Components: Generate RadioGroup

Summary

Generate RadioGroup with its canonical contract, implementation, tests, stories, platform validation, and explicit package export so mutually exclusive choices have coordinated state and accessibility. Radio now exists, but Label and the shared roving-focus model remain prerequisites.

Observed. radio-group is entry radio-group in the canonical catalog and has no
implementation in packages/agentic/components. Upstream describes it as: Grouping container for two to five Radio sub-components presenting mutually exclusive choices. Owns the group-level accessible name, single-select coordination, and roving tabindex; delegates per-option presentation to Radio.

Goal

Generate the radio-group component in
packages/agentic/components/src/components against the
canonical Fluent Flex spec, with the colocated coverage, stories, and platform
validation the package requires, and export it explicitly from
src/index.ts.

Stage

Stage 3 (production). Wave 3 - focus-orchestrated and composed.

Why it matters

Observed. The canonical catalog has 50 components and the agentic library has
generated 18. radio-group is one of the 32 in the gap, enumerated in
the missing-components index.

Inferred. It is a composed or focus-orchestrated component whose blockers resolve once the wave 2 overlay stack and the focus decision land, so it is a direct measure of whether those foundations were built correctly.

Observed current state

Canonical spec and source

Field	Value
Canonical spec	plugins/components/skills/radio-group/SKILL.md with usage.md, web/tokens.yaml, web/accessibility.md, and web/interaction.md, at x3-design/fluent-design @ d334acf
Upstream tier	molecule
Upstream catalog key	radio-group in component-map.yaml
Local staged spec	packages/agentic/components/specs/radio-group - SPEC.md, usage.md, interaction.md, accessibility.md, tokens.yaml, blockers.md
Local implementation	None. No src/components/radio-group directory exists

Closest existing FURN or V1 implementation

• packages/components/RadioGroup - RadioGroup/, Radio/, and a legacy/ subtree. RadioGroup.tsx renders FocusZone for keyboard coordination; the legacy subtree contains MacOSRadioButtonNativeComponent.ts, a macOS-specific native path that is not the cross-platform baseline.

Note on platform canon. Where a legacy package has platform-specific files, the
desktop path (.win32, .windows, .macos, or the unsuffixed cross-platform file)
is the compatibility reference. .ios, .android, and .mobile variants are not
treated as the cross-platform canonical implementation.

Local dependencies

• src/components/radio - Observed present (component).

Known blockers

Observed (blockers.md, blocker class A in the index).

Recorded as blocked because "the RadioGroup spec depends on a dedicated Radio child contract and a Label-style legend composition" and "packages/agentic/components currently only has button and icon source trees; there is no radio, label, or radio-group implementation to build on".

Reassessment.

Half stale. src/components/radio now exists with a runtime test, a type test, and a story. label does not. Inferred: unblocked by label plus the roving-focus decision in focus-zone.md.

Scope

• Ingest or promote the spec into the component directory in the generated shape
  (SPEC.md plus a spec/ folder), per
  spec-ingestion-and-generation.md.
• Establish the public contract first: variants, slots, native props, accessibility,
  interaction states, and platform behavior.
• Implement the required file structure from
  src/components/AGENTS.md:
  radio-group.types.ts, radio-group.styles.ts, use*.ts, use*Styles.ts, render*.tsx, and
  the assembly module with a displayName.
• Style from Flex tokens first and record genuine token gaps rather than substituting
  unrelated values.
• Export the component, its public types, and the state, style, and render stages
  explicitly from src/index.ts under component-qualified unstable names.
• Own single-select coordination at the group; the existing radio treats selection as externally driven, which matches.
• Implement the legend using label once it exists.
• Implement roving tabindex per the outcome of focus-zone.md.

Out of scope

• Changing the contract of an existing generated component, unless this task explicitly
  calls for it above; raise it as its own item.
• Building shared infrastructure owned by another task: overlay hosting
  (popover), roving focus (focus-zone.md), spec
  ingestion (spec-ingestion-and-generation.md),
  or the coverage contract (component-test-strategy.md).
• Migrating any legacy FURN consumer onto this component.
• Mobile platform support; the applicable upstream companion is web, retargeted to
  React Native for Windows and macOS.

Deliverables

• src/components/radio-group/ implementing the full required file structure.
• SPEC.md and a spec/ companion folder with provenance recording the upstream path,
  SHA, and ingestion date.
• Colocated radio-group.test.tsx, radio-group.types.test.ts, and radio-group.stories.tsx.
• Snapshot coverage where the visual contract is state-dependent, or a recorded
  exemption.
• Explicit exports from src/index.ts, with src/index.test.ts updated.
• A changeset, per the repository's version management rules.
• A recorded readiness level with its supporting evidence.

Required coverage, stories, and platform validation

Runtime and type coverage:

• Selection moves between options and only one is selected at a time.
• Arrow-key navigation selects and moves focus per the platform convention, with wrap behavior.
• Only one option is in the tab order.
• Group accessible name from the legend, and per-option checked state.
• Disabled group and disabled individual options.
• Both controlled and uncontrolled group value.
• User style and handler forwarding, and that user styles are applied after component
  styles.
• Compile-time slot acceptance and any mutually exclusive props.

Stories: a typed Meta titled Components/RadioGroup, an args-driven Default, a
grouped Overview when the contract has several axes, one focused story per variant
axis, and an interactive story for each stateful axis. Self-driving axes expose
default<State> as the control; externally driven selection is owned by the story with
React.useState. Identity-changing axes are not exposed as controls.

Platform validation: the component's stories render and behave correctly on macOS
and on React Native Windows Fabric through apps/storybook,
including hover, pressed, disabled, focus, optional-slot, and constrained-content
scenarios. Add a stable root testID if the component joins the automated smoke set.

Acceptance criteria

• src/components/radio-group exists with the required file structure and a displayName.
• Every variant axis in the component's SPEC.md is implemented, or a deviation is recorded in the spec.
• The component styles from Flex tokens, with any genuine token gap recorded rather than approximated.
• Focus feedback renders through FocusVisual; no outline* props and no React Native Windows native focus visuals are introduced.
• State precedence is explicit: disabled wins over pressed, which wins over hovered.
• Every stateful axis is classified as self-driving or externally driven, and both paths are covered.
• Runtime, type, and story coverage satisfy the package coverage contract.
• The component, its public types, and its pipeline stages are exported explicitly from src/index.ts, and src/index.test.ts covers them.
• The component's stories render on macOS and on React Native Windows Fabric.
• yarn workspace @fluentui-react-native/components format, lint, build, and test pass, and the Storybook bundle succeeds for both desktop platforms.
• A readiness level is recorded with its supporting evidence.
• A changeset is included.

Dependencies and ordering

• Blocked by wave 2, either directly or through a wave 2 dependency listed above.
• Gated by component-test-strategy.md: the component must satisfy the coverage contract on arrival rather than being retrofitted.
• Gated by readiness-model.md: the target readiness level must be stated before the work starts.
• Consumes the Infrastructure workstream (storybook-e2e.md) for automated macOS and Windows Fabric validation; until it lands, platform validation is manual.

Risks and open decisions

1. Arrow-key behavior for radio groups differs between macOS and Windows conventions; the chosen behavior must be stated in the spec and validated on both.
2. It depends on both label and the focus decision, so it cannot start until Stage 1 completes.

Evidence and references

• Upstream spec: plugins/components/skills/radio-group/SKILL.md and
  components-index/SKILL.md
  at x3-design/fluent-design @ d334acf.
• Local staged spec and blocker record: packages/agentic/components/specs/radio-group.
• Canonical implementations to follow: src/components/button
  for a styled higher-order component and src/primitives/icon
  for a primitive.
• Authoring rules: src/components/AGENTS.md and the
  agentic component authoring skill.
• Workstream context: Components README and
  the missing-components index.

Sources retrieved 2026-08-20.

Start with packages/agentic/components/specs/radio-group/SPEC.md and its spec files, then read src/components/AGENTS.md, the existing src/components/radio implementation, and the label and focus-zone blockers. Implement the required radio-group source, tests, stories, exports, and changeset only after those dependencies and decisions are available. Done means the acceptance criteria and runtime, type, story, platform, and export coverage are satisfied.