What does this skill do, and when should you use it?
DOCA Debug covers cross-cutting symptoms such as compilation failures, unresolved doca_* symbols, DOCA_ERROR_* returns, silent programs, packets not reaching the wire, and crash artifacts. It uses a layered path across installation, version, build, link, runtime, program, and driver checks. It also covers verbosity controls, trace builds, container observability, and state capture for the NVIDIA DOCA Developer Forum. The skill is a thin loader that points agents to CAPABILITIES.md, TASKS.md, and related skills.
Classifies a reported DOCA symptom by debugging layer; directs the agent to CAPABILITIES.md and TASKS.md; identifies observability surfaces such as --sdk-log-level, DOCA_LOG_LEVEL, and the doca-{lib}-trace build flavor; guides reproducible-state capture and Developer Forum routing; and hands off environment, program, or library-internal issues to doca-setup, doca-programming-guide, or the matching library skill.
- A DOCA developer sees undefined reference to doca_* and needs to identify the link layer.
- A DOCA Flow program is silent and the developer needs higher-verbosity logging.
- An engineer is debugging inside the NGC DOCA container and needs to understand what is observable there.
- A user is preparing a reproducible DOCA Developer Forum question or bug report.
- A developer has a DOCA stack trace, valgrind output, or core dump and needs to decide where to start.
How do you install this skill?
- Before executing advice involving gdb, valgrind, kernel modules, eswitch, hugepages, or mlxconfig, obtain explicit user confirmation, capture current state, and provide an environment-appropriate rollback procedure.
- Do not treat the percentages in BENCHMARK.md or skill-card.md as independent validation; the report states that the original evaluation dataset was unavailable.
- The core workflow depends on CAPABILITIES.md, TASKS.md, and several other skills, so it may be incomplete when installed alone or used with an unmatched DOCA version.
- The skill does not describe Chinese-language output behavior or establish that all external NVIDIA documentation and forum dependencies are reliably reachable from mainland China.
- Shell / CLI
- Local filesystem
DOCA installation at /opt/mellanox/doca for validation steps
Install the NVIDIA skills collection with the default skills CLI: npx skills add nvidia/skills. To browse and select the skill, use: npx skills add nvidia/skills --list. The source does not provide a separate prompt-free installation command for doca-debug.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/doca-debug" ~/.claude/skills/
rm -rf "$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:
- ld says undefined reference to doca_flow_init—which layer?
- My DOCA Flow program is silent—where do I get more log output?
After installation, describe the DOCA symptom directly, for example: “ld says undefined reference to doca_flow_init—which layer?” or “My DOCA Flow program is silent—where do I get more log output?” The skill then directs the investigation through the layered debug workflow.
What are this skill's strengths and limitations?
- Provides one cross-library path spanning installation, version, build, link, runtime, program, and driver layers.
- Clearly separates cross-cutting debugging from environment, program, and library-specific troubleshooting.
- Covers logging, tracing, container debugging, and issue-report state capture.
- The skill can be read without a DOCA installation.
- The substantive workflows depend on CAPABILITIES.md and TASKS.md, whose contents are not included in the supplied source.
- Validation steps require a live DOCA installation at /opt/mellanox/doca.
- It does not own DOCA_ERROR_* taxonomy, environment-only pkg-config or hugepages issues, or Flow, RDMA QP, and Comch-specific debugging.
- The source provides no platform test matrix or standalone test-suite evidence.
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 Debug this page ✓ NVIDIA · Official | 58 · Recommended | ★ 3.6k | 13d ago | Apache-2.0 |
| cmux Debugging Skill | 56 · Use with care | ★ 28k | 13d ago | NOASSERTION |
| DOCA SHA Acceleration ✓ NVIDIA · Official | 56 · Use with care | ★ 3.6k | 13d ago | Apache-2.0 |
| DOCA STA Storage Target Acceleration ✓ NVIDIA · Official | 55 · Use with care | ★ 3.6k | 13d ago | Apache-2.0 |
| NVIDIA DOCA Public Knowledge Map ✓ NVIDIA · Official | 54 · Use with care | ★ 3.6k | 13d ago | Apache-2.0 |
How did FollowSkills review this skill?
The skill emphasizes read-only capture, single-variable changes, public-source sanitization, change logging, and provides Apache-2.0 licensing and NVIDIA attribution; however, it lacks consistently explicit user confirmation, privilege boundaries, and verifiable rollback procedures for gdb, module reloads, eswitch changes, or other hardware-related mutations, and dependency security is lightly addressed. Points are deducted for these gaps.
The seven-layer ladder, five-phase loop, container constraints, and escalation behavior are detailed and mostly consistent; however, the workflow depends on cross-file routing, assumes particular platforms and installed tools, lacks key-path test coverage, and cites a benchmark whose dataset is unavailable. Under static-review calibration the score cannot exceed 10, so points are deducted.
Triggers, exclusions, environment classes, and container boundaries are clearly described, including routing among env-class, program-class, and library-specific debugging; nevertheless, there is no Chinese-language interaction guidance, and evidence for varied distributions, permissions, and hardware topologies is limited. Points are deducted accordingly.
The layered SKILL.md, CAPABILITIES.md, and TASKS.md structure is readable and supports progressive disclosure, with dependency notes, examples, limitations, license, owner, and version/evaluation records; however, the SKILL metadata omits an author, change history and formal maintenance/update responsibility remain incomplete, and the benchmark reports structural omissions. Points are deducted for these governance gaps.
The material offers directly usable layered diagnosis, logging controls, state capture, and escalation workflows for core cross-cutting DOCA debugging; however, much of the substance depends on companion files and other skills, some commands require version-specific adjustment, and representative outputs are not independently verifiable. The static ceiling is 7, so points are deducted.
Internal links, commands, version gates, and the evaluation report provide some traceability; however, the benchmark source dataset is unavailable, the report is not independently reproducible, and no real CI plus committed test suite covers the skill's key paths. Under static calibration the score is capped at 5, so points are deducted.
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 →