AutoGineer summary
Written by AutoGineer from the skill’s own files, not by the author
The doca-dms skill provides operational guidance for the NVIDIA DOCA Management Service (DMS). It covers the deployment, configuration, and management of the dmsd and dmspe services on host systems, BlueField DPUs, or Kubernetes pods. The skill focuses on using gNMI for configuration and gNOI for lifecycle management, while emphasizing strict safety protocols for destructive operations like reboots and factory resets. It is designed for operators managing networking platforms from a centralized control plane.
What it does
- Guides the configuration of dmsd and dmspe services for various deployment topologies.
- Supports gNMI Get/Set operations for managing device settings via YANG paths.
- Facilitates gNOI workflows including reboots, OS installs, and factory resets.
- Provides debugging procedures for frontend and backend service failures.
- Enforces mandatory safety confirmation protocols for destructive hardware operations.
- Assists in setting up authentication modes and managing user access boundaries.
When to use it
- When deploying or configuring the DMS daemon on a management endpoint.
- When performing remote management of BlueField or ConnectX devices via gRPC.
- When executing gNOI lifecycle operations like system reboots or firmware updates.
- When troubleshooting errors between the DMS frontend and backend components.
- When determining the appropriate authentication and security posture for DMS.
Advantages
- Provides a structured approach to managing complex DPU/SmartNIC environments.
- Ensures operational safety through explicit confirmation requirements for destructive tasks.
- Standardizes interaction with hardware via industry-standard gNMI/gNOI protocols.
- Clearly separates operational tasks from development or library-level programming.
Usage examples
- “How do I read a BlueField setting locally using dms-cli?”
- The agent will guide you to use dms-cli with the appropriate YANG path, suggesting the use of --tree to explore available paths before reading or modifying values.
- “Trigger a clean reboot of the target via gNOI.”
- The agent will identify the gNOI reboot operation as destructive, require target-bound explicit confirmation, and then provide the necessary gNOI command sequence.
- “My DMS request failed; how do I debug it?”
- The agent will walk you through the layered debug process to determine if the error originated in the dmsd frontend or the dmspe backend.
- “Which authentication mode should I pick for a multi-tenant environment?”
- The agent will reference the auth-mode trade-off table to help you select the appropriate security configuration for your specific environment.
Install
Download the skill as a ZIP:
https://mp-staging.autogineer.com/agent-skills/nvidia/doca-dms/downloadUnzip it into the cross-client skills folder of your project:
mkdir -p .agents/skills && unzip doca-dms.zip -d .agents/skills
Documentation by NVIDIA
DOCA Management Service (DMS)
Examples
"Read or change a BlueField setting locally." — use
dms-cli(DOCA 3.5+); explore the path with--treefirst."Manage a remote BlueField over gRPC." — this is in scope. Use
dmsd+dmspewith the chosen auth; do not drop to ad-hoc SSH for the management plane."gNOI reboot from orchestrator." — treat reboot / OS install / factory-reset as destructive: require target-bound explicit confirmation first.
Instructions
Follow Loading order below. Open the tasks companion for configure / gNMI / debug. Open the capabilities companion for auth and gNOI scope. Do not open every companion file.
⚠️ Destructive operations. The gNOI
reboot,OS install,factory-reset, and managed-file deletion operations are irreversible and service-impacting — they can take a production BlueField or ConnectX offline or wipe its configuration. Before issuing any of them the agent MUST: (1) verify the target device identity, and (2) obtain explicit confirmation bound to that target and action. In an interactive session this is an explicit user reply naming/accepting both; in unattended execution it must be an approved-system authorization artifact bound to both. Otherwise stop withconfirmation_required. Never invoke them speculatively or as a side effect of another task. See the public DMS guide's safety guidance for these operations.
Where to start: This skill is for operating DMS, not for
linking against a library. If the user wants to deploy or run
the daemon, open the tasks companion and start at
## configure. If the question is what shape
of service is DMS and what protocols does it speak, start at
the capabilities companion. If DOCA is not installed on the
management endpoint yet, route to
doca-setup/SKILL.md first.
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.
DOCA 3.5 and later: start with dms-cli. The DMS guide now centres on
/opt/mellanox/doca/services/dms/dms-cli, run directly on the host or
BlueField Arm with no server or daemon. One syntax per YANG path:
dms-cli -t pci/<BDF> /nvidia/mode/operating-mode reads a value, the same
path plus a value sets it, and action paths run operations. Explore before
writing: --tree /nvidia, --describe <path> and list-xpaths <path>;
script with --json; validate with --dry-run. VFs and SFs take
-t pci/<BDF>,pf<N>vf<M> or pf<N>sf<M>. On DOCA 3.5: as a normal
user it fails with backend fork/exec .../dmspe: permission denied — run it
with sudo or add the user to dmsgroup; some example paths do not exist on
every device (for example congestion-control slots are under
/nvidia/cc/algo/slot there), so confirm every path with --tree /
list-xpaths before using it. Firmware or BFB installs
(/nvidia/provisioning/install), resets and mode changes are
hardware-state changes: follow doca-hardware-safety. The dmsd + dmspe
gRPC (gNMI / gNOI) service below remains for remote and fleet management.
Example questions this skill answers well
Never invent the image tag. It comes from the NGC catalog's DOCA org or the DMS page on
docs.nvidia.com, not from memory — a plausible tag reads as authoritative and mismatches the installed DOCA. Cite where it came from, or send the user to look it up.
The CLASSES of DMS questions this skill is built to answer, each with one worked example. The class is the load-bearing piece; the worked example is one instance.
"Where should DMS run for my target topology?" — worked example: "I have a host without a DPU; can I still manage a remote ConnectX?". Answered by the deployment-shape decision in the capabilities companion
the tasks companion.
"Which authentication mode should I pick for my security posture?" — worked example: "a multi-tenant production env vs a single-tenant lab". Answered by the auth-mode trade-off table in the capabilities companion
the auth-wiring step in the tasks companion.
"How do I issue a gNMI
Get/Setagainst a modeled path?" — worked example: "set/interfaces/interface/config/mtuon a remote interface". Answered by the gNMI/gNOI surface in the capabilities companionthe request-shape workflow in the tasks companion.
"How do I run a gNOI operation (reboot, OS install, file transfer)?" — worked example: "trigger a clean reboot of the target via gNOI". Answered by the gNOI catalog in the capabilities companion
the run workflow in the tasks companion.
"My DMS request returned an error — was it the frontend or the backend?" — worked example: "
mlxconfigfailed under DMS but works on the shell". Answered by the frontend-vs-backend split in the capabilities companionthe layered ladder in the tasks companion.
"Where do I read DMS logs and how do I rotate them?" — worked example: "persistent log directory + journald + log-rotation policy". Answered by the logging surface in the capabilities companion
the log-rotation step in the tasks companion.
Audience
This skill serves external operators and platform teams who deploy and
operate DMS to manage NVIDIA® BlueField® networking platforms or
NVIDIA® ConnectX® SmartNICs from a centralized control plane. Concretely:
people running dmsd, integrating gNMI/gNOI clients against it, choosing
an authentication mode, or wiring DMS into a Kubernetes deployment.
It is not for NVIDIA developers contributing to DMS itself, and it
is not a programming guide for building applications on top of
DOCA libraries (that is doca-programming-guide plus the matching
library skill under libs/). DMS is a service, not a library: the
user invokes it as a daemon and talks to it over gRPC; they do not link
against libdms.so to write their own program.
Status note. Per the public DMS guide, DMS is currently in beta, with General Availability scoped to SPC-X use cases. The skill reflects the public guide's posture: prescribe the documented launch / auth / deployment paths, follow the documented security best practices, and defer roadmap and GA-scope questions to the live public guide rather than guessing.
When to load this skill
Load this skill when the user is doing hands-on DMS operation work against a BlueField or ConnectX target where DOCA is already installed on the management endpoint (host, DPU, or pod). Concretely:
Deciding where DMS should run (host non-DPU / BlueField Arm / Kubernetes pod) for a given target topology.
Bringing up the
dmsddaemon — choosing SystemD vs manual launch, selecting an authentication mode, wiring-allowed_users(the gRPC client authorization boundary) and, if needed,dmsgroup(thedmspebackend-helper Unix group).Issuing
gNMIGet/Setrequests against modeled paths (e.g./interfaces/interface/config/mtu).Issuing
gNOIoperations: OS install, reboot, file transfer, factory-reset,mlxconfig, containerz.Choosing an authentication mode (localhost / PAM / credentials / mTLS) and understanding the security trade-offs the public guide calls out.
Reading or rotating DMS logs, configuring config persistency, or recovering from a crashed daemon.
Debugging a DMS request that returned an error — separating "frontend rejected before reaching backend" from "backend executed and the underlying tool (e.g.
mlxconfig) failed".
Do not load this skill for general DOCA orientation, install of
DOCA itself, or library-API questions. For those, route via
doca-public-knowledge-map/SKILL.md,
doca-setup/SKILL.md, or the matching
libs/<library> skill.
What this skill provides
This is a thin loader. Substantive material lives in two companion files:
the capabilities companion — DMS architecture (frontend
dmsd/ privileged backenddmspe), management protocols (gNMI, gNOI), the YANG-based unified configuration dictionary, deployment shapes, authentication modes with their security trade-offs, the configuration-persistency model, the logging surface, and DMS's documented security posture.the tasks companion — step-by-step workflows for the in-scope DMS verbs:
configure,build,modify,run,test,debug, plus aDeferred task verbsblock routing out-of-scope questions.
The skill assumes a host where DOCA is already installed and the
operator has root / sudo access where the public guide says it is
required. 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 templates or sample-config bundle. To keep the boundary clean, it deliberately does not contain — and pull requests should not add:
Pre-baked DMS configuration files (YANG instance documents, full-stack example configs, ready-to-run
dmsdflag bundles) intended to be copy-pasted into production. Configs are deployment- specific and the safe answer for an external operator is to derive them from the public guide against their own target. The agent's job is to prescribe the procedure and quote the documented flags and paths, not to ship a config the user might run unmodified.Pre-written gNMI / gNOI client programs in any language. The client surface is standard gNMI / gNOI (publicly documented); the skill describes which paths and operations DMS supports, not how to build a gRPC client in language X.
TLS material, credentials, or PAM stanzas. These are user-environment artifacts; the skill points at the documented configuration knobs and the documented security best practices and stops there.
A
samples/,templates/, orreference/subtree of any kind. A mock or incomplete artifact in this skill's tree, even one labeled "reference", is misleading: operators will read it as production-ready.
Loading order
Read this skill first to confirm the user's question is in scope.
For the DMS architecture, deployment shapes, auth modes, protocol/path inventory, persistency, logging, and security posture, see the capabilities companion.
For step-by-step workflows — configure, build, modify, run, test, debug — see the tasks companion.
Apply the destructive-operation gate in every phase. Before any gNOI OS install, reboot, factory reset, or managed-file deletion — including a test or debug action — verify the exact target and obtain explicit confirmation for that specific operation using the interactive or approved-system mechanism defined in the warning above. No earlier confirmation or workflow phase carries authorization forward.
Related skills
doca-public-knowledge-map/SKILL.md— the routing table to the public DMS guide and the rest of the public DOCA documentation set.doca-setup/SKILL.md— env preparation and install verification on the host wheredmsdwill run, including the I have no install yet path via the public NGC DOCA container. This skill assumes its preconditions are satisfied at the management endpoint.doca-programming-guide/SKILL.md— general DOCA patterns. DMS is service-shaped not library-shaped, so the build / modify / first-app pattern there does not apply directly, but the cross-libraryDOCA_ERROR_*taxonomy and the layered-debug order remain useful when DMS reports errors that originated in a DOCA library it called.
Files
- ATTRIBUTION.mdAdded by AutoGineer882 B
- BENCHMARK.md9.8 KiB
- CAPABILITIES.md15.6 KiB
- LICENSELicence file added by AutoGineer11.2 KiB
- SKILL.md13.1 KiB
- SKILLCARD.yaml2.6 KiB
- TASKS.md20.1 KiB
- evals/evals.json2.2 KiB
- skill-card.md5.2 KiB
- skill.oms.sig5.1 KiB
BENCHMARK.md
Skill Benchmark: doca-dms
✅ Overall verdict: PASS — Recommended for publication
Publication Recommendation
Recommended for publication based on the completed evaluation evidence in this report.
Evaluation Metadata
Skill:
doca-dmsValidation status:
passedEvaluation date: 2026-10-28
Evaluator version:
1.6.2Agents: Claude Code (
aws/anthropic/bedrock-claude-opus-5-5), Codex (openai/openai/gpt-6.1-sol)Tasks: 4 evaluation tasks (3 positive, 1 negative)
Dataset digest:
sha256:e498e199e73e2d7f8252c201b21ccf39ea42f12447a4e3231ad0f2c7c04ebfb2(skill-evaluator-dataset-snapshot/1)Attempts per task: 1
Environment:
k8s-sandboxTier 2 evidence: required for publication
Tier 3 evidence: required for publication
Each task attempt ran in its own isolated sandbox pod.
Tier 1 LLM Provenance
SkillSpector provider, model, analyzer revision, and determinism evidence retained for this scan.
| Skill | Provider | Adapter | Analyzer | Model | Revision | Determinism |
|---|---|---|---|---|---|---|
| doca-dms | nv_inference | openai | mcp_tool_poisoning | aws/anthropic/bedrock-claude-opus-5-5 | 2.13.1 | nondeterministic / provider_defaults |
| doca-dms | nv_inference | openai | semantic_developer_intent | aws/anthropic/bedrock-claude-opus-5-5 | 2.13.1 | nondeterministic / provider_defaults |
| doca-dms | nv_inference | openai | semantic_quality_policy | aws/anthropic/bedrock-claude-opus-5-5 | 2.13.1 | nondeterministic / provider_defaults |
| doca-dms | nv_inference | openai | semantic_security_discovery | aws/anthropic/bedrock-claude-opus-5-5 | 2.13.1 | nondeterministic / provider_defaults |
| doca-dms | nv_inference | openai | meta_analyzer | aws/anthropic/bedrock-claude-opus-5-5 | 2.13.1 | nondeterministic / provider_defaults |
Controls for
doca-dms: temperature=not set (source=provider_default, forwarded=no); seed=not set (source=unset, forwarded=no); reasoning_effort=not set (source=provider_default, forwarded=no)
What This Report Answers
The three-tier evaluation checks whether the skill:
is safe to use;
produces correct answers;
is discovered and activated when needed;
helps the agent complete the user's goal and expected workflow; and
avoids wasted skill and tool usage.
Results at a Glance
| Measure | Claude Code (Baseline → Skill Uplift) | Codex (Baseline → Skill Uplift) |
|---|---|---|
| Overall | 90.3% — baseline ran, but no comparable score was available; uplift unavailable | 85.8% — baseline ran, but no comparable score was available; uplift unavailable |
| Security | 100.0% → 100.0% (±0.0 points) | 100.0% → 100.0% (±0.0 points) |
| Correctness | 55.0% → 90.0% (+35.0 points) | 80.0% → 90.0% (+10.0 points) |
| Discoverability | 100.0% — baseline ran, but no comparable score was available; uplift unavailable | 95.0% — baseline ran, but no comparable score was available; uplift unavailable |
| Effectiveness | 25.0% → 75.0% (+50.0 points) | 56.3% → 75.0% (+18.7 points) |
| Efficiency | 86.3% — baseline ran, but no comparable score was available; uplift unavailable | 69.1% — baseline ran, but no comparable score was available; uplift unavailable |
How to read this table: baseline is the same task attempted without the target skill. Scores are rounded to one decimal; threshold-adjacent values use additional precision so their displayed band matches the verdict. Uplift is derived from those displayed scores and shown in percentage points.
Example: 47.0% → 92.0% (+45.0 points) means the skill-assisted run scored 92.0%, 45.0 percentage points above its 47.0% no-skill baseline.
A partial dimension was calculated from only the available configured signals; review the detailed report before relying on it.
Comparison details
Shared-source comparisons use the same signals and weights on both conditions. They do not replace the full dimension scores above or change the publication verdict.
| Agent | Dimension | Basis | Baseline → With skill (uplift) | Explanation |
|---|---|---|---|---|
| claude-code | Discoverability | Dimension comparison | N/A | Discoverability is not applicable to the no-skill baseline; there is no installed skill to discover. |
| claude-code | Efficiency | Token-only comparison | 98.0% → 92.5% (-5.5 points) | Only Token Efficiency is comparable; the full dimension scores use different evaluator sources. |
| codex | Discoverability | Dimension comparison | N/A | Discoverability is not applicable to the no-skill baseline; there is no installed skill to discover. |
| codex | Efficiency | Token-only comparison | 84.4% → 83.5% (-0.9 points) | Only Token Efficiency is comparable; the full dimension scores use different evaluator sources. |
Token Usage
Actual Tier 3 execution usage is reported for every observed agent/case pair and both conditions.
| Agent | Dataset case | With skill | Without skill | Delta | Change | Coverage |
|---|---|---|---|---|---|---|
| claude-code | All cases | 668,430 | 432,847 | +235,583 | +54.43% | skill 4/4; base 4/4 |
| claude-code | doca-dms.explicit-positive.v1 | 209,769 | 108,334 | +101,435 | +93.63% | skill 1/1; base 1/1 |
| claude-code | doca-dms.implicit-positive.v1 | 209,527 | 144,069 | +65,458 | +45.44% | skill 1/1; base 1/1 |
| claude-code | doca-dms.implicit-positive.v2 | 212,578 | 72,386 | +140,192 | +193.67% | skill 1/1; base 1/1 |
| claude-code | doca-dms.negative.v1 | 36,556 | 108,058 | -71,502 | -66.17% | skill 1/1; base 1/1 |
| codex | All cases | 380,151 | 176,740 | +203,411 | +115.09% | skill 4/4; base 4/4 |
| codex | doca-dms.explicit-positive.v1 | 102,882 | 75,800 | +27,082 | +35.73% | skill 1/1; base 1/1 |
| codex | doca-dms.implicit-positive.v1 | 102,713 | 39,063 | +63,650 | +162.94% | skill 1/1; base 1/1 |
| codex | doca-dms.implicit-positive.v2 | 142,697 | 22,012 | +120,685 | +548.27% | skill 1/1; base 1/1 |
| codex | doca-dms.negative.v1 | 31,859 | 39,865 | -8,006 | -20.08% | skill 1/1; base 1/1 |
| ALL AGENTS | Dataset aggregate | 1,048,581 | 609,587 | +438,994 | +72.01% | skill 8/8; base 8/8 |
Prompt tokens include cached reads, so total tokens are prompt + completion (cached is not added twice). The Efficiency score uses (prompt - cached) + completion. Change is 100 * (with - without) / without; negative means fewer tokens. N/A comparisons include their reason in the coverage column; coverage is never estimated.
Tier Status
| Tier | Purpose | Status | Evidence |
|---|---|---|---|
| Tier 1 | Static validation | PASSED WITH OBSERVATIONS | 11 validator(s); 20 finding(s) |
| Tier 2 | Semantic deduplication | PASSED | 2 validator(s); 0 finding(s) |
| Tier 3 | Live agent evaluation | PASS | 2 agent(s); 4 task(s) |
Findings and Observations
<details> <summary>Show detailed findings and successful checks</summary>
MEDIUM SCHEMA/metadata_key_style: Metadata key 'doca_version' is not kebab-case (
skills/services/doca-dms/SKILL.md)MEDIUM SCHEMA/folder_hierarchy: Unexpected nesting depth for general skill (
skills/services/doca-dms)MEDIUM SECURITY/Session Persistence (RA2): Rogue Agent: SystemD unit, and install (
BENCHMARK.md:86)MEDIUM SECURITY/Session Persistence (RA2): Rogue Agent: SystemD unit, and install (
BENCHMARK.md:87)MEDIUM SECURITY/Session Persistence (RA2): Rogue Agent: SystemD unit, and install (
BENCHMARK.md:88)15 additional finding(s) are available in the full evaluation artifacts.
</details>
Scoring Methodology
<details> <summary>Show dimension definitions, source signals, and thresholds</summary>
| Dimension | Question | Scored signals |
|---|---|---|
| Security | Is it safe to use? | security (100%) |
| Correctness | Is the answer correct? | accuracy (100%) |
| Discoverability | Was the right skill loaded when needed? | skill_execution (100%) |
| Effectiveness | Did the skill help complete the task? | goal_accuracy (50%) + behavior_check (50%) |
| Efficiency | Did it avoid wasted tool calls and token usage? | skill_efficiency (50%) + token_efficiency (50%) |
Dimension bands: PASS at 50% or above; NEUTRAL from 40% to below 50%; FAIL below 40%.
Overall Tier 3 lift: PASS at +5 points or more; FAIL at -10 points or less; values between those bands are NEUTRAL.
Overall verdict: PASS only when every configured dimension passes for at least one supported agent. Lift is reported as diagnostic evidence and does not override this gate.
The 50% attempt pass threshold is a separate per-task gate; it is not the dimension pass threshold.
Effectiveness is the equal-weight mean of goal completion (
goal_accuracy) and expected workflow adherence (behavior_check).Efficiency is 50% tool-call productivity (the backward-compatible
skill_efficiencywire id) and 50%token_efficiency. Positive-case skill routing is scored under Discoverability, not Efficiency; a negative case without a routing target is N/A. N/A sources are omitted, remaining weights are renormalized, and the dimension is marked partial.
Signals present in this run:
security(Security): unsafe operations, secret leakage, and unauthorized access.skill_execution(Skill Execution): whether the expected skill was selected, decoys were avoided, and the workflow executed.skill_efficiency(Tool Productivity): tool-call productivity (legacy wire id; routing is scored under Discoverability).accuracy(Accuracy): final-answer correctness against the reference answer.goal_accuracy(Goal Accuracy): whether the user's goal was achieved.behavior_check(Behavior Check): whether the expected workflow behavior was followed.token_efficiency(Token Efficiency): actual uncached prompt plus completion usage (50% of Efficiency).
</details>
Freshness
Regenerate this benchmark when the skill, evaluation dataset, target agent/model, evaluator version, environment, or scoring policy changes.
Requirements and compatibility
Works with: Claude Code, Claude apps, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Kiro, OpenCode, Goose, Amp, Roo Code, JetBrains Junie, Factory Droid, Devin (ex-Windsurf), Mistral Vibe, OpenHands, Letta
“Works with” means the tool reads the open Agent Skills format; a skill written for one tool may use features others lack.
The author says: DOCA service shipped with the DOCA install at /opt/mellanox/doca on the management endpoint (x86 host (non-DPU), BlueField Arm, or Kubernetes pod) on Linux (Ubuntu 22.04/24.04 or RHEL/SLES); `dmsd` + `dmspe` run there, and DMS is also pulled as an NGC container image. Verify the public DMS guide version matches the installed DOCA release before quoting flags or YANG paths. The validation steps require a live DOCA install.
Security
What our scanner noticed· Medium risk
- Large base64-encoded blobs
A static scan is not a guarantee. Skills are third-party content: review a skill before you let an agent use it.
Attribution and licence
- Created by
- NVIDIA
- Publisher
- NVIDIA (Official)
- Source
- NVIDIA/skills · skills/doca-dms · commit fce3cca
- Licence
- Apache License 2.0 (redistribution allowed)
- LICENSE (added by AutoGineer from LICENSE-APACHE)
- ATTRIBUTION.md (written by AutoGineer)
- Synced
AutoGineer is not affiliated with or endorsed by NVIDIA.
Download ZIP
The ZIP holds the skill’s files, its licence files and an ATTRIBUTION.md. Download doca-dms.zip