What does this skill do, and when should you use it?
This skill serves external developers using the DOCA Compress C library. It covers DEFLATE compression, DEFLATE decompression, and LZ4 stream or block decompression on a BlueField DPU, ConnectX NIC, or DOCA-enabled host. It explains context setup, device capability and per-task buffer-limit queries, mmap permissions, completion events, and DOCA_ERROR_* diagnosis. It also helps decide whether bulk data should be offloaded or handled by CPU zlib/zstd. It does not install DOCA or ship buildable application source, bindings, fixtures, or standalone build manifests.
Guides users to inspect the local DOCA installation with pkg-config --modversion doca-compress and examine libraries, headers, samples, and applications under /opt/mellanox/doca. It covers doca_compress context setup, the compress-deflate and decompress-deflate task APIs, capability checks, source and destination sizing, mmap permissions, lifecycle, progress-engine completion events, and layered error diagnosis. It prescribes round-trip validation or decompression of a known DEFLATE fixture before bulk accelerator use.
- A developer compressing a multi-megabyte log buffer on BlueField who needs to compare DOCA offload with CPU zlib.
- A network service on ConnectX that only receives DEFLATE payloads and needs decompression without enabling compression.
- An application returning DOCA_ERROR_INVALID_VALUE from doca_compress_task_compress_deflate_alloc_init.
- A developer receiving DOCA_ERROR_NOT_PERMITTED and checking source and destination mmap access flags.
- A submitted task that produces no completion event and requires lifecycle, sizing, permission, and support checks.
- A Rust, Go, or Python wrapper that must preserve the Compress C ABI's capability, permission, and lifecycle rules.
How do you install this skill?
- Do not treat the PASS verdict or high benchmark scores as independently executed, reproducible evidence; the evaluation dataset is not supplied.
- Before real-device use, defer to local headers, pkg-config, doca_caps, and actual doca_devinfo capability queries, especially for LZ4 task APIs and version compatibility.
- The skill guides hardware accelerator, mmap, and task-queue operations; perform the minimal round-trip smoke only after confirming rollback, task draining, and mmap lifetime handling.
- Chinese users may need additional translation and terminology checking; core use depends on a local DOCA installation and the NVIDIA ecosystem.
- Shell / CLI
- Local filesystem
DOCA SDK installed at /opt/mellanox/docaLinuxBlueField DPU or ConnectX NICC library with pkg-config module doca-compress
Install the specific skill with the default skills CLI: npx skills add nvidia/skills --skill doca-compress --yes. The CLI installs it to a selected agent destination; the README does not document a separate package or manual-copy procedure. DOCA SDK must still be installed on Linux, with a BlueField DPU or ConnectX NIC attached.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/doca-compress" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- How do I offload this gzip compression task on BlueField with doca_compress?
- Why does doca_compress_task_compress_deflate return DOCA_ERROR_NOT_PERMITTED?
After installation, ask the agent an in-scope question such as: “How do I offload this gzip compression task on BlueField with doca_compress?” or “Why does doca_compress_task_compress_deflate return DOCA_ERROR_NOT_PERMITTED?” For configure, build, modify, run, test, or debug workflows, load TASKS.md. For capability and path-selection questions, load CAPABILITIES.md. If DOCA is not installed, use doca-setup first.
What are this skill's strengths and limitations?
- Covers practical task configuration, capability discovery, permissions, buffer sizing, and error diagnosis.
- Clearly defines the supported scope: bidirectional DEFLATE and LZ4 decompression, with LZ4 encoding excluded.
- Provides language-neutral guidance for C/C++ and FFI-based consumers.
- Includes round-trip smoke validation and progress-engine completion guidance.
- Requires a pre-existing DOCA SDK installation and BlueField DPU or ConnectX hardware.
- Excludes DOCA installation, DOCA Core lifecycle internals, and other compression algorithms.
- Ships no application source, language bindings, sample subtree, test fixtures, or standalone build manifest.
- The supplied source gives no concrete device-model matrix, numeric version matrix, or performance benchmarks.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| DOCA Compress Hardware Offload this page ✓ NVIDIA · Official | 58 · Recommended | ★ 3.8k | today | Apache-2.0 |
| DOCA AES-GCM Acceleration ✓ NVIDIA · Official | 57 · Use with care | ★ 3.8k | today | Apache-2.0 |
| DOCA SHA Acceleration ✓ NVIDIA · Official | 56 · Use with care | ★ 3.8k | today | Apache-2.0 |
| DOCA Flow Tune ✓ NVIDIA · Official | 54 · Use with care | ★ 3.8k | today | Apache-2.0 |
| DOCA Erasure Coding Development Skill ✓ NVIDIA · Official | 49 · Use with care | ★ 3.8k | today | Apache-2.0 |
The skill explicitly compares DOCA Compress with CPU zlib/zstd for bulk-input path selection. LZ4 encoding should use a CPU LZ4 library, while pure mmap-to-mmap copying belongs to doca-dma.
How did FollowSkills review this skill?
The documentation makes source/destination mmap permissions, capability discovery, user confirmation, data-flow direction, rollback, and task draining explicit, and states that credentials are not required; the main safety risks are visible. Deducted 5 points because dependency security, hardware privilege boundaries, and external-effect controls remain incomplete, while some capability claims lack independent evidence within the assessed path.
The configure/build/run/test/debug/rollback workflows are detailed and include error classification, progress-engine handling, and escalation routes. Deducted 12 points because key paths were not executed in this static review; LZ4 support is described inconsistently across SKILL.md, CAPABILITIES.md, and the skill card, and several conclusions depend on unavailable companion skills or local installation state.
Audience, trigger phrases, task inputs, non-fit ranges, DEFLATE/LZ4 directionality, and CPU fallbacks are clearly described. Deducted 4 points because the material is almost entirely English, Chinese interaction quality is not demonstrated, and the Linux/local DOCA/BlueField-or-ConnectX prerequisites plus mainland-China network reachability have limited evidence.
The SKILL.md loader, CAPABILITIES.md, and TASKS.md provide layered documentation with examples, dependencies, version checks, maintenance contacts, licensing, and evaluation material. Deducted 6 points for no clear author or changelog, inconsistencies between the skill-card license/repository metadata and the assessed material, TBD signature/version fields, and benchmark findings for missing recommended sections.
It can directly guide capability discovery, permission setup, buffer sizing, round-trip smoke validation, debugging, and rollback, while clearly stating that it does not ship buildable implementation code. Deducted 9 points because no implementation, fixtures, or independently reproducible tests are included; the modification renderer is deferred, leaving users to manually compare local samples, so result completeness is only partly supported.
The assessed path contains concrete APIs, commands, error mappings, evaluation assertions, and a benchmark report, providing some auditability. Deducted 6 points because the evaluation dataset is unavailable, the report provides no committed test suite or CI evidence covering key paths, and many DOCA behavior claims cannot be independently cross-checked from these files alone.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →