[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-doca-rdmi":3,"mdc-jlfedj-key":31,"related-repo-nvidia-doca-rdmi":1577,"related-org-nvidia-doca-rdmi":1683},{"slug":4,"name":4,"fn":5,"description":6,"org":7,"tags":11,"stars":20,"repoUrl":21,"updatedAt":22,"license":23,"forks":24,"topics":25,"repo":26,"sourceUrl":29,"mdContent":30},"doca-rdmi","program RDMA Initiator flows","Use this skill when the user is doing hands-on DOCA RDMI (RDMA Initiator) programming — picking doca-rdmi vs doca-rdma for an accelerator-initiated one-sided RDMA flow, standing up a doca_rdmi_connection or doca_rdmi_poster, attaching a doca_dpa_completion or doca_verbs_cq before doca_ctx_start(), retrieving the DPA-side handle for a DPA kernel, auditing whether a doca_rdmi_* symbol is EXPERIMENTAL on this DOCA, or debugging DOCA_ERROR_* returns from RDMI calls. Trigger even when the user does not say \"DOCA RDMI\" or \"initiator\" — implicit phrasings include \"my DPA kernel needs to post RDMA writes to a remote responder\", \"DPA kernel sees no completions\", \"function not found on doca_rdmi_* at link time\", \"DOCA_ERROR_BAD_STATE from completion attach\", or \"the DPA posted but the work request never arrived\". Refuse and route elsewhere for two-sided or host-CPU RDMA, the DPA programming model, GPU-side RDMA initiation, or general RDMA\u002FIB\u002FRoCE concepts — 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",[12,16,17],{"name":13,"slug":14,"type":15},"Networking","networking","tag",{"name":9,"slug":8,"type":15},{"name":18,"slug":19,"type":15},"Engineering","engineering",2473,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills","2026-07-30T05:28:22.437166","Apache-2.0",281,[],{"repoUrl":21,"stars":20,"forks":24,"topics":27,"description":28},[],"AI agent skills published by NVIDIA","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills\u002Ftree\u002FHEAD\u002Fskills\u002Fdoca-rdmi","---\nlicense: Apache-2.0\nname: doca-rdmi\ndescription: >\n  Use this skill when the user is doing hands-on DOCA RDMI (RDMA\n  Initiator) programming — picking doca-rdmi vs doca-rdma for an\n  accelerator-initiated one-sided RDMA flow, standing up a\n  doca_rdmi_connection or doca_rdmi_poster, attaching a\n  doca_dpa_completion or doca_verbs_cq before doca_ctx_start(),\n  retrieving the DPA-side handle for a DPA kernel, auditing whether\n  a doca_rdmi_* symbol is EXPERIMENTAL on this DOCA, or debugging\n  DOCA_ERROR_* returns from RDMI calls. Trigger even when the user\n  does not say \"DOCA RDMI\" or \"initiator\" — implicit phrasings\n  include \"my DPA kernel needs to post RDMA writes to a remote\n  responder\", \"DPA kernel sees no completions\", \"function not found\n  on doca_rdmi_* at link time\", \"DOCA_ERROR_BAD_STATE from\n  completion attach\", or \"the DPA posted but the work request never\n  arrived\". Refuse and route elsewhere for two-sided or host-CPU\n  RDMA, the DPA programming model, GPU-side RDMA initiation, or\n  general RDMA\u002FIB\u002FRoCE concepts — those belong to other skills.\nmetadata:\n  kind: library\ncompatibility: >\n  Requires DOCA SDK installed at \u002Fopt\u002Fmellanox\u002Fdoca on Linux\n  (Ubuntu 22.04\u002F24.04 or RHEL\u002FSLES) with a BlueField DPU or\n  ConnectX NIC attached, plus the DOCA DPA toolchain and a\n  DPA-capable BlueField when the data path runs on the DPA. Reads\n  the user's local install via `pkg-config doca-rdmi` (alongside\n  doca-common, doca-verbs, doca-dpa) and inspects\n  \u002Fopt\u002Fmellanox\u002Fdoca\u002F{lib,include,samples}.\n---\n\n# DOCA RDMA Initiator\n\n**Where to start:** This skill assumes DOCA is already installed and\nthe user is doing **hands-on RDMI work** on a host or BlueField with\nthe DOCA package set that ships the `doca-rdmi` library. Open\n[`TASKS.md`](TASKS.md) if the user wants to *do* something\n(install \u002F configure \u002F build \u002F modify \u002F run \u002F test \u002F debug \u002F use);\nopen [`CAPABILITIES.md`](CAPABILITIES.md) when the question is\n*what can RDMI express on this version* — the object model, the\nDPA-side handle types, the relationship to `doca-rdma`, the\nEXPERIMENTAL-tag policy, and the safety overlay.\n\n**End-to-end \"walk me through doca-rdmi\" questions are answerable\nentirely from this skill.** Go straight to\n[`TASKS.md ## end-to-end (quickref)`](TASKS.md#end-to-end-quickref),\nwhich carries the self-contained install-check → device\u002Fcap\ndiscovery → sample → `pkg-config` build → run → debug walkthrough\nwith the exact commands. You do **not** need to open `doca-setup` or\n`doca-programming-guide` to answer an RDMI build\u002Frun\u002Fdebug question.\nRoute to [`doca-setup`](..\u002F..\u002Fdoca-setup\u002FSKILL.md) when the required\nDOCA prerequisites are absent, partial, or version-mismatched.\n\n## Example questions this skill answers well\n\nThe CLASSES of RDMI questions this skill is built to answer, each\nwith one worked example. The agent should treat the *class* as the\nload-bearing piece — the worked example is a single instance.\n\n- **\"Should I use `doca-rdmi` or `doca-rdma` for this case?\"** —\n  worked example: *\"I have a DPA kernel that needs to fire 1 MB\n  RDMA writes at a remote responder; which library?\"*. Answered by\n  the *initiator-side vs general-purpose* selection rule in\n  [`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes)\n  surface-selection table + the routing back to\n  [`doca-rdma`](..\u002Fdoca-rdma\u002FSKILL.md) when the use case is\n  two-sided or host-CPU initiated.\n- **\"How do I bring up an RDMI connection on the DPA datapath?\"** —\n  worked example: *\"create a `doca_rdmi_connection`, attach a DPA\n  completion context, hand the DPA-side handle to my kernel\"*.\n  Answered by the connection-object lifecycle in\n  [`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes)\n  + the configure walk in\n  [`TASKS.md ## configure`](TASKS.md#configure).\n- **\"How do connection and poster relate — when do I need both?\"**\n  — worked example: *\"my application receives work requests AND\n  posts RDMA writes; do I need a `doca_rdmi_connection` plus a\n  `doca_rdmi_poster`, or one of them?\"*. Answered by the\n  two-object model in\n  [`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes)\n  + the modify-from-sample slot table in\n  [`TASKS.md ## modify`](TASKS.md#modify).\n- **\"How do I drive completions on the DPA side?\"** — worked\n  example: *\"hook the connection to a `doca_dpa_completion` so my\n  kernel polls completions directly\"*. Answered by the DPA-side\n  completion-attach pattern in\n  [`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes)\n  + the run-side wiring in\n  [`TASKS.md ## run`](TASKS.md#run), cross-linked into\n  [`doca-dpa`](..\u002Fdoca-dpa\u002FSKILL.md) for the DPA programming\n  surface itself.\n- **\"Is the symbol I want available — and is it stable enough to\n  ship?\"** — worked example: *\"is `doca_rdmi_poster_post` GA on\n  my installed DOCA, or still EXPERIMENTAL?\"*. Answered by the\n  EXPERIMENTAL-tag policy in\n  [`CAPABILITIES.md ## Version compatibility`](CAPABILITIES.md#version-compatibility)\n  + the version-discovery rule\n  (`pkg-config --modversion doca-rdmi`) pinned in\n  [`TASKS.md ## configure`](TASKS.md#configure).\n- **\"What does this `DOCA_ERROR_*` from a `doca_rdmi_*` call\n  mean?\"** — worked example: *\"`DOCA_ERROR_BAD_STATE` from\n  `doca_rdmi_connection_dpa_completion_attach`\"*. Answered by the\n  RDMI overlay on the cross-library taxonomy in\n  [`CAPABILITIES.md ## Error taxonomy`](CAPABILITIES.md#error-taxonomy)\n  + the layered ladder in [`TASKS.md ## debug`](TASKS.md#debug)\n  that escalates to [`doca-debug`](..\u002F..\u002Fdoca-debug\u002FSKILL.md).\n\n## Audience\n\nThis skill serves **external developers building DPA-resident DOCA\napplications that need to *initiate* one-sided RDMA operations\nagainst a remote responder** — i.e., users whose accelerator-side\ncode wants to post sends, writes, or reads directly from the\naccelerator without round-tripping through the host CPU. The\ncanonical caller is a DPA kernel that has been compiled with\n`doca-dpacc-compiler` and runs on the BlueField DPA datapath; a\nGPU-side caller that drives the DPU's RDMA queues is the\nsister case routed to [`doca-gpi`](..\u002Fdoca-gpi\u002FSKILL.md). This\nskill is *not* for NVIDIA developers contributing to DOCA RDMI\nitself, and it is not the right surface for general host-CPU\ntwo-sided RDMA — that belongs to\n[`doca-rdma`](..\u002Fdoca-rdma\u002FSKILL.md).\n\n## Language scope\n\nDOCA RDMI ships as a C library with the `pkg-config` module name\n`doca-rdmi`. The library's *host-side* surface\n(`doca_rdmi_connection_*`, `doca_rdmi_poster_*`) is C; the *DPA-side*\nsurface that the accelerator kernel uses is also C, compiled against\nthe DOCA DPA toolchain documented in\n[`doca-dpa`](..\u002Fdoca-dpa\u002FSKILL.md). Other-language consumers (Rust,\nGo, Python, …) consume the host-side `*.so` through FFI; the skill's\ncontribution in that case is to keep the connection \u002F poster\nlifecycle, the EXPERIMENTAL-tag policy, the DPA-side handoff rules,\nand the safety overlay language-neutral, and to route the agent to\nthe public C ABI as the authoritative surface that any wrapper will\neventually call. The DPA-side surface is *not* wrappable in another\nlanguage — it is compiled and linked into the DPA binary itself.\n\n## When to load this skill\n\nLoad this skill when the user is doing **hands-on DOCA RDMI work**\non a host or BlueField with DOCA installed. Concretely:\n\n- Deciding between `doca-rdmi` and `doca-rdma` for a new\n  one-sided RDMA workload that runs from the DPA datapath.\n- Creating a `doca_rdmi_connection` or `doca_rdmi_poster`,\n  attaching a `doca_dpa_completion` or a `doca_verbs_cq`, and\n  starting the context on the DPA datapath.\n- Wiring the DPA-side handle returned by\n  `doca_rdmi_connection_get_dpa_handle` \u002F\n  `doca_rdmi_poster_get_dpa_handle` into a DPA kernel that calls\n  the matching device-side header\n  (`doca_rdmi_dev_connection.h`, `doca_rdmi_dev_poster.h`,\n  `doca_rdmi_dev_cqe.h`).\n- Acknowledging receive completions on the host side via\n  `doca_rdmi_connection_recv_ack` after the DPA kernel consumed\n  them.\n- Auditing which RDMI symbols are EXPERIMENTAL on the installed\n  DOCA version, before declaring an RDMI-using component\n  production-stable.\n- Debugging a `DOCA_ERROR_*` returned by a `doca_rdmi_*` call and\n  deciding whether the cause is a configuration mistake, a\n  lifecycle ordering bug, an unsupported capability on this device,\n  or a layer below DOCA.\n\nDo **not** load this skill for general DOCA orientation, install of\nDOCA itself, or two-sided host-CPU RDMA questions. For those, use\n[`doca-public-knowledge-map`](..\u002F..\u002Fdoca-public-knowledge-map\u002FSKILL.md),\n[`doca-setup`](..\u002F..\u002Fdoca-setup\u002FSKILL.md), and\n[`doca-rdma`](..\u002Fdoca-rdma\u002FSKILL.md) respectively.\n\n## What this skill provides\n\nThis is a **thin loader**. The body keeps only the orientation\nneeded to pick the right next file. The substantive RDMI-specific\nmaterial lives in two companion files:\n\n- `CAPABILITIES.md` — what RDMI can express on this version: the\n  `doca_rdmi_connection` and `doca_rdmi_poster` object model, the\n  DPA-side completion-attach pattern, the relationship to\n  `doca-verbs` (RDMI builds on a `doca_verbs_context`) and to\n  `doca-dpa` (the accelerator-side datapath), the EXPERIMENTAL-tag\n  rule for version handling, the RDMI overlay on the cross-library\n  `DOCA_ERROR_*` taxonomy, the observability surface (completion\n  events on the PE \u002F DPA-side completions), and the safety policy\n  that gates posting work from an accelerator kernel into a remote\n  responder's memory.\n- `TASKS.md` — step-by-step workflows for the eight in-scope verbs:\n  `install`, `configure`, `build`, `modify`, `run`, `test`,\n  `debug`, `use`. Plus a `## rollback` overlay (RDMI-specific\n  five-step teardown for the verbs \u002F connection \u002F DPA-attach \u002F\n  MR stack) and the 5-phase universal debug-loop instantiation\n  appended to `## debug`. Plus a `Deferred task verbs` block\n  that points out-of-scope questions at the right next skill.\n\nThe skill assumes a host or BlueField where DOCA is already\ninstalled at the standard location and the user has the privileges\ntheir public install profile expects. It does not cover installing\nDOCA itself — that path goes through\n[`doca-setup`](..\u002F..\u002Fdoca-setup\u002FSKILL.md).\n\n## What this skill deliberately does not ship\n\nThis skill is **agent guidance**, not a samples or templates\nbundle. To keep the boundary clean, it deliberately does not\ncontain — and pull requests should not add:\n\n- **Pre-written DOCA RDMI application source code, in any\n  language.** The agent's job is to route the user to verified\n  reference code on the user's installed DOCA and to prescribe a\n  minimum-diff modification via the universal modify-a-sample\n  workflow in\n  [`doca-programming-guide`](..\u002F..\u002Fdoca-programming-guide\u002FSKILL.md),\n  layered with the RDMI-specific overrides in\n  [`TASKS.md ## modify`](TASKS.md#modify). Because every RDMI\n  symbol is EXPERIMENTAL at the time of writing, the skill\n  refuses to author RDMI source code from documentation prose —\n  the API can change between releases and the resulting code may\n  not even compile.\n- **Standalone build manifests** (`meson.build`, `CMakeLists.txt`,\n  `Cargo.toml`, …) parked inside the skill. The agent constructs\n  the build manifest *in the user's project directory* against the\n  user's installed DOCA, where `pkg-config --modversion doca-rdmi`\n  is the source of truth.\n- **DPA-side kernel templates.** The DPA-side surface is owned by\n  [`doca-dpa`](..\u002Fdoca-dpa\u002FSKILL.md); RDMI's DPA-side headers are\n  *consumed by* the DPA programming model documented there. This\n  skill names the RDMI-specific handoff (the DPA handle type, the\n  completion-attach call) but does not author DPA kernels.\n- **A `samples\u002F`, `bindings\u002F`, or `reference\u002F` subtree** of any\n  kind. A mock or incomplete artifact in this skill's tree, even\n  one labeled \"reference\", is misleading: users will read it as\n  buildable.\n\n## Loading order\n\n1. Read this `SKILL.md` first to confirm the user's question is in\n   scope.\n2. **For the RDMI object model, the DPA-side handoff pattern, the\n   EXPERIMENTAL-tag policy, the error taxonomy, observability, and\n   safety policy, see [CAPABILITIES.md](CAPABILITIES.md).**\n3. **For step-by-step workflows — install, configure, build,\n   modify, run, test, debug, use — see [TASKS.md](TASKS.md).**\n\nBoth companion files cross-link to each other and to\n[`doca-public-knowledge-map`](..\u002F..\u002Fdoca-public-knowledge-map\u002FSKILL.md)\nwhenever the right answer is \"look it up in the public docs or the\ninstalled package layout\" rather than \"RDMI-specific guidance\".\n\n## Related skills\n\n- [`doca-rdma`](..\u002Fdoca-rdma\u002FSKILL.md) — the higher-level RDMA\n  library covering two-sided and host-CPU-initiated RDMA. RDMI is\n  the focused initiator-side surface; doca-rdma is the right\n  answer for the majority of RDMA use cases. The selection table\n  in [`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes)\n  is the load-bearing decision aid.\n- [`doca-dpa`](..\u002Fdoca-dpa\u002FSKILL.md) — the DOCA DPA programming\n  surface. RDMI returns DPA-side handles\n  (`doca_dpa_dev_rdmi_connection_t`, `doca_rdmi_dev_poster_t`)\n  that the DPA kernel uses through the device-side headers\n  (`doca_rdmi_dev_connection.h`, `doca_rdmi_dev_poster.h`,\n  `doca_rdmi_dev_cqe.h`); the DPA toolchain, kernel build, and\n  execution model are owned by that skill.\n- [`doca-gpi`](..\u002Fdoca-gpi\u002FSKILL.md) — the GPU-side sister of\n  this skill. GPI is the channel\u002Fqueue surface a CUDA kernel\n  uses to initiate RDMA; RDMI is the DPA-side surface. Both\n  layer on the same DOCA RDMA \u002F verbs substrate; either may\n  apply depending on whether the initiator is on the DPA or on\n  the GPU.\n- [`doca-public-knowledge-map`](..\u002F..\u002Fdoca-public-knowledge-map\u002FSKILL.md) — the\n  routing table for every public DOCA documentation source and\n  the on-disk layout of an installed DOCA package.\n- [`doca-setup`](..\u002F..\u002Fdoca-setup\u002FSKILL.md) — env preparation,\n  install verification, and the *I have no install yet* path with\n  the public NGC DOCA container.\n- [`doca-programming-guide`](..\u002F..\u002Fdoca-programming-guide\u002FSKILL.md) —\n  general DOCA programming patterns shared by every library: the\n  canonical `pkg-config` + meson build pattern, the universal\n  modify-a-shipped-sample first-app workflow, the universal\n  Core-context lifecycle, the cross-library `DOCA_ERROR_*`\n  taxonomy. This skill layers RDMI specifics on top.\n- [`doca-debug`](..\u002F..\u002Fdoca-debug\u002FSKILL.md) — the cross-cutting\n  debug ladder (install \u002F version \u002F build \u002F link \u002F runtime \u002F\n  program \u002F driver). RDMI-specific debug overlays on top of it.\n- [`doca-hardware-safety`](..\u002F..\u002Fdoca-hardware-safety\u002FSKILL.md) —\n  the bundle-wide hardware-safety meta-policy. The `## Safety\n  policy` overlay in `CAPABILITIES.md` cross-links it.\n- [`doca-version`](..\u002F..\u002Fdoca-version\u002FSKILL.md) — the version\n  detection \u002F four-way match rule every per-artifact `##\n  Version compatibility` anchor builds on. This skill quotes\n  the RDMI-specific overlay only.\n- [`doca-structured-tools-contract`](..\u002F..\u002Fdoca-structured-tools-contract\u002FSKILL.md) —\n  the JSON-schema contracts for the agent-preferred structured\n  helpers (env probe, capability snapshot, version-matrix\n  lookup); the `## Command appendix` in `TASKS.md` defers to\n  them before falling back to the manual chain.\n\n",{"data":32,"body":36},{"license":23,"name":4,"description":6,"metadata":33,"compatibility":35},{"kind":34},"library","Requires DOCA SDK installed at \u002Fopt\u002Fmellanox\u002Fdoca on Linux (Ubuntu 22.04\u002F24.04 or RHEL\u002FSLES) with a BlueField DPU or ConnectX NIC attached, plus the DOCA DPA toolchain and a DPA-capable BlueField when the data path runs on the DPA. Reads the user's local install via `pkg-config doca-rdmi` (alongside doca-common, doca-verbs, doca-dpa) and inspects \u002Fopt\u002Fmellanox\u002Fdoca\u002F{lib,include,samples}.\n",{"type":37,"children":38},"root",[39,48,121,185,192,204,594,600,654,660,733,739,751,887,929,935,947,1093,1107,1113,1125,1265,1271,1314,1329,1335],{"type":40,"tag":41,"props":42,"children":44},"element","h1",{"id":43},"doca-rdma-initiator",[45],{"type":46,"value":47},"text","DOCA RDMA Initiator",{"type":40,"tag":49,"props":50,"children":51},"p",{},[52,58,60,65,67,73,75,85,87,93,95,104,106,111,113,119],{"type":40,"tag":53,"props":54,"children":55},"strong",{},[56],{"type":46,"value":57},"Where to start:",{"type":46,"value":59}," This skill assumes DOCA is already installed and\nthe user is doing ",{"type":40,"tag":53,"props":61,"children":62},{},[63],{"type":46,"value":64},"hands-on RDMI work",{"type":46,"value":66}," on a host or BlueField with\nthe DOCA package set that ships the ",{"type":40,"tag":68,"props":69,"children":71},"code",{"className":70},[],[72],{"type":46,"value":4},{"type":46,"value":74}," library. Open\n",{"type":40,"tag":76,"props":77,"children":79},"a",{"href":78},"TASKS.md",[80],{"type":40,"tag":68,"props":81,"children":83},{"className":82},[],[84],{"type":46,"value":78},{"type":46,"value":86}," if the user wants to ",{"type":40,"tag":88,"props":89,"children":90},"em",{},[91],{"type":46,"value":92},"do",{"type":46,"value":94}," something\n(install \u002F configure \u002F build \u002F modify \u002F run \u002F test \u002F debug \u002F use);\nopen ",{"type":40,"tag":76,"props":96,"children":98},{"href":97},"CAPABILITIES.md",[99],{"type":40,"tag":68,"props":100,"children":102},{"className":101},[],[103],{"type":46,"value":97},{"type":46,"value":105}," when the question is\n",{"type":40,"tag":88,"props":107,"children":108},{},[109],{"type":46,"value":110},"what can RDMI express on this version",{"type":46,"value":112}," — the object model, the\nDPA-side handle types, the relationship to ",{"type":40,"tag":68,"props":114,"children":116},{"className":115},[],[117],{"type":46,"value":118},"doca-rdma",{"type":46,"value":120},", the\nEXPERIMENTAL-tag policy, and the safety overlay.",{"type":40,"tag":49,"props":122,"children":123},{},[124,129,131,141,143,149,151,156,158,164,166,172,174,183],{"type":40,"tag":53,"props":125,"children":126},{},[127],{"type":46,"value":128},"End-to-end \"walk me through doca-rdmi\" questions are answerable\nentirely from this skill.",{"type":46,"value":130}," Go straight to\n",{"type":40,"tag":76,"props":132,"children":134},{"href":133},"TASKS.md#end-to-end-quickref",[135],{"type":40,"tag":68,"props":136,"children":138},{"className":137},[],[139],{"type":46,"value":140},"TASKS.md ## end-to-end (quickref)",{"type":46,"value":142},",\nwhich carries the self-contained install-check → device\u002Fcap\ndiscovery → sample → ",{"type":40,"tag":68,"props":144,"children":146},{"className":145},[],[147],{"type":46,"value":148},"pkg-config",{"type":46,"value":150}," build → run → debug walkthrough\nwith the exact commands. You do ",{"type":40,"tag":53,"props":152,"children":153},{},[154],{"type":46,"value":155},"not",{"type":46,"value":157}," need to open ",{"type":40,"tag":68,"props":159,"children":161},{"className":160},[],[162],{"type":46,"value":163},"doca-setup",{"type":46,"value":165}," or\n",{"type":40,"tag":68,"props":167,"children":169},{"className":168},[],[170],{"type":46,"value":171},"doca-programming-guide",{"type":46,"value":173}," to answer an RDMI build\u002Frun\u002Fdebug question.\nRoute to ",{"type":40,"tag":76,"props":175,"children":177},{"href":176},"..\u002F..\u002Fdoca-setup\u002FSKILL.md",[178],{"type":40,"tag":68,"props":179,"children":181},{"className":180},[],[182],{"type":46,"value":163},{"type":46,"value":184}," when the required\nDOCA prerequisites are absent, partial, or version-mismatched.",{"type":40,"tag":186,"props":187,"children":189},"h2",{"id":188},"example-questions-this-skill-answers-well",[190],{"type":46,"value":191},"Example questions this skill answers well",{"type":40,"tag":49,"props":193,"children":194},{},[195,197,202],{"type":46,"value":196},"The CLASSES of RDMI questions this skill is built to answer, each\nwith one worked example. The agent should treat the ",{"type":40,"tag":88,"props":198,"children":199},{},[200],{"type":46,"value":201},"class",{"type":46,"value":203}," as the\nload-bearing piece — the worked example is a single instance.",{"type":40,"tag":205,"props":206,"children":207},"ul",{},[208,270,322,381,446,506],{"type":40,"tag":209,"props":210,"children":211},"li",{},[212,231,233,238,240,245,247,257,259,268],{"type":40,"tag":53,"props":213,"children":214},{},[215,217,222,224,229],{"type":46,"value":216},"\"Should I use ",{"type":40,"tag":68,"props":218,"children":220},{"className":219},[],[221],{"type":46,"value":4},{"type":46,"value":223}," or ",{"type":40,"tag":68,"props":225,"children":227},{"className":226},[],[228],{"type":46,"value":118},{"type":46,"value":230}," for this case?\"",{"type":46,"value":232}," —\nworked example: ",{"type":40,"tag":88,"props":234,"children":235},{},[236],{"type":46,"value":237},"\"I have a DPA kernel that needs to fire 1 MB\nRDMA writes at a remote responder; which library?\"",{"type":46,"value":239},". Answered by\nthe ",{"type":40,"tag":88,"props":241,"children":242},{},[243],{"type":46,"value":244},"initiator-side vs general-purpose",{"type":46,"value":246}," selection rule in\n",{"type":40,"tag":76,"props":248,"children":250},{"href":249},"CAPABILITIES.md#capabilities-and-modes",[251],{"type":40,"tag":68,"props":252,"children":254},{"className":253},[],[255],{"type":46,"value":256},"CAPABILITIES.md ## Capabilities and modes",{"type":46,"value":258},"\nsurface-selection table + the routing back to\n",{"type":40,"tag":76,"props":260,"children":262},{"href":261},"..\u002Fdoca-rdma\u002FSKILL.md",[263],{"type":40,"tag":68,"props":264,"children":266},{"className":265},[],[267],{"type":46,"value":118},{"type":46,"value":269}," when the use case is\ntwo-sided or host-CPU initiated.",{"type":40,"tag":209,"props":271,"children":272},{},[273,278,279,292,294,302],{"type":40,"tag":53,"props":274,"children":275},{},[276],{"type":46,"value":277},"\"How do I bring up an RDMI connection on the DPA datapath?\"",{"type":46,"value":232},{"type":40,"tag":88,"props":280,"children":281},{},[282,284,290],{"type":46,"value":283},"\"create a ",{"type":40,"tag":68,"props":285,"children":287},{"className":286},[],[288],{"type":46,"value":289},"doca_rdmi_connection",{"type":46,"value":291},", attach a DPA\ncompletion context, hand the DPA-side handle to my kernel\"",{"type":46,"value":293},".\nAnswered by the connection-object lifecycle in\n",{"type":40,"tag":76,"props":295,"children":296},{"href":249},[297],{"type":40,"tag":68,"props":298,"children":300},{"className":299},[],[301],{"type":46,"value":256},{"type":40,"tag":205,"props":303,"children":304},{},[305],{"type":40,"tag":209,"props":306,"children":307},{},[308,310,320],{"type":46,"value":309},"the configure walk in\n",{"type":40,"tag":76,"props":311,"children":313},{"href":312},"TASKS.md#configure",[314],{"type":40,"tag":68,"props":315,"children":317},{"className":316},[],[318],{"type":46,"value":319},"TASKS.md ## configure",{"type":46,"value":321},".",{"type":40,"tag":209,"props":323,"children":324},{},[325,330,332,352,354,362],{"type":40,"tag":53,"props":326,"children":327},{},[328],{"type":46,"value":329},"\"How do connection and poster relate — when do I need both?\"",{"type":46,"value":331},"\n— worked example: ",{"type":40,"tag":88,"props":333,"children":334},{},[335,337,342,344,350],{"type":46,"value":336},"\"my application receives work requests AND\nposts RDMA writes; do I need a ",{"type":40,"tag":68,"props":338,"children":340},{"className":339},[],[341],{"type":46,"value":289},{"type":46,"value":343}," plus a\n",{"type":40,"tag":68,"props":345,"children":347},{"className":346},[],[348],{"type":46,"value":349},"doca_rdmi_poster",{"type":46,"value":351},", or one of them?\"",{"type":46,"value":353},". Answered by the\ntwo-object model in\n",{"type":40,"tag":76,"props":355,"children":356},{"href":249},[357],{"type":40,"tag":68,"props":358,"children":360},{"className":359},[],[361],{"type":46,"value":256},{"type":40,"tag":205,"props":363,"children":364},{},[365],{"type":40,"tag":209,"props":366,"children":367},{},[368,370,380],{"type":46,"value":369},"the modify-from-sample slot table in\n",{"type":40,"tag":76,"props":371,"children":373},{"href":372},"TASKS.md#modify",[374],{"type":40,"tag":68,"props":375,"children":377},{"className":376},[],[378],{"type":46,"value":379},"TASKS.md ## modify",{"type":46,"value":321},{"type":40,"tag":209,"props":382,"children":383},{},[384,389,391,404,406,414],{"type":40,"tag":53,"props":385,"children":386},{},[387],{"type":46,"value":388},"\"How do I drive completions on the DPA side?\"",{"type":46,"value":390}," — worked\nexample: ",{"type":40,"tag":88,"props":392,"children":393},{},[394,396,402],{"type":46,"value":395},"\"hook the connection to a ",{"type":40,"tag":68,"props":397,"children":399},{"className":398},[],[400],{"type":46,"value":401},"doca_dpa_completion",{"type":46,"value":403}," so my\nkernel polls completions directly\"",{"type":46,"value":405},". Answered by the DPA-side\ncompletion-attach pattern in\n",{"type":40,"tag":76,"props":407,"children":408},{"href":249},[409],{"type":40,"tag":68,"props":410,"children":412},{"className":411},[],[413],{"type":46,"value":256},{"type":40,"tag":205,"props":415,"children":416},{},[417],{"type":40,"tag":209,"props":418,"children":419},{},[420,422,432,434,444],{"type":46,"value":421},"the run-side wiring in\n",{"type":40,"tag":76,"props":423,"children":425},{"href":424},"TASKS.md#run",[426],{"type":40,"tag":68,"props":427,"children":429},{"className":428},[],[430],{"type":46,"value":431},"TASKS.md ## run",{"type":46,"value":433},", cross-linked into\n",{"type":40,"tag":76,"props":435,"children":437},{"href":436},"..\u002Fdoca-dpa\u002FSKILL.md",[438],{"type":40,"tag":68,"props":439,"children":441},{"className":440},[],[442],{"type":46,"value":443},"doca-dpa",{"type":46,"value":445}," for the DPA programming\nsurface itself.",{"type":40,"tag":209,"props":447,"children":448},{},[449,454,456,469,471,481],{"type":40,"tag":53,"props":450,"children":451},{},[452],{"type":46,"value":453},"\"Is the symbol I want available — and is it stable enough to\nship?\"",{"type":46,"value":455}," — worked example: ",{"type":40,"tag":88,"props":457,"children":458},{},[459,461,467],{"type":46,"value":460},"\"is ",{"type":40,"tag":68,"props":462,"children":464},{"className":463},[],[465],{"type":46,"value":466},"doca_rdmi_poster_post",{"type":46,"value":468}," GA on\nmy installed DOCA, or still EXPERIMENTAL?\"",{"type":46,"value":470},". Answered by the\nEXPERIMENTAL-tag policy in\n",{"type":40,"tag":76,"props":472,"children":474},{"href":473},"CAPABILITIES.md#version-compatibility",[475],{"type":40,"tag":68,"props":476,"children":478},{"className":477},[],[479],{"type":46,"value":480},"CAPABILITIES.md ## Version compatibility",{"type":40,"tag":205,"props":482,"children":483},{},[484],{"type":40,"tag":209,"props":485,"children":486},{},[487,489,495,497,505],{"type":46,"value":488},"the version-discovery rule\n(",{"type":40,"tag":68,"props":490,"children":492},{"className":491},[],[493],{"type":46,"value":494},"pkg-config --modversion doca-rdmi",{"type":46,"value":496},") pinned in\n",{"type":40,"tag":76,"props":498,"children":499},{"href":312},[500],{"type":40,"tag":68,"props":501,"children":503},{"className":502},[],[504],{"type":46,"value":319},{"type":46,"value":321},{"type":40,"tag":209,"props":507,"children":508},{},[509,530,531,551,553,563],{"type":40,"tag":53,"props":510,"children":511},{},[512,514,520,522,528],{"type":46,"value":513},"\"What does this ",{"type":40,"tag":68,"props":515,"children":517},{"className":516},[],[518],{"type":46,"value":519},"DOCA_ERROR_*",{"type":46,"value":521}," from a ",{"type":40,"tag":68,"props":523,"children":525},{"className":524},[],[526],{"type":46,"value":527},"doca_rdmi_*",{"type":46,"value":529}," call\nmean?\"",{"type":46,"value":455},{"type":40,"tag":88,"props":532,"children":533},{},[534,536,542,544,550],{"type":46,"value":535},"\"",{"type":40,"tag":68,"props":537,"children":539},{"className":538},[],[540],{"type":46,"value":541},"DOCA_ERROR_BAD_STATE",{"type":46,"value":543}," from\n",{"type":40,"tag":68,"props":545,"children":547},{"className":546},[],[548],{"type":46,"value":549},"doca_rdmi_connection_dpa_completion_attach",{"type":46,"value":535},{"type":46,"value":552},". Answered by the\nRDMI overlay on the cross-library taxonomy in\n",{"type":40,"tag":76,"props":554,"children":556},{"href":555},"CAPABILITIES.md#error-taxonomy",[557],{"type":40,"tag":68,"props":558,"children":560},{"className":559},[],[561],{"type":46,"value":562},"CAPABILITIES.md ## Error taxonomy",{"type":40,"tag":205,"props":564,"children":565},{},[566],{"type":40,"tag":209,"props":567,"children":568},{},[569,571,581,583,593],{"type":46,"value":570},"the layered ladder in ",{"type":40,"tag":76,"props":572,"children":574},{"href":573},"TASKS.md#debug",[575],{"type":40,"tag":68,"props":576,"children":578},{"className":577},[],[579],{"type":46,"value":580},"TASKS.md ## debug",{"type":46,"value":582},"\nthat escalates to ",{"type":40,"tag":76,"props":584,"children":586},{"href":585},"..\u002F..\u002Fdoca-debug\u002FSKILL.md",[587],{"type":40,"tag":68,"props":588,"children":590},{"className":589},[],[591],{"type":46,"value":592},"doca-debug",{"type":46,"value":321},{"type":40,"tag":186,"props":595,"children":597},{"id":596},"audience",[598],{"type":46,"value":599},"Audience",{"type":40,"tag":49,"props":601,"children":602},{},[603,605,617,619,625,627,637,639,643,645,653],{"type":46,"value":604},"This skill serves ",{"type":40,"tag":53,"props":606,"children":607},{},[608,610,615],{"type":46,"value":609},"external developers building DPA-resident DOCA\napplications that need to ",{"type":40,"tag":88,"props":611,"children":612},{},[613],{"type":46,"value":614},"initiate",{"type":46,"value":616}," one-sided RDMA operations\nagainst a remote responder",{"type":46,"value":618}," — i.e., users whose accelerator-side\ncode wants to post sends, writes, or reads directly from the\naccelerator without round-tripping through the host CPU. The\ncanonical caller is a DPA kernel that has been compiled with\n",{"type":40,"tag":68,"props":620,"children":622},{"className":621},[],[623],{"type":46,"value":624},"doca-dpacc-compiler",{"type":46,"value":626}," and runs on the BlueField DPA datapath; a\nGPU-side caller that drives the DPU's RDMA queues is the\nsister case routed to ",{"type":40,"tag":76,"props":628,"children":630},{"href":629},"..\u002Fdoca-gpi\u002FSKILL.md",[631],{"type":40,"tag":68,"props":632,"children":634},{"className":633},[],[635],{"type":46,"value":636},"doca-gpi",{"type":46,"value":638},". This\nskill is ",{"type":40,"tag":88,"props":640,"children":641},{},[642],{"type":46,"value":155},{"type":46,"value":644}," for NVIDIA developers contributing to DOCA RDMI\nitself, and it is not the right surface for general host-CPU\ntwo-sided RDMA — that belongs to\n",{"type":40,"tag":76,"props":646,"children":647},{"href":261},[648],{"type":40,"tag":68,"props":649,"children":651},{"className":650},[],[652],{"type":46,"value":118},{"type":46,"value":321},{"type":40,"tag":186,"props":655,"children":657},{"id":656},"language-scope",[658],{"type":46,"value":659},"Language scope",{"type":40,"tag":49,"props":661,"children":662},{},[663,665,670,672,677,679,684,686,692,694,700,702,707,709,717,719,725,727,731],{"type":46,"value":664},"DOCA RDMI ships as a C library with the ",{"type":40,"tag":68,"props":666,"children":668},{"className":667},[],[669],{"type":46,"value":148},{"type":46,"value":671}," module name\n",{"type":40,"tag":68,"props":673,"children":675},{"className":674},[],[676],{"type":46,"value":4},{"type":46,"value":678},". The library's ",{"type":40,"tag":88,"props":680,"children":681},{},[682],{"type":46,"value":683},"host-side",{"type":46,"value":685}," surface\n(",{"type":40,"tag":68,"props":687,"children":689},{"className":688},[],[690],{"type":46,"value":691},"doca_rdmi_connection_*",{"type":46,"value":693},", ",{"type":40,"tag":68,"props":695,"children":697},{"className":696},[],[698],{"type":46,"value":699},"doca_rdmi_poster_*",{"type":46,"value":701},") is C; the ",{"type":40,"tag":88,"props":703,"children":704},{},[705],{"type":46,"value":706},"DPA-side",{"type":46,"value":708},"\nsurface that the accelerator kernel uses is also C, compiled against\nthe DOCA DPA toolchain documented in\n",{"type":40,"tag":76,"props":710,"children":711},{"href":436},[712],{"type":40,"tag":68,"props":713,"children":715},{"className":714},[],[716],{"type":46,"value":443},{"type":46,"value":718},". Other-language consumers (Rust,\nGo, Python, …) consume the host-side ",{"type":40,"tag":68,"props":720,"children":722},{"className":721},[],[723],{"type":46,"value":724},"*.so",{"type":46,"value":726}," through FFI; the skill's\ncontribution in that case is to keep the connection \u002F poster\nlifecycle, the EXPERIMENTAL-tag policy, the DPA-side handoff rules,\nand the safety overlay language-neutral, and to route the agent to\nthe public C ABI as the authoritative surface that any wrapper will\neventually call. The DPA-side surface is ",{"type":40,"tag":88,"props":728,"children":729},{},[730],{"type":46,"value":155},{"type":46,"value":732}," wrappable in another\nlanguage — it is compiled and linked into the DPA binary itself.",{"type":40,"tag":186,"props":734,"children":736},{"id":735},"when-to-load-this-skill",[737],{"type":46,"value":738},"When to load this skill",{"type":40,"tag":49,"props":740,"children":741},{},[742,744,749],{"type":46,"value":743},"Load this skill when the user is doing ",{"type":40,"tag":53,"props":745,"children":746},{},[747],{"type":46,"value":748},"hands-on DOCA RDMI work",{"type":46,"value":750},"\non a host or BlueField with DOCA installed. Concretely:",{"type":40,"tag":205,"props":752,"children":753},{},[754,773,806,850,863,868],{"type":40,"tag":209,"props":755,"children":756},{},[757,759,764,766,771],{"type":46,"value":758},"Deciding between ",{"type":40,"tag":68,"props":760,"children":762},{"className":761},[],[763],{"type":46,"value":4},{"type":46,"value":765}," and ",{"type":40,"tag":68,"props":767,"children":769},{"className":768},[],[770],{"type":46,"value":118},{"type":46,"value":772}," for a new\none-sided RDMA workload that runs from the DPA datapath.",{"type":40,"tag":209,"props":774,"children":775},{},[776,778,783,784,789,791,796,798,804],{"type":46,"value":777},"Creating a ",{"type":40,"tag":68,"props":779,"children":781},{"className":780},[],[782],{"type":46,"value":289},{"type":46,"value":223},{"type":40,"tag":68,"props":785,"children":787},{"className":786},[],[788],{"type":46,"value":349},{"type":46,"value":790},",\nattaching a ",{"type":40,"tag":68,"props":792,"children":794},{"className":793},[],[795],{"type":46,"value":401},{"type":46,"value":797}," or a ",{"type":40,"tag":68,"props":799,"children":801},{"className":800},[],[802],{"type":46,"value":803},"doca_verbs_cq",{"type":46,"value":805},", and\nstarting the context on the DPA datapath.",{"type":40,"tag":209,"props":807,"children":808},{},[809,811,817,819,825,827,833,834,840,842,848],{"type":46,"value":810},"Wiring the DPA-side handle returned by\n",{"type":40,"tag":68,"props":812,"children":814},{"className":813},[],[815],{"type":46,"value":816},"doca_rdmi_connection_get_dpa_handle",{"type":46,"value":818}," \u002F\n",{"type":40,"tag":68,"props":820,"children":822},{"className":821},[],[823],{"type":46,"value":824},"doca_rdmi_poster_get_dpa_handle",{"type":46,"value":826}," into a DPA kernel that calls\nthe matching device-side header\n(",{"type":40,"tag":68,"props":828,"children":830},{"className":829},[],[831],{"type":46,"value":832},"doca_rdmi_dev_connection.h",{"type":46,"value":693},{"type":40,"tag":68,"props":835,"children":837},{"className":836},[],[838],{"type":46,"value":839},"doca_rdmi_dev_poster.h",{"type":46,"value":841},",\n",{"type":40,"tag":68,"props":843,"children":845},{"className":844},[],[846],{"type":46,"value":847},"doca_rdmi_dev_cqe.h",{"type":46,"value":849},").",{"type":40,"tag":209,"props":851,"children":852},{},[853,855,861],{"type":46,"value":854},"Acknowledging receive completions on the host side via\n",{"type":40,"tag":68,"props":856,"children":858},{"className":857},[],[859],{"type":46,"value":860},"doca_rdmi_connection_recv_ack",{"type":46,"value":862}," after the DPA kernel consumed\nthem.",{"type":40,"tag":209,"props":864,"children":865},{},[866],{"type":46,"value":867},"Auditing which RDMI symbols are EXPERIMENTAL on the installed\nDOCA version, before declaring an RDMI-using component\nproduction-stable.",{"type":40,"tag":209,"props":869,"children":870},{},[871,873,878,880,885],{"type":46,"value":872},"Debugging a ",{"type":40,"tag":68,"props":874,"children":876},{"className":875},[],[877],{"type":46,"value":519},{"type":46,"value":879}," returned by a ",{"type":40,"tag":68,"props":881,"children":883},{"className":882},[],[884],{"type":46,"value":527},{"type":46,"value":886}," call and\ndeciding whether the cause is a configuration mistake, a\nlifecycle ordering bug, an unsupported capability on this device,\nor a layer below DOCA.",{"type":40,"tag":49,"props":888,"children":889},{},[890,892,896,898,908,909,917,919,927],{"type":46,"value":891},"Do ",{"type":40,"tag":53,"props":893,"children":894},{},[895],{"type":46,"value":155},{"type":46,"value":897}," load this skill for general DOCA orientation, install of\nDOCA itself, or two-sided host-CPU RDMA questions. For those, use\n",{"type":40,"tag":76,"props":899,"children":901},{"href":900},"..\u002F..\u002Fdoca-public-knowledge-map\u002FSKILL.md",[902],{"type":40,"tag":68,"props":903,"children":905},{"className":904},[],[906],{"type":46,"value":907},"doca-public-knowledge-map",{"type":46,"value":841},{"type":40,"tag":76,"props":910,"children":911},{"href":176},[912],{"type":40,"tag":68,"props":913,"children":915},{"className":914},[],[916],{"type":46,"value":163},{"type":46,"value":918},", and\n",{"type":40,"tag":76,"props":920,"children":921},{"href":261},[922],{"type":40,"tag":68,"props":923,"children":925},{"className":924},[],[926],{"type":46,"value":118},{"type":46,"value":928}," respectively.",{"type":40,"tag":186,"props":930,"children":932},{"id":931},"what-this-skill-provides",[933],{"type":46,"value":934},"What this skill provides",{"type":40,"tag":49,"props":936,"children":937},{},[938,940,945],{"type":46,"value":939},"This is a ",{"type":40,"tag":53,"props":941,"children":942},{},[943],{"type":46,"value":944},"thin loader",{"type":46,"value":946},". The body keeps only the orientation\nneeded to pick the right next file. The substantive RDMI-specific\nmaterial lives in two companion files:",{"type":40,"tag":205,"props":948,"children":949},{},[950,1003],{"type":40,"tag":209,"props":951,"children":952},{},[953,958,960,965,966,971,973,979,981,987,989,994,996,1001],{"type":40,"tag":68,"props":954,"children":956},{"className":955},[],[957],{"type":46,"value":97},{"type":46,"value":959}," — what RDMI can express on this version: the\n",{"type":40,"tag":68,"props":961,"children":963},{"className":962},[],[964],{"type":46,"value":289},{"type":46,"value":765},{"type":40,"tag":68,"props":967,"children":969},{"className":968},[],[970],{"type":46,"value":349},{"type":46,"value":972}," object model, the\nDPA-side completion-attach pattern, the relationship to\n",{"type":40,"tag":68,"props":974,"children":976},{"className":975},[],[977],{"type":46,"value":978},"doca-verbs",{"type":46,"value":980}," (RDMI builds on a ",{"type":40,"tag":68,"props":982,"children":984},{"className":983},[],[985],{"type":46,"value":986},"doca_verbs_context",{"type":46,"value":988},") and to\n",{"type":40,"tag":68,"props":990,"children":992},{"className":991},[],[993],{"type":46,"value":443},{"type":46,"value":995}," (the accelerator-side datapath), the EXPERIMENTAL-tag\nrule for version handling, the RDMI overlay on the cross-library\n",{"type":40,"tag":68,"props":997,"children":999},{"className":998},[],[1000],{"type":46,"value":519},{"type":46,"value":1002}," taxonomy, the observability surface (completion\nevents on the PE \u002F DPA-side completions), and the safety policy\nthat gates posting work from an accelerator kernel into a remote\nresponder's memory.",{"type":40,"tag":209,"props":1004,"children":1005},{},[1006,1011,1013,1019,1020,1026,1027,1033,1034,1040,1041,1047,1048,1054,1055,1061,1062,1068,1070,1076,1078,1084,1085,1091],{"type":40,"tag":68,"props":1007,"children":1009},{"className":1008},[],[1010],{"type":46,"value":78},{"type":46,"value":1012}," — step-by-step workflows for the eight in-scope verbs:\n",{"type":40,"tag":68,"props":1014,"children":1016},{"className":1015},[],[1017],{"type":46,"value":1018},"install",{"type":46,"value":693},{"type":40,"tag":68,"props":1021,"children":1023},{"className":1022},[],[1024],{"type":46,"value":1025},"configure",{"type":46,"value":693},{"type":40,"tag":68,"props":1028,"children":1030},{"className":1029},[],[1031],{"type":46,"value":1032},"build",{"type":46,"value":693},{"type":40,"tag":68,"props":1035,"children":1037},{"className":1036},[],[1038],{"type":46,"value":1039},"modify",{"type":46,"value":693},{"type":40,"tag":68,"props":1042,"children":1044},{"className":1043},[],[1045],{"type":46,"value":1046},"run",{"type":46,"value":693},{"type":40,"tag":68,"props":1049,"children":1051},{"className":1050},[],[1052],{"type":46,"value":1053},"test",{"type":46,"value":841},{"type":40,"tag":68,"props":1056,"children":1058},{"className":1057},[],[1059],{"type":46,"value":1060},"debug",{"type":46,"value":693},{"type":40,"tag":68,"props":1063,"children":1065},{"className":1064},[],[1066],{"type":46,"value":1067},"use",{"type":46,"value":1069},". Plus a ",{"type":40,"tag":68,"props":1071,"children":1073},{"className":1072},[],[1074],{"type":46,"value":1075},"## rollback",{"type":46,"value":1077}," overlay (RDMI-specific\nfive-step teardown for the verbs \u002F connection \u002F DPA-attach \u002F\nMR stack) and the 5-phase universal debug-loop instantiation\nappended to ",{"type":40,"tag":68,"props":1079,"children":1081},{"className":1080},[],[1082],{"type":46,"value":1083},"## debug",{"type":46,"value":1069},{"type":40,"tag":68,"props":1086,"children":1088},{"className":1087},[],[1089],{"type":46,"value":1090},"Deferred task verbs",{"type":46,"value":1092}," block\nthat points out-of-scope questions at the right next skill.",{"type":40,"tag":49,"props":1094,"children":1095},{},[1096,1098,1106],{"type":46,"value":1097},"The skill assumes a host or BlueField where DOCA is already\ninstalled at the standard location and the user has the privileges\ntheir public install profile expects. It does not cover installing\nDOCA itself — that path goes through\n",{"type":40,"tag":76,"props":1099,"children":1100},{"href":176},[1101],{"type":40,"tag":68,"props":1102,"children":1104},{"className":1103},[],[1105],{"type":46,"value":163},{"type":46,"value":321},{"type":40,"tag":186,"props":1108,"children":1110},{"id":1109},"what-this-skill-deliberately-does-not-ship",[1111],{"type":46,"value":1112},"What this skill deliberately does not ship",{"type":40,"tag":49,"props":1114,"children":1115},{},[1116,1118,1123],{"type":46,"value":1117},"This skill is ",{"type":40,"tag":53,"props":1119,"children":1120},{},[1121],{"type":46,"value":1122},"agent guidance",{"type":46,"value":1124},", not a samples or templates\nbundle. To keep the boundary clean, it deliberately does not\ncontain — and pull requests should not add:",{"type":40,"tag":205,"props":1126,"children":1127},{},[1128,1159,1205,1232],{"type":40,"tag":209,"props":1129,"children":1130},{},[1131,1136,1138,1147,1149,1157],{"type":40,"tag":53,"props":1132,"children":1133},{},[1134],{"type":46,"value":1135},"Pre-written DOCA RDMI application source code, in any\nlanguage.",{"type":46,"value":1137}," The agent's job is to route the user to verified\nreference code on the user's installed DOCA and to prescribe a\nminimum-diff modification via the universal modify-a-sample\nworkflow in\n",{"type":40,"tag":76,"props":1139,"children":1141},{"href":1140},"..\u002F..\u002Fdoca-programming-guide\u002FSKILL.md",[1142],{"type":40,"tag":68,"props":1143,"children":1145},{"className":1144},[],[1146],{"type":46,"value":171},{"type":46,"value":1148},",\nlayered with the RDMI-specific overrides in\n",{"type":40,"tag":76,"props":1150,"children":1151},{"href":372},[1152],{"type":40,"tag":68,"props":1153,"children":1155},{"className":1154},[],[1156],{"type":46,"value":379},{"type":46,"value":1158},". Because every RDMI\nsymbol is EXPERIMENTAL at the time of writing, the skill\nrefuses to author RDMI source code from documentation prose —\nthe API can change between releases and the resulting code may\nnot even compile.",{"type":40,"tag":209,"props":1160,"children":1161},{},[1162,1167,1169,1175,1176,1182,1183,1189,1191,1196,1198,1203],{"type":40,"tag":53,"props":1163,"children":1164},{},[1165],{"type":46,"value":1166},"Standalone build manifests",{"type":46,"value":1168}," (",{"type":40,"tag":68,"props":1170,"children":1172},{"className":1171},[],[1173],{"type":46,"value":1174},"meson.build",{"type":46,"value":693},{"type":40,"tag":68,"props":1177,"children":1179},{"className":1178},[],[1180],{"type":46,"value":1181},"CMakeLists.txt",{"type":46,"value":841},{"type":40,"tag":68,"props":1184,"children":1186},{"className":1185},[],[1187],{"type":46,"value":1188},"Cargo.toml",{"type":46,"value":1190},", …) parked inside the skill. The agent constructs\nthe build manifest ",{"type":40,"tag":88,"props":1192,"children":1193},{},[1194],{"type":46,"value":1195},"in the user's project directory",{"type":46,"value":1197}," against the\nuser's installed DOCA, where ",{"type":40,"tag":68,"props":1199,"children":1201},{"className":1200},[],[1202],{"type":46,"value":494},{"type":46,"value":1204},"\nis the source of truth.",{"type":40,"tag":209,"props":1206,"children":1207},{},[1208,1213,1215,1223,1225,1230],{"type":40,"tag":53,"props":1209,"children":1210},{},[1211],{"type":46,"value":1212},"DPA-side kernel templates.",{"type":46,"value":1214}," The DPA-side surface is owned by\n",{"type":40,"tag":76,"props":1216,"children":1217},{"href":436},[1218],{"type":40,"tag":68,"props":1219,"children":1221},{"className":1220},[],[1222],{"type":46,"value":443},{"type":46,"value":1224},"; RDMI's DPA-side headers are\n",{"type":40,"tag":88,"props":1226,"children":1227},{},[1228],{"type":46,"value":1229},"consumed by",{"type":46,"value":1231}," the DPA programming model documented there. This\nskill names the RDMI-specific handoff (the DPA handle type, the\ncompletion-attach call) but does not author DPA kernels.",{"type":40,"tag":209,"props":1233,"children":1234},{},[1235,1263],{"type":40,"tag":53,"props":1236,"children":1237},{},[1238,1240,1246,1247,1253,1255,1261],{"type":46,"value":1239},"A ",{"type":40,"tag":68,"props":1241,"children":1243},{"className":1242},[],[1244],{"type":46,"value":1245},"samples\u002F",{"type":46,"value":693},{"type":40,"tag":68,"props":1248,"children":1250},{"className":1249},[],[1251],{"type":46,"value":1252},"bindings\u002F",{"type":46,"value":1254},", or ",{"type":40,"tag":68,"props":1256,"children":1258},{"className":1257},[],[1259],{"type":46,"value":1260},"reference\u002F",{"type":46,"value":1262}," subtree",{"type":46,"value":1264}," of any\nkind. A mock or incomplete artifact in this skill's tree, even\none labeled \"reference\", is misleading: users will read it as\nbuildable.",{"type":40,"tag":186,"props":1266,"children":1268},{"id":1267},"loading-order",[1269],{"type":46,"value":1270},"Loading order",{"type":40,"tag":1272,"props":1273,"children":1274},"ol",{},[1275,1288,1301],{"type":40,"tag":209,"props":1276,"children":1277},{},[1278,1280,1286],{"type":46,"value":1279},"Read this ",{"type":40,"tag":68,"props":1281,"children":1283},{"className":1282},[],[1284],{"type":46,"value":1285},"SKILL.md",{"type":46,"value":1287}," first to confirm the user's question is in\nscope.",{"type":40,"tag":209,"props":1289,"children":1290},{},[1291],{"type":40,"tag":53,"props":1292,"children":1293},{},[1294,1296,1300],{"type":46,"value":1295},"For the RDMI object model, the DPA-side handoff pattern, the\nEXPERIMENTAL-tag policy, the error taxonomy, observability, and\nsafety policy, see ",{"type":40,"tag":76,"props":1297,"children":1298},{"href":97},[1299],{"type":46,"value":97},{"type":46,"value":321},{"type":40,"tag":209,"props":1302,"children":1303},{},[1304],{"type":40,"tag":53,"props":1305,"children":1306},{},[1307,1309,1313],{"type":46,"value":1308},"For step-by-step workflows — install, configure, build,\nmodify, run, test, debug, use — see ",{"type":40,"tag":76,"props":1310,"children":1311},{"href":78},[1312],{"type":46,"value":78},{"type":46,"value":321},{"type":40,"tag":49,"props":1315,"children":1316},{},[1317,1319,1327],{"type":46,"value":1318},"Both companion files cross-link to each other and to\n",{"type":40,"tag":76,"props":1320,"children":1321},{"href":900},[1322],{"type":40,"tag":68,"props":1323,"children":1325},{"className":1324},[],[1326],{"type":46,"value":907},{"type":46,"value":1328},"\nwhenever the right answer is \"look it up in the public docs or the\ninstalled package layout\" rather than \"RDMI-specific guidance\".",{"type":40,"tag":186,"props":1330,"children":1332},{"id":1331},"related-skills",[1333],{"type":46,"value":1334},"Related skills",{"type":40,"tag":205,"props":1336,"children":1337},{},[1338,1361,1408,1421,1434,1454,1481,1494,1524,1547],{"type":40,"tag":209,"props":1339,"children":1340},{},[1341,1349,1351,1359],{"type":40,"tag":76,"props":1342,"children":1343},{"href":261},[1344],{"type":40,"tag":68,"props":1345,"children":1347},{"className":1346},[],[1348],{"type":46,"value":118},{"type":46,"value":1350}," — the higher-level RDMA\nlibrary covering two-sided and host-CPU-initiated RDMA. RDMI is\nthe focused initiator-side surface; doca-rdma is the right\nanswer for the majority of RDMA use cases. The selection table\nin ",{"type":40,"tag":76,"props":1352,"children":1353},{"href":249},[1354],{"type":40,"tag":68,"props":1355,"children":1357},{"className":1356},[],[1358],{"type":46,"value":256},{"type":46,"value":1360},"\nis the load-bearing decision aid.",{"type":40,"tag":209,"props":1362,"children":1363},{},[1364,1372,1374,1380,1381,1387,1389,1394,1395,1400,1401,1406],{"type":40,"tag":76,"props":1365,"children":1366},{"href":436},[1367],{"type":40,"tag":68,"props":1368,"children":1370},{"className":1369},[],[1371],{"type":46,"value":443},{"type":46,"value":1373}," — the DOCA DPA programming\nsurface. RDMI returns DPA-side handles\n(",{"type":40,"tag":68,"props":1375,"children":1377},{"className":1376},[],[1378],{"type":46,"value":1379},"doca_dpa_dev_rdmi_connection_t",{"type":46,"value":693},{"type":40,"tag":68,"props":1382,"children":1384},{"className":1383},[],[1385],{"type":46,"value":1386},"doca_rdmi_dev_poster_t",{"type":46,"value":1388},")\nthat the DPA kernel uses through the device-side headers\n(",{"type":40,"tag":68,"props":1390,"children":1392},{"className":1391},[],[1393],{"type":46,"value":832},{"type":46,"value":693},{"type":40,"tag":68,"props":1396,"children":1398},{"className":1397},[],[1399],{"type":46,"value":839},{"type":46,"value":841},{"type":40,"tag":68,"props":1402,"children":1404},{"className":1403},[],[1405],{"type":46,"value":847},{"type":46,"value":1407},"); the DPA toolchain, kernel build, and\nexecution model are owned by that skill.",{"type":40,"tag":209,"props":1409,"children":1410},{},[1411,1419],{"type":40,"tag":76,"props":1412,"children":1413},{"href":629},[1414],{"type":40,"tag":68,"props":1415,"children":1417},{"className":1416},[],[1418],{"type":46,"value":636},{"type":46,"value":1420}," — the GPU-side sister of\nthis skill. GPI is the channel\u002Fqueue surface a CUDA kernel\nuses to initiate RDMA; RDMI is the DPA-side surface. Both\nlayer on the same DOCA RDMA \u002F verbs substrate; either may\napply depending on whether the initiator is on the DPA or on\nthe GPU.",{"type":40,"tag":209,"props":1422,"children":1423},{},[1424,1432],{"type":40,"tag":76,"props":1425,"children":1426},{"href":900},[1427],{"type":40,"tag":68,"props":1428,"children":1430},{"className":1429},[],[1431],{"type":46,"value":907},{"type":46,"value":1433}," — the\nrouting table for every public DOCA documentation source and\nthe on-disk layout of an installed DOCA package.",{"type":40,"tag":209,"props":1435,"children":1436},{},[1437,1445,1447,1452],{"type":40,"tag":76,"props":1438,"children":1439},{"href":176},[1440],{"type":40,"tag":68,"props":1441,"children":1443},{"className":1442},[],[1444],{"type":46,"value":163},{"type":46,"value":1446}," — env preparation,\ninstall verification, and the ",{"type":40,"tag":88,"props":1448,"children":1449},{},[1450],{"type":46,"value":1451},"I have no install yet",{"type":46,"value":1453}," path with\nthe public NGC DOCA container.",{"type":40,"tag":209,"props":1455,"children":1456},{},[1457,1465,1467,1472,1474,1479],{"type":40,"tag":76,"props":1458,"children":1459},{"href":1140},[1460],{"type":40,"tag":68,"props":1461,"children":1463},{"className":1462},[],[1464],{"type":46,"value":171},{"type":46,"value":1466}," —\ngeneral DOCA programming patterns shared by every library: the\ncanonical ",{"type":40,"tag":68,"props":1468,"children":1470},{"className":1469},[],[1471],{"type":46,"value":148},{"type":46,"value":1473}," + meson build pattern, the universal\nmodify-a-shipped-sample first-app workflow, the universal\nCore-context lifecycle, the cross-library ",{"type":40,"tag":68,"props":1475,"children":1477},{"className":1476},[],[1478],{"type":46,"value":519},{"type":46,"value":1480},"\ntaxonomy. This skill layers RDMI specifics on top.",{"type":40,"tag":209,"props":1482,"children":1483},{},[1484,1492],{"type":40,"tag":76,"props":1485,"children":1486},{"href":585},[1487],{"type":40,"tag":68,"props":1488,"children":1490},{"className":1489},[],[1491],{"type":46,"value":592},{"type":46,"value":1493}," — the cross-cutting\ndebug ladder (install \u002F version \u002F build \u002F link \u002F runtime \u002F\nprogram \u002F driver). RDMI-specific debug overlays on top of it.",{"type":40,"tag":209,"props":1495,"children":1496},{},[1497,1507,1509,1515,1517,1522],{"type":40,"tag":76,"props":1498,"children":1500},{"href":1499},"..\u002F..\u002Fdoca-hardware-safety\u002FSKILL.md",[1501],{"type":40,"tag":68,"props":1502,"children":1504},{"className":1503},[],[1505],{"type":46,"value":1506},"doca-hardware-safety",{"type":46,"value":1508}," —\nthe bundle-wide hardware-safety meta-policy. The ",{"type":40,"tag":68,"props":1510,"children":1512},{"className":1511},[],[1513],{"type":46,"value":1514},"## Safety policy",{"type":46,"value":1516}," overlay in ",{"type":40,"tag":68,"props":1518,"children":1520},{"className":1519},[],[1521],{"type":46,"value":97},{"type":46,"value":1523}," cross-links it.",{"type":40,"tag":209,"props":1525,"children":1526},{},[1527,1537,1539,1545],{"type":40,"tag":76,"props":1528,"children":1530},{"href":1529},"..\u002F..\u002Fdoca-version\u002FSKILL.md",[1531],{"type":40,"tag":68,"props":1532,"children":1534},{"className":1533},[],[1535],{"type":46,"value":1536},"doca-version",{"type":46,"value":1538}," — the version\ndetection \u002F four-way match rule every per-artifact ",{"type":40,"tag":68,"props":1540,"children":1542},{"className":1541},[],[1543],{"type":46,"value":1544},"## Version compatibility",{"type":46,"value":1546}," anchor builds on. This skill quotes\nthe RDMI-specific overlay only.",{"type":40,"tag":209,"props":1548,"children":1549},{},[1550,1560,1562,1568,1570,1575],{"type":40,"tag":76,"props":1551,"children":1553},{"href":1552},"..\u002F..\u002Fdoca-structured-tools-contract\u002FSKILL.md",[1554],{"type":40,"tag":68,"props":1555,"children":1557},{"className":1556},[],[1558],{"type":46,"value":1559},"doca-structured-tools-contract",{"type":46,"value":1561}," —\nthe JSON-schema contracts for the agent-preferred structured\nhelpers (env probe, capability snapshot, version-matrix\nlookup); the ",{"type":40,"tag":68,"props":1563,"children":1565},{"className":1564},[],[1566],{"type":46,"value":1567},"## Command appendix",{"type":46,"value":1569}," in ",{"type":40,"tag":68,"props":1571,"children":1573},{"className":1572},[],[1574],{"type":46,"value":78},{"type":46,"value":1576}," defers to\nthem before falling back to the manual chain.",{"items":1578,"total":1682},[1579,1596,1610,1624,1636,1653,1668],{"slug":1580,"name":1580,"fn":1581,"description":1582,"org":1583,"tags":1584,"stars":20,"repoUrl":21,"updatedAt":1595},"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},[1585,1588,1591,1592],{"name":1586,"slug":1587,"type":15},"Data Analysis","data-analysis",{"name":1589,"slug":1590,"type":15},"Data Engineering","data-engineering",{"name":9,"slug":8,"type":15},{"name":1593,"slug":1594,"type":15},"Performance","performance","2026-07-14T05:28:43.176466",{"slug":1597,"name":1597,"fn":1598,"description":1599,"org":1600,"tags":1601,"stars":20,"repoUrl":21,"updatedAt":1609},"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},[1602,1605,1608],{"name":1603,"slug":1604,"type":15},"Deployment","deployment",{"name":1606,"slug":1607,"type":15},"Infrastructure","infrastructure",{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":1611,"name":1611,"fn":1612,"description":1613,"org":1614,"tags":1615,"stars":20,"repoUrl":21,"updatedAt":1623},"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},[1616,1619,1620],{"name":1617,"slug":1618,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":1621,"slug":1622,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":1625,"name":1625,"fn":1626,"description":1627,"org":1628,"tags":1629,"stars":20,"repoUrl":21,"updatedAt":1635},"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},[1630,1631,1632],{"name":1586,"slug":1587,"type":15},{"name":9,"slug":8,"type":15},{"name":1633,"slug":1634,"type":15},"Testing","testing","2026-07-17T05:29:03.913266",{"slug":1637,"name":1637,"fn":1638,"description":1639,"org":1640,"tags":1641,"stars":20,"repoUrl":21,"updatedAt":1652},"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},[1642,1645,1648,1649],{"name":1643,"slug":1644,"type":15},"Automation","automation",{"name":1646,"slug":1647,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":1650,"slug":1651,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":1654,"name":1654,"fn":1655,"description":1656,"org":1657,"tags":1658,"stars":20,"repoUrl":21,"updatedAt":1667},"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},[1659,1660,1663,1664],{"name":1603,"slug":1604,"type":15},{"name":1661,"slug":1662,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":1665,"slug":1666,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":1669,"name":1669,"fn":1670,"description":1671,"org":1672,"tags":1673,"stars":20,"repoUrl":21,"updatedAt":1681},"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},[1674,1675,1678],{"name":9,"slug":8,"type":15},{"name":1676,"slug":1677,"type":15},"Quantum Computing","quantum-computing",{"name":1679,"slug":1680,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305,{"items":1684,"total":1835},[1685,1703,1719,1730,1742,1756,1769,1783,1794,1803,1817,1826],{"slug":1686,"name":1686,"fn":1687,"description":1688,"org":1689,"tags":1690,"stars":1700,"repoUrl":1701,"updatedAt":1702},"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},[1691,1694,1697],{"name":1692,"slug":1693,"type":15},"Documentation","documentation",{"name":1695,"slug":1696,"type":15},"MCP","mcp",{"name":1698,"slug":1699,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":1704,"name":1704,"fn":1705,"description":1706,"org":1707,"tags":1708,"stars":1716,"repoUrl":1717,"updatedAt":1718},"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},[1709,1712,1713],{"name":1710,"slug":1711,"type":15},"Containers","containers",{"name":1603,"slug":1604,"type":15},{"name":1714,"slug":1715,"type":15},"Python","python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":1720,"name":1720,"fn":1721,"description":1722,"org":1723,"tags":1724,"stars":1716,"repoUrl":1717,"updatedAt":1729},"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},[1725,1728],{"name":1726,"slug":1727,"type":15},"CI\u002FCD","ci-cd",{"name":1603,"slug":1604,"type":15},"2026-07-14T05:25:59.97109",{"slug":1731,"name":1731,"fn":1732,"description":1733,"org":1734,"tags":1735,"stars":1716,"repoUrl":1717,"updatedAt":1741},"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},[1736,1737,1738],{"name":1726,"slug":1727,"type":15},{"name":1603,"slug":1604,"type":15},{"name":1739,"slug":1740,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":1743,"name":1743,"fn":1744,"description":1745,"org":1746,"tags":1747,"stars":1716,"repoUrl":1717,"updatedAt":1755},"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},[1748,1751,1752],{"name":1749,"slug":1750,"type":15},"Debugging","debugging",{"name":1739,"slug":1740,"type":15},{"name":1753,"slug":1754,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":1757,"name":1757,"fn":1758,"description":1759,"org":1760,"tags":1761,"stars":1716,"repoUrl":1717,"updatedAt":1768},"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},[1762,1765],{"name":1763,"slug":1764,"type":15},"Best Practices","best-practices",{"name":1766,"slug":1767,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":1770,"name":1770,"fn":1771,"description":1772,"org":1773,"tags":1774,"stars":1716,"repoUrl":1717,"updatedAt":1782},"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},[1775,1778,1781],{"name":1776,"slug":1777,"type":15},"Machine Learning","machine-learning",{"name":1779,"slug":1780,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":1784,"name":1784,"fn":1785,"description":1786,"org":1787,"tags":1788,"stars":1716,"repoUrl":1717,"updatedAt":1793},"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},[1789,1792],{"name":1790,"slug":1791,"type":15},"QA","qa",{"name":1633,"slug":1634,"type":15},"2026-07-14T05:25:53.673039",{"slug":1795,"name":1795,"fn":1796,"description":1797,"org":1798,"tags":1799,"stars":1716,"repoUrl":1717,"updatedAt":1802},"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},[1800,1801],{"name":1603,"slug":1604,"type":15},{"name":1606,"slug":1607,"type":15},"2026-07-14T05:25:49.362534",{"slug":1804,"name":1804,"fn":1805,"description":1806,"org":1807,"tags":1808,"stars":1716,"repoUrl":1717,"updatedAt":1816},"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},[1809,1812,1813],{"name":1810,"slug":1811,"type":15},"Code Review","code-review",{"name":1739,"slug":1740,"type":15},{"name":1814,"slug":1815,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":1818,"name":1818,"fn":1819,"description":1820,"org":1821,"tags":1822,"stars":1716,"repoUrl":1717,"updatedAt":1825},"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},[1823,1824],{"name":1790,"slug":1791,"type":15},{"name":1633,"slug":1634,"type":15},"2026-07-14T05:25:54.928983",{"slug":1827,"name":1827,"fn":1828,"description":1829,"org":1830,"tags":1831,"stars":1716,"repoUrl":1717,"updatedAt":1834},"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},[1832,1833],{"name":1643,"slug":1644,"type":15},{"name":1726,"slug":1727,"type":15},"2026-07-30T05:29:03.275638",496]