Doca Telemetry Utils Agent Skill

DOCA Telemetry Utils

Where to start: This is a tool skill for invoking

doca_telemetry_utils — the documented host-side CLI that

supports a DOCA Telemetry exporter / collector pipeline by

discovering the counter schema, translating counter names ↔

Data IDs, and probing per-device counter support. Open

TASKS.md and start at

## install for the host-side

prerequisites and ## run for the

three documented invocation classes

(enumerate / name→ID / ID→name). Open

CAPABILITIES.md when the question is

*what does this tool actually discover about the telemetry

schema*, *how does it pair with the developer-side

doca-telemetry and

exporter libraries*, *how do I confirm a device supports a

counter before committing an exporter config to it*, or

*why does my exporter pipeline silently drop a metric*.

This skill is the operator-side support tool for a

DOCA Telemetry deployment. It is NOT the developer-side

collector library (that is

doca-telemetry),

NOT the developer-side publisher library (that is

doca-telemetry-exporter — see

doca-telemetry ## Related skills),

and NOT a DOCA Telemetry Service (DTS) deployment guide

(route via

doca-public-knowledge-map).

Three separate surfaces; conflating them is the most

common telemetry first-touch error.

Example questions this skill answers well

The CLASSES of doca_telemetry_utils 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.

• **"My exporter says it's emitting port_rx_bytes but

nothing shows up downstream — what did I get wrong?"** —

worked example: *"my exporter config has a counter name

string and my collector sees no events with that

name"*. Answered by the name ↔ Data ID translation

step in

CAPABILITIES.md ## Capabilities and modes

+ the per-device-support probe in

TASKS.md ## test: the exporter

ships a Data ID, not a name; a name in the config that

resolves to a Data ID the device does not support is

silently dropped.

• **"Which DOCA diagnostic counters does this BlueField

actually expose?"** — worked example: *"enumerate the

full counter schema for my BlueField-3 before I write

the exporter config"*. Answered by the schema-discovery

invocation class in

CAPABILITIES.md ## Capabilities and modes

(doca_telemetry_utils get-counters lists every

counter name the diagnostic-data surface knows about;

pair with a per-device probe to confirm support).

• **"I have a Data ID in a captured log — what counter

was that?"** — worked example: *"a downstream consumer

emitted Data ID: 0x1160000600030201 — translate it

back so I can correlate against the public guide"*.

Answered by the reverse-resolve invocation class in

CAPABILITIES.md ## Capabilities and modes

+ the

ID-encodes-properties rule in

TASKS.md ## use (a Data ID carries

the counter's property dimensions; the reverse-resolve

reports them).

• **"What property dimensions does this counter take and

what values are valid?"** — worked example: *"I know

the counter is pcie_link_write_stalled_time_* — what

do node / pcie_index / depth mean and what

values does the device accept?"*. Answered by the

property-dimension table in

CAPABILITIES.md ## Capabilities and modes

+ the <name> invocation without arguments which

prints the documented property options + units +

unit-specific axes.

• **"Is this counter supported on this device before I

commit it to the exporter config?"** — worked example:

*"validate that port_rx_bytes with node=1 is

exposed on the BlueField at PCIe address X before I

write the exporter config"*. Answered by the

per-device-support probe (`<device PCI> <name>

[properties]`) in

CAPABILITIES.md ## Capabilities and modes

+ the gate-before-commit rule in

TASKS.md ## use.

• **"Is doca_telemetry_utils on my install, and is it

paired with the matching doca-telemetry library

version?"** — worked example: *"is the diagnostic-

data counter set my exporter targets on this DOCA

version?"*. Answered by the version-overlay in

CAPABILITIES.md ## Version compatibility,

which redirects to the canonical

doca-version chain

and adds the *tool ↔ doca-telemetry library

schema-version* match rule.

Audience

This skill serves **external operators, developers, and

AI agents standing up or debugging a DOCA Telemetry

exporter / collector pipeline who need to confirm the

counter schema, validate per-device support, or

translate between human-readable counter names and the

binary Data IDs the exporter actually ships**.

Concretely:

• A platform operator standing up a new DOCA Telemetry

exporter on a BlueField fleet who needs to confirm

which counters the target devices actually expose

before committing the exporter config.

• A developer of a downstream consumer (a collector

app linking doca-telemetry,

or a third-party aggregator consuming via DTS) who

has a captured Data ID stream and needs to translate

IDs back to counter names + properties.

• An operator debugging a *"nothing is shipping

downstream"* / *"this metric is silently missing"*

report against a deployed exporter — the

schema-discovery + per-device-support probes are the

canonical *"is the counter even supposed to work on

this device"* first step before suspecting the

collector or the network path.

• An AI agent producing a *"validated exporter config

for this BlueField + this DOCA version"* answer

honestly — with each counter resolved to its Data ID

and each Data ID confirmed against the per-device

capability probe.

It is not for users debugging the doca_telemetry_utils

binary itself, not a substitute for the live public

DOCA Telemetry guides, and not the right place for

learning how to *write* an exporter application (that

audience belongs in

doca-telemetry-exporter skill

when present, or via

doca-public-knowledge-map).

The tool is shipped as a CLI binary under

/opt/mellanox/doca/tools/, not a library you link

against. The skill uses the same kind: tool

three-file shape as the rest of the bundle so the

agent's task-verb contract is uniform across libraries,

services, and tools.

Language scope

doca_telemetry_utils is a C host-side CLI that uses

the documented doca_telemetry_diag surface to list

counter types and probe per-device support. Its inputs

are command-line arguments (counter name, property

values, Data ID, optional device PCI address); its

outputs are human-readable Data ID + properties

mappings. The skill keeps workflow guidance language-

neutral; downstream consumers in any language can use

the resolved Data IDs against their own collector code.

When to load this skill

Load this skill when the user is — or the agent needs

to — invoke doca_telemetry_utils on a real host with

DOCA installed, alongside an exporter / collector

pipeline that needs schema discovery, per-device

support validation, or Data ID translation.

Concretely:

• Enumerating the full counter schema before writing

a fresh DOCA Telemetry exporter config (get-counters).

• Validating that a chosen counter + property set

resolves to a Data ID the target device actually

supports, before committing the config.

• Reverse-resolving a Data ID captured from a

downstream consumer back to a counter name +

properties for correlation against the public guide.

• Debugging a *"exporter ships, collector receives

nothing"* report (the canonical schema-mismatch /

unsupported-counter failure mode).

• Producing a *"validated counter set for this

BlueField + this DOCA version"* artifact as part of

a structured exporter-config baseline.

• Migrating an exporter pipeline across DOCA versions

and confirming each previously-supported counter

still resolves cleanly on the new version.

Do not load this skill for general DOCA orientation,

collector / exporter library programming, DTS

deployment, or DOCA install. For those, route to

doca-public-knowledge-map,

doca-telemetry,

or doca-setup.

What this skill provides

This is a thin loader. Substantive material lives

in two companion files:

• CAPABILITIES.md — what doca_telemetry_utils

discovers + how it pairs with the developer-side

surfaces: the three documented invocation classes

(enumerate / name→ID / ID→name), the property-

dimension model (counters carry property axes such

as node, pcie_index, depth, plus per-unit

axes), the optional per-device capability probe

(<device PCI> <name> runs the resolved counter

against the device), the operator-side support

role (this is NOT the developer-side library; it

exists to make exporter / collector setups

honest), the version overlay (tool ↔

doca-telemetry library schema version pairing),

the layered error taxonomy (install / parse /

unknown-counter / unknown-data-id / property-out-

of-range / device-not-supported / version /

cross-cutting), the observability surface

(stdout-only), and the safety policy (the tool is

read-only; mistakes appear downstream as silent

metric drops, not crashes).

• TASKS.md — step-by-step workflows for the

in-scope task verbs: install (host-side DOCA +

telemetry component prerequisites), configure

(axis decisions: which invocation class, which

device, which counter), build (route to install

— the binary is shipped), modify (refuse — do

not patch the binary; modify the invocation and

the exporter / collector config that consumes the

resolved Data IDs), run (the three documented

invocations), test (round-trip a chosen

counter through name → Data ID → per-device

probe → exporter config → collector receipt),

debug (walk the error taxonomy), use (the

hand-off into the developer-side doca-telemetry

collector / exporter pipeline), plus a `Deferred

task verbs` block.

The skill assumes a host where DOCA is already

installed with the telemetry component, a BlueField

the operator is targeting is visible to DOCA, and the

exporter / collector pipeline that consumes the

resolved Data IDs is either being authored or has

already been authored against the matching

doca-telemetry

library version.

What this skill deliberately does not ship

This skill is agent guidance, not a samples or

scripts bundle. To keep the boundary clean, it

deliberately does not contain — and pull requests

should not add:

• **A verbatim counter inventory or Data ID

table.** The counter set evolves across DOCA

versions and BlueField generations; an

inventory pinned in this skill would silently rot.

doca_telemetry_utils get-counters on the

installed binary is the authoritative source.

• Pre-baked exporter / collector configs. Configs

are install-, device-, and use-case-specific; a

packaged config in this skill would mislead

operators on a different setup.

• A DTS deployment recipe. DTS is a separate DOCA

service with its own public guide; route via

doca-public-knowledge-map.

This skill *resolves the counter names + IDs* a

DTS pipeline consumes; it does not configure DTS.

• Wrappers, parsers, or scripts in any language

that consume the tool's output. The output is a

simple text mapping; users who want to script

against it should read the live guide and write

the parser against their installed version.

• A samples/ or reference/ subtree. This is

a thin loader for a shipped CLI; substantive

material lives on the public page, in --help,

and in doca-telemetry.

Loading order

• Read this SKILL.md first to confirm the user's

question is in scope (operator-side support

tooling for a telemetry pipeline, not the

developer-side library programming).

• **For the three invocation classes, the

property-dimension model, the per-device support

probe, the version overlay, the error taxonomy,

the observability surface, and the safety

policy, see CAPABILITIES.md.**

• **For the documented invocations and the

discover → resolve → validate → consume

workflow — install, configure, build,

modify, run, test, debug, use — see

TASKS.md.**

Related skills

• doca-telemetry

— the developer-side collector library whose

schema this tool helps the operator discover. Pair

them in every exporter-pipeline triage session.

The collector library skill teaches the

collector-vs-exporter rule, the schema-must-match

contract with the publisher, and the consumer-

queue-full back-pressure rule; this tool's role

is to make the *schema* half of that contract

inspectable from the operator side. Conflating

the library and the tool is the most common

telemetry first-touch error.

• doca-public-knowledge-map

— routing to the public DOCA Telemetry guide and

the DOCA Telemetry Service (DTS) page on

docs.nvidia.com, plus the on-disk install

layout for the tool.

• doca-version —

canonical DOCA version-handling rules. The

## Version compatibility section in

CAPABILITIES.md is a concise

overlay that redirects here for the body and

adds the *tool ↔ doca-telemetry library

schema-version pairing* rule.

• doca-setup — env

preparation, install verification, and the *I

have no install yet* path with the public NGC

DOCA container. This skill assumes its

preconditions are satisfied (DOCA installed

with the telemetry component, BlueField visible

to DOCA).

• doca-debug — the

cross-cutting debug ladder. Telemetry-utils

feeds the cross-cutting ladder by surfacing the

counter schema + per-device support truth at the

runtime layer; exporter-pipeline regressions

often resolve at the schema layer before

touching the collector / network path.

• doca-structured-tools-contract

— the bundle's detect → prefer → fall back →

report contract for structured helper tools. The

command appendix in TASKS.md

honors this contract.

• doca-programming-guide

— general DOCA programming patterns shared by

every library / tool surface, including the

cross-library DOCA_ERROR_* taxonomy this

tool's host-side error layer overlays on top of

when downstream collector / exporter code

surfaces a related error.