DOCA CollectX Telemetry Deployment
Deploy, operate, and debug CollectX telemetry collectors so counters reach downstream exporters.
The files explicitly constrain the skill to read-only collection, require device-support checks, prohibit invented configuration details, and route hardware mutations to a safety skill. Data flows and exporter classes are disclosed, and no credential requirement is identified. Points are deducted because stepwise user confirmation, least-privilege detail, sensitive-metric handling, and an executable rollback plan are incomplete.
SKILL.md, CAPABILITIES.md, and TASKS.md provide a consistent loading order, layered failure taxonomy, prerequisites, and diagnostic feedback rules. Because this is a static review with no executable key-path tests or real CI coverage, the static cap is 10; dependency availability and edge-case behavior still depend on the live installation and external documentation, so points are deducted.
The audience, trigger scenarios, exclusions, four-surface boundary, and routing conditions are clearly described. The skill deliberately omits concrete configuration names, parameters, and commands, requiring a live DOCA installation and public documentation; there is no evidence of Chinese-language support or mainland-China reachability, while core references depend on docs.nvidia.com, so points are deducted.
The layered SKILL.md, CAPABILITIES.md, TASKS.md, and references/details.md structure supplies dependencies, example questions, limitations, related skills, version guidance, and some ownership signals. Points are deducted for missing author metadata, inconsistent license wording between the skill metadata and skill card, and the absence of a clear changelog and explicit maintenance/update process.
The skill offers directly relevant scope decomposition, pre-configuration gating, runtime verification, end-to-end smoke testing, and layered debugging guidance. Points are deducted because it supplies no directly runnable provider, schema, flag, path, or sample configuration, leaving substantial site-specific lookup and review; representative outputs are not independently verified. The static cap is therefore 7.
Evidence consists mainly of structured files within the selected skill directory, cross-references, and a benchmark report. The report claims evaluation success, but its dataset is unavailable and no reproducible test materials are committed. This supports limited auditability but not independent reproduction or strong cross-source corroboration, so points are deducted.
- Do not use this skill for the productized DTS container, reader/publisher APIs, DOCA installation, or hardware-state changes; first confirm that the request concerns the CollectX runtime.
- Do not infer provider names, schema fields, exporter flags, or configuration paths from the skill; verify them against the target live configuration, --help output, or authoritative public documentation.
- The skill does not provide rollback commands or a detailed permission model; before changing configuration, the operator should confirm authorization, preserve the current state, and prepare a recoverable deployment snapshot.
- For mainland-China users, separately verify reachability of docs.nvidia.com and the selected exporter sink.
What does this skill do, and when should you use it?
This skill is for operators running an owned CollectX (clx)-based telemetry collector on a host or NVIDIA BlueField with DOCA already installed. It covers wiring providers and counters into a schema, running the collection daemon, and shaping Prometheus, Fluent Bit, NetFlow, file, or IPC exporters. It explicitly separates the CollectX collection mechanism from the DOCA Telemetry reader library, the DOCA Telemetry Exporter publisher library, and the productized DTS container. Concrete provider names, schema fields, flags, and configuration paths must come from the live installation or NVIDIA public documentation, making this a fit for collector deployment and diagnosis rather than DOCA installation or DTS operations.
Determines whether a request concerns the CollectX collector runtime; guides provider and counter wiring into a schema; runs the collection daemon; configures Prometheus pull, Fluent Bit push, NetFlow, file, or IPC export; checks that schema rows are produced and metrics reach downstream consumers; diagnoses failures in layers from daemon startup through providers, schema matching, exporters, downstream skew, and transport; routes reader-library, publisher-library, productized DTS, DOCA setup, and hardware-state-change questions to their stated owners.
- A network operator deploys an owned CollectX collector on a host or BlueField and needs to expose counters to Prometheus.
- A platform engineer needs to push collector output to Fluent Bit or deliver it through NetFlow, file, or IPC.
- An operator investigates a daemon that starts successfully but produces no schema rows or downstream metrics.
- An engineer verifies that the target device exposes a provider or counter family before committing configuration.
- A user asks about the productized DOCA Telemetry Service, the reader API, or the publisher API and needs correct scope routing.
What are this skill's strengths and limitations?
- Covers the operational path from provider counters to downstream exporters.
- Maintains a clear boundary between CollectX, the two telemetry libraries, and productized DTS.
- Provides entry points for configure, run, test, and debug workflows.
- Addresses Prometheus, Fluent Bit, NetFlow, file, and IPC export classes.
- Requires an existing, validated DOCA installation on the target host or BlueField.
- Does not supply concrete provider names, schema fields, flags, or configuration paths.
- Does not cover DOCA installation, reader or publisher APIs, or the productized DTS container.
- The source provides no concrete command examples, platform test results, or standalone test suite evidence.
How do you install this skill?
Install the specific skill from NVIDIA/skills with the repository's documented skills CLI flow:
npx skills add nvidia/skills --skill doca-collectx-deployment --yes
The README does not specify additional manual installation steps. The skill becomes available when the agent next loads skills for a relevant task.
How do you use this skill?
Use it on a host or BlueField with a healthy DOCA installation at /opt/mellanox/doca, access to exporter sinks, and access to NVIDIA's public DOCA Telemetry and DTS guides. Example prompt: “Deploy an owned CollectX telemetry collector on this DOCA-enabled BlueField, export counters through Prometheus, and diagnose why no metrics reach the downstream consumer.” Confirm concrete provider names, fields, flags, and paths from the live installation or public documentation.