Components: Generate Dialog

[JasonVMo]

Summary

Generate Dialog with its canonical contract, implementation, tests, stories, platform validation, and explicit package export so modal and alert workflows are available in the agentic library. The task first needs a local spec and then the shared overlay, focus-trap, dismissal, and focus-return foundations.

Observed. dialog is entry dialog in the canonical catalog and has no
implementation in packages/agentic/components. Upstream describes it as: Modal surface that interrupts the workflow to communicate critical information or request a decision the user must resolve before continuing. Covers type (Modal/Non-modal/Alert), the scrim backdrop, header/body/footer anatomy, the right-aligned footer action group, and dependency delegation to Button.

Goal

Generate the dialog 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 2 (beta). Wave 2 - overlay and popover stack.

Why it matters

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

Inferred. It belongs to the overlay stack, which is the largest single cluster of blocked components. Delivering the cluster converts a broad "not implementable" state into ordinary component work.

Observed current state

Canonical spec and source

Field	Value
Canonical spec	plugins/components/skills/dialog/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	dialog in component-map.yaml
Local staged spec	None. No specs/dialog folder exists; the spec must be ingested first
Local implementation	None. No src/components/dialog directory exists

Closest existing FURN or V1 implementation

• No dialog exists in any FURN catalog. The nearest infrastructure is packages/native/Callout for a native floating surface and packages/experimental/Popover for a JavaScript one; neither provides a scrim or a modal focus trap.

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/button - Observed present (component).
• src/components/divider - Observed present (component).

Known blockers

Observed. There is no recorded blocker, because there is no local spec folder and therefore no
blockers.md. The absence of an ingested spec is itself the first blocker; see
spec-ingestion-and-generation.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:
  dialog.types.ts, dialog.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.
• Depend on the overlay host from popover for surfacing, and on the focus-trap decision in focus-zone.md for Modal and Alert types.
• Implement the scrim as part of the overlay host so drawer can reuse it.
• Handle the upstream narrow-viewport rule: footer actions stack full-width with the primary action on top below the documented width.

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/dialog/ implementing the full required file structure.
• SPEC.md and a spec/ companion folder with provenance recording the upstream path,
  SHA, and ingestion date.
• Colocated dialog.test.tsx, dialog.types.test.ts, and dialog.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:

• All three types, with Modal and Alert trapping focus and Non-modal not.
• Escape dismisses where the type allows it and does not where it does not.
• Focus moves into the dialog on open and returns to the opener on close.
• Header, body, and footer slots, including the right-aligned action group and the stacked narrow layout.
• Dialog role and accessible name announced on open.
• 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/Dialog, 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/dialog 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 spec-ingestion-and-generation.md: the spec must be pulled from the pinned upstream revision before generation can start.
• Blocked by popover for the overlay host, anchoring, light dismiss, and focus return.
• 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. Modal focus trapping is the strongest dependency on the focus decision; a JavaScript roving-focus hook may not be sufficient for a true trap.
2. React Native Modal behavior on macOS and on Windows Fabric differs in window ownership and z-ordering, which affects whether the scrim can cover the whole surface.

Evidence and references

• Upstream spec: plugins/components/skills/dialog/SKILL.md and
  components-index/SKILL.md
  at x3-design/fluent-design @ d334acf.
• 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.