the public DOCA Telemetry and DTS guides on\n  `docs.nvidia.com` as the authoritative source for any concrete\n  provider name, schema field, flag, or config path.\n\nIt is **not** for:\n\n- developers writing the hardware-counter reader API (route to\n  [`doca-telemetry`](..\u002Flibs\u002Fdoca-telemetry\u002FSKILL.md)) or the\n  publisher API (route to\n  [`doca-telemetry-exporter`](..\u002Flibs\u002Fdoca-telemetry-exporter\u002FSKILL.md)),\n- operators deploying \u002F configuring the **productized DTS\n  container** as a turnkey service — that is externally\n  productized (Non-goal #7); route to the public DTS guide,\n- fresh-no-install users — those belong on\n  [`doca-setup ## no-install`](..\u002Fdoca-setup\u002FTASKS.md#no-install).\n\nThe skill teaches the agent the *procedure and the scope\nboundary*; it does not invent clx symbol names, provider names,\nschema field names, exporter flag names, or config paths from\nmemory — those come from the live install and the public docs via\n[`doca-public-knowledge-map`](..\u002Fdoca-public-knowledge-map\u002FSKILL.md).\n\n## When to load this skill\n\nLoad this skill when the user is doing hands-on **deployment or\noperation of a CollectX-based telemetry collector** and the\nquestion is about the collector runtime shape, not a library API.\nConcretely:\n\n- Standing up a collector that gathers provider counters into a\n  schema and ships them out — and deciding which export backend\n  (Prometheus pull, Fluent Bit push, NetFlow, file \u002F IPC) fits\n  the downstream consumer.\n- Wiring a provider \u002F counter family into the collector and\n  confirming the device actually exposes it before the config\n  commits (the gate-before-commit rule, shared with\n  [`doca-telemetry-utils`](..\u002Ftools\u002Fdoca-telemetry-utils\u002FSKILL.md)).\n- Turning on \u002F shaping an exporter so the metrics actually leave\n  the box, and confirming the downstream consumer receives them\n  end-to-end (not just \"the daemon is running\").\n- Diagnosing a collector that starts but produces no schema rows,\n  or ships nothing downstream, or whose exporter endpoint is\n  silent — walking the layered ladder rather than guessing.\n- Recognising when the user is actually asking about the\n  productized DTS container (route to public docs, Non-goal #7),\n  the reader library (route to `doca-telemetry`), or the\n  publisher library (route to `doca-telemetry-exporter`) instead\n  of the collection mechanism this skill owns.\n\nDo **not** load this skill for: the hardware-counter reader API\n(use [`doca-telemetry`](..\u002Flibs\u002Fdoca-telemetry\u002FSKILL.md)); the\npublisher API (use\n[`doca-telemetry-exporter`](..\u002Flibs\u002Fdoca-telemetry-exporter\u002FSKILL.md));\noperating the productized DTS container (route via\n[`doca-public-knowledge-map`](..\u002Fdoca-public-knowledge-map\u002FSKILL.md)\nNon-goal #7); installing DOCA or preparing the env (use\n[`doca-setup`](..\u002Fdoca-setup\u002FSKILL.md)); or any hardware-state\nchange (use\n[`doca-hardware-safety`](..\u002Fdoca-hardware-safety\u002FSKILL.md)).\n\n## What this skill provides\n\nThis is a **thin loader**. The substantive material lives in two\ncompanion files:\n\n- `CAPABILITIES.md` — the collector deployment contract as a\n  class: the four-surface decomposition (clx collection\n  mechanism vs reader library vs publisher library vs productized\n  DTS), the collection pipeline shape (providers \u002F counters →\n  schema → collector daemon → exporters), the export-backend\n  surface (Prometheus pull, Fluent Bit push, NetFlow, file \u002F\n  IPC) at class level, the version overlay on\n  [`doca-version`](..\u002Fdoca-version\u002FSKILL.md), the error taxonomy\n  (collector won't start → no provider rows → schema mismatch →\n  exporter silent → downstream skew → transport), the\n  observability surface, and the safety policy (gate provider\n  support before commit; collector is read-only against the\n  device; route any mutating step to\n  [`doca-hardware-safety`](..\u002Fdoca-hardware-safety\u002FSKILL.md); do\n  not invent clx names \u002F paths).\n- `TASKS.md` — step-by-step workflows for the deployment verbs:\n  `configure`, `build` (routing stub), `modify`, `run`, `test`,\n  `debug`, plus a `Deferred task verbs` block that routes\n  out-of-scope questions (the two libraries, the productized DTS\n  container, env prep, hardware-state change) to their owners.\n\nThe skill assumes a host or BlueField where DOCA is already\ninstalled and healthy (per\n[`doca-setup ## test`](..\u002Fdoca-setup\u002FTASKS.md#test)) and the\noperator can run the collector and reach its exporter sinks. It\ndoes not cover installing DOCA — that path goes through\n[`doca-setup`](..\u002Fdoca-setup\u002FSKILL.md) — and it does not cover\nthe reader \u002F publisher library APIs or the productized DTS\ncontainer.\n\n## Loading order\n\n1. Read this `SKILL.md` first to confirm the user's question is\n   in scope (operating a CollectX-based collector and its\n   exporters — NOT the reader \u002F publisher library APIs, NOT the\n   productized DTS container).\n2. **For the four-surface decomposition, the collection pipeline\n   shape, the export-backend class surface, the version overlay,\n   the error taxonomy, the observability surface, and the safety\n   policy, see [CAPABILITIES.md](CAPABILITIES.md).**\n3. **For step-by-step workflows — `configure`, `build` (routing\n   stub), `modify`, `run`, `test`, `debug`, and the\n   `Deferred task verbs` block — see [TASKS.md](TASKS.md).**\n\nBoth companion files cross-link to each other,\n[`doca-version`](..\u002Fdoca-version\u002FSKILL.md) for the canonical\nversion-handling rules, and\n[`doca-public-knowledge-map`](..\u002Fdoca-public-knowledge-map\u002FSKILL.md)\nwhenever the right answer is \"read the live config \u002F public docs\"\nrather than collector-specific guidance.\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":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 a deployment \u002F operation overlay over the DOCA telemetry libraries and the CollectX collection mechanism). The hands-on steps DO require a live DOCA install at \u002Fopt\u002Fmellanox\u002Fdoca on a host or BlueField, an operator account that can run the collector and reach its exporter sinks, and the public DOCA Telemetry \u002F DTS guides on docs.nvidia.com for any concrete provider name, schema field, flag, or config path (this skill never invents those).\n",{"type":40,"children":41},"root",[42,51,157,164,211,319,345,351,363,416,428,485,507,513,525,579,641,647,659,758,783,789,877,902,908,924,930,944,950],{"type":43,"tag":44,"props":45,"children":47},"element","h1",{"id":46},"doca-collectx-telemetry-deployment",[48],{"type":49,"value":50},"text","DOCA CollectX telemetry deployment",{"type":43,"tag":52,"props":53,"children":54},"p",{},[55,61,63,69,71,83,85,95,97,102,104,113,115,125,127,132,134,143,145,155],{"type":43,"tag":56,"props":57,"children":58},"strong",{},[59],{"type":49,"value":60},"Where to start:",{"type":49,"value":62}," This skill is the bundle's home for\n",{"type":43,"tag":64,"props":65,"children":66},"em",{},[67],{"type":49,"value":68},"operating a CollectX (clx) based telemetry collector",{"type":49,"value":70}," — the\ncollection framework that gathers provider counters into a\nschema and ships them out through one or more exporters. It is a\ndeployment \u002F operation skill, parallel to\n",{"type":43,"tag":72,"props":73,"children":75},"a",{"href":74},"..\u002Fdoca-bare-metal-deployment\u002FSKILL.md",[76],{"type":43,"tag":77,"props":78,"children":80},"code",{"className":79},[],[81],{"type":49,"value":82},"doca-bare-metal-deployment",{"type":49,"value":84},"\nand ",{"type":43,"tag":72,"props":86,"children":88},{"href":87},"..\u002Fdoca-container-deployment\u002FSKILL.md",[89],{"type":43,"tag":77,"props":90,"children":92},{"className":91},[],[93],{"type":49,"value":94},"doca-container-deployment",{"type":49,"value":96},":\nit owns the ",{"type":43,"tag":64,"props":98,"children":99},{},[100],{"type":49,"value":101},"runtime shape",{"type":49,"value":103}," of a telemetry collector on the\noperator's host or BlueField, not the library APIs the operator's\nown program calls. If the user wants to stand up, wire, or debug\na collector and its exporters, open ",{"type":43,"tag":72,"props":105,"children":107},{"href":106},"TASKS.md",[108],{"type":43,"tag":77,"props":109,"children":111},{"className":110},[],[112],{"type":49,"value":106},{"type":49,"value":114}," and\nstart at ",{"type":43,"tag":72,"props":116,"children":118},{"href":117},"TASKS.md#configure",[119],{"type":43,"tag":77,"props":120,"children":122},{"className":121},[],[123],{"type":49,"value":124},"## configure",{"type":49,"value":126},". If the question is\n",{"type":43,"tag":64,"props":128,"children":129},{},[130],{"type":49,"value":131},"what surfaces does the collector even have and where is the\nscope boundary",{"type":49,"value":133},", start at ",{"type":43,"tag":72,"props":135,"children":137},{"href":136},"CAPABILITIES.md",[138],{"type":43,"tag":77,"props":139,"children":141},{"className":140},[],[142],{"type":49,"value":136},{"type":49,"value":144},".\nIf the user has not installed DOCA yet, route to\n",{"type":43,"tag":72,"props":146,"children":148},{"href":147},"..\u002Fdoca-setup\u002FSKILL.md",[149],{"type":43,"tag":77,"props":150,"children":152},{"className":151},[],[153],{"type":49,"value":154},"doca-setup",{"type":49,"value":156}," first.",{"type":43,"tag":158,"props":159,"children":161},"h2",{"id":160},"the-scope-boundary-read-this-before-anything-else",[162],{"type":49,"value":163},"The scope boundary (read this before anything else)",{"type":43,"tag":52,"props":165,"children":166},{},[167,169,174,176,181,183,188,190,195,197,209],{"type":49,"value":168},"CollectX (clx) is NVIDIA's telemetry ",{"type":43,"tag":56,"props":170,"children":171},{},[172],{"type":49,"value":173},"collection",{"type":49,"value":175}," framework. It\nunderpins the ",{"type":43,"tag":56,"props":177,"children":178},{},[179],{"type":49,"value":180},"DOCA Telemetry Service (DTS)",{"type":49,"value":182}," — and DTS\n",{"type":43,"tag":64,"props":184,"children":185},{},[186],{"type":49,"value":187},"as-deployed",{"type":49,"value":189}," (the productized, NGC-shipped \u002F kubelet-started\nservice container) is ",{"type":43,"tag":56,"props":191,"children":192},{},[193],{"type":49,"value":194},"out of scope",{"type":49,"value":196}," for this bundle per\n",{"type":43,"tag":72,"props":198,"children":200},{"href":199},"..\u002F..\u002FAGENTS.md#non-goals-questions-the-agent-should-recognize-and-refuse-politely",[201,207],{"type":43,"tag":77,"props":202,"children":204},{"className":203},[],[205],{"type":49,"value":206},"AGENTS.md",{"type":49,"value":208}," Non-goal #7",{"type":49,"value":210},".\nThis skill therefore draws a hard line and the agent MUST state\nit up front:",{"type":43,"tag":212,"props":213,"children":214},"ul",{},[215,240,288],{"type":43,"tag":216,"props":217,"children":218},"li",{},[219,224,226,231,233,238],{"type":43,"tag":56,"props":220,"children":221},{},[222],{"type":49,"value":223},"In scope here:",{"type":49,"value":225}," the CollectX ",{"type":43,"tag":64,"props":227,"children":228},{},[229],{"type":49,"value":230},"collection mechanism",{"type":49,"value":232}," as a\nclass (providers \u002F counters → schema → collector daemon →\nexporters), and the operator deploying \u002F running \u002F debugging a\ncollector that they own, plus the operator's own usage of the\ntwo in-bundle telemetry ",{"type":43,"tag":56,"props":234,"children":235},{},[236],{"type":49,"value":237},"libraries",{"type":49,"value":239}," when those feed or\nconsume the collector.",{"type":43,"tag":216,"props":241,"children":242},{},[243,248,250,255,257,267,269,274,276,286],{"type":43,"tag":56,"props":244,"children":245},{},[246],{"type":49,"value":247},"Routed to the DOCA telemetry libraries:",{"type":49,"value":249}," the\nhardware-counter ",{"type":43,"tag":56,"props":251,"children":252},{},[253],{"type":49,"value":254},"reader",{"type":49,"value":256}," API is owned by\n",{"type":43,"tag":72,"props":258,"children":260},{"href":259},"..\u002Flibs\u002Fdoca-telemetry\u002FSKILL.md",[261],{"type":43,"tag":77,"props":262,"children":264},{"className":263},[],[265],{"type":49,"value":266},"doca-telemetry",{"type":49,"value":268},"; the\napplication-side ",{"type":43,"tag":56,"props":270,"children":271},{},[272],{"type":49,"value":273},"publisher",{"type":49,"value":275}," API (emit counters \u002F events from\na DOCA program) is owned by\n",{"type":43,"tag":72,"props":277,"children":279},{"href":278},"..\u002Flibs\u002Fdoca-telemetry-exporter\u002FSKILL.md",[280],{"type":43,"tag":77,"props":281,"children":283},{"className":282},[],[284],{"type":49,"value":285},"doca-telemetry-exporter",{"type":49,"value":287},".\nThis skill does not re-document either API surface.",{"type":43,"tag":216,"props":289,"children":290},{},[291,296,298,303,305,317],{"type":43,"tag":56,"props":292,"children":293},{},[294],{"type":49,"value":295},"Routed to public docs (Non-goal #7):",{"type":49,"value":297}," the productized\n",{"type":43,"tag":56,"props":299,"children":300},{},[301],{"type":49,"value":302},"DTS container",{"type":49,"value":304}," — its packaged config schema, its built-in\nprovider set, its kubelet manifest, its NGC image — is\nexternally productized. Route every \"operate the DTS service\"\nquestion to the\n",{"type":43,"tag":72,"props":306,"children":308},{"href":307},"..\u002Fdoca-public-knowledge-map\u002FSKILL.md#externally-productized-doca-software--not-in-this-bundle-but-here-is-where-to-route",[309,315],{"type":43,"tag":77,"props":310,"children":312},{"className":311},[],[313],{"type":49,"value":314},"doca-public-knowledge-map",{"type":49,"value":316}," externally-productized routing row",{"type":49,"value":318},"\nand the public DTS guide it points at. The agent must NOT\nsynthesize DTS config file names, provider knob names, or\npaths from memory.",{"type":43,"tag":52,"props":320,"children":321},{},[322,324,329,331,336,338,343],{"type":49,"value":323},"The load-bearing first-touch failure this skill exists to\nprevent is ",{"type":43,"tag":56,"props":325,"children":326},{},[327],{"type":49,"value":328},"collapsing these four surfaces into \"DOCA\ntelemetry\"",{"type":49,"value":330},": the clx collection mechanism, the ",{"type":43,"tag":77,"props":332,"children":334},{"className":333},[],[335],{"type":49,"value":266},{"type":49,"value":337},"\nreader library, the ",{"type":43,"tag":77,"props":339,"children":341},{"className":340},[],[342],{"type":49,"value":285},{"type":49,"value":344}," publisher library,\nand the productized DTS container are four different things with\nfour different owners. The agent surfaces the decomposition\nBEFORE any config-level guidance.",{"type":43,"tag":158,"props":346,"children":348},{"id":347},"audience",[349],{"type":49,"value":350},"Audience",{"type":43,"tag":52,"props":352,"children":353},{},[354,356,361],{"type":49,"value":355},"This skill serves ",{"type":43,"tag":56,"props":357,"children":358},{},[359],{"type":49,"value":360},"external operators standing up or running a\nCollectX-based telemetry collector",{"type":49,"value":362}," on a host or BlueField they\nadminister — people who already have:",{"type":43,"tag":212,"props":364,"children":365},{},[366,383,403],{"type":43,"tag":216,"props":367,"children":368},{},[369,371,381],{"type":49,"value":370},"a DOCA install on the side they are collecting from (host x86\nor BlueField Arm), verified per\n",{"type":43,"tag":72,"props":372,"children":374},{"href":373},"..\u002Fdoca-setup\u002FTASKS.md#test",[375],{"type":43,"tag":77,"props":376,"children":378},{"className":377},[],[379],{"type":49,"value":380},"doca-setup ## test",{"type":49,"value":382},",",{"type":43,"tag":216,"props":384,"children":385},{},[386,388,393,395,401],{"type":49,"value":387},"a goal of ",{"type":43,"tag":64,"props":389,"children":390},{},[391],{"type":49,"value":392},"getting counters off the box",{"type":49,"value":394}," through a collector +\nexporter, not of writing the reader \u002F publisher library code\n(that is the two ",{"type":43,"tag":77,"props":396,"children":398},{"className":397},[],[399],{"type":49,"value":400},"libs\u002F",{"type":49,"value":402}," skills above), and",{"type":43,"tag":216,"props":404,"children":405},{},[406,408,414],{"type":49,"value":407},"access to the public DOCA Telemetry and DTS guides on\n",{"type":43,"tag":77,"props":409,"children":411},{"className":410},[],[412],{"type":49,"value":413},"docs.nvidia.com",{"type":49,"value":415}," as the authoritative source for any concrete\nprovider name, schema field, flag, or config path.",{"type":43,"tag":52,"props":417,"children":418},{},[419,421,426],{"type":49,"value":420},"It is ",{"type":43,"tag":56,"props":422,"children":423},{},[424],{"type":49,"value":425},"not",{"type":49,"value":427}," for:",{"type":43,"tag":212,"props":429,"children":430},{},[431,456,468],{"type":43,"tag":216,"props":432,"children":433},{},[434,436,444,446,454],{"type":49,"value":435},"developers writing the hardware-counter reader API (route to\n",{"type":43,"tag":72,"props":437,"children":438},{"href":259},[439],{"type":43,"tag":77,"props":440,"children":442},{"className":441},[],[443],{"type":49,"value":266},{"type":49,"value":445},") or the\npublisher API (route to\n",{"type":43,"tag":72,"props":447,"children":448},{"href":278},[449],{"type":43,"tag":77,"props":450,"children":452},{"className":451},[],[453],{"type":49,"value":285},{"type":49,"value":455},"),",{"type":43,"tag":216,"props":457,"children":458},{},[459,461,466],{"type":49,"value":460},"operators deploying \u002F configuring the ",{"type":43,"tag":56,"props":462,"children":463},{},[464],{"type":49,"value":465},"productized DTS\ncontainer",{"type":49,"value":467}," as a turnkey service — that is externally\nproductized (Non-goal #7); route to the public DTS guide,",{"type":43,"tag":216,"props":469,"children":470},{},[471,473,483],{"type":49,"value":472},"fresh-no-install users — those belong on\n",{"type":43,"tag":72,"props":474,"children":476},{"href":475},"..\u002Fdoca-setup\u002FTASKS.md#no-install",[477],{"type":43,"tag":77,"props":478,"children":480},{"className":479},[],[481],{"type":49,"value":482},"doca-setup ## no-install",{"type":49,"value":484},".",{"type":43,"tag":52,"props":486,"children":487},{},[488,490,495,497,506],{"type":49,"value":489},"The skill teaches the agent the ",{"type":43,"tag":64,"props":491,"children":492},{},[493],{"type":49,"value":494},"procedure and the scope\nboundary",{"type":49,"value":496},"; it does not invent clx symbol names, provider names,\nschema field names, exporter flag names, or config paths from\nmemory — those come from the live install and the public docs via\n",{"type":43,"tag":72,"props":498,"children":500},{"href":499},"..\u002Fdoca-public-knowledge-map\u002FSKILL.md",[501],{"type":43,"tag":77,"props":502,"children":504},{"className":503},[],[505],{"type":49,"value":314},{"type":49,"value":484},{"type":43,"tag":158,"props":508,"children":510},{"id":509},"when-to-load-this-skill",[511],{"type":49,"value":512},"When to load this skill",{"type":43,"tag":52,"props":514,"children":515},{},[516,518,523],{"type":49,"value":517},"Load this skill when the user is doing hands-on ",{"type":43,"tag":56,"props":519,"children":520},{},[521],{"type":49,"value":522},"deployment or\noperation of a CollectX-based telemetry collector",{"type":49,"value":524}," and the\nquestion is about the collector runtime shape, not a library API.\nConcretely:",{"type":43,"tag":212,"props":526,"children":527},{},[528,533,550,555,560],{"type":43,"tag":216,"props":529,"children":530},{},[531],{"type":49,"value":532},"Standing up a collector that gathers provider counters into a\nschema and ships them out — and deciding which export backend\n(Prometheus pull, Fluent Bit push, NetFlow, file \u002F IPC) fits\nthe downstream consumer.",{"type":43,"tag":216,"props":534,"children":535},{},[536,538,548],{"type":49,"value":537},"Wiring a provider \u002F counter family into the collector and\nconfirming the device actually exposes it before the config\ncommits (the gate-before-commit rule, shared with\n",{"type":43,"tag":72,"props":539,"children":541},{"href":540},"..\u002Ftools\u002Fdoca-telemetry-utils\u002FSKILL.md",[542],{"type":43,"tag":77,"props":543,"children":545},{"className":544},[],[546],{"type":49,"value":547},"doca-telemetry-utils",{"type":49,"value":549},").",{"type":43,"tag":216,"props":551,"children":552},{},[553],{"type":49,"value":554},"Turning on \u002F shaping an exporter so the metrics actually leave\nthe box, and confirming the downstream consumer receives them\nend-to-end (not just \"the daemon is running\").",{"type":43,"tag":216,"props":556,"children":557},{},[558],{"type":49,"value":559},"Diagnosing a collector that starts but produces no schema rows,\nor ships nothing downstream, or whose exporter endpoint is\nsilent — walking the layered ladder rather than guessing.",{"type":43,"tag":216,"props":561,"children":562},{},[563,565,570,572,577],{"type":49,"value":564},"Recognising when the user is actually asking about the\nproductized DTS container (route to public docs, Non-goal #7),\nthe reader library (route to ",{"type":43,"tag":77,"props":566,"children":568},{"className":567},[],[569],{"type":49,"value":266},{"type":49,"value":571},"), or the\npublisher library (route to ",{"type":43,"tag":77,"props":573,"children":575},{"className":574},[],[576],{"type":49,"value":285},{"type":49,"value":578},") instead\nof the collection mechanism this skill owns.",{"type":43,"tag":52,"props":580,"children":581},{},[582,584,588,590,598,600,608,610,618,620,628,630,640],{"type":49,"value":583},"Do ",{"type":43,"tag":56,"props":585,"children":586},{},[587],{"type":49,"value":425},{"type":49,"value":589}," load this skill for: the hardware-counter reader API\n(use ",{"type":43,"tag":72,"props":591,"children":592},{"href":259},[593],{"type":43,"tag":77,"props":594,"children":596},{"className":595},[],[597],{"type":49,"value":266},{"type":49,"value":599},"); the\npublisher API (use\n",{"type":43,"tag":72,"props":601,"children":602},{"href":278},[603],{"type":43,"tag":77,"props":604,"children":606},{"className":605},[],[607],{"type":49,"value":285},{"type":49,"value":609},");\noperating the productized DTS container (route via\n",{"type":43,"tag":72,"props":611,"children":612},{"href":499},[613],{"type":43,"tag":77,"props":614,"children":616},{"className":615},[],[617],{"type":49,"value":314},{"type":49,"value":619},"\nNon-goal #7); installing DOCA or preparing the env (use\n",{"type":43,"tag":72,"props":621,"children":622},{"href":147},[623],{"type":43,"tag":77,"props":624,"children":626},{"className":625},[],[627],{"type":49,"value":154},{"type":49,"value":629},"); or any hardware-state\nchange (use\n",{"type":43,"tag":72,"props":631,"children":633},{"href":632},"..\u002Fdoca-hardware-safety\u002FSKILL.md",[634],{"type":43,"tag":77,"props":635,"children":637},{"className":636},[],[638],{"type":49,"value":639},"doca-hardware-safety",{"type":49,"value":549},{"type":43,"tag":158,"props":642,"children":644},{"id":643},"what-this-skill-provides",[645],{"type":49,"value":646},"What this skill provides",{"type":43,"tag":52,"props":648,"children":649},{},[650,652,657],{"type":49,"value":651},"This is a ",{"type":43,"tag":56,"props":653,"children":654},{},[655],{"type":49,"value":656},"thin loader",{"type":49,"value":658},". The substantive material lives in two\ncompanion files:",{"type":43,"tag":212,"props":660,"children":661},{},[662,694],{"type":43,"tag":216,"props":663,"children":664},{},[665,670,672,682,684,692],{"type":43,"tag":77,"props":666,"children":668},{"className":667},[],[669],{"type":49,"value":136},{"type":49,"value":671}," — the collector deployment contract as a\nclass: the four-surface decomposition (clx collection\nmechanism vs reader library vs publisher library vs productized\nDTS), the collection pipeline shape (providers \u002F counters →\nschema → collector daemon → exporters), the export-backend\nsurface (Prometheus pull, Fluent Bit push, NetFlow, file \u002F\nIPC) at class level, the version overlay on\n",{"type":43,"tag":72,"props":673,"children":675},{"href":674},"..\u002Fdoca-version\u002FSKILL.md",[676],{"type":43,"tag":77,"props":677,"children":679},{"className":678},[],[680],{"type":49,"value":681},"doca-version",{"type":49,"value":683},", the error taxonomy\n(collector won't start → no provider rows → schema mismatch →\nexporter silent → downstream skew → transport), the\nobservability surface, and the safety policy (gate provider\nsupport before commit; collector is read-only against the\ndevice; route any mutating step to\n",{"type":43,"tag":72,"props":685,"children":686},{"href":632},[687],{"type":43,"tag":77,"props":688,"children":690},{"className":689},[],[691],{"type":49,"value":639},{"type":49,"value":693},"; do\nnot invent clx names \u002F paths).",{"type":43,"tag":216,"props":695,"children":696},{},[697,702,704,710,712,718,720,726,727,733,734,740,742,748,750,756],{"type":43,"tag":77,"props":698,"children":700},{"className":699},[],[701],{"type":49,"value":106},{"type":49,"value":703}," — step-by-step workflows for the deployment verbs:\n",{"type":43,"tag":77,"props":705,"children":707},{"className":706},[],[708],{"type":49,"value":709},"configure",{"type":49,"value":711},", ",{"type":43,"tag":77,"props":713,"children":715},{"className":714},[],[716],{"type":49,"value":717},"build",{"type":49,"value":719}," (routing stub), ",{"type":43,"tag":77,"props":721,"children":723},{"className":722},[],[724],{"type":49,"value":725},"modify",{"type":49,"value":711},{"type":43,"tag":77,"props":728,"children":730},{"className":729},[],[731],{"type":49,"value":732},"run",{"type":49,"value":711},{"type":43,"tag":77,"props":735,"children":737},{"className":736},[],[738],{"type":49,"value":739},"test",{"type":49,"value":741},",\n",{"type":43,"tag":77,"props":743,"children":745},{"className":744},[],[746],{"type":49,"value":747},"debug",{"type":49,"value":749},", plus a ",{"type":43,"tag":77,"props":751,"children":753},{"className":752},[],[754],{"type":49,"value":755},"Deferred task verbs",{"type":49,"value":757}," block that routes\nout-of-scope questions (the two libraries, the productized DTS\ncontainer, env prep, hardware-state change) to their owners.",{"type":43,"tag":52,"props":759,"children":760},{},[761,763,771,773,781],{"type":49,"value":762},"The skill assumes a host or BlueField where DOCA is already\ninstalled and healthy (per\n",{"type":43,"tag":72,"props":764,"children":765},{"href":373},[766],{"type":43,"tag":77,"props":767,"children":769},{"className":768},[],[770],{"type":49,"value":380},{"type":49,"value":772},") and the\noperator can run the collector and reach its exporter sinks. It\ndoes not cover installing DOCA — that path goes through\n",{"type":43,"tag":72,"props":774,"children":775},{"href":147},[776],{"type":43,"tag":77,"props":777,"children":779},{"className":778},[],[780],{"type":49,"value":154},{"type":49,"value":782}," — and it does not cover\nthe reader \u002F publisher library APIs or the productized DTS\ncontainer.",{"type":43,"tag":158,"props":784,"children":786},{"id":785},"loading-order",[787],{"type":49,"value":788},"Loading order",{"type":43,"tag":790,"props":791,"children":792},"ol",{},[793,806,819],{"type":43,"tag":216,"props":794,"children":795},{},[796,798,804],{"type":49,"value":797},"Read this ",{"type":43,"tag":77,"props":799,"children":801},{"className":800},[],[802],{"type":49,"value":803},"SKILL.md",{"type":49,"value":805}," first to confirm the user's question is\nin scope (operating a CollectX-based collector and its\nexporters — NOT the reader \u002F publisher library APIs, NOT the\nproductized DTS container).",{"type":43,"tag":216,"props":807,"children":808},{},[809],{"type":43,"tag":56,"props":810,"children":811},{},[812,814,818],{"type":49,"value":813},"For the four-surface decomposition, the collection pipeline\nshape, the export-backend class surface, the version overlay,\nthe error taxonomy, the observability surface, and the safety\npolicy, see ",{"type":43,"tag":72,"props":815,"children":816},{"href":136},[817],{"type":49,"value":136},{"type":49,"value":484},{"type":43,"tag":216,"props":820,"children":821},{},[822],{"type":43,"tag":56,"props":823,"children":824},{},[825,827,832,833,838,840,845,846,851,852,857,858,863,865,870,872,876],{"type":49,"value":826},"For step-by-step workflows — ",{"type":43,"tag":77,"props":828,"children":830},{"className":829},[],[831],{"type":49,"value":709},{"type":49,"value":711},{"type":43,"tag":77,"props":834,"children":836},{"className":835},[],[837],{"type":49,"value":717},{"type":49,"value":839}," (routing\nstub), ",{"type":43,"tag":77,"props":841,"children":843},{"className":842},[],[844],{"type":49,"value":725},{"type":49,"value":711},{"type":43,"tag":77,"props":847,"children":849},{"className":848},[],[850],{"type":49,"value":732},{"type":49,"value":711},{"type":43,"tag":77,"props":853,"children":855},{"className":854},[],[856],{"type":49,"value":739},{"type":49,"value":711},{"type":43,"tag":77,"props":859,"children":861},{"className":860},[],[862],{"type":49,"value":747},{"type":49,"value":864},", and the\n",{"type":43,"tag":77,"props":866,"children":868},{"className":867},[],[869],{"type":49,"value":755},{"type":49,"value":871}," block — see ",{"type":43,"tag":72,"props":873,"children":874},{"href":106},[875],{"type":49,"value":106},{"type":49,"value":484},{"type":43,"tag":52,"props":878,"children":879},{},[880,882,890,892,900],{"type":49,"value":881},"Both companion files cross-link to each other,\n",{"type":43,"tag":72,"props":883,"children":884},{"href":674},[885],{"type":43,"tag":77,"props":886,"children":888},{"className":887},[],[889],{"type":49,"value":681},{"type":49,"value":891}," for the canonical\nversion-handling rules, and\n",{"type":43,"tag":72,"props":893,"children":894},{"href":499},[895],{"type":43,"tag":77,"props":896,"children":898},{"className":897},[],[899],{"type":49,"value":314},{"type":49,"value":901},"\nwhenever the right answer is \"read the live config \u002F public docs\"\nrather than collector-specific guidance.",{"type":43,"tag":158,"props":903,"children":905},{"id":904},"example-questions-this-skill-answers-well",[906],{"type":49,"value":907},"Example questions this skill answers well",{"type":43,"tag":52,"props":909,"children":910},{},[911,913,923],{"type":49,"value":912},"See ",{"type":43,"tag":72,"props":914,"children":916},{"href":915},"references\u002Fdetails.md#example-questions-this-skill-answers-well",[917],{"type":43,"tag":77,"props":918,"children":920},{"className":919},[],[921],{"type":49,"value":922},"references\u002Fdetails.md",{"type":49,"value":484},{"type":43,"tag":158,"props":925,"children":927},{"id":926},"what-this-skill-deliberately-does-not-ship",[928],{"type":49,"value":929},"What this skill deliberately does not ship",{"type":43,"tag":52,"props":931,"children":932},{},[933,934,943],{"type":49,"value":912},{"type":43,"tag":72,"props":935,"children":937},{"href":936},"references\u002Fdetails.md#what-this-skill-deliberately-does-not-ship",[938],{"type":43,"tag":77,"props":939,"children":941},{"className":940},[],[942],{"type":49,"value":922},{"type":49,"value":484},{"type":43,"tag":158,"props":945,"children":947},{"id":946},"related-skills",[948],{"type":49,"value":949},"Related skills",{"type":43,"tag":52,"props":951,"children":952},{},[953,954,963],{"type":49,"value":912},{"type":43,"tag":72,"props":955,"children":957},{"href":956},"references\u002Fdetails.md#related-skills",[958],{"type":43,"tag":77,"props":959,"children":961},{"className":960},[],[962],{"type":49,"value":922},{"type":49,"value":484},{"items":965,"total":1067},[966,983,995,1009,1021,1038,1053],{"slug":967,"name":967,"fn":968,"description":969,"org":970,"tags":971,"stars":23,"repoUrl":24,"updatedAt":982},"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},[972,975,978,979],{"name":973,"slug":974,"type":15},"Data Analysis","data-analysis",{"name":976,"slug":977,"type":15},"Data Engineering","data-engineering",{"name":9,"slug":8,"type":15},{"name":980,"slug":981,"type":15},"Performance","performance","2026-07-14T05:28:43.176466",{"slug":984,"name":984,"fn":985,"description":986,"org":987,"tags":988,"stars":23,"repoUrl":24,"updatedAt":994},"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},[989,990,993],{"name":20,"slug":21,"type":15},{"name":991,"slug":992,"type":15},"Infrastructure","infrastructure",{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":996,"name":996,"fn":997,"description":998,"org":999,"tags":1000,"stars":23,"repoUrl":24,"updatedAt":1008},"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},[1001,1004,1005],{"name":1002,"slug":1003,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":1006,"slug":1007,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":1010,"name":1010,"fn":1011,"description":1012,"org":1013,"tags":1014,"stars":23,"repoUrl":24,"updatedAt":1020},"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},[1015,1016,1017],{"name":973,"slug":974,"type":15},{"name":9,"slug":8,"type":15},{"name":1018,"slug":1019,"type":15},"Testing","testing","2026-07-17T05:29:03.913266",{"slug":1022,"name":1022,"fn":1023,"description":1024,"org":1025,"tags":1026,"stars":23,"repoUrl":24,"updatedAt":1037},"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},[1027,1030,1033,1034],{"name":1028,"slug":1029,"type":15},"Automation","automation",{"name":1031,"slug":1032,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":1035,"slug":1036,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":1039,"name":1039,"fn":1040,"description":1041,"org":1042,"tags":1043,"stars":23,"repoUrl":24,"updatedAt":1052},"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},[1044,1045,1048,1049],{"name":20,"slug":21,"type":15},{"name":1046,"slug":1047,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":1050,"slug":1051,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":1054,"name":1054,"fn":1055,"description":1056,"org":1057,"tags":1058,"stars":23,"repoUrl":24,"updatedAt":1066},"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},[1059,1060,1063],{"name":9,"slug":8,"type":15},{"name":1061,"slug":1062,"type":15},"Quantum Computing","quantum-computing",{"name":1064,"slug":1065,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305,{"items":1069,"total":1220},[1070,1088,1104,1115,1127,1141,1154,1168,1179,1188,1202,1211],{"slug":1071,"name":1071,"fn":1072,"description":1073,"org":1074,"tags":1075,"stars":1085,"repoUrl":1086,"updatedAt":1087},"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},[1076,1079,1082],{"name":1077,"slug":1078,"type":15},"Documentation","documentation",{"name":1080,"slug":1081,"type":15},"MCP","mcp",{"name":1083,"slug":1084,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":1089,"name":1089,"fn":1090,"description":1091,"org":1092,"tags":1093,"stars":1101,"repoUrl":1102,"updatedAt":1103},"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},[1094,1097,1098],{"name":1095,"slug":1096,"type":15},"Containers","containers",{"name":20,"slug":21,"type":15},{"name":1099,"slug":1100,"type":15},"Python","python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":1105,"name":1105,"fn":1106,"description":1107,"org":1108,"tags":1109,"stars":1101,"repoUrl":1102,"updatedAt":1114},"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},[1110,1113],{"name":1111,"slug":1112,"type":15},"CI\u002FCD","ci-cd",{"name":20,"slug":21,"type":15},"2026-07-14T05:25:59.97109",{"slug":1116,"name":1116,"fn":1117,"description":1118,"org":1119,"tags":1120,"stars":1101,"repoUrl":1102,"updatedAt":1126},"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},[1121,1122,1123],{"name":1111,"slug":1112,"type":15},{"name":20,"slug":21,"type":15},{"name":1124,"slug":1125,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":1128,"name":1128,"fn":1129,"description":1130,"org":1131,"tags":1132,"stars":1101,"repoUrl":1102,"updatedAt":1140},"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},[1133,1136,1137],{"name":1134,"slug":1135,"type":15},"Debugging","debugging",{"name":1124,"slug":1125,"type":15},{"name":1138,"slug":1139,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":1142,"name":1142,"fn":1143,"description":1144,"org":1145,"tags":1146,"stars":1101,"repoUrl":1102,"updatedAt":1153},"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},[1147,1150],{"name":1148,"slug":1149,"type":15},"Best Practices","best-practices",{"name":1151,"slug":1152,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":1155,"name":1155,"fn":1156,"description":1157,"org":1158,"tags":1159,"stars":1101,"repoUrl":1102,"updatedAt":1167},"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},[1160,1163,1166],{"name":1161,"slug":1162,"type":15},"Machine Learning","machine-learning",{"name":1164,"slug":1165,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":1169,"name":1169,"fn":1170,"description":1171,"org":1172,"tags":1173,"stars":1101,"repoUrl":1102,"updatedAt":1178},"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},[1174,1177],{"name":1175,"slug":1176,"type":15},"QA","qa",{"name":1018,"slug":1019,"type":15},"2026-07-14T05:25:53.673039",{"slug":1180,"name":1180,"fn":1181,"description":1182,"org":1183,"tags":1184,"stars":1101,"repoUrl":1102,"updatedAt":1187},"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},[1185,1186],{"name":20,"slug":21,"type":15},{"name":991,"slug":992,"type":15},"2026-07-14T05:25:49.362534",{"slug":1189,"name":1189,"fn":1190,"description":1191,"org":1192,"tags":1193,"stars":1101,"repoUrl":1102,"updatedAt":1201},"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},[1194,1197,1198],{"name":1195,"slug":1196,"type":15},"Code Review","code-review",{"name":1124,"slug":1125,"type":15},{"name":1199,"slug":1200,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":1203,"name":1203,"fn":1204,"description":1205,"org":1206,"tags":1207,"stars":1101,"repoUrl":1102,"updatedAt":1210},"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},[1208,1209],{"name":1175,"slug":1176,"type":15},{"name":1018,"slug":1019,"type":15},"2026-07-14T05:25:54.928983",{"slug":1212,"name":1212,"fn":1213,"description":1214,"org":1215,"tags":1216,"stars":1101,"repoUrl":1102,"updatedAt":1219},"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},[1217,1218],{"name":1028,"slug":1029,"type":15},{"name":1111,"slug":1112,"type":15},"2026-07-30T05:29:03.275638",496]