[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-doca-compress":3,"mdc--jleej7-key":34,"related-repo-nvidia-doca-compress":1366,"related-org-nvidia-doca-compress":1470},{"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-compress","program DOCA hardware-accelerated compression","Use this skill for hands-on DOCA Compress programming on a BlueField DPU, ConnectX NIC, or host with DOCA — enabling compress-deflate, decompress-deflate, decompress-lz4-stream, or decompress-lz4-block tasks on a doca_compress context (the hardware supports DEFLATE both directions plus LZ4 decompress; LZ4 encode is NOT supported), sizing source \u002F destination doca_buf against the per-task cap query, setting mmap permissions, deciding offload vs CPU zlib \u002F zstd, validating with a round-trip smoke, or debugging DOCA_ERROR_* from a Compress call. Trigger on phrasings like \"offload this gzip\", \"decompress incoming network data\", \"compress task returns INVALID_VALUE on alloc_init\", \"submitted a task but no completion arrives\", or \"decompress LZ4 on the BlueField.\" Refuse and route elsewhere for non-DEFLATE \u002F non-LZ4 algorithms (zstd \u002F Snappy \u002F brotli), LZ4 encode (route to a CPU LZ4 library), pure mmap-to-mmap copies (doca-dma), or DOCA Core lifecycle internals.\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},"Hardware","hardware",{"name":9,"slug":8,"type":15},{"name":21,"slug":22,"type":15},"Engineering","engineering",2473,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills","2026-07-30T05:28:44.504776","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-compress","---\nlicense: Apache-2.0\nname: doca-compress\ndescription: >\n  Use this skill for hands-on DOCA Compress programming on a\n  BlueField DPU, ConnectX NIC, or host with DOCA — enabling\n  compress-deflate, decompress-deflate, decompress-lz4-stream,\n  or decompress-lz4-block tasks on a doca_compress context\n  (the hardware supports DEFLATE both directions plus LZ4\n  decompress; LZ4 encode is NOT supported), sizing source \u002F\n  destination doca_buf against the per-task cap query, setting\n  mmap permissions, deciding offload vs CPU zlib \u002F zstd,\n  validating with a round-trip smoke, or debugging\n  DOCA_ERROR_* from a Compress call. Trigger on phrasings\n  like \"offload this gzip\", \"decompress incoming network\n  data\", \"compress task returns INVALID_VALUE on alloc_init\",\n  \"submitted a task but no completion arrives\", or \"decompress\n  LZ4 on the BlueField.\" Refuse and route elsewhere for\n  non-DEFLATE \u002F non-LZ4 algorithms (zstd \u002F Snappy \u002F brotli),\n  LZ4 encode (route to a CPU LZ4 library), pure mmap-to-mmap\n  copies (doca-dma), or DOCA Core lifecycle internals.\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. Reads the user's local install via\n  `pkg-config doca-compress` and inspects\n  \u002Fopt\u002Fmellanox\u002Fdoca\u002F{lib,include,samples,applications}.\n---\n\n# DOCA Compress\n\n**Where to start:** This skill assumes DOCA is already installed and\nthe user is doing **hands-on Compress work** (bulk DEFLATE\ncompression or decompression) on a BlueField \u002F ConnectX \u002F host\nwith DOCA. Open [`TASKS.md`](TASKS.md) if the user wants to *do*\nsomething (configure \u002F build \u002F modify \u002F run \u002F test \u002F debug); open\n[`CAPABILITIES.md`](CAPABILITIES.md) when the question is *what can\nDOCA Compress express* on this version. If the user has not\ninstalled DOCA yet, route to\n[`doca-setup`](..\u002F..\u002Fdoca-setup\u002FSKILL.md) first. If the user is\nasking *\"should I even offload this compression to the\naccelerator?\"*, the size-threshold path-selection rule in\n[`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes)\nis the first stop — bulk compress is the canonical fit, tiny\none-shot is not.\n\n## Example questions this skill answers well\n\nThe CLASSES of DOCA Compress questions this skill is built to\nanswer, each with one worked example. The agent should treat the\n*class* as the load-bearing piece — the worked example is a single\ninstance.\n\n- **\"Should I offload this compression to DOCA Compress, or just\n  zlib\u002Fzstd on the CPU?\"** — worked example: *\"I am compressing a\n  4 MiB log buffer before writing it to storage; is doca-compress\n  worth the setup vs zlib on the CPU?\"*. Answered by the\n  size-threshold path-selection rule in\n  [`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes)\n  + the *\"when NOT to use doca-compress\"* bullets in\n  [`CAPABILITIES.md ## Safety policy`](CAPABILITIES.md#safety-policy).\n- **\"Does my device support the compress (or decompress) task I\n  want, and how big a buffer can it move per submission?\"** —\n  worked example: *\"is `doca_compress_task_compress_deflate` on\n  this BlueField, and what is the max source size per task?\"*.\n  Answered by the per-task capability-query rule\n  (`doca_compress_cap_task_compress_deflate_is_supported`,\n  `_decompress_deflate_is_supported`, the matching\n  `_get_max_buf_size` queries) in\n  [`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes)\n  + the discovery step in\n  [`TASKS.md ## configure`](TASKS.md#configure).\n- **\"I just need to decompress incoming network data — is that a\n  valid standalone use of doca-compress?\"** — worked example:\n  *\"my client receives DEFLATE-compressed payloads and never\n  produces any compressed output of its own\"*. Answered by the\n  *\"decompression-only\"* note in\n  [`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes)\n  task-type table + the per-task configuration matrix in\n  [`TASKS.md ## configure`](TASKS.md#configure) step 5.\n- **\"What permissions does the source \u002F destination mmap need?\"** —\n  worked example: *\"my `doca_compress_task_compress_deflate`\n  returns `DOCA_ERROR_NOT_PERMITTED`\"*. Answered by the permission\n  matrix in\n  [`CAPABILITIES.md ## Safety policy`](CAPABILITIES.md#safety-policy)\n  + the mmap-set-permissions checklist in\n  [`TASKS.md ## test`](TASKS.md#test).\n- **\"Is this DOCA Compress API available on my installed DOCA\n  version?\"** — worked example: *\"is\n  `doca_compress_task_decompress_deflate` in the DOCA I have\n  installed?\"*. Answered by the version-compatibility overlay in\n  [`CAPABILITIES.md ## Version compatibility`](CAPABILITIES.md#version-compatibility),\n  which cross-links the canonical detection chain in\n  [`doca-version`](..\u002F..\u002Fdoca-version\u002FSKILL.md) and adds the\n  Compress-specific *\"discover per-task support, do not assume\"*\n  bullets.\n- **\"What does this `DOCA_ERROR_*` from a Compress call mean and\n  which layer caused it?\"** — worked example: *\"`DOCA_ERROR_INVALID_VALUE`\n  on `doca_compress_task_compress_deflate_alloc_init`\"*. Answered\n  by the Compress overlay on the cross-library taxonomy in\n  [`CAPABILITIES.md ## Error taxonomy`](CAPABILITIES.md#error-taxonomy)\n  + the layered ladder in\n  [`TASKS.md ## debug`](TASKS.md#debug) that escalates to\n  [`doca-debug`](..\u002F..\u002Fdoca-debug\u002FSKILL.md).\n\n## Audience\n\nThis skill serves **external developers building applications that\nconsume the DOCA Compress library** — i.e., users whose code calls\n`doca_compress_*` (directly in C\u002FC++, or through FFI\u002Fbindings from\nanother language) to offload bulk DEFLATE compression or\ndecompression onto a BlueField DPU or ConnectX accelerator. It is\n*not* for NVIDIA developers contributing to DOCA Compress itself.\n\n**Language scope.** DOCA Compress ships as a C library with\n`pkg-config` module name `doca-compress`. The shipped samples are\nwritten in C. C and C++ consumers are the canonical case and the\nworked examples in `TASKS.md` assume that path. Other-language\nconsumers (Rust, Go, Python, …) consume the same `*.so` through\nFFI or language-specific bindings; the skill's contribution in\nthat case is to keep the lifecycle, capability-discovery,\npermission, error-taxonomy, and compress-vs-decompress guidance\nlanguage-neutral, and to route the agent to the public C ABI as\nthe authoritative surface that any wrapper will eventually call.\n\n## When to load this skill\n\nLoad this skill when the user is doing hands-on DOCA Compress\nwork, in any language. Concretely:\n\n- Initializing a `doca_compress` context on a `doca_dev` and\n  configuring at least one task type\n  (`doca_compress_task_compress_deflate` and\u002For\n  `doca_compress_task_decompress_deflate`) before\n  `doca_ctx_start()`.\n- Choosing between the **compress** task and the **decompress**\n  task for the user's data flow. The two task types are\n  independent: a consumer that only decompresses inbound DEFLATE\n  payloads is a valid shape and does not need to enable the\n  compress task.\n- Setting permissions on `doca_mmap` correctly for the source\n  buffer (`DOCA_ACCESS_FLAG_LOCAL_READ_ONLY` at minimum) and the\n  destination buffer (`DOCA_ACCESS_FLAG_LOCAL_READ_WRITE`).\n- Sizing the source buffer against the per-task\n  `doca_compress_cap_task_*_get_max_buf_size(devinfo)` ceiling\n  and the destination buffer against the worst-case output size\n  the algorithm can produce on this input.\n- Checking which task types this device's accelerator advertises\n  via `doca_compress_cap_task_compress_deflate_is_supported` and\n  `doca_compress_cap_task_decompress_deflate_is_supported`\n  against the active `doca_devinfo`.\n- Validating against a round-trip (compress → decompress, or\n  decompress → compare against a known DEFLATE-encoded fixture)\n  before pushing bulk input through the accelerator.\n- Deciding *whether to offload at all* — the size-threshold rule\n  in [`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes)\n  says doca-compress is the right answer only for bulk inputs\n  (rule of thumb: ≥ a few KiB); below that, CPU compression beats\n  the DMA-to-accelerator round-trip.\n- Debugging a `DOCA_ERROR_*` returned from a Compress call\n  (lifecycle vs. buffer-sizing vs. permission vs.\n  unsupported-task) and the task-completion event on the progress\n  engine.\n- Designing or extending non-C bindings (Rust, Go, Python, …)\n  that wrap the Compress C ABI — for the lifecycle, permission,\n  capability, and compress-vs-decompress rules the wrapper must\n  honor.\n\nDo **not** load this skill for general DOCA orientation, install\nof DOCA itself, non-DEFLATE compression libraries on CPU (use\nzlib \u002F zstd \u002F similar), or other DOCA libraries. For those, use\n[`doca-public-knowledge-map`](..\u002F..\u002Fdoca-public-knowledge-map\u002FSKILL.md).\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\nCompress-specific material lives in two companion files:\n\n- `CAPABILITIES.md` — what DOCA Compress can express on this\n  version: the four task types (compress-deflate,\n  decompress-deflate, decompress-lz4-stream, and\n  decompress-lz4-block, each independently capability-gated), the\n  per-task capability-query surface (`doca_compress_cap_*` for\n  task support and per-submission max buffer size), the Compress\n  error taxonomy (mapped onto the cross-library `DOCA_ERROR_*`\n  set), the observability surface (per-task completion events on\n  the progress engine), the safety policy that gates source \u002F\n  destination mmap permission decisions, and the size-threshold\n  path-selection rule (when to use doca-compress versus CPU\n  compression or `doca-dma` for the no-compression copy case).\n- `TASKS.md` — step-by-step workflows for the six in-scope\n  Compress verbs: `configure`, `build`, `modify`, `run`, `test`,\n  `debug`. Plus a `## rollback` overlay (Compress-specific\n  five-step teardown that drains in-flight tasks before\n  `doca_ctx_stop`, unregisters mmap regions in reverse-register\n  order, and re-verifies the device path with the round-trip\n  smoke) 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\nprivileges their public install profile expects. It does not\ncover installing DOCA — 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 Compress application source code, in any\n  language.** The verified Compress source code is the shipped C\n  samples at `\u002Fopt\u002Fmellanox\u002Fdoca\u002Fsamples\u002Fdoca_compress\u002F`, plus\n  the File Compression reference application linked from the\n  public DOCA Compress guide. The agent's job is to route the\n  user to those files and prescribe a minimum-diff modification\n  on them via the universal modify-a-sample workflow in\n  [`doca-programming-guide`](..\u002F..\u002Fdoca-programming-guide\u002FSKILL.md),\n  layered with the Compress-specific overrides in\n  [`TASKS.md ## modify`](TASKS.md#modify).\n- **Pre-encoded DEFLATE fixtures for arbitrary inputs.** The\n  skill tells the agent to use a *round-trip* validation\n  (compress → decompress and compare to the original, or\n  decompress a known DEFLATE blob produced by a published CPU\n  encoder like `zlib`) as the known-good smoke; it does not ship\n  a fixture bank of its own.\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\n  the user's installed DOCA, where `pkg-config --modversion\n  doca-compress` is the source of truth.\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 Compress capability matrix, the compress \u002F decompress\n   task split (including decompression-only as a standalone\n   shape), per-task capability-query rules, permission matrix,\n   error taxonomy, observability, and safety \u002F path-selection\n   policy, see [CAPABILITIES.md](CAPABILITIES.md).**\n3. **For step-by-step workflows — configure, build, modify, run,\n   test, debug — see [TASKS.md](TASKS.md).**\n\nBoth companion files cross-link to each other,\n[`doca-version`](..\u002F..\u002Fdoca-version\u002FSKILL.md) for the canonical\nversion-handling rules, and\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\nthe installed package layout\" rather than \"Compress-specific\nguidance\".\n\n## Related skills\n\n- [`doca-public-knowledge-map`](..\u002F..\u002Fdoca-public-knowledge-map\u002FSKILL.md) —\n  the routing table for every public DOCA documentation source\n  and the on-disk layout of an installed DOCA package. The DOCA\n  Compress page lives at\n  `docs.nvidia.com\u002Fdoca\u002Fsdk\u002FDOCA-Compress\u002F`; the File Compression\n  reference application is the canonical worked example.\n- [`doca-setup`](..\u002F..\u002Fdoca-setup\u002FSKILL.md) — env preparation,\n  install verification, and the *I have no install yet* path\n  with the public NGC DOCA container. This skill assumes its\n  preconditions are satisfied.\n- [`doca-version`](..\u002F..\u002Fdoca-version\u002FSKILL.md) — canonical DOCA\n  version-handling rules. This skill's\n  [`## Version compatibility`](CAPABILITIES.md#version-compatibility)\n  cross-links the four-way match rule and adds only the\n  Compress-specific *\"discover per-task support + per-task max\n  buffer size via cap query\"* overlay.\n- [`doca-structured-tools-contract`](..\u002F..\u002Fdoca-structured-tools-contract\u002FSKILL.md) —\n  the bundle's structured-tools precedence rule (detect \u002F prefer\n  \u002F fall back \u002F report). The Command appendix in\n  [TASKS.md](TASKS.md) honors this contract.\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  lifecycle, the cross-library `DOCA_ERROR_*` taxonomy, and the\n  program-side debug order. This skill layers Compress specifics\n  on top.\n- [`doca-dma`](..\u002Fdoca-dma\u002FSKILL.md) — the right library when\n  the data flow is a pure mmap-to-mmap copy with no compression\n  required. This skill's path-selection rule routes to DMA when\n  Compress is *not* the answer (e.g. the user only wants to move\n  bytes, not encode them).\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). Compress-specific debug (task-not-supported,\n  source-buffer-too-large, destination-buffer-too-small) overlays\n  on top of that ladder.\n",{"data":35,"body":39},{"license":26,"name":4,"description":6,"metadata":36,"compatibility":38},{"kind":37},"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. Reads the user's local install via `pkg-config doca-compress` and inspects \u002Fopt\u002Fmellanox\u002Fdoca\u002F{lib,include,samples,applications}.\n",{"type":40,"children":41},"root",[42,50,139,146,158,527,533,560,600,606,611,788,810,816,828,949,963,969,981,1130,1136,1179,1204,1210],{"type":43,"tag":44,"props":45,"children":46},"element","h1",{"id":4},[47],{"type":48,"value":49},"text","DOCA Compress",{"type":43,"tag":51,"props":52,"children":53},"p",{},[54,60,62,67,69,80,82,88,90,99,101,106,108,118,120,125,127,137],{"type":43,"tag":55,"props":56,"children":57},"strong",{},[58],{"type":48,"value":59},"Where to start:",{"type":48,"value":61}," This skill assumes DOCA is already installed and\nthe user is doing ",{"type":43,"tag":55,"props":63,"children":64},{},[65],{"type":48,"value":66},"hands-on Compress work",{"type":48,"value":68}," (bulk DEFLATE\ncompression or decompression) on a BlueField \u002F ConnectX \u002F host\nwith DOCA. Open ",{"type":43,"tag":70,"props":71,"children":73},"a",{"href":72},"TASKS.md",[74],{"type":43,"tag":75,"props":76,"children":78},"code",{"className":77},[],[79],{"type":48,"value":72},{"type":48,"value":81}," if the user wants to ",{"type":43,"tag":83,"props":84,"children":85},"em",{},[86],{"type":48,"value":87},"do",{"type":48,"value":89},"\nsomething (configure \u002F build \u002F modify \u002F run \u002F test \u002F debug); open\n",{"type":43,"tag":70,"props":91,"children":93},{"href":92},"CAPABILITIES.md",[94],{"type":43,"tag":75,"props":95,"children":97},{"className":96},[],[98],{"type":48,"value":92},{"type":48,"value":100}," when the question is ",{"type":43,"tag":83,"props":102,"children":103},{},[104],{"type":48,"value":105},"what can\nDOCA Compress express",{"type":48,"value":107}," on this version. If the user has not\ninstalled DOCA yet, route to\n",{"type":43,"tag":70,"props":109,"children":111},{"href":110},"..\u002F..\u002Fdoca-setup\u002FSKILL.md",[112],{"type":43,"tag":75,"props":113,"children":115},{"className":114},[],[116],{"type":48,"value":117},"doca-setup",{"type":48,"value":119}," first. If the user is\nasking ",{"type":43,"tag":83,"props":121,"children":122},{},[123],{"type":48,"value":124},"\"should I even offload this compression to the\naccelerator?\"",{"type":48,"value":126},", the size-threshold path-selection rule in\n",{"type":43,"tag":70,"props":128,"children":130},{"href":129},"CAPABILITIES.md#capabilities-and-modes",[131],{"type":43,"tag":75,"props":132,"children":134},{"className":133},[],[135],{"type":48,"value":136},"CAPABILITIES.md ## Capabilities and modes",{"type":48,"value":138},"\nis the first stop — bulk compress is the canonical fit, tiny\none-shot is not.",{"type":43,"tag":140,"props":141,"children":143},"h2",{"id":142},"example-questions-this-skill-answers-well",[144],{"type":48,"value":145},"Example questions this skill answers well",{"type":43,"tag":51,"props":147,"children":148},{},[149,151,156],{"type":48,"value":150},"The CLASSES of DOCA Compress questions this skill is built to\nanswer, each with one worked example. The agent should treat the\n",{"type":43,"tag":83,"props":152,"children":153},{},[154],{"type":48,"value":155},"class",{"type":48,"value":157}," as the load-bearing piece — the worked example is a single\ninstance.",{"type":43,"tag":159,"props":160,"children":161},"ul",{},[162,215,291,335,393,448],{"type":43,"tag":163,"props":164,"children":165},"li",{},[166,171,173,178,180,188],{"type":43,"tag":55,"props":167,"children":168},{},[169],{"type":48,"value":170},"\"Should I offload this compression to DOCA Compress, or just\nzlib\u002Fzstd on the CPU?\"",{"type":48,"value":172}," — worked example: ",{"type":43,"tag":83,"props":174,"children":175},{},[176],{"type":48,"value":177},"\"I am compressing a\n4 MiB log buffer before writing it to storage; is doca-compress\nworth the setup vs zlib on the CPU?\"",{"type":48,"value":179},". Answered by the\nsize-threshold path-selection rule in\n",{"type":43,"tag":70,"props":181,"children":182},{"href":129},[183],{"type":43,"tag":75,"props":184,"children":186},{"className":185},[],[187],{"type":48,"value":136},{"type":43,"tag":159,"props":189,"children":190},{},[191],{"type":43,"tag":163,"props":192,"children":193},{},[194,196,201,203,213],{"type":48,"value":195},"the ",{"type":43,"tag":83,"props":197,"children":198},{},[199],{"type":48,"value":200},"\"when NOT to use doca-compress\"",{"type":48,"value":202}," bullets in\n",{"type":43,"tag":70,"props":204,"children":206},{"href":205},"CAPABILITIES.md#safety-policy",[207],{"type":43,"tag":75,"props":208,"children":210},{"className":209},[],[211],{"type":48,"value":212},"CAPABILITIES.md ## Safety policy",{"type":48,"value":214},".",{"type":43,"tag":163,"props":216,"children":217},{},[218,223,225,238,240,246,248,254,256,262,264,272],{"type":43,"tag":55,"props":219,"children":220},{},[221],{"type":48,"value":222},"\"Does my device support the compress (or decompress) task I\nwant, and how big a buffer can it move per submission?\"",{"type":48,"value":224}," —\nworked example: ",{"type":43,"tag":83,"props":226,"children":227},{},[228,230,236],{"type":48,"value":229},"\"is ",{"type":43,"tag":75,"props":231,"children":233},{"className":232},[],[234],{"type":48,"value":235},"doca_compress_task_compress_deflate",{"type":48,"value":237}," on\nthis BlueField, and what is the max source size per task?\"",{"type":48,"value":239},".\nAnswered by the per-task capability-query rule\n(",{"type":43,"tag":75,"props":241,"children":243},{"className":242},[],[244],{"type":48,"value":245},"doca_compress_cap_task_compress_deflate_is_supported",{"type":48,"value":247},",\n",{"type":43,"tag":75,"props":249,"children":251},{"className":250},[],[252],{"type":48,"value":253},"_decompress_deflate_is_supported",{"type":48,"value":255},", the matching\n",{"type":43,"tag":75,"props":257,"children":259},{"className":258},[],[260],{"type":48,"value":261},"_get_max_buf_size",{"type":48,"value":263}," queries) in\n",{"type":43,"tag":70,"props":265,"children":266},{"href":129},[267],{"type":43,"tag":75,"props":268,"children":270},{"className":269},[],[271],{"type":48,"value":136},{"type":43,"tag":159,"props":273,"children":274},{},[275],{"type":43,"tag":163,"props":276,"children":277},{},[278,280,290],{"type":48,"value":279},"the discovery step in\n",{"type":43,"tag":70,"props":281,"children":283},{"href":282},"TASKS.md#configure",[284],{"type":43,"tag":75,"props":285,"children":287},{"className":286},[],[288],{"type":48,"value":289},"TASKS.md ## configure",{"type":48,"value":214},{"type":43,"tag":163,"props":292,"children":293},{},[294,299,301,306,308,313,315,323,325,333],{"type":43,"tag":55,"props":295,"children":296},{},[297],{"type":48,"value":298},"\"I just need to decompress incoming network data — is that a\nvalid standalone use of doca-compress?\"",{"type":48,"value":300}," — worked example:\n",{"type":43,"tag":83,"props":302,"children":303},{},[304],{"type":48,"value":305},"\"my client receives DEFLATE-compressed payloads and never\nproduces any compressed output of its own\"",{"type":48,"value":307},". Answered by the\n",{"type":43,"tag":83,"props":309,"children":310},{},[311],{"type":48,"value":312},"\"decompression-only\"",{"type":48,"value":314}," note in\n",{"type":43,"tag":70,"props":316,"children":317},{"href":129},[318],{"type":43,"tag":75,"props":319,"children":321},{"className":320},[],[322],{"type":48,"value":136},{"type":48,"value":324},"\ntask-type table + the per-task configuration matrix in\n",{"type":43,"tag":70,"props":326,"children":327},{"href":282},[328],{"type":43,"tag":75,"props":329,"children":331},{"className":330},[],[332],{"type":48,"value":289},{"type":48,"value":334}," step 5.",{"type":43,"tag":163,"props":336,"children":337},{},[338,343,344,364,366,374],{"type":43,"tag":55,"props":339,"children":340},{},[341],{"type":48,"value":342},"\"What permissions does the source \u002F destination mmap need?\"",{"type":48,"value":224},{"type":43,"tag":83,"props":345,"children":346},{},[347,349,354,356,362],{"type":48,"value":348},"\"my ",{"type":43,"tag":75,"props":350,"children":352},{"className":351},[],[353],{"type":48,"value":235},{"type":48,"value":355},"\nreturns ",{"type":43,"tag":75,"props":357,"children":359},{"className":358},[],[360],{"type":48,"value":361},"DOCA_ERROR_NOT_PERMITTED",{"type":48,"value":363},"\"",{"type":48,"value":365},". Answered by the permission\nmatrix in\n",{"type":43,"tag":70,"props":367,"children":368},{"href":205},[369],{"type":43,"tag":75,"props":370,"children":372},{"className":371},[],[373],{"type":48,"value":212},{"type":43,"tag":159,"props":375,"children":376},{},[377],{"type":43,"tag":163,"props":378,"children":379},{},[380,382,392],{"type":48,"value":381},"the mmap-set-permissions checklist in\n",{"type":43,"tag":70,"props":383,"children":385},{"href":384},"TASKS.md#test",[386],{"type":43,"tag":75,"props":387,"children":389},{"className":388},[],[390],{"type":48,"value":391},"TASKS.md ## test",{"type":48,"value":214},{"type":43,"tag":163,"props":394,"children":395},{},[396,401,402,415,417,427,429,439,441,446],{"type":43,"tag":55,"props":397,"children":398},{},[399],{"type":48,"value":400},"\"Is this DOCA Compress API available on my installed DOCA\nversion?\"",{"type":48,"value":172},{"type":43,"tag":83,"props":403,"children":404},{},[405,407,413],{"type":48,"value":406},"\"is\n",{"type":43,"tag":75,"props":408,"children":410},{"className":409},[],[411],{"type":48,"value":412},"doca_compress_task_decompress_deflate",{"type":48,"value":414}," in the DOCA I have\ninstalled?\"",{"type":48,"value":416},". Answered by the version-compatibility overlay in\n",{"type":43,"tag":70,"props":418,"children":420},{"href":419},"CAPABILITIES.md#version-compatibility",[421],{"type":43,"tag":75,"props":422,"children":424},{"className":423},[],[425],{"type":48,"value":426},"CAPABILITIES.md ## Version compatibility",{"type":48,"value":428},",\nwhich cross-links the canonical detection chain in\n",{"type":43,"tag":70,"props":430,"children":432},{"href":431},"..\u002F..\u002Fdoca-version\u002FSKILL.md",[433],{"type":43,"tag":75,"props":434,"children":436},{"className":435},[],[437],{"type":48,"value":438},"doca-version",{"type":48,"value":440}," and adds the\nCompress-specific ",{"type":43,"tag":83,"props":442,"children":443},{},[444],{"type":48,"value":445},"\"discover per-task support, do not assume\"",{"type":48,"value":447},"\nbullets.",{"type":43,"tag":163,"props":449,"children":450},{},[451,464,465,484,486,496],{"type":43,"tag":55,"props":452,"children":453},{},[454,456,462],{"type":48,"value":455},"\"What does this ",{"type":43,"tag":75,"props":457,"children":459},{"className":458},[],[460],{"type":48,"value":461},"DOCA_ERROR_*",{"type":48,"value":463}," from a Compress call mean and\nwhich layer caused it?\"",{"type":48,"value":172},{"type":43,"tag":83,"props":466,"children":467},{},[468,469,475,477,483],{"type":48,"value":363},{"type":43,"tag":75,"props":470,"children":472},{"className":471},[],[473],{"type":48,"value":474},"DOCA_ERROR_INVALID_VALUE",{"type":48,"value":476},"\non ",{"type":43,"tag":75,"props":478,"children":480},{"className":479},[],[481],{"type":48,"value":482},"doca_compress_task_compress_deflate_alloc_init",{"type":48,"value":363},{"type":48,"value":485},". Answered\nby the Compress overlay on the cross-library taxonomy in\n",{"type":43,"tag":70,"props":487,"children":489},{"href":488},"CAPABILITIES.md#error-taxonomy",[490],{"type":43,"tag":75,"props":491,"children":493},{"className":492},[],[494],{"type":48,"value":495},"CAPABILITIES.md ## Error taxonomy",{"type":43,"tag":159,"props":497,"children":498},{},[499],{"type":43,"tag":163,"props":500,"children":501},{},[502,504,514,516,526],{"type":48,"value":503},"the layered ladder in\n",{"type":43,"tag":70,"props":505,"children":507},{"href":506},"TASKS.md#debug",[508],{"type":43,"tag":75,"props":509,"children":511},{"className":510},[],[512],{"type":48,"value":513},"TASKS.md ## debug",{"type":48,"value":515}," that escalates to\n",{"type":43,"tag":70,"props":517,"children":519},{"href":518},"..\u002F..\u002Fdoca-debug\u002FSKILL.md",[520],{"type":43,"tag":75,"props":521,"children":523},{"className":522},[],[524],{"type":48,"value":525},"doca-debug",{"type":48,"value":214},{"type":43,"tag":140,"props":528,"children":530},{"id":529},"audience",[531],{"type":48,"value":532},"Audience",{"type":43,"tag":51,"props":534,"children":535},{},[536,538,543,545,551,553,558],{"type":48,"value":537},"This skill serves ",{"type":43,"tag":55,"props":539,"children":540},{},[541],{"type":48,"value":542},"external developers building applications that\nconsume the DOCA Compress library",{"type":48,"value":544}," — i.e., users whose code calls\n",{"type":43,"tag":75,"props":546,"children":548},{"className":547},[],[549],{"type":48,"value":550},"doca_compress_*",{"type":48,"value":552}," (directly in C\u002FC++, or through FFI\u002Fbindings from\nanother language) to offload bulk DEFLATE compression or\ndecompression onto a BlueField DPU or ConnectX accelerator. It is\n",{"type":43,"tag":83,"props":554,"children":555},{},[556],{"type":48,"value":557},"not",{"type":48,"value":559}," for NVIDIA developers contributing to DOCA Compress itself.",{"type":43,"tag":51,"props":561,"children":562},{},[563,568,570,576,578,583,585,590,592,598],{"type":43,"tag":55,"props":564,"children":565},{},[566],{"type":48,"value":567},"Language scope.",{"type":48,"value":569}," DOCA Compress ships as a C library with\n",{"type":43,"tag":75,"props":571,"children":573},{"className":572},[],[574],{"type":48,"value":575},"pkg-config",{"type":48,"value":577}," module name ",{"type":43,"tag":75,"props":579,"children":581},{"className":580},[],[582],{"type":48,"value":4},{"type":48,"value":584},". The shipped samples are\nwritten in C. C and C++ consumers are the canonical case and the\nworked examples in ",{"type":43,"tag":75,"props":586,"children":588},{"className":587},[],[589],{"type":48,"value":72},{"type":48,"value":591}," assume that path. Other-language\nconsumers (Rust, Go, Python, …) consume the same ",{"type":43,"tag":75,"props":593,"children":595},{"className":594},[],[596],{"type":48,"value":597},"*.so",{"type":48,"value":599}," through\nFFI or language-specific bindings; the skill's contribution in\nthat case is to keep the lifecycle, capability-discovery,\npermission, error-taxonomy, and compress-vs-decompress guidance\nlanguage-neutral, and to route the agent to the public C ABI as\nthe authoritative surface that any wrapper will eventually call.",{"type":43,"tag":140,"props":601,"children":603},{"id":602},"when-to-load-this-skill",[604],{"type":48,"value":605},"When to load this skill",{"type":43,"tag":51,"props":607,"children":608},{},[609],{"type":48,"value":610},"Load this skill when the user is doing hands-on DOCA Compress\nwork, in any language. Concretely:",{"type":43,"tag":159,"props":612,"children":613},{},[614,656,675,704,717,744,749,771,783],{"type":43,"tag":163,"props":615,"children":616},{},[617,619,625,627,633,635,640,642,647,649,655],{"type":48,"value":618},"Initializing a ",{"type":43,"tag":75,"props":620,"children":622},{"className":621},[],[623],{"type":48,"value":624},"doca_compress",{"type":48,"value":626}," context on a ",{"type":43,"tag":75,"props":628,"children":630},{"className":629},[],[631],{"type":48,"value":632},"doca_dev",{"type":48,"value":634}," and\nconfiguring at least one task type\n(",{"type":43,"tag":75,"props":636,"children":638},{"className":637},[],[639],{"type":48,"value":235},{"type":48,"value":641}," and\u002For\n",{"type":43,"tag":75,"props":643,"children":645},{"className":644},[],[646],{"type":48,"value":412},{"type":48,"value":648},") before\n",{"type":43,"tag":75,"props":650,"children":652},{"className":651},[],[653],{"type":48,"value":654},"doca_ctx_start()",{"type":48,"value":214},{"type":43,"tag":163,"props":657,"children":658},{},[659,661,666,668,673],{"type":48,"value":660},"Choosing between the ",{"type":43,"tag":55,"props":662,"children":663},{},[664],{"type":48,"value":665},"compress",{"type":48,"value":667}," task and the ",{"type":43,"tag":55,"props":669,"children":670},{},[671],{"type":48,"value":672},"decompress",{"type":48,"value":674},"\ntask for the user's data flow. The two task types are\nindependent: a consumer that only decompresses inbound DEFLATE\npayloads is a valid shape and does not need to enable the\ncompress task.",{"type":43,"tag":163,"props":676,"children":677},{},[678,680,686,688,694,696,702],{"type":48,"value":679},"Setting permissions on ",{"type":43,"tag":75,"props":681,"children":683},{"className":682},[],[684],{"type":48,"value":685},"doca_mmap",{"type":48,"value":687}," correctly for the source\nbuffer (",{"type":43,"tag":75,"props":689,"children":691},{"className":690},[],[692],{"type":48,"value":693},"DOCA_ACCESS_FLAG_LOCAL_READ_ONLY",{"type":48,"value":695}," at minimum) and the\ndestination buffer (",{"type":43,"tag":75,"props":697,"children":699},{"className":698},[],[700],{"type":48,"value":701},"DOCA_ACCESS_FLAG_LOCAL_READ_WRITE",{"type":48,"value":703},").",{"type":43,"tag":163,"props":705,"children":706},{},[707,709,715],{"type":48,"value":708},"Sizing the source buffer against the per-task\n",{"type":43,"tag":75,"props":710,"children":712},{"className":711},[],[713],{"type":48,"value":714},"doca_compress_cap_task_*_get_max_buf_size(devinfo)",{"type":48,"value":716}," ceiling\nand the destination buffer against the worst-case output size\nthe algorithm can produce on this input.",{"type":43,"tag":163,"props":718,"children":719},{},[720,722,727,729,735,737,743],{"type":48,"value":721},"Checking which task types this device's accelerator advertises\nvia ",{"type":43,"tag":75,"props":723,"children":725},{"className":724},[],[726],{"type":48,"value":245},{"type":48,"value":728}," and\n",{"type":43,"tag":75,"props":730,"children":732},{"className":731},[],[733],{"type":48,"value":734},"doca_compress_cap_task_decompress_deflate_is_supported",{"type":48,"value":736},"\nagainst the active ",{"type":43,"tag":75,"props":738,"children":740},{"className":739},[],[741],{"type":48,"value":742},"doca_devinfo",{"type":48,"value":214},{"type":43,"tag":163,"props":745,"children":746},{},[747],{"type":48,"value":748},"Validating against a round-trip (compress → decompress, or\ndecompress → compare against a known DEFLATE-encoded fixture)\nbefore pushing bulk input through the accelerator.",{"type":43,"tag":163,"props":750,"children":751},{},[752,754,759,761,769],{"type":48,"value":753},"Deciding ",{"type":43,"tag":83,"props":755,"children":756},{},[757],{"type":48,"value":758},"whether to offload at all",{"type":48,"value":760}," — the size-threshold rule\nin ",{"type":43,"tag":70,"props":762,"children":763},{"href":129},[764],{"type":43,"tag":75,"props":765,"children":767},{"className":766},[],[768],{"type":48,"value":136},{"type":48,"value":770},"\nsays doca-compress is the right answer only for bulk inputs\n(rule of thumb: ≥ a few KiB); below that, CPU compression beats\nthe DMA-to-accelerator round-trip.",{"type":43,"tag":163,"props":772,"children":773},{},[774,776,781],{"type":48,"value":775},"Debugging a ",{"type":43,"tag":75,"props":777,"children":779},{"className":778},[],[780],{"type":48,"value":461},{"type":48,"value":782}," returned from a Compress call\n(lifecycle vs. buffer-sizing vs. permission vs.\nunsupported-task) and the task-completion event on the progress\nengine.",{"type":43,"tag":163,"props":784,"children":785},{},[786],{"type":48,"value":787},"Designing or extending non-C bindings (Rust, Go, Python, …)\nthat wrap the Compress C ABI — for the lifecycle, permission,\ncapability, and compress-vs-decompress rules the wrapper must\nhonor.",{"type":43,"tag":51,"props":789,"children":790},{},[791,793,797,799,809],{"type":48,"value":792},"Do ",{"type":43,"tag":55,"props":794,"children":795},{},[796],{"type":48,"value":557},{"type":48,"value":798}," load this skill for general DOCA orientation, install\nof DOCA itself, non-DEFLATE compression libraries on CPU (use\nzlib \u002F zstd \u002F similar), or other DOCA libraries. For those, use\n",{"type":43,"tag":70,"props":800,"children":802},{"href":801},"..\u002F..\u002Fdoca-public-knowledge-map\u002FSKILL.md",[803],{"type":43,"tag":75,"props":804,"children":806},{"className":805},[],[807],{"type":48,"value":808},"doca-public-knowledge-map",{"type":48,"value":214},{"type":43,"tag":140,"props":811,"children":813},{"id":812},"what-this-skill-provides",[814],{"type":48,"value":815},"What this skill provides",{"type":43,"tag":51,"props":817,"children":818},{},[819,821,826],{"type":48,"value":820},"This is a ",{"type":43,"tag":55,"props":822,"children":823},{},[824],{"type":48,"value":825},"thin loader",{"type":48,"value":827},". The body keeps only the orientation\nneeded to pick the right next file. The substantive\nCompress-specific material lives in two companion files:",{"type":43,"tag":159,"props":829,"children":830},{},[831,864],{"type":43,"tag":163,"props":832,"children":833},{},[834,839,841,847,849,854,856,862],{"type":43,"tag":75,"props":835,"children":837},{"className":836},[],[838],{"type":48,"value":92},{"type":48,"value":840}," — what DOCA Compress can express on this\nversion: the four task types (compress-deflate,\ndecompress-deflate, decompress-lz4-stream, and\ndecompress-lz4-block, each independently capability-gated), the\nper-task capability-query surface (",{"type":43,"tag":75,"props":842,"children":844},{"className":843},[],[845],{"type":48,"value":846},"doca_compress_cap_*",{"type":48,"value":848}," for\ntask support and per-submission max buffer size), the Compress\nerror taxonomy (mapped onto the cross-library ",{"type":43,"tag":75,"props":850,"children":852},{"className":851},[],[853],{"type":48,"value":461},{"type":48,"value":855},"\nset), the observability surface (per-task completion events on\nthe progress engine), the safety policy that gates source \u002F\ndestination mmap permission decisions, and the size-threshold\npath-selection rule (when to use doca-compress versus CPU\ncompression or ",{"type":43,"tag":75,"props":857,"children":859},{"className":858},[],[860],{"type":48,"value":861},"doca-dma",{"type":48,"value":863}," for the no-compression copy case).",{"type":43,"tag":163,"props":865,"children":866},{},[867,872,874,880,882,888,889,895,896,902,903,909,910,916,918,924,926,932,934,940,941,947],{"type":43,"tag":75,"props":868,"children":870},{"className":869},[],[871],{"type":48,"value":72},{"type":48,"value":873}," — step-by-step workflows for the six in-scope\nCompress verbs: ",{"type":43,"tag":75,"props":875,"children":877},{"className":876},[],[878],{"type":48,"value":879},"configure",{"type":48,"value":881},", ",{"type":43,"tag":75,"props":883,"children":885},{"className":884},[],[886],{"type":48,"value":887},"build",{"type":48,"value":881},{"type":43,"tag":75,"props":890,"children":892},{"className":891},[],[893],{"type":48,"value":894},"modify",{"type":48,"value":881},{"type":43,"tag":75,"props":897,"children":899},{"className":898},[],[900],{"type":48,"value":901},"run",{"type":48,"value":881},{"type":43,"tag":75,"props":904,"children":906},{"className":905},[],[907],{"type":48,"value":908},"test",{"type":48,"value":247},{"type":43,"tag":75,"props":911,"children":913},{"className":912},[],[914],{"type":48,"value":915},"debug",{"type":48,"value":917},". Plus a ",{"type":43,"tag":75,"props":919,"children":921},{"className":920},[],[922],{"type":48,"value":923},"## rollback",{"type":48,"value":925}," overlay (Compress-specific\nfive-step teardown that drains in-flight tasks before\n",{"type":43,"tag":75,"props":927,"children":929},{"className":928},[],[930],{"type":48,"value":931},"doca_ctx_stop",{"type":48,"value":933},", unregisters mmap regions in reverse-register\norder, and re-verifies the device path with the round-trip\nsmoke) and the 5-phase universal debug-loop instantiation\nappended to ",{"type":43,"tag":75,"props":935,"children":937},{"className":936},[],[938],{"type":48,"value":939},"## debug",{"type":48,"value":917},{"type":43,"tag":75,"props":942,"children":944},{"className":943},[],[945],{"type":48,"value":946},"Deferred task verbs",{"type":48,"value":948}," block\nthat points out-of-scope questions at the right next skill.",{"type":43,"tag":51,"props":950,"children":951},{},[952,954,962],{"type":48,"value":953},"The skill assumes a host or BlueField where DOCA is already\ninstalled at the standard location and the user has the\nprivileges their public install profile expects. It does not\ncover installing DOCA — that path goes through\n",{"type":43,"tag":70,"props":955,"children":956},{"href":110},[957],{"type":43,"tag":75,"props":958,"children":960},{"className":959},[],[961],{"type":48,"value":117},{"type":48,"value":214},{"type":43,"tag":140,"props":964,"children":966},{"id":965},"what-this-skill-deliberately-does-not-ship",[967],{"type":48,"value":968},"What this skill deliberately does not ship",{"type":43,"tag":51,"props":970,"children":971},{},[972,974,979],{"type":48,"value":973},"This skill is ",{"type":43,"tag":55,"props":975,"children":976},{},[977],{"type":48,"value":978},"agent guidance",{"type":48,"value":980},", not a samples or templates\nbundle. To keep the boundary clean, it deliberately does not\ncontain — and pull requests should not add:",{"type":43,"tag":159,"props":982,"children":983},{},[984,1025,1050,1097],{"type":43,"tag":163,"props":985,"children":986},{},[987,992,994,1000,1002,1012,1014,1024],{"type":43,"tag":55,"props":988,"children":989},{},[990],{"type":48,"value":991},"Pre-written DOCA Compress application source code, in any\nlanguage.",{"type":48,"value":993}," The verified Compress source code is the shipped C\nsamples at ",{"type":43,"tag":75,"props":995,"children":997},{"className":996},[],[998],{"type":48,"value":999},"\u002Fopt\u002Fmellanox\u002Fdoca\u002Fsamples\u002Fdoca_compress\u002F",{"type":48,"value":1001},", plus\nthe File Compression reference application linked from the\npublic DOCA Compress guide. The agent's job is to route the\nuser to those files and prescribe a minimum-diff modification\non them via the universal modify-a-sample workflow in\n",{"type":43,"tag":70,"props":1003,"children":1005},{"href":1004},"..\u002F..\u002Fdoca-programming-guide\u002FSKILL.md",[1006],{"type":43,"tag":75,"props":1007,"children":1009},{"className":1008},[],[1010],{"type":48,"value":1011},"doca-programming-guide",{"type":48,"value":1013},",\nlayered with the Compress-specific overrides in\n",{"type":43,"tag":70,"props":1015,"children":1017},{"href":1016},"TASKS.md#modify",[1018],{"type":43,"tag":75,"props":1019,"children":1021},{"className":1020},[],[1022],{"type":48,"value":1023},"TASKS.md ## modify",{"type":48,"value":214},{"type":43,"tag":163,"props":1026,"children":1027},{},[1028,1033,1035,1040,1042,1048],{"type":43,"tag":55,"props":1029,"children":1030},{},[1031],{"type":48,"value":1032},"Pre-encoded DEFLATE fixtures for arbitrary inputs.",{"type":48,"value":1034}," The\nskill tells the agent to use a ",{"type":43,"tag":83,"props":1036,"children":1037},{},[1038],{"type":48,"value":1039},"round-trip",{"type":48,"value":1041}," validation\n(compress → decompress and compare to the original, or\ndecompress a known DEFLATE blob produced by a published CPU\nencoder like ",{"type":43,"tag":75,"props":1043,"children":1045},{"className":1044},[],[1046],{"type":48,"value":1047},"zlib",{"type":48,"value":1049},") as the known-good smoke; it does not ship\na fixture bank of its own.",{"type":43,"tag":163,"props":1051,"children":1052},{},[1053,1058,1060,1066,1067,1073,1074,1080,1082,1087,1089,1095],{"type":43,"tag":55,"props":1054,"children":1055},{},[1056],{"type":48,"value":1057},"Standalone build manifests",{"type":48,"value":1059}," (",{"type":43,"tag":75,"props":1061,"children":1063},{"className":1062},[],[1064],{"type":48,"value":1065},"meson.build",{"type":48,"value":881},{"type":43,"tag":75,"props":1068,"children":1070},{"className":1069},[],[1071],{"type":48,"value":1072},"CMakeLists.txt",{"type":48,"value":247},{"type":43,"tag":75,"props":1075,"children":1077},{"className":1076},[],[1078],{"type":48,"value":1079},"Cargo.toml",{"type":48,"value":1081},", …) parked inside the skill. The agent constructs\nthe build manifest ",{"type":43,"tag":83,"props":1083,"children":1084},{},[1085],{"type":48,"value":1086},"in the user's project directory",{"type":48,"value":1088}," against\nthe user's installed DOCA, where ",{"type":43,"tag":75,"props":1090,"children":1092},{"className":1091},[],[1093],{"type":48,"value":1094},"pkg-config --modversion doca-compress",{"type":48,"value":1096}," is the source of truth.",{"type":43,"tag":163,"props":1098,"children":1099},{},[1100,1128],{"type":43,"tag":55,"props":1101,"children":1102},{},[1103,1105,1111,1112,1118,1120,1126],{"type":48,"value":1104},"A ",{"type":43,"tag":75,"props":1106,"children":1108},{"className":1107},[],[1109],{"type":48,"value":1110},"samples\u002F",{"type":48,"value":881},{"type":43,"tag":75,"props":1113,"children":1115},{"className":1114},[],[1116],{"type":48,"value":1117},"bindings\u002F",{"type":48,"value":1119},", or ",{"type":43,"tag":75,"props":1121,"children":1123},{"className":1122},[],[1124],{"type":48,"value":1125},"reference\u002F",{"type":48,"value":1127}," subtree",{"type":48,"value":1129}," 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":43,"tag":140,"props":1131,"children":1133},{"id":1132},"loading-order",[1134],{"type":48,"value":1135},"Loading order",{"type":43,"tag":1137,"props":1138,"children":1139},"ol",{},[1140,1153,1166],{"type":43,"tag":163,"props":1141,"children":1142},{},[1143,1145,1151],{"type":48,"value":1144},"Read this ",{"type":43,"tag":75,"props":1146,"children":1148},{"className":1147},[],[1149],{"type":48,"value":1150},"SKILL.md",{"type":48,"value":1152}," first to confirm the user's question is in\nscope.",{"type":43,"tag":163,"props":1154,"children":1155},{},[1156],{"type":43,"tag":55,"props":1157,"children":1158},{},[1159,1161,1165],{"type":48,"value":1160},"For the Compress capability matrix, the compress \u002F decompress\ntask split (including decompression-only as a standalone\nshape), per-task capability-query rules, permission matrix,\nerror taxonomy, observability, and safety \u002F path-selection\npolicy, see ",{"type":43,"tag":70,"props":1162,"children":1163},{"href":92},[1164],{"type":48,"value":92},{"type":48,"value":214},{"type":43,"tag":163,"props":1167,"children":1168},{},[1169],{"type":43,"tag":55,"props":1170,"children":1171},{},[1172,1174,1178],{"type":48,"value":1173},"For step-by-step workflows — configure, build, modify, run,\ntest, debug — see ",{"type":43,"tag":70,"props":1175,"children":1176},{"href":72},[1177],{"type":48,"value":72},{"type":48,"value":214},{"type":43,"tag":51,"props":1180,"children":1181},{},[1182,1184,1192,1194,1202],{"type":48,"value":1183},"Both companion files cross-link to each other,\n",{"type":43,"tag":70,"props":1185,"children":1186},{"href":431},[1187],{"type":43,"tag":75,"props":1188,"children":1190},{"className":1189},[],[1191],{"type":48,"value":438},{"type":48,"value":1193}," for the canonical\nversion-handling rules, and\n",{"type":43,"tag":70,"props":1195,"children":1196},{"href":801},[1197],{"type":43,"tag":75,"props":1198,"children":1200},{"className":1199},[],[1201],{"type":48,"value":808},{"type":48,"value":1203},"\nwhenever the right answer is \"look it up in the public docs or\nthe installed package layout\" rather than \"Compress-specific\nguidance\".",{"type":43,"tag":140,"props":1205,"children":1207},{"id":1206},"related-skills",[1208],{"type":48,"value":1209},"Related skills",{"type":43,"tag":159,"props":1211,"children":1212},{},[1213,1234,1254,1285,1306,1333,1353],{"type":43,"tag":163,"props":1214,"children":1215},{},[1216,1224,1226,1232],{"type":43,"tag":70,"props":1217,"children":1218},{"href":801},[1219],{"type":43,"tag":75,"props":1220,"children":1222},{"className":1221},[],[1223],{"type":48,"value":808},{"type":48,"value":1225}," —\nthe routing table for every public DOCA documentation source\nand the on-disk layout of an installed DOCA package. The DOCA\nCompress page lives at\n",{"type":43,"tag":75,"props":1227,"children":1229},{"className":1228},[],[1230],{"type":48,"value":1231},"docs.nvidia.com\u002Fdoca\u002Fsdk\u002FDOCA-Compress\u002F",{"type":48,"value":1233},"; the File Compression\nreference application is the canonical worked example.",{"type":43,"tag":163,"props":1235,"children":1236},{},[1237,1245,1247,1252],{"type":43,"tag":70,"props":1238,"children":1239},{"href":110},[1240],{"type":43,"tag":75,"props":1241,"children":1243},{"className":1242},[],[1244],{"type":48,"value":117},{"type":48,"value":1246}," — env preparation,\ninstall verification, and the ",{"type":43,"tag":83,"props":1248,"children":1249},{},[1250],{"type":48,"value":1251},"I have no install yet",{"type":48,"value":1253}," path\nwith the public NGC DOCA container. This skill assumes its\npreconditions are satisfied.",{"type":43,"tag":163,"props":1255,"children":1256},{},[1257,1265,1267,1276,1278,1283],{"type":43,"tag":70,"props":1258,"children":1259},{"href":431},[1260],{"type":43,"tag":75,"props":1261,"children":1263},{"className":1262},[],[1264],{"type":48,"value":438},{"type":48,"value":1266}," — canonical DOCA\nversion-handling rules. This skill's\n",{"type":43,"tag":70,"props":1268,"children":1269},{"href":419},[1270],{"type":43,"tag":75,"props":1271,"children":1273},{"className":1272},[],[1274],{"type":48,"value":1275},"## Version compatibility",{"type":48,"value":1277},"\ncross-links the four-way match rule and adds only the\nCompress-specific ",{"type":43,"tag":83,"props":1279,"children":1280},{},[1281],{"type":48,"value":1282},"\"discover per-task support + per-task max\nbuffer size via cap query\"",{"type":48,"value":1284}," overlay.",{"type":43,"tag":163,"props":1286,"children":1287},{},[1288,1298,1300,1304],{"type":43,"tag":70,"props":1289,"children":1291},{"href":1290},"..\u002F..\u002Fdoca-structured-tools-contract\u002FSKILL.md",[1292],{"type":43,"tag":75,"props":1293,"children":1295},{"className":1294},[],[1296],{"type":48,"value":1297},"doca-structured-tools-contract",{"type":48,"value":1299}," —\nthe bundle's structured-tools precedence rule (detect \u002F prefer\n\u002F fall back \u002F report). The Command appendix in\n",{"type":43,"tag":70,"props":1301,"children":1302},{"href":72},[1303],{"type":48,"value":72},{"type":48,"value":1305}," honors this contract.",{"type":43,"tag":163,"props":1307,"children":1308},{},[1309,1317,1319,1324,1326,1331],{"type":43,"tag":70,"props":1310,"children":1311},{"href":1004},[1312],{"type":43,"tag":75,"props":1313,"children":1315},{"className":1314},[],[1316],{"type":48,"value":1011},{"type":48,"value":1318}," —\ngeneral DOCA programming patterns shared by every library: the\ncanonical ",{"type":43,"tag":75,"props":1320,"children":1322},{"className":1321},[],[1323],{"type":48,"value":575},{"type":48,"value":1325}," + meson build pattern, the universal\nmodify-a-shipped-sample first-app workflow, the universal\nlifecycle, the cross-library ",{"type":43,"tag":75,"props":1327,"children":1329},{"className":1328},[],[1330],{"type":48,"value":461},{"type":48,"value":1332}," taxonomy, and the\nprogram-side debug order. This skill layers Compress specifics\non top.",{"type":43,"tag":163,"props":1334,"children":1335},{},[1336,1345,1347,1351],{"type":43,"tag":70,"props":1337,"children":1339},{"href":1338},"..\u002Fdoca-dma\u002FSKILL.md",[1340],{"type":43,"tag":75,"props":1341,"children":1343},{"className":1342},[],[1344],{"type":48,"value":861},{"type":48,"value":1346}," — the right library when\nthe data flow is a pure mmap-to-mmap copy with no compression\nrequired. This skill's path-selection rule routes to DMA when\nCompress is ",{"type":43,"tag":83,"props":1348,"children":1349},{},[1350],{"type":48,"value":557},{"type":48,"value":1352}," the answer (e.g. the user only wants to move\nbytes, not encode them).",{"type":43,"tag":163,"props":1354,"children":1355},{},[1356,1364],{"type":43,"tag":70,"props":1357,"children":1358},{"href":518},[1359],{"type":43,"tag":75,"props":1360,"children":1362},{"className":1361},[],[1363],{"type":48,"value":525},{"type":48,"value":1365}," — the cross-cutting\ndebug ladder (install \u002F version \u002F build \u002F link \u002F runtime \u002F\nprogram \u002F driver). Compress-specific debug (task-not-supported,\nsource-buffer-too-large, destination-buffer-too-small) overlays\non top of that ladder.",{"items":1367,"total":1469},[1368,1383,1397,1411,1423,1440,1455],{"slug":1369,"name":1369,"fn":1370,"description":1371,"org":1372,"tags":1373,"stars":23,"repoUrl":24,"updatedAt":1382},"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},[1374,1377,1380,1381],{"name":1375,"slug":1376,"type":15},"Data Analysis","data-analysis",{"name":1378,"slug":1379,"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":1384,"name":1384,"fn":1385,"description":1386,"org":1387,"tags":1388,"stars":23,"repoUrl":24,"updatedAt":1396},"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},[1389,1392,1395],{"name":1390,"slug":1391,"type":15},"Deployment","deployment",{"name":1393,"slug":1394,"type":15},"Infrastructure","infrastructure",{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":1398,"name":1398,"fn":1399,"description":1400,"org":1401,"tags":1402,"stars":23,"repoUrl":24,"updatedAt":1410},"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},[1403,1406,1407],{"name":1404,"slug":1405,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":1408,"slug":1409,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":1412,"name":1412,"fn":1413,"description":1414,"org":1415,"tags":1416,"stars":23,"repoUrl":24,"updatedAt":1422},"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},[1417,1418,1419],{"name":1375,"slug":1376,"type":15},{"name":9,"slug":8,"type":15},{"name":1420,"slug":1421,"type":15},"Testing","testing","2026-07-17T05:29:03.913266",{"slug":1424,"name":1424,"fn":1425,"description":1426,"org":1427,"tags":1428,"stars":23,"repoUrl":24,"updatedAt":1439},"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},[1429,1432,1435,1436],{"name":1430,"slug":1431,"type":15},"Automation","automation",{"name":1433,"slug":1434,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":1437,"slug":1438,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":1441,"name":1441,"fn":1442,"description":1443,"org":1444,"tags":1445,"stars":23,"repoUrl":24,"updatedAt":1454},"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},[1446,1447,1450,1451],{"name":1390,"slug":1391,"type":15},{"name":1448,"slug":1449,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":1452,"slug":1453,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":1456,"name":1456,"fn":1457,"description":1458,"org":1459,"tags":1460,"stars":23,"repoUrl":24,"updatedAt":1468},"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},[1461,1462,1465],{"name":9,"slug":8,"type":15},{"name":1463,"slug":1464,"type":15},"Quantum Computing","quantum-computing",{"name":1466,"slug":1467,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305,{"items":1471,"total":1622},[1472,1490,1506,1517,1529,1543,1556,1570,1581,1590,1604,1613],{"slug":1473,"name":1473,"fn":1474,"description":1475,"org":1476,"tags":1477,"stars":1487,"repoUrl":1488,"updatedAt":1489},"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},[1478,1481,1484],{"name":1479,"slug":1480,"type":15},"Documentation","documentation",{"name":1482,"slug":1483,"type":15},"MCP","mcp",{"name":1485,"slug":1486,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":1491,"name":1491,"fn":1492,"description":1493,"org":1494,"tags":1495,"stars":1503,"repoUrl":1504,"updatedAt":1505},"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},[1496,1499,1500],{"name":1497,"slug":1498,"type":15},"Containers","containers",{"name":1390,"slug":1391,"type":15},{"name":1501,"slug":1502,"type":15},"Python","python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":1507,"name":1507,"fn":1508,"description":1509,"org":1510,"tags":1511,"stars":1503,"repoUrl":1504,"updatedAt":1516},"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},[1512,1515],{"name":1513,"slug":1514,"type":15},"CI\u002FCD","ci-cd",{"name":1390,"slug":1391,"type":15},"2026-07-14T05:25:59.97109",{"slug":1518,"name":1518,"fn":1519,"description":1520,"org":1521,"tags":1522,"stars":1503,"repoUrl":1504,"updatedAt":1528},"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},[1523,1524,1525],{"name":1513,"slug":1514,"type":15},{"name":1390,"slug":1391,"type":15},{"name":1526,"slug":1527,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":1530,"name":1530,"fn":1531,"description":1532,"org":1533,"tags":1534,"stars":1503,"repoUrl":1504,"updatedAt":1542},"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},[1535,1538,1539],{"name":1536,"slug":1537,"type":15},"Debugging","debugging",{"name":1526,"slug":1527,"type":15},{"name":1540,"slug":1541,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":1544,"name":1544,"fn":1545,"description":1546,"org":1547,"tags":1548,"stars":1503,"repoUrl":1504,"updatedAt":1555},"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},[1549,1552],{"name":1550,"slug":1551,"type":15},"Best Practices","best-practices",{"name":1553,"slug":1554,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":1557,"name":1557,"fn":1558,"description":1559,"org":1560,"tags":1561,"stars":1503,"repoUrl":1504,"updatedAt":1569},"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},[1562,1565,1568],{"name":1563,"slug":1564,"type":15},"Machine Learning","machine-learning",{"name":1566,"slug":1567,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":1571,"name":1571,"fn":1572,"description":1573,"org":1574,"tags":1575,"stars":1503,"repoUrl":1504,"updatedAt":1580},"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},[1576,1579],{"name":1577,"slug":1578,"type":15},"QA","qa",{"name":1420,"slug":1421,"type":15},"2026-07-14T05:25:53.673039",{"slug":1582,"name":1582,"fn":1583,"description":1584,"org":1585,"tags":1586,"stars":1503,"repoUrl":1504,"updatedAt":1589},"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},[1587,1588],{"name":1390,"slug":1391,"type":15},{"name":1393,"slug":1394,"type":15},"2026-07-14T05:25:49.362534",{"slug":1591,"name":1591,"fn":1592,"description":1593,"org":1594,"tags":1595,"stars":1503,"repoUrl":1504,"updatedAt":1603},"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},[1596,1599,1600],{"name":1597,"slug":1598,"type":15},"Code Review","code-review",{"name":1526,"slug":1527,"type":15},{"name":1601,"slug":1602,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":1605,"name":1605,"fn":1606,"description":1607,"org":1608,"tags":1609,"stars":1503,"repoUrl":1504,"updatedAt":1612},"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},[1610,1611],{"name":1577,"slug":1578,"type":15},{"name":1420,"slug":1421,"type":15},"2026-07-14T05:25:54.928983",{"slug":1614,"name":1614,"fn":1615,"description":1616,"org":1617,"tags":1618,"stars":1503,"repoUrl":1504,"updatedAt":1621},"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},[1619,1620],{"name":1430,"slug":1431,"type":15},{"name":1513,"slug":1514,"type":15},"2026-07-30T05:29:03.275638",496]