К основному содержанию
скилл CX Codex

doca-sha — скилл для Claude Code: SHA и установка

Скилл doca-sha для Claude Code помогает работать с аппаратным ускорением SHA на BlueField-2 через DOCA. Выбирает режим хеширования и разбирает ошибки.

Apache-2.0 131 установок ★ 4 106 обновлён 2026-10-23

Установить в Claude Code · Через Skills CLI (Vercel) Все способы и платформы ↓
npx skills add NVIDIA/skills --skill doca-sha -a claude-code
Открыть на GitHub

Что делает doca-sha

Скилл ориентирован на приложения, которые вызывают библиотеку doca_sha_* напрямую или через FFI. Он помогает выбрать однократный doca_sha_task_hash для одного буфера либо инкрементальный doca_sha_task_partial_hash для данных, поступающих частями. Также разбирает проверку алгоритмов, размеров буферов и выравнивания через doca_sha_cap_*.

Перед работой нужен установленный DOCA SDK в /opt/mellanox/doca на Linux с подключённым BlueField-2 DPU. Проверка установки использует pkg-config doca-sha и каталоги SDK. В комплекте описаны сценарии для C и C++; другие языки подключают ту же C ABI через FFI.

BlueField-3 не имеет SHA-движка, поэтому ускорение через этот скилл на нём не сработает. Для установки SDK и подготовки BFB скилл не предназначен: такие задачи нужно передавать doca-setup.

Какие задачи решает

  • выбрать однократное или частичное хеширование
  • проверить поддержку SHA-1, SHA-256 или SHA-512
  • рассчитать минимальный размер буфера дайджеста
  • разобрать ошибку DOCA_ERROR_BAD_STATE
  • проверить права mmap для исходного и целевого буфера

Примеры запросов агенту

  • Проверь, поддерживает ли мой BlueField-2 SHA-256 и какой размер выходного буфера нужен.
  • Выбери между doca_sha_task_hash и doca_sha_task_partial_hash для входа, который приходит частями.
  • Разбери DOCA_ERROR_BAD_STATE при продолжении частичного хеширования и проверь установку DOCA через pkg-config doca-sha.

Как установить doca-sha

1Через Skills CLI (Vercel)

npx skills add NVIDIA/skills --skill doca-sha -a claude-code

2Вручную: положить файлы

~/.claude/skills/doca-sha/SKILL.md

Папка скилла целиком; для одного проекта — .claude/skills/doca-sha/

По шагам, с командами для каждой платформы: Как установить скилл в Claude Code: пять способов

Вопросы о doca-sha

Как установить doca-sha в Claude Code?

Выполните npx skills add NVIDIA/skills --skill doca-sha -a claude-code, затем используйте файл ~/.claude/skills/doca-sha/SKILL.md.

Что делает doca-sha?

Скилл помогает использовать SHA-1, SHA-256 и SHA-512 через doca_sha на BlueField-2. Он объясняет выбор между однократным и частичным хешированием, проверку возможностей устройства и разбор ошибок DOCA.

Что нужно для работы doca-sha?

Нужны DOCA SDK в /opt/mellanox/doca, Linux и подключённый BlueField-2 DPU. BlueField-3 не имеет SHA-движка.

Исходный текст

SKILL.md · Apache-2.0
---
license: Apache-2.0
name: doca-sha
description: >
  Use this skill when the user needs SHA-1, SHA-256, or SHA-512 digest
  acceleration on BlueField-2 silicon via doca_sha (absent on BF-3) —
  selecting one-shot `doca_sha_task_hash` versus incremental
  `doca_sha_task_partial_hash` for chunked input, consulting
  `doca_sha_cap_*` for per-algorithm digest-size limits and input
  alignment, or decoding DOCA_ERROR_* statuses such as BAD_STATE on a
  resumed partial hash. Also load it for implied asks like "accelerate
  SHA-256 on the NIC", "streaming digest over chunks", "digest output
  size mismatch", or "is SHA-512 present on this card". Refuse and
  route to the proper owner for cryptographic-hash theory (collision
  resistance, SHA-3 selection), the AES-GCM / Compress / DMA
  libraries, or SDK install and BFB bring-up.
metadata:
  author: NVIDIA <[email protected]>
  kind: library
  doca_version: "3.5"
  tags:
    - doca
    - sha
    - hash
compatibility: >
  Requires DOCA SDK installed at /opt/mellanox/doca on Linux (Ubuntu
  22.04/24.04 or RHEL/SLES) with a BlueField-2 DPU attached
  (BlueField-3 has no SHA engine). Reads the user's local install via `pkg-config doca-sha` and inspects
  /opt/mellanox/doca/{lib,include,samples,applications}.
  The validation steps require a live DOCA install.
---

# DOCA SHA

## Examples

- **"Offload SHA-256 to the BlueField."** — this is in scope. Use the
  shipped SHA sample and `doca_sha_cap_*` before picking SHA-1 /
  SHA-256 / SHA-512.
- **"Partial hash returns BAD_STATE."** — stay on the incremental
  `doca_sha_task_partial_hash` lifecycle; do not switch to a CPU
  hash library mid-stream.

## Instructions

Follow Loading order below. Open the tasks companion for configure /
build / run. Open the capabilities companion for one-shot vs
partial hash and algorithm support. Do not open every companion file.

**Where to start:** This skill assumes DOCA is already installed and
the user is doing **hands-on SHA-acceleration work** on a BlueField-2
DPU (Arm side or its host) with DOCA. BlueField-3 has no SHA engine. Open the tasks companion if the user
wants to *do* something (configure / build / modify / run / test /
debug); open the capabilities companion when the question
is *what can DOCA SHA express* on this version. If the user has not
installed DOCA yet, route to
`doca-setup/SKILL.md` first. If the user is
asking *"should I even use the accelerator for this hash?"*, the
path-selection rule in
## Capabilities and modes
is the first stop.

**How to use this skill.** Read this file, then search the companion file
for the `## <name>` heading the question needs and read that whole section.
Answer completely from it, quoting only its symbols and commands. Do not
browse docs; use the entry points in `doca-public-knowledge-map`. Give
environment checks as commands for the user's DOCA host. For out-of-scope
requests, name the skill that owns them and stop.

## Example questions this skill answers well

The CLASSES of DOCA SHA questions this skill is built to answer,
each with one worked example. The agent should treat the *class* as
the load-bearing piece — the worked example is a single instance.

- **"Should I offload this hash to DOCA SHA, or just compute it on
  the CPU?"** — worked example: *"I am verifying file integrity on
  a 4 GiB image; is doca-sha worth the setup vs OpenSSL on the
  CPU?"*. Answered by the path-selection table in
  ## Capabilities and modes
  + the *"when NOT to use doca-sha"* bullets in
  ## Safety policy.
- **"Does my device support the SHA algorithm I want?"** — worked
  example: *"is SHA-256 in the accelerator on this BlueField, and
  what is the minimum destination buffer size for it?"*. Answered
  by the algorithm + buffer-sizing capability-query rule
  (`doca_sha_cap_task_hash_get_supported(devinfo, algorithm)`
  for the one-shot path; `_task_partial_hash_get_supported(devinfo, algorithm)`
  for the partial-hash path; `doca_sha_cap_get_min_dst_buf_size`,
  `doca_sha_cap_get_max_src_buf_size`) in
  ## Capabilities and modes
  + the discovery step in
  ## configure.
- **"How do I pick between the one-shot and the partial / incremental
  hash task?"** — worked example: *"my input is 1 GiB and the device
  cap says max source buffer is 64 MiB"*. Answered by the
  one-shot-vs-partial table in
  ## Capabilities and modes
  + the task-config workflow in
  ## modify.
- **"What permissions does the source / destination mmap need?"** —
  worked example: *"my `doca_sha_task_hash` returns
  `DOCA_ERROR_NOT_PERMITTED`"*. Answered by the permission matrix
  in ## Safety policy
  + the mmap-set-permissions checklist in
  ## test.
- **"Is this DOCA SHA API available on my installed DOCA version?"**
  — worked example: *"is `doca_sha_task_partial_hash` in the DOCA
  I have installed?"*. Answered by the version-compatibility
  overlay in
  ## Version compatibility,
  which cross-links the canonical detection chain in
  `doca-version/SKILL.md` and adds the
  SHA-specific *"discover, do not assume"* bullets.
- **"What does this `DOCA_ERROR_*` from a SHA call mean and which
  layer caused it?"** — worked example: *"`DOCA_ERROR_INVALID_VALUE`
  on `doca_sha_task_hash_alloc_init`"*. Answered by the SHA overlay
  on the cross-library taxonomy in
  ## Error taxonomy
  + the layered ladder in
  ## debug that escalates to
  `doca-debug/SKILL.md`.

## Audience

This skill serves **external developers building applications that
consume the DOCA SHA library** — i.e., users whose code calls
`doca_sha_*` (directly in C/C++, or through FFI/bindings from
another language) to offload SHA hashing onto the BlueField-2 SHA
engine (BlueField-3 has none). It is *not* for NVIDIA developers contributing
to DOCA SHA itself.

**Language scope.** DOCA SHA ships as a C library with `pkg-config`
module name `doca-sha`. The shipped samples are written in C. C and
C++ consumers are the canonical case and the worked examples in
the tasks companion assume that path. Other-language consumers (Rust, Go,
Python, …) consume the same `*.so` through FFI or language-specific
bindings; the skill's contribution in that case is to keep the
lifecycle, capability-discovery, permission, error-taxonomy, and
one-shot-vs-partial guidance language-neutral, and to route the
agent to the public C ABI as the authoritative surface that any
wrapper will eventually call.

## When to load this skill

Load this skill when the user is doing hands-on DOCA SHA work, in
any language. Concretely:

- Initializing a `doca_sha` context on a `doca_dev` and configuring
  at least one task type (`doca_sha_task_hash` and/or
  `doca_sha_task_partial_hash`) before `doca_ctx_start()`.
- Choosing between the **one-shot** task (`doca_sha_task_hash` —
  input fits in a single source buffer, output digest lands in a
  single destination buffer) and the **partial / incremental** task
  (`doca_sha_task_partial_hash` — input streamed in chunks, finalized
  separately) for the user's data shape.
- Setting permissions on `doca_mmap` correctly for the source buffer
  (`DOCA_ACCESS_FLAG_LOCAL_READ_ONLY` at minimum) and the destination
  buffer (`DOCA_ACCESS_FLAG_LOCAL_READ_WRITE`).
- Sizing the destination buffer against
  `doca_sha_cap_get_min_dst_buf_size(devinfo, algorithm)` and
  the source buffer against
  `doca_sha_cap_get_max_src_buf_size(devinfo)`.
- Checking which SHA algorithm enums
  (`DOCA_SHA_ALGORITHM_SHA1`, `DOCA_SHA_ALGORITHM_SHA256`,
  `DOCA_SHA_ALGORITHM_SHA512`) the active device's accelerator
  advertises, via
  `doca_sha_cap_task_hash_get_supported(devinfo, algorithm)` and
  `doca_sha_cap_task_partial_hash_get_supported(devinfo, algorithm)` —
  both fold task-support and algorithm-support into one call.
- Validating a digest against a published test vector before pushing
  bulk input through the accelerator.
- Debugging a `DOCA_ERROR_*` returned from a SHA call (lifecycle vs.
  buffer-sizing vs. permission vs. unsupported-algorithm) and the
  task-completion event on the progress engine.
- Designing or extending non-C bindings (Rust, Go, Python, …) that
  wrap the SHA C ABI — for the lifecycle, permission, capability,
  and one-shot-vs-partial rules the wrapper must honor.

Do **not** load this skill for general DOCA orientation, install of
DOCA itself, non-SHA hashing libraries on CPU (use OpenSSL or
similar), or other DOCA libraries. For those, use
`doca-public-knowledge-map/SKILL.md`.

## What this skill provides

This is a **thin loader**. The body keeps only the orientation
needed to pick the right next file. The substantive SHA-specific
material lives in two companion files:

- the capabilities companion — what DOCA SHA can express on this version:
  the two task types (one-shot hash and partial / incremental
  hash), the three algorithm enums, the capability-query surface
  (`doca_sha_cap_*` for algorithm support and buffer sizing), the
  SHA error taxonomy (mapped onto the cross-library `DOCA_ERROR_*`
  set), the observability surface (per-task completion events on
  the progress engine), the safety policy that gates source /
  destination mmap permission decisions, and the path-selection
  rule (when to use doca-sha versus a CPU hash or a different
  DOCA crypto library).
- the tasks companion — step-by-step workflows for the six in-scope SHA
  verbs: `configure`, `build`, `modify`, `run`, `test`, `debug`.
  Plus a `Deferred task verbs` block that points out-of-scope
  questions at the right next skill.

The skill assumes a host or BlueField where DOCA is already
installed at the standard location and the user has the privileges
their public install profile expects. It does not cover installing
DOCA — that path goes through
`doca-setup/SKILL.md`.

## What this skill deliberately does not ship

This skill is **agent guidance**, not a samples or templates
bundle. To keep the boundary clean, it deliberately does not
contain — and pull requests should not add:

- **Pre-written DOCA SHA application source code, in any
  language.** The verified SHA source code is the shipped C samples
  at `/opt/mellanox/doca/samples/doca_sha/`, plus the File Integrity
  reference application linked from the public DOCA SHA guide. The
  agent's job is to route the user to those files and prescribe a
  minimum-diff modification on them via the universal
  modify-a-sample workflow in
  `doca-programming-guide/SKILL.md`,
  layered with the SHA-specific overrides in
  ## modify.
- **Pre-computed digest tables for arbitrary inputs.** The skill
  tells the agent to use a *published* test vector (e.g. the NIST
  SHA test vectors for the empty string, "abc", and the
  million-`a` input) as the known-vector smoke; it does not ship a
  vector bank of its own.
- **Standalone build manifests** (`meson.build`, `CMakeLists.txt`,
  `Cargo.toml`, …) parked inside the skill. The agent constructs
  the build manifest *in the user's project directory* against the
  user's installed DOCA, where `pkg-config --modversion doca-sha`
  is the source of truth.
- **A `samples/`, `bindings/`, or `reference/` subtree** of any
  kind. A mock or incomplete artifact in this skill's tree, even
  one labeled "reference", is misleading: users will read it as
  buildable.

## Loading order

1. Read this skill first to confirm the user's question is in
   scope.
2. **For the SHA capability matrix, algorithm enums, one-shot vs
   partial task split, capability-query rules, permission matrix,
   error taxonomy, observability, and safety / path-selection
   policy, see the capabilities companion.**
3. **For step-by-step workflows — configure, build, modify, run,
   test, debug — see the tasks companion.**

Both companion files cross-link to each other,
`doca-version/SKILL.md` for the canonical
version-handling rules, and
`doca-public-knowledge-map/SKILL.md`
whenever the right answer is "look it up in the public docs or the
installed package layout" rather than "SHA-specific guidance".

## Related skills

- `doca-public-knowledge-map/SKILL.md` —
  the routing table for every public DOCA documentation source and
  the on-disk layout of an installed DOCA package. The DOCA SHA
  page lives at `docs.nvidia.com/doca/sdk/DOCA-SHA/`; the File
  Integrity reference application is the canonical worked example.
- `doca-setup/SKILL.md` — env preparation,
  install verification, and the *I have no install yet* path with
  the public NGC DOCA container. This skill assumes its
  preconditions are satisfied.
- `doca-version/SKILL.md` — canonical DOCA
  version-handling rules. This skill's
  `## Version compatibility`
  cross-links the four-way match rule and adds only the SHA-specific
  *"discover algorithms + buffer sizes via cap query"* overlay.
- `doca-structured-tools-contract/SKILL.md` —
  the bundle's structured-tools precedence rule (detect / prefer
  / fall back / report). The Command appendix in
  the tasks companion honors this contract.
- `doca-programming-guide/SKILL.md` —
  general DOCA programming patterns shared by every library: the
  canonical `pkg-config` + meson build pattern, the universal
  modify-a-shipped-sample first-app workflow, the universal
  lifecycle, the cross-library `DOCA_ERROR_*` taxonomy, and the
  program-side debug order. This skill layers SHA specifics on
  top.
- `doca-debug/SKILL.md` — the cross-cutting
  debug ladder (install / version / build / link / runtime /
  program / driver). SHA-specific debug (algorithm-not-supported,
  destination-buffer-too-small, partial-hash-out-of-order) overlays
  on top of that ladder.

© NVIDIA · Apache-2.0 · NVIDIA/skills

Похожие скиллы

Все скиллы по теме «Настройка агента: память, контекст, субагенты» →
nemo-guardrails Orchestra Research Скилл nemo-guardrails для Claude Code, Codex, Cursor и других платформ добавляет runtime-проверки безопасности LLM-приложений: детекцию jailbreak, фильтрацию PII, проверку на галлюцинации и токсичность. CX Codex +1 1 100установок openshell-cli NVIDIA Скилл openshell-cli для Claude Code и других агентов: проводит по командам openshell — гейтвеи, песочницы, провайдеры, профили, политики, BYOC. Поднимает песочницу руками агента без опоры на память. CX Codex +1 790установок debug-openshell-cluster NVIDIA Скилл debug-openshell-cluster помогает диагностировать сбои шлюза OpenShell, проблемы со средами Docker, Podman и Kubernetes, а также ошибки запуска песочниц. CX Codex +1 739установок aiq-deploy официальный NVIDIA AI-Q Blueprint Team Скилл aiq-deploy разворачивает и проверяет инфраструктуру NVIDIA AI-Q Blueprint, а также помогает остановить сервисы и разобрать сбои. CX Codex GP ChatGPT +1 ★ 8 052звёзд репо spark-training-gotchas Seth Hobson Скилл spark-training-gotchas диагностирует 10 частых сбоев при обучении моделей на NVIDIA DGX Spark. Помогает найти утечки памяти, троттлинг и ошибки CUDA перед долгими запусками. CX Codex +1 2 500установок spark-memory-thermal-ops Seth Hobson Скилл spark-memory-thermal-ops помогает планировать память и отслеживать нагрев NVIDIA DGX Spark во время долгих ML-запусков. Пригодится при OOM и падении производительности. CX Codex +1 2 500установок

Из того же набора312

Весь набор →
nemo-relay-plugin-observability Скилл nemo-relay-plugin-observability для Claude Code помогает выбрать и настроить экспортеры и подписчики для телеметрии NeMo Relay версий 0.6 и 0.7. CX Codex +1 140установок doca-compress Скилл doca-compress для Claude Code помогает переносить сжатие DEFLATE и распаковку DEFLATE или LZ4 на DOCA Compress. Он подключается при настройке задач сжатия и разборе ошибок аппаратного ускорителя. CX Codex +1 140установок doca-dms Скилл doca-dms для Claude Code управляет NVIDIA DOCA Management Service на BlueField, хосте или в Kubernetes. Выполняет локальные операции через dms-cli и удалённые gNMI/gNOI-операции. CX Codex +1 138установок doca-eth Помогает настроить очереди пакетов DOCA Ethernet на сетевых картах ConnectX и DPU BlueField, выбрать тип RX-очереди и разобрать ошибки передачи. CX Codex +1 137установок doca-version Скилл doca-version для Claude Code сверяет четыре источника версии DOCA и выявляет расхождения между сборкой и рантаймом. CX Codex +1 130установок doca-mgmt Скилл doca-mgmt для программирования управления устройствами NVIDIA BlueField DPU и ConnectX NIC прямо из кода. CX Codex +1 128установок doca-pcc-counters Скилл doca-pcc-counters для Claude Code и Codex, который помогает вооружить и считать аппаратные PCC-счётчики (CNP, RTT, WRED-drop) на устройствах ConnectX / BlueField через pcc_counters.sh. CX Codex +1 128установок nvidia-skills Плагин подбирает NVIDIA-скиллы для CUDA, GPU, AI-агентов, обучения моделей, робототехники и симуляции. ★ 4 106звёзд репо

Спонсоры проекта

Компании, которые помогают держать аналитику открытой