create-config-field

name	create-config-field
description	Add a new configuration field to the Datadog Agent (datadog.yaml) by declaring it in the config schema
allowed-tools	Bash, Read, Write, Edit, Glob, Grep, AskUserQuestion
argument-hint	[config.key.name]
model	sonnet

Add a new configuration setting to a Datadog Agent config file by declaring it in the configuration schema.

The schema is the single source of truth for every Agent setting: its type, default, documentation, environment variables, validation rules and visibility all live in one YAML node. Do not add BindEnvAndSetDefault calls by hand — the pkg/config/setup/*_settings.go files are generated from the schema, and so are datadog.yaml.example / system-probe.yaml.example, the JSON Schema published to SchemaStore, and the runtime config validation.

Full reference: docs/public/agent-schema/ — index, keywords, examples, cli, faq.

Where settings live

One schema per config file, all under pkg/config/schema/yaml/:

Config file	Schema	--schema value
datadog.yaml	pkg/config/schema/yaml/core_schema.yaml	core (default)
system-probe.yaml	pkg/config/schema/yaml/system-probe_schema.yaml	system-probe

Large top-level sections are split into sibling files referenced via $ref, so the node for apm_config.enabled lives in apm_config.yaml, not in core_schema.yaml.

Never grep for the file by hand — dda inv -- schema.locate resolves the $ref for you (see Step 2).

Instructions

Step 1: Gather information from the user

Use AskUserQuestion to collect the following. If $ARGUMENTS provides the setting path, skip that question.

1. Target schema: datadog.yaml (core) or system-probe.yaml?
2. Setting path (dot-separated, e.g. my_feature.enabled).
3. Type: boolean, string, number, integer, array, or object. For array, also the element type (items.type is mandatory).
4. Default: a single default, or per-platform platform_default.
5. Visibility: public (appears in the generated *.yaml.example and public docs) or undocumented (the default — internal, no keyword emitted).
6. Description: mandatory for public, strongly encouraged otherwise. Written for users, not Agent developers. This should explain what the settings does and how to use it.
7. Description for each ancestor section: if a setting is public, each parent section must be public too with their own description. Ask for a description per section newly made public, separately from the setting's. Never reuse or copy the setting's description into its parent section — a section describes what the group of settings is for, a setting describes its own value.
8. Comment: an optional description aimed at developers.

Step 2: Check the setting does not already exist

dda inv -- schema.locate my_feature.enabled          # exact path
dda inv -- schema.locate '.*my_feature'              # pattern (regex/glob)

This also tells you which file to edit. See the locate-config-setting skill for the full flag set.

Step 3: Add the node to the schema

Preferred — the interactive wizard, which routes split sections to the right sub-file, preserves the file's hand-curated ordering, makes ancestor sections public when needed, and lints at the end:

dda inv schema.add-setting                       # core schema
dda inv schema.add-setting --schema=system-probe # system-probe schema

The wizard is interactive (it reads from stdin), so when you cannot drive a TTY, edit the YAML directly instead. Read a neighbouring node first and match its style:

my_feature:
  node_type: section
  type: object
  visibility: public
  description: Configuration for my feature.
  properties:

    enabled:
      node_type: setting
      type: boolean
      default: false
      description: Enables my feature.
      visibility: public

Rules that the linter enforces:

• Every node needs node_type: section or node_type: setting.
• Every setting needs a type and exactly one of default / platform_default.
• platform_default must cover every platform — list linux, windows, darwin, aix explicitly, or add an other catch-all. container / fargate are optional and fall back to linux then other.
• An array setting must declare items.type.
• A public node needs a non-empty description, and every ancestor section must also be public with a description — its own description, gathered in Step 1, not a copy of the child setting's.
• A section needs at least one child; a public section needs at least one direct public child.
• Set node_type: setting — not section — when the value is an object (e.g. docker_labels_as_tags: type: object, default: {}). A section is only for grouping child settings.

Placement matters: the generated config examples follow schema order, so insert the node where it belongs logically, not at the end of the file.

Step 4: Lint and preview

dda inv schema.lint

Keyword quick reference

Full up-to-date details in docs/public/agent-schema/keywords.md.

Keyword	Where	Notes
node_type	all	section or setting. Mandatory.
type	setting	boolean, number, integer, string, array, object.
default	setting	Must match type. Mutually exclusive with platform_default.
platform_default	setting	Keys: linux, windows, darwin, aix, container, fargate, other.
description	all	Mandatory when public. Use the | block scalar for multi-line.
visibility	all	public or undocumented (default).
env_vars	setting	Overrides the derived DD_* name; first match wins.
env_parser	setting	comma_separated, space_separated, json. Needed for complex types.
sensitive	setting	Scrubs the value from logs, flare and Fleet Automation.
items	setting	Mandatory for array.
properties	section / object setting	Child settings on a section; value sub-schema on an object setting.
title	section	Banner heading in the generated example.
comment	all	Developer-only note; never rendered to users.
example	setting	Overrides the value shown on the rendered example line.
tags	all	See below.

Relative defaults: use ${conf_path}, ${install_path}, ${log_path}, ${run_path} with / separators rather than hardcoding per-OS paths — e.g. default: "${conf_path}/conf.d". Valid in default and platform_default.

Tags

Three are usable for new settings:

• template_section:<name> — selects which config-example flavors include the setting. Omit it and the setting renders in every build type.
• platform_only:<os>[,<os>] — restricts the setting to the listed OSes (windows, linux, darwin); it is dropped from the examples generated for any other --os-target.
• generate_const:<Name> — emits a Go constant <Name> in pkg/config/setup holding this setting's default. Use it instead of hardcoding a default (port, timeout, path) in Go code, so the two can never drift.

golang_type:*, no-env and the legacy env_parser values (comma_and_space_separated, traces_span, csv_comma_separated, comma_then_space_separated, json_list_or_*) exist only to support existing settings — do not use them for new ones.

Reading the setting from Go

pkgconfigsetup.Datadog().GetBool("my_feature.enabled")
pkgconfigsetup.SystemProbe().GetInt("system_probe_config.max_conns")

In components, prefer the injected config.Component over the global accessor.

Related

• locate-config-setting — find where an existing setting is defined.
• create-release-note — a user-visible new setting needs a reno note.

Usage

• /create-config-field — interactive: prompts for all details
• /create-config-field my_feature.enabled — pre-fills the setting path