create-core-check
SkillMonitoring & opsCreate a new Go core check that collects metrics and sends them to Datadog
Available today. Use it from your connected AI after setup.
No other account needed.
Connect ahel once, and every AI you use reads what you have installed.
Then ask your AI: use the create-core-check skill
What this skill tells your AI
The instructions your AI receives, as published by datadog/datadog-agent in .agents/skills/create-core-check/SKILL.md and read by ahel’s review.
Create a new Go-based core check for the Datadog Agent. Core checks collect metrics, service checks, or events and send them to Datadog at regular intervals.
Instructions
Step 1: Gather information from the user
Use AskUserQuestion to collect the following. If $ARGUMENTS provides the check name, skip that question.
-
Check name: The identifier for the check (e.g.
uptime,memory,ntp). Used as the package name, registration key, and config directory name. -
Check category: Where should the check live under
pkg/collector/corechecks/?system/— System-level checks (CPU, memory, uptime, disk)net/— Network checks (NTP, DNS)containers/— Container-related checksebpf/— eBPF-based checks (these are more complex, seepkg/collector/corechecks/ebpf/AGENTS.md)embed/— Embedded service checks- Top-level under
corechecks/— For standalone checks
-
What does it collect?: Describe the metrics, service checks, or events it produces.
-
Configuration: Does it need instance-level configuration?
- No config — Single instance, no user parameters (like
uptime) - Simple config — A few YAML parameters (like
memorywithcollect_memory_pressure) - Multi-instance — Supports multiple configured instances (like
ntpwith different servers)
- No config — Single instance, no user parameters (like
-
Component dependencies: Does the check need injected components?
- None — Simple check, no external dependencies
- Tagger — Needs to tag metrics with container/host tags
- WorkloadMeta — Needs access to workload metadata store
- Other — Specify which components
-
Long-running?: Does the check run continuously in the background?
- No (default) —
Run()is called at regular intervals (default 15s) - Yes —
Run()never returns, processes events in a loop
- No (default) —
-
Platform restrictions: Does the check only work on certain platforms?
- All platforms (default)
- Linux only
- Windows only
- Linux + macOS (not Windows)
Step 2: Read reference examples from the codebase
Before writing any code, read the appropriate reference files based on the check type determined in Step 1. Follow the patterns found in these files exactly.
| Check type | Reference file to read |
|---|---|
| Simple, no config | pkg/collector/corechecks/system/uptime/uptime.go |
| Simple with config | pkg/collector/corechecks/system/memory/memory.go |
| Multi-instance with config | pkg/collector/corechecks/net/ntp/ntp.go |
| With component dependencies | pkg/collector/corechecks/containerimage/check.go |
| Long-running | Read the NewLongRunningCheckWrapper usage in pkg/collector/corechecks/containerimage/check.go |
| Platform-specific stubs | Find a _no*.go or _stub.go file alongside a platform-specific check in pkg/collector/corechecks/system/ |
Also read these files for registration and test patterns:
pkg/commonchecks/corechecks.go— to see how checks are registered (import alias convention,RegisterCheckcalls)- The
_test.gofile alongside whichever reference check you read — to see mock sender patterns
Step 3: Create the check package
Directory: pkg/collector/corechecks/<category>/<checkname>/
Create the check implementation file following the patterns from the reference files read in Step 2. Key structural elements that every check needs:
CheckNameconstant — string identifier for the checkCheckstruct — embedscore.CheckBase, plus any config or component fieldsFactory()function — returnsoption.Option[func() check.Check]. Components are injected as Factory parameters.Configure()method — callsCommonConfigure, thenFinalizeCheckServiceTag, then parses instance config if neededRun()method — collects data, calls sender methods, ends withsender.Commit()
Key rules to follow:
- For multi-instance checks: call
c.BuildID(integrationConfigDigest, rawInstance, rawInitConfig)beforeCommonConfigure() - For long-running checks: wrap with
core.NewLongRunningCheckWrapper()in Factory, return0fromInterval(), implementStop() - For platform-specific checks: add
//go:build <platform>tag and create a stub file for other platforms that returnsoption.None[func() check.Check]()
Step 4: Register the check
Edit pkg/commonchecks/corechecks.go:
- Add an import for the check package using the standard alias convention visible in the file (typically the check name)
- Add a
corecheckLoader.RegisterCheck()call inRegisterChecks(), matching the Factory signature to available component parameters
Step 5: Create the default configuration
File: cmd/agent/dist/conf.d/<checkname>.d/conf.yaml.default
Look at an existing example in cmd/agent/dist/conf.d/ for the format. At minimum:
init_config:
instances:
- {}
For checks with configuration, use @param annotations following the same format as other conf.yaml.default files in the tree.
Step 6: Write tests
File: pkg/collector/corechecks/<category>/<checkname>/<checkname>_test.go
Follow the test patterns from the reference file read in Step 2. The standard test flow is:
- Create a
mocksender.NewMockSender("") - Set up
mockSender.On("FinalizeCheckServiceTag").Return() - Create and
Configurethe check withmockSender.GetSenderManager() - Call
mocksender.SetSender(mockSender, check.ID()) - Set expectations on the mock sender for expected metrics
- Call
Run()and assert expectations
Step 7: Verify
-
Run the check tests:
dda inv test --targets=./pkg/collector/corechecks/<category>/<checkname> -
Build the agent:
dda inv agent.build --build-exclude=systemd -
Run the linter:
dda inv linter.go -
Report the results to the user.
Sender Methods Reference
The sender (c.GetSender()) provides these methods for submitting data:
| Method | Description |
|---|---|
Gauge(metric, value, hostname, tags) | Submit a gauge metric |
Rate(metric, value, hostname, tags) | Submit a rate metric |
Count(metric, value, hostname, tags) | Submit a count metric |
MonotonicCount(metric, value, hostname, tags) | Submit a monotonic count |
Histogram(metric, value, hostname, tags) | Submit a histogram metric |
Distribution(metric, value, hostname, tags) | Submit a distribution metric |
ServiceCheck(name, status, hostname, tags, message) | Submit a service check |
Event(event) | Submit an event |
Commit() | Flush all submitted data — must be called at end of Run() |
- Pass
""for hostname to use the agent's default hostname. - Pass
nilfor tags if no tags are needed. - Service check statuses:
servicecheck.ServiceCheckOK,ServiceCheckWarning,ServiceCheckCritical,ServiceCheckUnknown(frompkg/metrics/servicecheck).
Important Notes
CheckBaseprovides default implementations for mostCheckinterface methods. You only need to overrideRun()and optionallyConfigure(),Stop(), andInterval().CommonConfigurehandles standard configuration: collection interval (min_collection_interval), custom tags, service tag, etc.FinalizeCheckServiceTag()must be called afterCommonConfigureto apply the service tag to the sender.- Always call
sender.Commit()at the end ofRun()to flush data. - For multi-instance checks,
BuildID()must be called beforeCommonConfigure(). - The
option.None[func() check.Check]()pattern is used for platform stubs — the loader skips checks with no factory. integration.FakeConfigHashis the constant to use in tests for the config digest parameter.
Usage
/create-core-check— Interactive: prompts for all details/create-core-check my_check— Pre-fills the check name
Signals
- GitHub stars
- 4k
- Forks
- 1k
- Last commit
- Sep 2026
Advanced
- Catalog kind
- skill
- Gateway key
create-core-check- Source
- github.com/datadog/datadog-agent