DOCA Telemetry Utils
Discover telemetry counters, translate Data IDs, and verify BlueField support before committing exporter configuration.
The skill clearly limits the tool to read-only use, requires device-support probing and re-resolution across versions, warns against invented Data IDs, and discloses possible operational-data sensitivity. It lacks explicit user confirmation, rollback guidance, detailed least-privilege handling, and pinned external sources, so points are deducted.
SKILL.md, CAPABILITIES.md, and TASKS.md provide detailed invocation classes, layered errors, and failure routing. However, this is a static review: the CLI was not executed, and there is no committed coverage-oriented test suite or reproducible abnormal-output evidence, so the score remains conservative.
Audience, trigger examples, supported scenarios, and out-of-scope boundaries are unusually explicit. The skill nevertheless requires Linux, a DOCA installation, BlueField hardware, and a fixed filesystem layout; Chinese-language operation is not addressed, and mainland-China reachability of external NVIDIA documentation is unverified, so points are deducted.
The documentation has strong structure, loading order, task verbs, related-skill routing, and limitation disclosures, with Apache-2.0 stated in frontmatter and issue intake documented. However, license metadata conflicts across supplied files, signatures and versions remain TBD, and no clear changelog is present, so points are deducted.
The core outcomes—counter enumeration, name/Data ID translation, device-support validation, and troubleshooting handoff—are clearly specified with workflows and an example. There are no scripts or structured outputs, actual results depend on local DOCA/device versions and external documentation, and the benchmark dataset is unavailable; direct usability therefore cannot be fully established statically.
The files include a concrete Data ID example, an auditable evidence-tuple requirement, error taxonomy, and a benchmark report. The benchmark dataset is absent, no real CI plus committed key-path test suite is shown, and external references were not independently verified from the supplied material, limiting the score.
- Before use, confirm the local DOCA version, Telemetry component, BlueField PCI address, and firmware; the documented command shapes do not replace --help from the installed version.
- Do not commit exporter configuration from schema resolution alone; probe every target device and retain stdout, version, firmware, and Data ID evidence.
- License metadata is inconsistent: SKILL.md states Apache-2.0, while SKILLCARD.yaml and skill-card.md state Apache-2.0 AND CC-BY-4.0; reconcile this before publication.
- The benchmark dataset is unavailable, so its PASS verdict and percentages should not be treated as independently reproducible evidence.
What does this skill do, and when should you use it?
This NVIDIA skill guides operators in using the DOCA Telemetry diagnostic CLI on a BlueField host. It discovers the counter schema, translates counter names to binary Data IDs and back, and probes whether a specific device supports a counter and its properties. It is suited to diagnosing exporters that appear to emit metrics while collectors receive nothing, including silent drops caused by unsupported counters. It is not a guide to developing collector or exporter libraries, deploying DTS, or installing or repairing DOCA.
It guides an agent to invoke /opt/mellanox/doca/tools/doca_telemetry_utils on a Linux host: run get-counters to enumerate diagnostic counters; resolve names and properties to Data IDs; reverse-resolve captured Data IDs into counter names and properties; and probe support with a device PCI address, counter name, and optional properties. It also directs the agent to TASKS.md and CAPABILITIES.md for prerequisite checks, invocation selection, testing, debugging, version matching, and the read-only safety model. The tool produces human-readable Data ID and property mappings.
- A platform operator is preparing a DOCA Telemetry exporter configuration for a BlueField fleet and needs to enumerate counters and verify device support.
- A downstream collector developer has a captured Data ID and needs to correlate it with a counter name and property dimensions.
- An operator is investigating a report that an exporter emits metrics but the collector receives no events.
- A developer is migrating an exporter pipeline across DOCA versions and must confirm that previously supported counters still resolve against the matching schema version.
- An AI agent must produce an exporter counter set validated for a particular BlueField, device address, and DOCA version.
What are this skill's strengths and limitations?
- Covers schema enumeration, name-to-ID resolution, ID-to-name reverse resolution, and per-device support probing.
- Clearly separates operator-side CLI support from the doca-telemetry collector library, exporter development, and DTS deployment.
- Uses the installed tool as the authoritative counter source instead of shipping a version-sensitive static inventory.
- Documents version pairing, layered error handling, silent metric-drop diagnosis, and a read-only safety policy.
- Requires an installed and compatible DOCA SDK, the Telemetry component, and a DOCA-visible BlueField DPU.
- Per-device probing typically requires elevated privileges.
- It does not ship a fixed counter table, exporter or collector configurations, wrappers, parsers, scripts, or DTS deployment recipes.
- The source provides no standalone automated test suite or evidence of validation across platforms.
How do you install this skill?
Install the individual skill with the repository's documented skills CLI flow: npx skills add nvidia/skills --skill doca-telemetry-utils --yes. Alternatively, run npx skills add nvidia/skills interactively and select the skill. The skill becomes available when the agent next loads skills for a relevant task. Runtime use additionally requires Linux, the DOCA SDK with its Telemetry optional component installed at /opt/mellanox/doca, and a BlueField DPU visible to DOCA. The source does not provide a separate doca_telemetry_utils installation command.
How do you use this skill?
Ask the agent an operator-side question such as “enumerate the DOCA diagnostic counters on this BlueField,” “reverse-resolve this captured Data ID,” “validate port_rx_bytes with node=1 on PCI address X,” or “the exporter emits but the collector receives nothing; check counter support.” The agent should read SKILL.md first, then consult TASKS.md and CAPABILITIES.md as needed, using the get-counters, name-to-ID, ID-to-name, and per-device probe workflows.
How does this skill compare with similar options?
It complements the developer-side doca-telemetry collector skill: this skill makes the counter schema and device support inspectable, while doca-telemetry covers collector-side development and pipeline contracts. It is distinct from doca-telemetry-exporter, doca-public-knowledge-map, and doca-setup, which cover exporter programming, public DOCA/DTS routing, and DOCA installation or environment preparation.