[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-doca-sha-offload-engine":3,"mdc--7svqkx-key":34,"related-repo-nvidia-doca-sha-offload-engine":1589,"related-org-nvidia-doca-sha-offload-engine":1693},{"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-sha-offload-engine","offload SHA hashing to DOCA hardware","Use this skill when wiring the DOCA SHA Offload Engine (an OpenSSL ENGINE) into an existing OpenSSL pipeline to offload one-shot SHA-1, SHA-256, or SHA-512 (EVP_Digest) onto DOCA SHA hardware without rewriting against doca-sha. Covers engine load mechanics (`openssl engine dynamic`, `set_pci_addr` ctrl, `-engine_impl`), the SHA-224 negative test that proves offload engaged, the message-size window where offload beats CPU SHA, and engine-vs-library selection. Trigger even when the user does not say \"DOCA SHA Offload Engine\" or \"OpenSSL ENGINE\" — typical implicit phrasings: \"speed up openssl SHA on BlueField\", \"offload SHA without code changes\", \"is openssl using the accelerator or falling back to software\", \"prove DOCA SHA actually ran\", \"openssl dgst hashed but I'm not sure it was offloaded\". Refuse and route elsewhere for new SHA pipelines (use doca-sha), MD5 \u002F SHA-3 \u002F SHA-224 \u002F HMAC-SHA offload, incremental hashing via chained `EVP_DigestUpdate`, or OpenSSL PROVIDER authoring.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},"nvidia","NVIDIA","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fnvidia.png",[12,16,19,20],{"name":13,"slug":14,"type":15},"Performance","performance","tag",{"name":17,"slug":18,"type":15},"Cryptography","cryptography",{"name":9,"slug":8,"type":15},{"name":21,"slug":22,"type":15},"Engineering","engineering",2473,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills","2026-07-30T05:28:37.427547","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-sha-offload-engine","---\nlicense: Apache-2.0\nname: doca-sha-offload-engine\ndescription: >\n  Use this skill when wiring the DOCA SHA Offload Engine\n  (an OpenSSL ENGINE) into an existing OpenSSL pipeline\n  to offload one-shot SHA-1, SHA-256, or SHA-512\n  (EVP_Digest) onto DOCA SHA hardware without rewriting\n  against doca-sha. Covers engine load mechanics\n  (`openssl engine dynamic`, `set_pci_addr` ctrl,\n  `-engine_impl`), the SHA-224 negative test that proves\n  offload engaged, the message-size window where offload\n  beats CPU SHA, and engine-vs-library selection.\n  Trigger even when the user does not say \"DOCA SHA\n  Offload Engine\" or \"OpenSSL ENGINE\" — typical implicit\n  phrasings: \"speed up openssl SHA on BlueField\",\n  \"offload SHA without code changes\", \"is openssl using\n  the accelerator or falling back to software\", \"prove\n  DOCA SHA actually ran\", \"openssl dgst hashed but I'm\n  not sure it was offloaded\". Refuse and route elsewhere\n  for new SHA pipelines (use doca-sha), MD5 \u002F SHA-3 \u002F\n  SHA-224 \u002F HMAC-SHA offload, incremental hashing via\n  chained `EVP_DigestUpdate`, or OpenSSL PROVIDER authoring.\nmetadata:\n  kind: tool\ncompatibility: >\n  Requires DOCA SDK installed at \u002Fopt\u002Fmellanox\u002Fdoca on\n  Linux (Ubuntu 22.04\u002F24.04 or RHEL\u002FSLES) with a\n  BlueField DPU or ConnectX NIC attached, plus OpenSSL\n  ≥ 1.1.1 and libssl-dev (or distro equivalent) on the\n  build host. The engine ships under\n  \u002Fopt\u002Fmellanox\u002Fdoca\u002Ftools\u002Fdoca_sha_offload_engine\u002F as\n  libdoca_sha_offload_engine.so; reads the user's local\n  install via `pkg-config doca-sha` and inspects\n  \u002Fopt\u002Fmellanox\u002Fdoca\u002F{lib,include,samples,applications}.\n---\n\n# DOCA SHA Offload Engine\n\n**Where to start:** This is a tool skill for the OpenSSL\nENGINE shipped in the DOCA SOURCE tree under\n`doca\u002Ftools\u002Fsha_offload_engine\u002F` and INSTALLED on the host\nunder `${DOCA_DIR}\u002Ftools\u002Fdoca_sha_offload_engine\u002F` as\n`libdoca_sha_offload_engine.so`. The directory-name shift\n(`sha_offload_engine` in the source layout vs\n`doca_sha_offload_engine` in the install layout) is an\nNVIDIA packaging convention, not a bundle inconsistency;\nboth forms appear below and are the same artifact at\ndifferent lifecycle stages — quote whichever the prompt is\nabout (build-from-source vs runtime-load). It is **not a CLI** —\nit is a shared object loaded by an OpenSSL-based\napplication or by `openssl` itself, that re-routes SHA-1 \u002F\nSHA-256 \u002F SHA-512 (one-shot only, via the `EVP_Digest`\ninterface) onto the DOCA SHA hardware path. Open\n[`TASKS.md`](TASKS.md) and start at\n[`## configure`](TASKS.md#configure) for the PCIe-address\nconfiguration and the OpenSSL prerequisites; jump to\n[`## run`](TASKS.md#run) for the *\"load the engine and\nprove it actually runs\"* flow. Open\n[`CAPABILITIES.md`](CAPABILITIES.md) when the question is\n*what the engine actually offloads vs falls back to*,\n*when the engine is a perf win vs not*, or *how to verify\noffload actually engaged*. If DOCA is not installed yet,\nroute to [`doca-setup`](..\u002F..\u002Fdoca-setup\u002FSKILL.md) first.\nIf the user is building a new SHA pipeline from scratch\n(not wrapping an existing OpenSSL-based one), this skill\nis the wrong surface — route to\n[`..\u002F..\u002Flibs\u002Fdoca-sha\u002FSKILL.md`](..\u002F..\u002Flibs\u002Fdoca-sha\u002FSKILL.md)\ninstead.\n\n## Example questions this skill answers well\n\nThe CLASSES of `doca-sha-offload-engine` questions this\nskill is built to answer, each with one worked example.\nThe class is the load-bearing piece; the worked example\nis one instance.\n\n- **\"I have an existing OpenSSL-based pipeline doing SHA;\n  can I offload the SHA to DOCA without rewriting the\n  app?\"** — worked example: *\"the app uses\n  `EVP_DigestInit_ex` \u002F `EVP_DigestUpdate` \u002F\n  `EVP_DigestFinal_ex` against `EVP_sha256()`; can I\n  drop in DOCA-SHA offload via an engine load?\"*.\n  Answered by the *\"when this engine is the right\n  surface\"* rule in\n  [`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes)\n  + the engine-load mechanics in\n  [`TASKS.md ## configure`](TASKS.md#configure).\n- **\"Did the engine actually load? Or did OpenSSL just\n  fall back to software SHA?\"** — worked example: *\"my\n  `openssl dgst` invocation completed; how do I know\n  DOCA SHA actually did the work and OpenSSL did not\n  silently use the software path?\"*. Answered by the\n  *\"prove the engine actually ran\"* pattern in\n  [`CAPABILITIES.md ## Observability`](CAPABILITIES.md#observability)\n  (the SHA-224 negative test and the `-engine_impl`\n  flag) + the verification flow in\n  [`TASKS.md ## test`](TASKS.md#test).\n- **\"For what message-size range is the engine offload a\n  win vs CPU SHA?\"** — worked example: *\"my pipeline\n  hashes 4 KB blocks at a time; is the engine a win\n  there, or does the round-trip to DOCA SHA cost more\n  than the CPU hash itself?\"*. Answered by the\n  message-size-window rule in\n  [`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes)\n  + the `openssl speed` perf-comparison pattern in\n  [`TASKS.md ## test`](TASKS.md#test).\n- **\"What SHA algorithms does the engine actually\n  support — and what happens for the ones it does not?\"**\n  — worked example: *\"my pipeline mixes SHA-1, SHA-256,\n  SHA-512, and SHA-224 — what does the engine do for\n  each?\"*. Answered by the algorithm-coverage matrix in\n  [`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes).\n- **\"Should I use this engine, or call doca-sha\n  directly?\"** — worked example: *\"I am writing a new\n  service from scratch; does the engine save me work or\n  does it add complexity I do not need?\"*. Answered by\n  the *\"engine vs library\"* selection table in\n  [`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes).\n- **\"What OpenSSL versions does the engine require, and\n  what about OpenSSL 3.x's deprecation of the ENGINE\n  API?\"** — worked example: *\"my host has OpenSSL 3.0;\n  will the engine still load?\"*. Answered by the version\n  overlay in\n  [`CAPABILITIES.md ## Version compatibility`](CAPABILITIES.md#version-compatibility)\n  (verified surface: the engine is documented for\n  OpenSSL 1.1.1f on Ubuntu 20.04 and OpenSSL 3.0.2 on\n  Ubuntu 22.04 per the shipped `readme.md`).\n\n## Audience\n\nThis skill serves **external developers and operators who\nhave an existing OpenSSL-based pipeline doing SHA hashing\nand want to offload SHA onto DOCA SHA without rewriting\ntheir application against the doca-sha C API**.\nConcretely:\n\n- A developer integrating DOCA SHA acceleration into a\n  service that already uses `EVP_Digest` for SHA-1 \u002F\n  SHA-256 \u002F SHA-512.\n- An operator deploying an `openssl dgst` \u002F `openssl\n  speed` based pipeline and wanting to measure the win\n  from DOCA SHA offload without changing the pipeline's\n  invocation surface.\n- An SRE \u002F performance engineer producing a *\"this is\n  the engine-offload win vs CPU SHA on this message-size\n  mix\"* artifact to inform a code-change decision (e.g.\n  *\"should we adopt the engine as-is, or rewrite to\n  doca-sha for finer control?\"*).\n- An AI agent answering *\"can I drop DOCA SHA into this\n  OpenSSL-based app without code changes\"* honestly —\n  with the verified algorithm-coverage matrix, the\n  message-size window, and the verification pattern that\n  proves the engine actually ran.\n\nIt is **not** for users building a new SHA pipeline from\nscratch (route to\n[`..\u002F..\u002Flibs\u002Fdoca-sha\u002FSKILL.md`](..\u002F..\u002Flibs\u002Fdoca-sha\u002FSKILL.md)),\n**not** for users wanting MD5 \u002F SHA-2-224 \u002F SHA-3 \u002F\nHMAC-SHA offload (the engine does not implement those —\nthe verified surface per the engine's source is one-shot\nSHA-1, SHA-256, SHA-512 via `EVP_Digest`), and **not** a\nsubstitute for the public DOCA SHA programming guide.\n\n## Language scope\n\nThe DOCA SHA Offload Engine is shipped as a **C dynamic\nshared object** (`libdoca_sha_offload_engine.so`) built\nfrom `doca\u002Ftools\u002Fsha_offload_engine\u002F{engine\u002Fdoca_sha_offload_engine.c,\nlib\u002Fdoca_sha_offload_lib.{c,h}}` via `meson`. Its\n*consumer* interface is OpenSSL's ENGINE API; any\nlanguage that calls OpenSSL (C, C++, Rust via `openssl`\ncrate, Python via `cryptography` and `pyca\u002Fcryptography`'s\nbackend, Node via `node:crypto`) can therefore *use* the\nengine, provided the language binding either calls\n`ENGINE_load_dynamic` \u002F `ENGINE_by_id` directly or honors\nan OpenSSL `engines` config that loads it. The skill's\nlanguage-neutral contribution is the engine-load mechanics\nand the verification pattern; the OpenSSL ENGINE API is\nthe contract.\n\n## When to load this skill\n\nLoad this skill when the user is — or the agent needs to\n— deploy the DOCA SHA Offload Engine into an OpenSSL-based\npipeline on a host with DOCA installed and a SHA-capable\ndevice. Concretely:\n\n- Wiring the engine into an existing OpenSSL-based\n  application or `openssl` CLI invocation, with the\n  intent of no source-level code changes (or only the\n  minimum `ENGINE_load_dynamic` \u002F `ENGINE_by_id` block\n  shown in the verified `readme.md`).\n- Verifying the engine actually engaged for the\n  hot-path digests (the *\"is offload real or did OpenSSL\n  fall back to software\"* question).\n- Characterizing the message-size range over which the\n  engine is a perf win vs the CPU `sha1` \u002F `sha256` \u002F\n  `sha512` paths.\n- Deciding between *\"adopt the engine\"* (existing\n  pipeline, minimal change) and *\"rewrite to doca-sha\"*\n  (new pipeline, fine-grained control needed).\n\nDo **not** load this skill for users building a new\nSHA-pipeline from scratch — route to\n[`..\u002F..\u002Flibs\u002Fdoca-sha\u002FSKILL.md`](..\u002F..\u002Flibs\u002Fdoca-sha\u002FSKILL.md).\nDo not load this skill for users wanting algorithms the\nengine does not implement (MD5, SHA-3, SHA-224, HMAC-SHA,\nstreaming\u002Fincremental SHA via `EVP_DigestUpdate` chains\nthat the engine treats as one-shot only).\n\n## What this skill provides\n\nThis is a **thin loader**. Substantive material lives in\ntwo companion files:\n\n- `CAPABILITIES.md` — what the engine offloads (verified:\n  one-shot SHA-1, SHA-256, SHA-512 via the OpenSSL\n  `EVP_Digest` high-level interface), what it does NOT\n  offload (anything else — including SHA-224, MD5,\n  SHA-3, HMAC-SHA, and any chained\n  `EVP_DigestInit_ex` \u002F `EVP_DigestUpdate` \u002F\n  `EVP_DigestFinal_ex` pattern that the engine implements\n  by buffering and then calling DOCA SHA in one shot at\n  `Final`), the engine-vs-library selection rule, the\n  message-size-window rule for when offload is a perf\n  win, the verified ctrl-cmd surface (`set_pci_addr`),\n  the version overlay (OpenSSL ≥ 1.1.1; the engine's\n  shipped tests cover OpenSSL 1.1.1f and OpenSSL 3.0.2\n  per the `readme.md`; OpenSSL 3.x deprecates the ENGINE\n  API but still supports it via the legacy code path),\n  the layered error taxonomy, the observability surface\n  (the *\"prove offload engaged\"* pattern using the\n  SHA-224 negative test and `-engine_impl`), and the\n  safety overlay.\n- `TASKS.md` — step-by-step workflows for the in-scope\n  task verbs: `install`, `configure` (PCIe address\n  selection — the engine defaults to `03:00.0` and\n  exposes the `set_pci_addr` ctrl-cmd plus a build-time\n  override per the shipped `test_cmdline_mode\u002Freadme.md`),\n  `build` (the `meson` flow), `modify` (refuses source\n  patching; modify the load-time invocation and the\n  PCIe address instead), `run` (load the engine via\n  `openssl engine dynamic`; the verification pattern;\n  the OpenSSL programmatic `ENGINE_load_dynamic` block),\n  `test` (the *\"prove the engine ran\"* SHA-224 negative\n  test; the `openssl speed` perf comparison; the\n  message-size window characterization), `debug` (walk\n  the error taxonomy layer by layer), `use` (the\n  engine-vs-library decision for the user's specific\n  pipeline), plus a `Deferred task verbs` block.\n\nThe skill assumes a host where DOCA is already installed,\nOpenSSL ≥ 1.1.1 is present (`libssl-dev` or equivalent),\nand the deploying user can access the selected DOCA SHA PCIe device. Verify\ndevice visibility and run the engine-load smoke as that same user before\nintegration; if either fails with a permission error, stop and route to\n`doca-setup` rather than guessing a group, ACL, or `sudo` policy.\n\n## What this skill deliberately does not ship\n\nThis skill is **agent guidance**, not a samples or\nscripts bundle. It deliberately does not contain — and\npull requests should not add:\n\n- **Pre-written OpenSSL ENGINE application source code**\n  beyond the verbatim verified block in the shipped\n  `readme.md` (the `ENGINE_load_dynamic` \u002F\n  `ENGINE_by_id` \u002F `ENGINE_ctrl_cmd_string` \u002F\n  `ENGINE_init` \u002F `ENGINE_set_default_digests` sequence,\n  cross-referenced into [`TASKS.md ## run`](TASKS.md#run)).\n  The shipped readme is the worked example.\n- **A `samples\u002F`, `bindings\u002F`, or `reference\u002F`\n  subtree.** This is a thin loader for a shipped\n  shared-object; substantive material lives in the\n  shipped `readme.md` and the doca-sha library docs.\n- **Performance numbers from agent memory.** The\n  message-size-window where the engine wins is\n  device-, firmware-, OpenSSL-version-, and\n  workload-specific. The `openssl speed` comparison\n  pattern in [`TASKS.md ## test`](TASKS.md#test) is the\n  way to capture it on the user's actual hardware;\n  quoting a number from memory is the cross-platform\n  failure mode this skill exists to prevent.\n- **Wrappers, parsers, or scripts** in any language\n  that consume the engine's stdout or the `openssl\n  speed` output format.\n\n## Loading order\n\n1. Read this `SKILL.md` first to confirm the user's\n   question is in scope (the user actually has an\n   existing OpenSSL-based pipeline and wants to offload\n   to DOCA SHA *without* code changes; if the user is\n   building from scratch, route to\n   [`..\u002F..\u002Flibs\u002Fdoca-sha\u002F`](..\u002F..\u002Flibs\u002Fdoca-sha\u002FSKILL.md)).\n2. **For what the engine offloads vs falls back, the\n   engine-vs-library selection rule, the message-size-\n   window rule, the version overlay, the error taxonomy,\n   the observability surface (and the prove-offload-\n   actually-engaged pattern), and the safety overlay,\n   see [CAPABILITIES.md](CAPABILITIES.md).**\n3. **For step-by-step workflows — `install`,\n   `configure`, `build`, `modify`, `run`, `test`,\n   `debug`, `use` — see [TASKS.md](TASKS.md).**\n\n## Related skills\n\n- [`..\u002F..\u002Flibs\u002Fdoca-sha\u002FSKILL.md`](..\u002F..\u002Flibs\u002Fdoca-sha\u002FSKILL.md) —\n  the underlying DOCA SHA library. The engine is a\n  thin OpenSSL-ENGINE wrapper around doca-sha; when the\n  user needs fine-grained control over the SHA task\n  surface (partial-hash, custom buffer permissions,\n  cap-query for unusual message sizes), the library is\n  the right answer. The engine wraps the *one-shot*\n  task; the library exposes both one-shot and partial-\n  hash per\n  [`..\u002F..\u002Flibs\u002Fdoca-sha\u002FCAPABILITIES.md#capabilities-and-modes`](..\u002F..\u002Flibs\u002Fdoca-sha\u002FCAPABILITIES.md#capabilities-and-modes).\n- [`doca-version`](..\u002F..\u002Fdoca-version\u002FSKILL.md) — the\n  canonical version-detection chain. The engine has a\n  TWO-axis version overlay (DOCA-side and OpenSSL-side);\n  the version skill carries the four-way match rule\n  this skill layers on top of.\n- [`doca-debug`](..\u002F..\u002Fdoca-debug\u002FSKILL.md) — the\n  cross-cutting debug ladder. The engine surfaces its\n  own error taxonomy; when the cause is below DOCA\n  (driver, firmware), the taxonomy hands off here.\n- [`doca-setup`](..\u002F..\u002Fdoca-setup\u002FSKILL.md) — env\n  preparation, install verification, the `libssl-dev`\n  install path, and the NGC DOCA container path.\n- [`doca-public-knowledge-map`](..\u002F..\u002Fdoca-public-knowledge-map\u002FSKILL.md) —\n  routing to the public DOCA SHA documentation set on\n  `docs.nvidia.com\u002Fdoca\u002Fsdk\u002F` and to the OpenSSL\n  ENGINE \u002F `openssl-engine` upstream documentation on\n  `openssl.org`.\n- [`doca-hardware-safety`](..\u002F..\u002Fdoca-hardware-safety\u002FSKILL.md) —\n  the bundle-wide hardware-safety meta-policy. The\n  engine binds to a specific PCIe device; the\n  `set_pci_addr` ctrl is the artifact-specific overlay,\n  but any host-side change underneath (firmware burn,\n  BFB reflash) runs through the meta-policy.\n",{"data":35,"body":39},{"license":26,"name":4,"description":6,"metadata":36,"compatibility":38},{"kind":37},"tool","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 OpenSSL ≥ 1.1.1 and libssl-dev (or distro equivalent) on the build host. The engine ships under \u002Fopt\u002Fmellanox\u002Fdoca\u002Ftools\u002Fdoca_sha_offload_engine\u002F as libdoca_sha_offload_engine.so; reads the user's local install via `pkg-config doca-sha` and inspects \u002Fopt\u002Fmellanox\u002Fdoca\u002F{lib,include,samples,applications}.\n",{"type":40,"children":41},"root",[42,50,225,232,244,539,545,557,620,661,667,769,775,780,872,900,906,918,1132,1160,1166,1178,1325,1331,1440,1446],{"type":43,"tag":44,"props":45,"children":46},"element","h1",{"id":4},[47],{"type":48,"value":49},"text","DOCA SHA Offload Engine",{"type":43,"tag":51,"props":52,"children":53},"p",{},[54,60,62,69,71,77,79,85,87,93,95,101,103,108,110,116,118,124,126,136,138,148,150,160,162,168,170,179,181,186,188,193,195,200,202,212,214,223],{"type":43,"tag":55,"props":56,"children":57},"strong",{},[58],{"type":48,"value":59},"Where to start:",{"type":48,"value":61}," This is a tool skill for the OpenSSL\nENGINE shipped in the DOCA SOURCE tree under\n",{"type":43,"tag":63,"props":64,"children":66},"code",{"className":65},[],[67],{"type":48,"value":68},"doca\u002Ftools\u002Fsha_offload_engine\u002F",{"type":48,"value":70}," and INSTALLED on the host\nunder ",{"type":43,"tag":63,"props":72,"children":74},{"className":73},[],[75],{"type":48,"value":76},"${DOCA_DIR}\u002Ftools\u002Fdoca_sha_offload_engine\u002F",{"type":48,"value":78}," as\n",{"type":43,"tag":63,"props":80,"children":82},{"className":81},[],[83],{"type":48,"value":84},"libdoca_sha_offload_engine.so",{"type":48,"value":86},". The directory-name shift\n(",{"type":43,"tag":63,"props":88,"children":90},{"className":89},[],[91],{"type":48,"value":92},"sha_offload_engine",{"type":48,"value":94}," in the source layout vs\n",{"type":43,"tag":63,"props":96,"children":98},{"className":97},[],[99],{"type":48,"value":100},"doca_sha_offload_engine",{"type":48,"value":102}," in the install layout) is an\nNVIDIA packaging convention, not a bundle inconsistency;\nboth forms appear below and are the same artifact at\ndifferent lifecycle stages — quote whichever the prompt is\nabout (build-from-source vs runtime-load). It is ",{"type":43,"tag":55,"props":104,"children":105},{},[106],{"type":48,"value":107},"not a CLI",{"type":48,"value":109}," —\nit is a shared object loaded by an OpenSSL-based\napplication or by ",{"type":43,"tag":63,"props":111,"children":113},{"className":112},[],[114],{"type":48,"value":115},"openssl",{"type":48,"value":117}," itself, that re-routes SHA-1 \u002F\nSHA-256 \u002F SHA-512 (one-shot only, via the ",{"type":43,"tag":63,"props":119,"children":121},{"className":120},[],[122],{"type":48,"value":123},"EVP_Digest",{"type":48,"value":125},"\ninterface) onto the DOCA SHA hardware path. Open\n",{"type":43,"tag":127,"props":128,"children":130},"a",{"href":129},"TASKS.md",[131],{"type":43,"tag":63,"props":132,"children":134},{"className":133},[],[135],{"type":48,"value":129},{"type":48,"value":137}," and start at\n",{"type":43,"tag":127,"props":139,"children":141},{"href":140},"TASKS.md#configure",[142],{"type":43,"tag":63,"props":143,"children":145},{"className":144},[],[146],{"type":48,"value":147},"## configure",{"type":48,"value":149}," for the PCIe-address\nconfiguration and the OpenSSL prerequisites; jump to\n",{"type":43,"tag":127,"props":151,"children":153},{"href":152},"TASKS.md#run",[154],{"type":43,"tag":63,"props":155,"children":157},{"className":156},[],[158],{"type":48,"value":159},"## run",{"type":48,"value":161}," for the ",{"type":43,"tag":163,"props":164,"children":165},"em",{},[166],{"type":48,"value":167},"\"load the engine and\nprove it actually runs\"",{"type":48,"value":169}," flow. Open\n",{"type":43,"tag":127,"props":171,"children":173},{"href":172},"CAPABILITIES.md",[174],{"type":43,"tag":63,"props":175,"children":177},{"className":176},[],[178],{"type":48,"value":172},{"type":48,"value":180}," when the question is\n",{"type":43,"tag":163,"props":182,"children":183},{},[184],{"type":48,"value":185},"what the engine actually offloads vs falls back to",{"type":48,"value":187},",\n",{"type":43,"tag":163,"props":189,"children":190},{},[191],{"type":48,"value":192},"when the engine is a perf win vs not",{"type":48,"value":194},", or ",{"type":43,"tag":163,"props":196,"children":197},{},[198],{"type":48,"value":199},"how to verify\noffload actually engaged",{"type":48,"value":201},". If DOCA is not installed yet,\nroute to ",{"type":43,"tag":127,"props":203,"children":205},{"href":204},"..\u002F..\u002Fdoca-setup\u002FSKILL.md",[206],{"type":43,"tag":63,"props":207,"children":209},{"className":208},[],[210],{"type":48,"value":211},"doca-setup",{"type":48,"value":213}," first.\nIf the user is building a new SHA pipeline from scratch\n(not wrapping an existing OpenSSL-based one), this skill\nis the wrong surface — route to\n",{"type":43,"tag":127,"props":215,"children":217},{"href":216},"..\u002F..\u002Flibs\u002Fdoca-sha\u002FSKILL.md",[218],{"type":43,"tag":63,"props":219,"children":221},{"className":220},[],[222],{"type":48,"value":216},{"type":48,"value":224},"\ninstead.",{"type":43,"tag":226,"props":227,"children":229},"h2",{"id":228},"example-questions-this-skill-answers-well",[230],{"type":48,"value":231},"Example questions this skill answers well",{"type":43,"tag":51,"props":233,"children":234},{},[235,237,242],{"type":48,"value":236},"The CLASSES of ",{"type":43,"tag":63,"props":238,"children":240},{"className":239},[],[241],{"type":48,"value":4},{"type":48,"value":243}," questions this\nskill is built to answer, each with one worked example.\nThe class is the load-bearing piece; the worked example\nis one instance.",{"type":43,"tag":245,"props":246,"children":247},"ul",{},[248,334,396,445,471,503],{"type":43,"tag":249,"props":250,"children":251},"li",{},[252,257,259,296,298,303,305,315],{"type":43,"tag":55,"props":253,"children":254},{},[255],{"type":48,"value":256},"\"I have an existing OpenSSL-based pipeline doing SHA;\ncan I offload the SHA to DOCA without rewriting the\napp?\"",{"type":48,"value":258}," — worked example: ",{"type":43,"tag":163,"props":260,"children":261},{},[262,264,270,272,278,280,286,288,294],{"type":48,"value":263},"\"the app uses\n",{"type":43,"tag":63,"props":265,"children":267},{"className":266},[],[268],{"type":48,"value":269},"EVP_DigestInit_ex",{"type":48,"value":271}," \u002F ",{"type":43,"tag":63,"props":273,"children":275},{"className":274},[],[276],{"type":48,"value":277},"EVP_DigestUpdate",{"type":48,"value":279}," \u002F\n",{"type":43,"tag":63,"props":281,"children":283},{"className":282},[],[284],{"type":48,"value":285},"EVP_DigestFinal_ex",{"type":48,"value":287}," against ",{"type":43,"tag":63,"props":289,"children":291},{"className":290},[],[292],{"type":48,"value":293},"EVP_sha256()",{"type":48,"value":295},"; can I\ndrop in DOCA-SHA offload via an engine load?\"",{"type":48,"value":297},".\nAnswered by the ",{"type":43,"tag":163,"props":299,"children":300},{},[301],{"type":48,"value":302},"\"when this engine is the right\nsurface\"",{"type":48,"value":304}," rule in\n",{"type":43,"tag":127,"props":306,"children":308},{"href":307},"CAPABILITIES.md#capabilities-and-modes",[309],{"type":43,"tag":63,"props":310,"children":312},{"className":311},[],[313],{"type":48,"value":314},"CAPABILITIES.md ## Capabilities and modes",{"type":43,"tag":245,"props":316,"children":317},{},[318],{"type":43,"tag":249,"props":319,"children":320},{},[321,323,332],{"type":48,"value":322},"the engine-load mechanics in\n",{"type":43,"tag":127,"props":324,"children":325},{"href":140},[326],{"type":43,"tag":63,"props":327,"children":329},{"className":328},[],[330],{"type":48,"value":331},"TASKS.md ## configure",{"type":48,"value":333},".",{"type":43,"tag":249,"props":335,"children":336},{},[337,342,343,356,358,363,365,375,377,383,385,395],{"type":43,"tag":55,"props":338,"children":339},{},[340],{"type":48,"value":341},"\"Did the engine actually load? Or did OpenSSL just\nfall back to software SHA?\"",{"type":48,"value":258},{"type":43,"tag":163,"props":344,"children":345},{},[346,348,354],{"type":48,"value":347},"\"my\n",{"type":43,"tag":63,"props":349,"children":351},{"className":350},[],[352],{"type":48,"value":353},"openssl dgst",{"type":48,"value":355}," invocation completed; how do I know\nDOCA SHA actually did the work and OpenSSL did not\nsilently use the software path?\"",{"type":48,"value":357},". Answered by the\n",{"type":43,"tag":163,"props":359,"children":360},{},[361],{"type":48,"value":362},"\"prove the engine actually ran\"",{"type":48,"value":364}," pattern in\n",{"type":43,"tag":127,"props":366,"children":368},{"href":367},"CAPABILITIES.md#observability",[369],{"type":43,"tag":63,"props":370,"children":372},{"className":371},[],[373],{"type":48,"value":374},"CAPABILITIES.md ## Observability",{"type":48,"value":376},"\n(the SHA-224 negative test and the ",{"type":43,"tag":63,"props":378,"children":380},{"className":379},[],[381],{"type":48,"value":382},"-engine_impl",{"type":48,"value":384},"\nflag) + the verification flow in\n",{"type":43,"tag":127,"props":386,"children":388},{"href":387},"TASKS.md#test",[389],{"type":43,"tag":63,"props":390,"children":392},{"className":391},[],[393],{"type":48,"value":394},"TASKS.md ## test",{"type":48,"value":333},{"type":43,"tag":249,"props":397,"children":398},{},[399,404,405,410,412,420],{"type":43,"tag":55,"props":400,"children":401},{},[402],{"type":48,"value":403},"\"For what message-size range is the engine offload a\nwin vs CPU SHA?\"",{"type":48,"value":258},{"type":43,"tag":163,"props":406,"children":407},{},[408],{"type":48,"value":409},"\"my pipeline\nhashes 4 KB blocks at a time; is the engine a win\nthere, or does the round-trip to DOCA SHA cost more\nthan the CPU hash itself?\"",{"type":48,"value":411},". Answered by the\nmessage-size-window rule in\n",{"type":43,"tag":127,"props":413,"children":414},{"href":307},[415],{"type":43,"tag":63,"props":416,"children":418},{"className":417},[],[419],{"type":48,"value":314},{"type":43,"tag":245,"props":421,"children":422},{},[423],{"type":43,"tag":249,"props":424,"children":425},{},[426,428,434,436,444],{"type":48,"value":427},"the ",{"type":43,"tag":63,"props":429,"children":431},{"className":430},[],[432],{"type":48,"value":433},"openssl speed",{"type":48,"value":435}," perf-comparison pattern in\n",{"type":43,"tag":127,"props":437,"children":438},{"href":387},[439],{"type":43,"tag":63,"props":440,"children":442},{"className":441},[],[443],{"type":48,"value":394},{"type":48,"value":333},{"type":43,"tag":249,"props":446,"children":447},{},[448,453,455,460,462,470],{"type":43,"tag":55,"props":449,"children":450},{},[451],{"type":48,"value":452},"\"What SHA algorithms does the engine actually\nsupport — and what happens for the ones it does not?\"",{"type":48,"value":454},"\n— worked example: ",{"type":43,"tag":163,"props":456,"children":457},{},[458],{"type":48,"value":459},"\"my pipeline mixes SHA-1, SHA-256,\nSHA-512, and SHA-224 — what does the engine do for\neach?\"",{"type":48,"value":461},". Answered by the algorithm-coverage matrix in\n",{"type":43,"tag":127,"props":463,"children":464},{"href":307},[465],{"type":43,"tag":63,"props":466,"children":468},{"className":467},[],[469],{"type":48,"value":314},{"type":48,"value":333},{"type":43,"tag":249,"props":472,"children":473},{},[474,479,480,485,487,492,494,502],{"type":43,"tag":55,"props":475,"children":476},{},[477],{"type":48,"value":478},"\"Should I use this engine, or call doca-sha\ndirectly?\"",{"type":48,"value":258},{"type":43,"tag":163,"props":481,"children":482},{},[483],{"type":48,"value":484},"\"I am writing a new\nservice from scratch; does the engine save me work or\ndoes it add complexity I do not need?\"",{"type":48,"value":486},". Answered by\nthe ",{"type":43,"tag":163,"props":488,"children":489},{},[490],{"type":48,"value":491},"\"engine vs library\"",{"type":48,"value":493}," selection table in\n",{"type":43,"tag":127,"props":495,"children":496},{"href":307},[497],{"type":43,"tag":63,"props":498,"children":500},{"className":499},[],[501],{"type":48,"value":314},{"type":48,"value":333},{"type":43,"tag":249,"props":504,"children":505},{},[506,511,512,517,519,529,531,537],{"type":43,"tag":55,"props":507,"children":508},{},[509],{"type":48,"value":510},"\"What OpenSSL versions does the engine require, and\nwhat about OpenSSL 3.x's deprecation of the ENGINE\nAPI?\"",{"type":48,"value":258},{"type":43,"tag":163,"props":513,"children":514},{},[515],{"type":48,"value":516},"\"my host has OpenSSL 3.0;\nwill the engine still load?\"",{"type":48,"value":518},". Answered by the version\noverlay in\n",{"type":43,"tag":127,"props":520,"children":522},{"href":521},"CAPABILITIES.md#version-compatibility",[523],{"type":43,"tag":63,"props":524,"children":526},{"className":525},[],[527],{"type":48,"value":528},"CAPABILITIES.md ## Version compatibility",{"type":48,"value":530},"\n(verified surface: the engine is documented for\nOpenSSL 1.1.1f on Ubuntu 20.04 and OpenSSL 3.0.2 on\nUbuntu 22.04 per the shipped ",{"type":43,"tag":63,"props":532,"children":534},{"className":533},[],[535],{"type":48,"value":536},"readme.md",{"type":48,"value":538},").",{"type":43,"tag":226,"props":540,"children":542},{"id":541},"audience",[543],{"type":48,"value":544},"Audience",{"type":43,"tag":51,"props":546,"children":547},{},[548,550,555],{"type":48,"value":549},"This skill serves ",{"type":43,"tag":55,"props":551,"children":552},{},[553],{"type":48,"value":554},"external developers and operators who\nhave an existing OpenSSL-based pipeline doing SHA hashing\nand want to offload SHA onto DOCA SHA without rewriting\ntheir application against the doca-sha C API",{"type":48,"value":556},".\nConcretely:",{"type":43,"tag":245,"props":558,"children":559},{},[560,572,590,608],{"type":43,"tag":249,"props":561,"children":562},{},[563,565,570],{"type":48,"value":564},"A developer integrating DOCA SHA acceleration into a\nservice that already uses ",{"type":43,"tag":63,"props":566,"children":568},{"className":567},[],[569],{"type":48,"value":123},{"type":48,"value":571}," for SHA-1 \u002F\nSHA-256 \u002F SHA-512.",{"type":43,"tag":249,"props":573,"children":574},{},[575,577,582,583,588],{"type":48,"value":576},"An operator deploying an ",{"type":43,"tag":63,"props":578,"children":580},{"className":579},[],[581],{"type":48,"value":353},{"type":48,"value":271},{"type":43,"tag":63,"props":584,"children":586},{"className":585},[],[587],{"type":48,"value":433},{"type":48,"value":589}," based pipeline and wanting to measure the win\nfrom DOCA SHA offload without changing the pipeline's\ninvocation surface.",{"type":43,"tag":249,"props":591,"children":592},{},[593,595,600,602,607],{"type":48,"value":594},"An SRE \u002F performance engineer producing a ",{"type":43,"tag":163,"props":596,"children":597},{},[598],{"type":48,"value":599},"\"this is\nthe engine-offload win vs CPU SHA on this message-size\nmix\"",{"type":48,"value":601}," artifact to inform a code-change decision (e.g.\n",{"type":43,"tag":163,"props":603,"children":604},{},[605],{"type":48,"value":606},"\"should we adopt the engine as-is, or rewrite to\ndoca-sha for finer control?\"",{"type":48,"value":538},{"type":43,"tag":249,"props":609,"children":610},{},[611,613,618],{"type":48,"value":612},"An AI agent answering ",{"type":43,"tag":163,"props":614,"children":615},{},[616],{"type":48,"value":617},"\"can I drop DOCA SHA into this\nOpenSSL-based app without code changes\"",{"type":48,"value":619}," honestly —\nwith the verified algorithm-coverage matrix, the\nmessage-size window, and the verification pattern that\nproves the engine actually ran.",{"type":43,"tag":51,"props":621,"children":622},{},[623,625,630,632,640,642,646,648,653,655,659],{"type":48,"value":624},"It is ",{"type":43,"tag":55,"props":626,"children":627},{},[628],{"type":48,"value":629},"not",{"type":48,"value":631}," for users building a new SHA pipeline from\nscratch (route to\n",{"type":43,"tag":127,"props":633,"children":634},{"href":216},[635],{"type":43,"tag":63,"props":636,"children":638},{"className":637},[],[639],{"type":48,"value":216},{"type":48,"value":641},"),\n",{"type":43,"tag":55,"props":643,"children":644},{},[645],{"type":48,"value":629},{"type":48,"value":647}," for users wanting MD5 \u002F SHA-2-224 \u002F SHA-3 \u002F\nHMAC-SHA offload (the engine does not implement those —\nthe verified surface per the engine's source is one-shot\nSHA-1, SHA-256, SHA-512 via ",{"type":43,"tag":63,"props":649,"children":651},{"className":650},[],[652],{"type":48,"value":123},{"type":48,"value":654},"), and ",{"type":43,"tag":55,"props":656,"children":657},{},[658],{"type":48,"value":629},{"type":48,"value":660}," a\nsubstitute for the public DOCA SHA programming guide.",{"type":43,"tag":226,"props":662,"children":664},{"id":663},"language-scope",[665],{"type":48,"value":666},"Language scope",{"type":43,"tag":51,"props":668,"children":669},{},[670,672,677,679,684,686,692,694,700,702,707,709,714,716,721,723,729,731,737,739,744,746,752,753,759,761,767],{"type":48,"value":671},"The DOCA SHA Offload Engine is shipped as a ",{"type":43,"tag":55,"props":673,"children":674},{},[675],{"type":48,"value":676},"C dynamic\nshared object",{"type":48,"value":678}," (",{"type":43,"tag":63,"props":680,"children":682},{"className":681},[],[683],{"type":48,"value":84},{"type":48,"value":685},") built\nfrom ",{"type":43,"tag":63,"props":687,"children":689},{"className":688},[],[690],{"type":48,"value":691},"doca\u002Ftools\u002Fsha_offload_engine\u002F{engine\u002Fdoca_sha_offload_engine.c, lib\u002Fdoca_sha_offload_lib.{c,h}}",{"type":48,"value":693}," via ",{"type":43,"tag":63,"props":695,"children":697},{"className":696},[],[698],{"type":48,"value":699},"meson",{"type":48,"value":701},". Its\n",{"type":43,"tag":163,"props":703,"children":704},{},[705],{"type":48,"value":706},"consumer",{"type":48,"value":708}," interface is OpenSSL's ENGINE API; any\nlanguage that calls OpenSSL (C, C++, Rust via ",{"type":43,"tag":63,"props":710,"children":712},{"className":711},[],[713],{"type":48,"value":115},{"type":48,"value":715},"\ncrate, Python via ",{"type":43,"tag":63,"props":717,"children":719},{"className":718},[],[720],{"type":48,"value":18},{"type":48,"value":722}," and ",{"type":43,"tag":63,"props":724,"children":726},{"className":725},[],[727],{"type":48,"value":728},"pyca\u002Fcryptography",{"type":48,"value":730},"'s\nbackend, Node via ",{"type":43,"tag":63,"props":732,"children":734},{"className":733},[],[735],{"type":48,"value":736},"node:crypto",{"type":48,"value":738},") can therefore ",{"type":43,"tag":163,"props":740,"children":741},{},[742],{"type":48,"value":743},"use",{"type":48,"value":745}," the\nengine, provided the language binding either calls\n",{"type":43,"tag":63,"props":747,"children":749},{"className":748},[],[750],{"type":48,"value":751},"ENGINE_load_dynamic",{"type":48,"value":271},{"type":43,"tag":63,"props":754,"children":756},{"className":755},[],[757],{"type":48,"value":758},"ENGINE_by_id",{"type":48,"value":760}," directly or honors\nan OpenSSL ",{"type":43,"tag":63,"props":762,"children":764},{"className":763},[],[765],{"type":48,"value":766},"engines",{"type":48,"value":768}," config that loads it. The skill's\nlanguage-neutral contribution is the engine-load mechanics\nand the verification pattern; the OpenSSL ENGINE API is\nthe contract.",{"type":43,"tag":226,"props":770,"children":772},{"id":771},"when-to-load-this-skill",[773],{"type":48,"value":774},"When to load this skill",{"type":43,"tag":51,"props":776,"children":777},{},[778],{"type":48,"value":779},"Load this skill when the user is — or the agent needs to\n— deploy the DOCA SHA Offload Engine into an OpenSSL-based\npipeline on a host with DOCA installed and a SHA-capable\ndevice. Concretely:",{"type":43,"tag":245,"props":781,"children":782},{},[783,814,826,853],{"type":43,"tag":249,"props":784,"children":785},{},[786,788,793,795,800,801,806,808,813],{"type":48,"value":787},"Wiring the engine into an existing OpenSSL-based\napplication or ",{"type":43,"tag":63,"props":789,"children":791},{"className":790},[],[792],{"type":48,"value":115},{"type":48,"value":794}," CLI invocation, with the\nintent of no source-level code changes (or only the\nminimum ",{"type":43,"tag":63,"props":796,"children":798},{"className":797},[],[799],{"type":48,"value":751},{"type":48,"value":271},{"type":43,"tag":63,"props":802,"children":804},{"className":803},[],[805],{"type":48,"value":758},{"type":48,"value":807}," block\nshown in the verified ",{"type":43,"tag":63,"props":809,"children":811},{"className":810},[],[812],{"type":48,"value":536},{"type":48,"value":538},{"type":43,"tag":249,"props":815,"children":816},{},[817,819,824],{"type":48,"value":818},"Verifying the engine actually engaged for the\nhot-path digests (the ",{"type":43,"tag":163,"props":820,"children":821},{},[822],{"type":48,"value":823},"\"is offload real or did OpenSSL\nfall back to software\"",{"type":48,"value":825}," question).",{"type":43,"tag":249,"props":827,"children":828},{},[829,831,837,838,844,845,851],{"type":48,"value":830},"Characterizing the message-size range over which the\nengine is a perf win vs the CPU ",{"type":43,"tag":63,"props":832,"children":834},{"className":833},[],[835],{"type":48,"value":836},"sha1",{"type":48,"value":271},{"type":43,"tag":63,"props":839,"children":841},{"className":840},[],[842],{"type":48,"value":843},"sha256",{"type":48,"value":279},{"type":43,"tag":63,"props":846,"children":848},{"className":847},[],[849],{"type":48,"value":850},"sha512",{"type":48,"value":852}," paths.",{"type":43,"tag":249,"props":854,"children":855},{},[856,858,863,865,870],{"type":48,"value":857},"Deciding between ",{"type":43,"tag":163,"props":859,"children":860},{},[861],{"type":48,"value":862},"\"adopt the engine\"",{"type":48,"value":864}," (existing\npipeline, minimal change) and ",{"type":43,"tag":163,"props":866,"children":867},{},[868],{"type":48,"value":869},"\"rewrite to doca-sha\"",{"type":48,"value":871},"\n(new pipeline, fine-grained control needed).",{"type":43,"tag":51,"props":873,"children":874},{},[875,877,881,883,891,893,898],{"type":48,"value":876},"Do ",{"type":43,"tag":55,"props":878,"children":879},{},[880],{"type":48,"value":629},{"type":48,"value":882}," load this skill for users building a new\nSHA-pipeline from scratch — route to\n",{"type":43,"tag":127,"props":884,"children":885},{"href":216},[886],{"type":43,"tag":63,"props":887,"children":889},{"className":888},[],[890],{"type":48,"value":216},{"type":48,"value":892},".\nDo not load this skill for users wanting algorithms the\nengine does not implement (MD5, SHA-3, SHA-224, HMAC-SHA,\nstreaming\u002Fincremental SHA via ",{"type":43,"tag":63,"props":894,"children":896},{"className":895},[],[897],{"type":48,"value":277},{"type":48,"value":899}," chains\nthat the engine treats as one-shot only).",{"type":43,"tag":226,"props":901,"children":903},{"id":902},"what-this-skill-provides",[904],{"type":48,"value":905},"What this skill provides",{"type":43,"tag":51,"props":907,"children":908},{},[909,911,916],{"type":48,"value":910},"This is a ",{"type":43,"tag":55,"props":912,"children":913},{},[914],{"type":48,"value":915},"thin loader",{"type":48,"value":917},". Substantive material lives in\ntwo companion files:",{"type":43,"tag":245,"props":919,"children":920},{},[921,994],{"type":43,"tag":249,"props":922,"children":923},{},[924,929,931,936,938,943,944,949,950,955,957,963,965,971,973,978,980,985,987,992],{"type":43,"tag":63,"props":925,"children":927},{"className":926},[],[928],{"type":48,"value":172},{"type":48,"value":930}," — what the engine offloads (verified:\none-shot SHA-1, SHA-256, SHA-512 via the OpenSSL\n",{"type":43,"tag":63,"props":932,"children":934},{"className":933},[],[935],{"type":48,"value":123},{"type":48,"value":937}," high-level interface), what it does NOT\noffload (anything else — including SHA-224, MD5,\nSHA-3, HMAC-SHA, and any chained\n",{"type":43,"tag":63,"props":939,"children":941},{"className":940},[],[942],{"type":48,"value":269},{"type":48,"value":271},{"type":43,"tag":63,"props":945,"children":947},{"className":946},[],[948],{"type":48,"value":277},{"type":48,"value":279},{"type":43,"tag":63,"props":951,"children":953},{"className":952},[],[954],{"type":48,"value":285},{"type":48,"value":956}," pattern that the engine implements\nby buffering and then calling DOCA SHA in one shot at\n",{"type":43,"tag":63,"props":958,"children":960},{"className":959},[],[961],{"type":48,"value":962},"Final",{"type":48,"value":964},"), the engine-vs-library selection rule, the\nmessage-size-window rule for when offload is a perf\nwin, the verified ctrl-cmd surface (",{"type":43,"tag":63,"props":966,"children":968},{"className":967},[],[969],{"type":48,"value":970},"set_pci_addr",{"type":48,"value":972},"),\nthe version overlay (OpenSSL ≥ 1.1.1; the engine's\nshipped tests cover OpenSSL 1.1.1f and OpenSSL 3.0.2\nper the ",{"type":43,"tag":63,"props":974,"children":976},{"className":975},[],[977],{"type":48,"value":536},{"type":48,"value":979},"; OpenSSL 3.x deprecates the ENGINE\nAPI but still supports it via the legacy code path),\nthe layered error taxonomy, the observability surface\n(the ",{"type":43,"tag":163,"props":981,"children":982},{},[983],{"type":48,"value":984},"\"prove offload engaged\"",{"type":48,"value":986}," pattern using the\nSHA-224 negative test and ",{"type":43,"tag":63,"props":988,"children":990},{"className":989},[],[991],{"type":48,"value":382},{"type":48,"value":993},"), and the\nsafety overlay.",{"type":43,"tag":249,"props":995,"children":996},{},[997,1002,1004,1010,1012,1018,1020,1026,1028,1033,1035,1041,1042,1048,1050,1055,1057,1063,1065,1071,1073,1079,1081,1086,1088,1094,1095,1100,1102,1107,1109,1115,1117,1122,1124,1130],{"type":43,"tag":63,"props":998,"children":1000},{"className":999},[],[1001],{"type":48,"value":129},{"type":48,"value":1003}," — step-by-step workflows for the in-scope\ntask verbs: ",{"type":43,"tag":63,"props":1005,"children":1007},{"className":1006},[],[1008],{"type":48,"value":1009},"install",{"type":48,"value":1011},", ",{"type":43,"tag":63,"props":1013,"children":1015},{"className":1014},[],[1016],{"type":48,"value":1017},"configure",{"type":48,"value":1019}," (PCIe address\nselection — the engine defaults to ",{"type":43,"tag":63,"props":1021,"children":1023},{"className":1022},[],[1024],{"type":48,"value":1025},"03:00.0",{"type":48,"value":1027}," and\nexposes the ",{"type":43,"tag":63,"props":1029,"children":1031},{"className":1030},[],[1032],{"type":48,"value":970},{"type":48,"value":1034}," ctrl-cmd plus a build-time\noverride per the shipped ",{"type":43,"tag":63,"props":1036,"children":1038},{"className":1037},[],[1039],{"type":48,"value":1040},"test_cmdline_mode\u002Freadme.md",{"type":48,"value":641},{"type":43,"tag":63,"props":1043,"children":1045},{"className":1044},[],[1046],{"type":48,"value":1047},"build",{"type":48,"value":1049}," (the ",{"type":43,"tag":63,"props":1051,"children":1053},{"className":1052},[],[1054],{"type":48,"value":699},{"type":48,"value":1056}," flow), ",{"type":43,"tag":63,"props":1058,"children":1060},{"className":1059},[],[1061],{"type":48,"value":1062},"modify",{"type":48,"value":1064}," (refuses source\npatching; modify the load-time invocation and the\nPCIe address instead), ",{"type":43,"tag":63,"props":1066,"children":1068},{"className":1067},[],[1069],{"type":48,"value":1070},"run",{"type":48,"value":1072}," (load the engine via\n",{"type":43,"tag":63,"props":1074,"children":1076},{"className":1075},[],[1077],{"type":48,"value":1078},"openssl engine dynamic",{"type":48,"value":1080},"; the verification pattern;\nthe OpenSSL programmatic ",{"type":43,"tag":63,"props":1082,"children":1084},{"className":1083},[],[1085],{"type":48,"value":751},{"type":48,"value":1087}," block),\n",{"type":43,"tag":63,"props":1089,"children":1091},{"className":1090},[],[1092],{"type":48,"value":1093},"test",{"type":48,"value":1049},{"type":43,"tag":163,"props":1096,"children":1097},{},[1098],{"type":48,"value":1099},"\"prove the engine ran\"",{"type":48,"value":1101}," SHA-224 negative\ntest; the ",{"type":43,"tag":63,"props":1103,"children":1105},{"className":1104},[],[1106],{"type":48,"value":433},{"type":48,"value":1108}," perf comparison; the\nmessage-size window characterization), ",{"type":43,"tag":63,"props":1110,"children":1112},{"className":1111},[],[1113],{"type":48,"value":1114},"debug",{"type":48,"value":1116}," (walk\nthe error taxonomy layer by layer), ",{"type":43,"tag":63,"props":1118,"children":1120},{"className":1119},[],[1121],{"type":48,"value":743},{"type":48,"value":1123}," (the\nengine-vs-library decision for the user's specific\npipeline), plus a ",{"type":43,"tag":63,"props":1125,"children":1127},{"className":1126},[],[1128],{"type":48,"value":1129},"Deferred task verbs",{"type":48,"value":1131}," block.",{"type":43,"tag":51,"props":1133,"children":1134},{},[1135,1137,1143,1145,1150,1152,1158],{"type":48,"value":1136},"The skill assumes a host where DOCA is already installed,\nOpenSSL ≥ 1.1.1 is present (",{"type":43,"tag":63,"props":1138,"children":1140},{"className":1139},[],[1141],{"type":48,"value":1142},"libssl-dev",{"type":48,"value":1144}," or equivalent),\nand the deploying user can access the selected DOCA SHA PCIe device. Verify\ndevice visibility and run the engine-load smoke as that same user before\nintegration; if either fails with a permission error, stop and route to\n",{"type":43,"tag":63,"props":1146,"children":1148},{"className":1147},[],[1149],{"type":48,"value":211},{"type":48,"value":1151}," rather than guessing a group, ACL, or ",{"type":43,"tag":63,"props":1153,"children":1155},{"className":1154},[],[1156],{"type":48,"value":1157},"sudo",{"type":48,"value":1159}," policy.",{"type":43,"tag":226,"props":1161,"children":1163},{"id":1162},"what-this-skill-deliberately-does-not-ship",[1164],{"type":48,"value":1165},"What this skill deliberately does not ship",{"type":43,"tag":51,"props":1167,"children":1168},{},[1169,1171,1176],{"type":48,"value":1170},"This skill is ",{"type":43,"tag":55,"props":1172,"children":1173},{},[1174],{"type":48,"value":1175},"agent guidance",{"type":48,"value":1177},", not a samples or\nscripts bundle. It deliberately does not contain — and\npull requests should not add:",{"type":43,"tag":245,"props":1179,"children":1180},{},[1181,1242,1281,1308],{"type":43,"tag":249,"props":1182,"children":1183},{},[1184,1189,1191,1196,1197,1202,1203,1208,1209,1215,1216,1222,1223,1229,1231,1240],{"type":43,"tag":55,"props":1185,"children":1186},{},[1187],{"type":48,"value":1188},"Pre-written OpenSSL ENGINE application source code",{"type":48,"value":1190},"\nbeyond the verbatim verified block in the shipped\n",{"type":43,"tag":63,"props":1192,"children":1194},{"className":1193},[],[1195],{"type":48,"value":536},{"type":48,"value":1049},{"type":43,"tag":63,"props":1198,"children":1200},{"className":1199},[],[1201],{"type":48,"value":751},{"type":48,"value":279},{"type":43,"tag":63,"props":1204,"children":1206},{"className":1205},[],[1207],{"type":48,"value":758},{"type":48,"value":271},{"type":43,"tag":63,"props":1210,"children":1212},{"className":1211},[],[1213],{"type":48,"value":1214},"ENGINE_ctrl_cmd_string",{"type":48,"value":279},{"type":43,"tag":63,"props":1217,"children":1219},{"className":1218},[],[1220],{"type":48,"value":1221},"ENGINE_init",{"type":48,"value":271},{"type":43,"tag":63,"props":1224,"children":1226},{"className":1225},[],[1227],{"type":48,"value":1228},"ENGINE_set_default_digests",{"type":48,"value":1230}," sequence,\ncross-referenced into ",{"type":43,"tag":127,"props":1232,"children":1233},{"href":152},[1234],{"type":43,"tag":63,"props":1235,"children":1237},{"className":1236},[],[1238],{"type":48,"value":1239},"TASKS.md ## run",{"type":48,"value":1241},").\nThe shipped readme is the worked example.",{"type":43,"tag":249,"props":1243,"children":1244},{},[1245,1272,1274,1279],{"type":43,"tag":55,"props":1246,"children":1247},{},[1248,1250,1256,1257,1263,1264,1270],{"type":48,"value":1249},"A ",{"type":43,"tag":63,"props":1251,"children":1253},{"className":1252},[],[1254],{"type":48,"value":1255},"samples\u002F",{"type":48,"value":1011},{"type":43,"tag":63,"props":1258,"children":1260},{"className":1259},[],[1261],{"type":48,"value":1262},"bindings\u002F",{"type":48,"value":194},{"type":43,"tag":63,"props":1265,"children":1267},{"className":1266},[],[1268],{"type":48,"value":1269},"reference\u002F",{"type":48,"value":1271},"\nsubtree.",{"type":48,"value":1273}," This is a thin loader for a shipped\nshared-object; substantive material lives in the\nshipped ",{"type":43,"tag":63,"props":1275,"children":1277},{"className":1276},[],[1278],{"type":48,"value":536},{"type":48,"value":1280}," and the doca-sha library docs.",{"type":43,"tag":249,"props":1282,"children":1283},{},[1284,1289,1291,1296,1298,1306],{"type":43,"tag":55,"props":1285,"children":1286},{},[1287],{"type":48,"value":1288},"Performance numbers from agent memory.",{"type":48,"value":1290}," The\nmessage-size-window where the engine wins is\ndevice-, firmware-, OpenSSL-version-, and\nworkload-specific. The ",{"type":43,"tag":63,"props":1292,"children":1294},{"className":1293},[],[1295],{"type":48,"value":433},{"type":48,"value":1297}," comparison\npattern in ",{"type":43,"tag":127,"props":1299,"children":1300},{"href":387},[1301],{"type":43,"tag":63,"props":1302,"children":1304},{"className":1303},[],[1305],{"type":48,"value":394},{"type":48,"value":1307}," is the\nway to capture it on the user's actual hardware;\nquoting a number from memory is the cross-platform\nfailure mode this skill exists to prevent.",{"type":43,"tag":249,"props":1309,"children":1310},{},[1311,1316,1318,1323],{"type":43,"tag":55,"props":1312,"children":1313},{},[1314],{"type":48,"value":1315},"Wrappers, parsers, or scripts",{"type":48,"value":1317}," in any language\nthat consume the engine's stdout or the ",{"type":43,"tag":63,"props":1319,"children":1321},{"className":1320},[],[1322],{"type":48,"value":433},{"type":48,"value":1324}," output format.",{"type":43,"tag":226,"props":1326,"children":1328},{"id":1327},"loading-order",[1329],{"type":48,"value":1330},"Loading order",{"type":43,"tag":1332,"props":1333,"children":1334},"ol",{},[1335,1365,1378],{"type":43,"tag":249,"props":1336,"children":1337},{},[1338,1340,1346,1348,1353,1355,1364],{"type":48,"value":1339},"Read this ",{"type":43,"tag":63,"props":1341,"children":1343},{"className":1342},[],[1344],{"type":48,"value":1345},"SKILL.md",{"type":48,"value":1347}," first to confirm the user's\nquestion is in scope (the user actually has an\nexisting OpenSSL-based pipeline and wants to offload\nto DOCA SHA ",{"type":43,"tag":163,"props":1349,"children":1350},{},[1351],{"type":48,"value":1352},"without",{"type":48,"value":1354}," code changes; if the user is\nbuilding from scratch, route to\n",{"type":43,"tag":127,"props":1356,"children":1357},{"href":216},[1358],{"type":43,"tag":63,"props":1359,"children":1361},{"className":1360},[],[1362],{"type":48,"value":1363},"..\u002F..\u002Flibs\u002Fdoca-sha\u002F",{"type":48,"value":538},{"type":43,"tag":249,"props":1366,"children":1367},{},[1368],{"type":43,"tag":55,"props":1369,"children":1370},{},[1371,1373,1377],{"type":48,"value":1372},"For what the engine offloads vs falls back, the\nengine-vs-library selection rule, the message-size-\nwindow rule, the version overlay, the error taxonomy,\nthe observability surface (and the prove-offload-\nactually-engaged pattern), and the safety overlay,\nsee ",{"type":43,"tag":127,"props":1374,"children":1375},{"href":172},[1376],{"type":48,"value":172},{"type":48,"value":333},{"type":43,"tag":249,"props":1379,"children":1380},{},[1381],{"type":43,"tag":55,"props":1382,"children":1383},{},[1384,1386,1391,1392,1397,1398,1403,1404,1409,1410,1415,1416,1421,1422,1427,1428,1433,1435,1439],{"type":48,"value":1385},"For step-by-step workflows — ",{"type":43,"tag":63,"props":1387,"children":1389},{"className":1388},[],[1390],{"type":48,"value":1009},{"type":48,"value":187},{"type":43,"tag":63,"props":1393,"children":1395},{"className":1394},[],[1396],{"type":48,"value":1017},{"type":48,"value":1011},{"type":43,"tag":63,"props":1399,"children":1401},{"className":1400},[],[1402],{"type":48,"value":1047},{"type":48,"value":1011},{"type":43,"tag":63,"props":1405,"children":1407},{"className":1406},[],[1408],{"type":48,"value":1062},{"type":48,"value":1011},{"type":43,"tag":63,"props":1411,"children":1413},{"className":1412},[],[1414],{"type":48,"value":1070},{"type":48,"value":1011},{"type":43,"tag":63,"props":1417,"children":1419},{"className":1418},[],[1420],{"type":48,"value":1093},{"type":48,"value":187},{"type":43,"tag":63,"props":1423,"children":1425},{"className":1424},[],[1426],{"type":48,"value":1114},{"type":48,"value":1011},{"type":43,"tag":63,"props":1429,"children":1431},{"className":1430},[],[1432],{"type":48,"value":743},{"type":48,"value":1434}," — see ",{"type":43,"tag":127,"props":1436,"children":1437},{"href":129},[1438],{"type":48,"value":129},{"type":48,"value":333},{"type":43,"tag":226,"props":1441,"children":1443},{"id":1442},"related-skills",[1444],{"type":48,"value":1445},"Related skills",{"type":43,"tag":245,"props":1447,"children":1448},{},[1449,1479,1494,1509,1529,1567],{"type":43,"tag":249,"props":1450,"children":1451},{},[1452,1460,1462,1467,1469,1478],{"type":43,"tag":127,"props":1453,"children":1454},{"href":216},[1455],{"type":43,"tag":63,"props":1456,"children":1458},{"className":1457},[],[1459],{"type":48,"value":216},{"type":48,"value":1461}," —\nthe underlying DOCA SHA library. The engine is a\nthin OpenSSL-ENGINE wrapper around doca-sha; when the\nuser needs fine-grained control over the SHA task\nsurface (partial-hash, custom buffer permissions,\ncap-query for unusual message sizes), the library is\nthe right answer. The engine wraps the ",{"type":43,"tag":163,"props":1463,"children":1464},{},[1465],{"type":48,"value":1466},"one-shot",{"type":48,"value":1468},"\ntask; the library exposes both one-shot and partial-\nhash per\n",{"type":43,"tag":127,"props":1470,"children":1472},{"href":1471},"..\u002F..\u002Flibs\u002Fdoca-sha\u002FCAPABILITIES.md#capabilities-and-modes",[1473],{"type":43,"tag":63,"props":1474,"children":1476},{"className":1475},[],[1477],{"type":48,"value":1471},{"type":48,"value":333},{"type":43,"tag":249,"props":1480,"children":1481},{},[1482,1492],{"type":43,"tag":127,"props":1483,"children":1485},{"href":1484},"..\u002F..\u002Fdoca-version\u002FSKILL.md",[1486],{"type":43,"tag":63,"props":1487,"children":1489},{"className":1488},[],[1490],{"type":48,"value":1491},"doca-version",{"type":48,"value":1493}," — the\ncanonical version-detection chain. The engine has a\nTWO-axis version overlay (DOCA-side and OpenSSL-side);\nthe version skill carries the four-way match rule\nthis skill layers on top of.",{"type":43,"tag":249,"props":1495,"children":1496},{},[1497,1507],{"type":43,"tag":127,"props":1498,"children":1500},{"href":1499},"..\u002F..\u002Fdoca-debug\u002FSKILL.md",[1501],{"type":43,"tag":63,"props":1502,"children":1504},{"className":1503},[],[1505],{"type":48,"value":1506},"doca-debug",{"type":48,"value":1508}," — the\ncross-cutting debug ladder. The engine surfaces its\nown error taxonomy; when the cause is below DOCA\n(driver, firmware), the taxonomy hands off here.",{"type":43,"tag":249,"props":1510,"children":1511},{},[1512,1520,1522,1527],{"type":43,"tag":127,"props":1513,"children":1514},{"href":204},[1515],{"type":43,"tag":63,"props":1516,"children":1518},{"className":1517},[],[1519],{"type":48,"value":211},{"type":48,"value":1521}," — env\npreparation, install verification, the ",{"type":43,"tag":63,"props":1523,"children":1525},{"className":1524},[],[1526],{"type":48,"value":1142},{"type":48,"value":1528},"\ninstall path, and the NGC DOCA container path.",{"type":43,"tag":249,"props":1530,"children":1531},{},[1532,1542,1544,1550,1552,1558,1560,1566],{"type":43,"tag":127,"props":1533,"children":1535},{"href":1534},"..\u002F..\u002Fdoca-public-knowledge-map\u002FSKILL.md",[1536],{"type":43,"tag":63,"props":1537,"children":1539},{"className":1538},[],[1540],{"type":48,"value":1541},"doca-public-knowledge-map",{"type":48,"value":1543}," —\nrouting to the public DOCA SHA documentation set on\n",{"type":43,"tag":63,"props":1545,"children":1547},{"className":1546},[],[1548],{"type":48,"value":1549},"docs.nvidia.com\u002Fdoca\u002Fsdk\u002F",{"type":48,"value":1551}," and to the OpenSSL\nENGINE \u002F ",{"type":43,"tag":63,"props":1553,"children":1555},{"className":1554},[],[1556],{"type":48,"value":1557},"openssl-engine",{"type":48,"value":1559}," upstream documentation on\n",{"type":43,"tag":63,"props":1561,"children":1563},{"className":1562},[],[1564],{"type":48,"value":1565},"openssl.org",{"type":48,"value":333},{"type":43,"tag":249,"props":1568,"children":1569},{},[1570,1580,1582,1587],{"type":43,"tag":127,"props":1571,"children":1573},{"href":1572},"..\u002F..\u002Fdoca-hardware-safety\u002FSKILL.md",[1574],{"type":43,"tag":63,"props":1575,"children":1577},{"className":1576},[],[1578],{"type":48,"value":1579},"doca-hardware-safety",{"type":48,"value":1581}," —\nthe bundle-wide hardware-safety meta-policy. The\nengine binds to a specific PCIe device; the\n",{"type":43,"tag":63,"props":1583,"children":1585},{"className":1584},[],[1586],{"type":48,"value":970},{"type":48,"value":1588}," ctrl is the artifact-specific overlay,\nbut any host-side change underneath (firmware burn,\nBFB reflash) runs through the meta-policy.",{"items":1590,"total":1692},[1591,1606,1620,1634,1646,1663,1678],{"slug":1592,"name":1592,"fn":1593,"description":1594,"org":1595,"tags":1596,"stars":23,"repoUrl":24,"updatedAt":1605},"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},[1597,1600,1603,1604],{"name":1598,"slug":1599,"type":15},"Data Analysis","data-analysis",{"name":1601,"slug":1602,"type":15},"Data Engineering","data-engineering",{"name":9,"slug":8,"type":15},{"name":13,"slug":14,"type":15},"2026-07-14T05:28:43.176466",{"slug":1607,"name":1607,"fn":1608,"description":1609,"org":1610,"tags":1611,"stars":23,"repoUrl":24,"updatedAt":1619},"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},[1612,1615,1618],{"name":1613,"slug":1614,"type":15},"Deployment","deployment",{"name":1616,"slug":1617,"type":15},"Infrastructure","infrastructure",{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":1621,"name":1621,"fn":1622,"description":1623,"org":1624,"tags":1625,"stars":23,"repoUrl":24,"updatedAt":1633},"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},[1626,1629,1630],{"name":1627,"slug":1628,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":1631,"slug":1632,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":1635,"name":1635,"fn":1636,"description":1637,"org":1638,"tags":1639,"stars":23,"repoUrl":24,"updatedAt":1645},"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},[1640,1641,1642],{"name":1598,"slug":1599,"type":15},{"name":9,"slug":8,"type":15},{"name":1643,"slug":1644,"type":15},"Testing","testing","2026-07-17T05:29:03.913266",{"slug":1647,"name":1647,"fn":1648,"description":1649,"org":1650,"tags":1651,"stars":23,"repoUrl":24,"updatedAt":1662},"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},[1652,1655,1658,1659],{"name":1653,"slug":1654,"type":15},"Automation","automation",{"name":1656,"slug":1657,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":1660,"slug":1661,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":1664,"name":1664,"fn":1665,"description":1666,"org":1667,"tags":1668,"stars":23,"repoUrl":24,"updatedAt":1677},"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},[1669,1670,1673,1674],{"name":1613,"slug":1614,"type":15},{"name":1671,"slug":1672,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":1675,"slug":1676,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":1679,"name":1679,"fn":1680,"description":1681,"org":1682,"tags":1683,"stars":23,"repoUrl":24,"updatedAt":1691},"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},[1684,1685,1688],{"name":9,"slug":8,"type":15},{"name":1686,"slug":1687,"type":15},"Quantum Computing","quantum-computing",{"name":1689,"slug":1690,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305,{"items":1694,"total":1845},[1695,1713,1729,1740,1752,1766,1779,1793,1804,1813,1827,1836],{"slug":1696,"name":1696,"fn":1697,"description":1698,"org":1699,"tags":1700,"stars":1710,"repoUrl":1711,"updatedAt":1712},"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},[1701,1704,1707],{"name":1702,"slug":1703,"type":15},"Documentation","documentation",{"name":1705,"slug":1706,"type":15},"MCP","mcp",{"name":1708,"slug":1709,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":1714,"name":1714,"fn":1715,"description":1716,"org":1717,"tags":1718,"stars":1726,"repoUrl":1727,"updatedAt":1728},"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},[1719,1722,1723],{"name":1720,"slug":1721,"type":15},"Containers","containers",{"name":1613,"slug":1614,"type":15},{"name":1724,"slug":1725,"type":15},"Python","python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":1730,"name":1730,"fn":1731,"description":1732,"org":1733,"tags":1734,"stars":1726,"repoUrl":1727,"updatedAt":1739},"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},[1735,1738],{"name":1736,"slug":1737,"type":15},"CI\u002FCD","ci-cd",{"name":1613,"slug":1614,"type":15},"2026-07-14T05:25:59.97109",{"slug":1741,"name":1741,"fn":1742,"description":1743,"org":1744,"tags":1745,"stars":1726,"repoUrl":1727,"updatedAt":1751},"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},[1746,1747,1748],{"name":1736,"slug":1737,"type":15},{"name":1613,"slug":1614,"type":15},{"name":1749,"slug":1750,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":1753,"name":1753,"fn":1754,"description":1755,"org":1756,"tags":1757,"stars":1726,"repoUrl":1727,"updatedAt":1765},"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},[1758,1761,1762],{"name":1759,"slug":1760,"type":15},"Debugging","debugging",{"name":1749,"slug":1750,"type":15},{"name":1763,"slug":1764,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":1767,"name":1767,"fn":1768,"description":1769,"org":1770,"tags":1771,"stars":1726,"repoUrl":1727,"updatedAt":1778},"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},[1772,1775],{"name":1773,"slug":1774,"type":15},"Best Practices","best-practices",{"name":1776,"slug":1777,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":1780,"name":1780,"fn":1781,"description":1782,"org":1783,"tags":1784,"stars":1726,"repoUrl":1727,"updatedAt":1792},"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},[1785,1788,1791],{"name":1786,"slug":1787,"type":15},"Machine Learning","machine-learning",{"name":1789,"slug":1790,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":1794,"name":1794,"fn":1795,"description":1796,"org":1797,"tags":1798,"stars":1726,"repoUrl":1727,"updatedAt":1803},"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},[1799,1802],{"name":1800,"slug":1801,"type":15},"QA","qa",{"name":1643,"slug":1644,"type":15},"2026-07-14T05:25:53.673039",{"slug":1805,"name":1805,"fn":1806,"description":1807,"org":1808,"tags":1809,"stars":1726,"repoUrl":1727,"updatedAt":1812},"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},[1810,1811],{"name":1613,"slug":1614,"type":15},{"name":1616,"slug":1617,"type":15},"2026-07-14T05:25:49.362534",{"slug":1814,"name":1814,"fn":1815,"description":1816,"org":1817,"tags":1818,"stars":1726,"repoUrl":1727,"updatedAt":1826},"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},[1819,1822,1823],{"name":1820,"slug":1821,"type":15},"Code Review","code-review",{"name":1749,"slug":1750,"type":15},{"name":1824,"slug":1825,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":1828,"name":1828,"fn":1829,"description":1830,"org":1831,"tags":1832,"stars":1726,"repoUrl":1727,"updatedAt":1835},"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},[1833,1834],{"name":1800,"slug":1801,"type":15},{"name":1643,"slug":1644,"type":15},"2026-07-14T05:25:54.928983",{"slug":1837,"name":1837,"fn":1838,"description":1839,"org":1840,"tags":1841,"stars":1726,"repoUrl":1727,"updatedAt":1844},"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},[1842,1843],{"name":1653,"slug":1654,"type":15},{"name":1736,"slug":1737,"type":15},"2026-07-30T05:29:03.275638",496]