kandev/apps/web/AGENTS.md

Frontend (Vite/React SPA) — architecture and conventions

Scoped guidance for apps/web/. Repo-wide rules (commit format, code-quality limits, etc.) live in the root AGENTS.md.

Plugin authoring

For plugin UI work, begin with the canonical plugin authoring guide. Follow: choose recipe → edit manifest.yaml → implement → validate → package → smoke test. The independently consumable author contract is @kandev/plugin-sdk in ../packages/plugin-sdk; ../../docs/plans/plugins/PLUGIN-API.md and lib/plugins/types.ts document and implement host compatibility. Concrete shared Host UI exports are in lib/plugins/host-api.ts, and registration/cleanup behavior is in lib/plugins/registry.ts and lib/plugins/host.ts. New and official plugins use typed host.context reads and never copy/import private AppState or Zustand slice shapes. Extend the SDK, host implementation, contract docs, and exact-consumer compatibility test together.

UI Components

Shadcn Components: Import from @kandev/ui package:

import { Badge } from "@kandev/ui/badge";
import { Button } from "@kandev/ui/button";
import { Dialog } from "@kandev/ui/dialog";

Do NOT import from @/components/ui/* - always use @kandev/ui package.

• Always prefer native shadcn components over custom implementations. For async-status UI, do not rely on a changing button label as a live announcement: use one translated role="status" region, keep the button's accessible name stable, and override @kandev/ui/spinner's default English aria-label with t(...) when it is announced (or use aria-hidden when another live status covers it).
• Check apps/packages/ui/src/ for available components (pagination, table, dialog, etc.).
• For data tables, use @kandev/ui/table with TanStack Table; use shadcn Pagination components.
• Only create custom components when shadcn doesn't provide what's needed.

Responsive and touch surfaces

• Use hooks/use-responsive-breakpoint.ts for application layout decisions. Its mobile boundary matches the sidebar's 768px md boundary, and it also models tablet, compact desktop, full desktop, and pointer precision; do not substitute the UI package's generic useIsMobile hook. Tablet is a coarse-pointer fallback between md and lg, so no fine-pointer width reports it — a spec that needs the tablet layout has to emulate touch.
• When a Tailwind visibility class gates the same surface the hook picks, both must use the same boundary. A sm: class paired with a hook-driven mobile branch leaves 640-767px in a state neither side renders.
• CSS visibility classes can leave the other viewport tree mounted. Gate live effects such as polling, subscriptions, timers, and measurements with the same responsive or pointer condition, and test both initialization orders when readiness can precede or follow hydration.
• Use useTouchDrawer when a hover/popover disclosure needs a coarse-pointer Drawer alternative. Width-based phone composition and pointer-based disclosure behavior are related but not interchangeable. Apply the 44px minimum to coarse-pointer hit areas, touch rows, and mobile controls only; keep fine-pointer desktop controls at the surrounding design-system density, using 28px for ordinary buttons, inputs, and selectors and 24px only for deliberate compact inline controls, and do not reuse a touch-sized h-11 class as the shared visual button size. See the sizing guide for exceptions.
• Below 640px, shared Radix DropdownMenu/ContextMenu use inset, safe-area-aware bottom sheets in app/globals.css. Open mobile-menu-root content owns one decorative positioner backdrop matching Drawer dimming and blur; fade it with the sheet's exit motion. Never mark submenus; keep the backdrop outside scrolling content and pointer-transparent so Radix owns dismissal and non-modal interaction. Reuse these primitives and cover long/nested menus instead of adding parallel mobile menus.
• Mobile capability parity does not require desktop layout parity. Load /mobile-parity for the Kandev surface decision guide, mobile design contract, and verification requirements; read confirmation guidance when adopting phone confirmation surfaces.
• Phone hamburger buttons use AppNavSheet for app navigation across listings, task workbenches, and page shells. The task title opens the task picker; ResponsiveTaskPicker owns its dialogs above responsive layout branches, and TaskSheetSelectionProvider shares cancellation with the embedded task sidebar. MobileTaskNavigationProvider retains the inline sidebar controller and action dialogs above responsive page headers, rendering the body into the shared menu outlet. Tasks collapses in place and shares the menu scroller. KanbanHeaderMobile opens listing-only MobileMenuSheet options from the Kanban/Threads/List title dropdown; Threads saved views render inline inside that surface. MobileListingMenuActions supplies search in listing options. Phone app navigation groups main-toolbar, sidebar workspace, and task plugin controls in MobilePluginNavSection, in one wrapping group. When task controls exist, each plugin uses its task toolbar instead of a second workspace toolbar; workspace-only plugins and sidebar actions remain available. Fallback system metrics follow navigation before Utilities. Keep activity/connection cues on AppNavTrigger and restore the actual opener unless focus moves into a launched surface. Tablet/desktop composition stays separate. Phone navigation uses Home for all listing modes; Tasks/Threads remain routes and palette destinations. The inline Tasks heading retains its independent plus. Phone Automations reuses workspace automation reads only while expanded, and Integrations retains a workspace settings entry even without configured links. Phone app navigation groups the built-in Quick Chat/Quick terminal actions below Home. MobileQuickActions shares launch and focus behavior; plugin actions and metrics retain separate slots. MobileIntegrationsSection opts into a local, initially collapsed disclosure only for phone app navigation; wider consumers retain expanded rendering.
• Saved phone sidebar layouts retain Home/quick actions before the task outlet and put optional tools/groups afterward in saved relative order. Built-in Automations, Canvases, and Integrations use labelled phone disclosures; icon strips belong to custom shortcut groups. The built-in integration disclosure excludes plugin links rendered as independently customizable nodes. Never persist this phone composition over the desktop layout.

Data Flow Pattern (Critical)

Go Boot Payload -> Hydrate Store -> Components Read Store -> Hooks Subscribe

Never fetch data directly in components.

Browser capability boundaries

• Use generateUUID() for client-only non-security IDs, not crypto.randomUUID(); use copyToClipboard() for copy actions, not navigator.clipboard.writeText(). Keep fallbacks non-security; test missing or rejected capabilities with an rg audit.

Store Structure (Domain Slices)

lib/state/
├── store.ts                        # Root composition
├── default-state.ts                # Default state + initial state merge
├── slices/                         # Domain slices
│   ├── kanban/                    # boards, tasks, columns
│   ├── session/                   # sessions, messages, turns, worktrees
│   ├── session-runtime/           # shell, processes, git, context
│   ├── workspace/                 # workspaces, repos, branches
│   ├── settings/                  # executors, agents, editors, prompts (incl. userSettings)
│   ├── comments/                  # code review diff comments
│   ├── github/                    # GitHub PRs, reviews
│   ├── gitlab/                    # GitLab MRs, watches, MR automation options
│   └── ui/                        # preview, connection, active state, sidebar views
├── hydration/                     # SSR merge strategies
hooks/domains/{kanban,session,workspace,settings,comments,github,gitlab}/  # Domain-organized hooks
lib/api/domains/                    # API clients
├── kanban-api, session-api, workspace-api, settings-api, process-api
├── plan-api, queue-api, workflow-api, stats-api, github-api
├── user-shell-api, debug-api, secrets-api, sprites-api, vscode-api
├── health-api, utility-api

Key State Paths:

• messages.bySession[sessionId], shell.outputs[taskEnvironmentId], gitStatus.byEnvironmentId[taskEnvironmentId], gitStatus.byEnvironmentRepo[taskEnvironmentId][repositoryName], and sessionCommits.byEnvironmentId[taskEnvironmentId]; code/tests starting from a session ID must resolve environmentIdBySessionId[sessionId] before direct environment-scoped reads/writes, and setGitStatus takes the environment ID.
• tasks.activeTaskId, tasks.activeSessionId, workspaces.activeId
• repositories.byWorkspace, repositoryBranches.byRepository

chatMotion owns per-device chat animation preview and persistence; useChatMotion applies OS reduced motion. Keep it separate from richOutputMotion and transcript auto-scroll. Quick Chat stores server conversations in quickChat.sessions and browser-local terminals in quickChat.terminalTabs; activeKind and terminal IDs track selection. quick-terminal-actions.ts owns lifecycle/fallback; terminal descriptors never enter conversation APIs or get lost in reconciliation.

Hydration: Go injects window.__KANDEV_BOOT_PAYLOAD__ into the SPA shell before React mounts. lib/state/hydration/merge-strategies.ts has deepMerge(), mergeSessionMap(), mergeLoadingState() to avoid overwriting live client state. Pass activeSessionId to protect active sessions.

For rebasing or finishing PRs written against the old Next.js runtime, follow docs/nextjs-spa-migration.md.

Hooks Pattern: Hooks in hooks/domains/ encapsulate WS subscription + store selection. WS client deduplicates subscriptions automatically.

WebSockets

Format: {id, type, action, payload, timestamp}.

Use subscription hooks only; the WS client auto-deduplicates. Task overview vs. session detail: Shared task rows read Task.statusSummary and task.status_summary.updated; rich streams stay session-detail-only. Extend the bounded projection per the spec and ADR. Branch-scoped task state: For live worktree/session state plus task_prs, key by (repository, checked-out branch), not task/repository alone. branch_switched invalidates prior status/commits; reject late results with a generation/identity guard and preserve siblings. Historical PRs affect Changes only when repository_id and normalized head_branch match; Review/PR history may still show them. Test single/multi-repo cases and desktop/mobile Changes behavior. HTTP/WS cache races: When HTTP hydrates a cache also updated by WebSockets, resolve every freshness-sensitive displayed field from one accepted live/HTTP projection, not a mixture of sources; this includes workflow placement and step badge name/color as well as activity/status. Guard responses with per-scope revision and request/workspace generation; discard or refresh stale responses and cover a newer live move plus stale-live rejection. For React StrictMode or multiple hydration effects, keep the mocked response deferred, assert request count while it is pending, then resolve it for cleanup. Action-level pending markers spanning status refreshes clear from the initiating request's settle path, not unrelated status changes; scope them by repository/path for concurrent operations. useEnsureTaskSession re-fires session.ensure when an open task page sees zero sessions after a prior ensure, so deleting the last session from that page spawns a replacement; test zero-session states through the backend/API instead.

• Wire timestamps: Use the shared parseTurnTimestamp parser before formatting or comparing untrusted RFC3339 values; Date.parse accepts or normalizes values such as "0" and February 30. Test syntactically malformed and Date-normalized malformed inputs, then follow the consumer's omit-or-reject contract.

When changing task lifecycle WS handlers (task.updated, task.deleted, task.state_changed), check both kanban and Office surfaces. Archive/delete events may need to update kanban caches, tasks.activeTaskId / session pin state, recent/sidebar prefs, Office refetch triggers such as setOfficeRefetchTrigger("tasks"), and route redirects for /t/:id, /tasks/:id, and /office/tasks/:id. Add focused tests for every affected surface.

Component conventions

• Framework adapters during Next removal: Client components should import links, router hooks, dynamic imports, images, and theme hooks from the local adapter modules (components/routing/*, lib/routing/*, components/theme/app-theme) instead of importing next/* or next-themes directly. The routing/image/dynamic adapters now provide browser-native behavior for the Vite SPA while legacy Next entrypoints are phased out.
• Task links: lib/links.ts::linkToTask is the only /t/:taskId builder; pass raw IDs, use TaskLink or AppLink, and use linkToTask for router pushes. Keep compatibility /tasks/:id, Office/API paths, and route-recognition prefixes separate.
• Components stay under 200 lines; extract domain components. Hooks belong in hooks/domains/ and encapsulate subscription plus selection.
• Code-host dashboards: GitHub, GitLab, and plugin code-host pages must use the provider-neutral primitives in components/integrations/ for change-request lists, rows, toolbars, scope controls, task preset menus, and linked-task indicators. Use the shared semantic IntegrationIcon glyphs instead of copying first-party SVG paths. Keep provider API/state logic in adapters; do not fork row anatomy or add dashboard review/launch flows outside the native task dialog and registered review surface. Plugins may override create transport only through an authenticated action; host verifies repository, session, and worktree branch authority.
• Code-host task status: registered providers publish ReviewItemSummary.taskStatus. Host chrome renders shared topbar, composer, popover/drawer, eager linked-row summaries, hover/focus refresh, and semantic colors. Initial refreshes are leased and deduplicated; do not add provider color fields, visual slots, or pollers; use change-request-* anatomy.
• Code-host review detail: GitHub and compatible review providers render components/integrations/change-request-detail.tsx (also exposed as host.ui.ChangeRequestDetail). Providers own normalized data/capabilities/actions, not parallel headers, review/check/comment sections, scroll containers, or mobile geometry.
• Code-host task links: keep one-field pull/merge-request linking in components/integrations/task-change-request-link-form.tsx. First-party providers compose it directly; plugins call host.openTaskLinkDialog. Link submenu children name the target only (for example, Bitbucket Pull Request) and preserve their registered provider icon.
• Interactivity: all buttons and links with actions must have cursor-pointer class.
• Self-documenting settings: give each setting a short, plain-language description of its effect. Use SettingsInfo for optional scope and implementation details: hover/focus on desktop, a drawer on touch devices. Keep active errors, permissions, managed values, and essential input constraints visible.
• Settings composition: use SettingsGroup for bordered groups and SettingsRow for simple preferences; keep one domain owner/save contributor and attach discovery to actual controls. Keep sections expanded and use existing header tabs for larger pages. Preserve specialized editor/table/diagnostic/credential layouts.
• Settings save coordination: settings surfaces with local unsaved state must register a contributor with useSettingsSaveContributor (or use SettingsPageTemplate) so the shared floating Save changes control, navigation guard, and discard flow own persistence. Do not add page-local Save/Cancel controls. Contributor save callbacks must reject on failure so the coordinator can report an error; discard must restore the contributor's authoritative baseline.
• Settings tabs: use components/settings/settings-tabs.tsx in SettingsPageHeader; preserve drafts with validated URL tab state, map discovery fragments to the owning tab, and use 44px controls on phones and coarse pointers.
• Dialog Enter-to-confirm: the base @kandev/ui DialogContent / AlertDialogContent activate the dialog's semantic action on plain Enter (packages/ui/src/lib/dialog-default-action.ts), so per-dialog "submit on Enter" input handlers are unnecessary — let the base own it. Resolution: AlertDialogAction → an explicit data-dialog-default-action button → the single primary (type="submit" or data-variant="default"|"destructive") button in DialogFooter. More than one primary candidate (counting disabled ones), a disabled resolved action, or one inside a hidden/aria-hidden subtree → no-op (never guesses). Left alone: textarea/contenteditable, Shift/Cmd/Ctrl/Alt+Enter, event.repeat auto-repeat, mid-IME composition (isComposing or keyCode 229), already-preventDefaulted events, and Enter fired from a focused interactive control that owns Enter (any action button — including outline/secondary like Copy/Back — <select>, combobox, or a listbox option / menu item). Only a slot-marked alert-dialog-cancel / dialog-close is treated as a dismiss control and overridden (the Radix-focuses-Cancel case). A plain single-line <input> is not exempt — type-to-confirm dialogs rely on Enter firing the primary. Pass enterConfirms={false} to opt a dialog out; mark the intended button with data-dialog-default-action when a footer has several action buttons.
• Radix tooltip on disabled buttons: disabled buttons do not receive pointer/focus events, so wrap the disabled Button in a focusable span and put TooltipTrigger asChild on that span:

  <Tooltip>
    <TooltipTrigger asChild>
      <span tabIndex={disabled ? 0 : -1} className="inline-flex">
        <Button disabled={disabled}>Run</Button>
      </span>
    </TooltipTrigger>
    <TooltipContent>{disabledReason}</TooltipContent>
  </Tooltip>
  

  Keep the wrapper focusable only while disabled; when enabled, the button itself owns focus.
• Interactive help inside Radix tooltips: do not nest a Tooltip root inside another TooltipContent when the inner trigger must remain interactive. Tooltip roots under one provider coordinate open state, so the inner tooltip can close and unmount its parent before the pointer reaches it. Render the secondary help inline in the existing content or use a disclosure primitive with independent open state. Touch-pinned help must close on a second trigger tap, outside interaction, and Escape; verify desktop pointer and mobile-sized touch flows.
• Renaming a data-testid: use data-legacy-testid for the old id while migrating specs; JSX and Playwright only support one data-testid attribute.
• Dockview session activation: audit pointer/keyboard tabs, shortcuts, reopen/menu actions, and close controls; combine store state with api.isActive, clear same-session intent, and treat default-tab close as delete rather than session switching.
• Conditional review panels: show pr-detail only for active tasks with a linked PR/MR; default layouts only provide preferred placement. Hydrated review loss removes canonical panels, while restoration/maximized and offered/dismissed markers suppress insertion; existing panels sync identity without moving.
• Dockview environment switching: reconcile ephemeral panels before restoring views; correlate ID-less groups by stable ID or position. Setup can precede asynchronous layout restoration. Reconcile dependent state after restoration settles. Dispose added subscriptions. Test late restoration and mixed valid or stale candidates before ambiguity checks. Treat chat/session:* as semantic only with a non-null activeSessionId.
• GitHub PR status UI: use the shared pr-task-icon.tsx display helpers and isPRReadyToMerge; aggregate counts are display-only and cannot enable merges. Update pr-task-icon.test.ts and pr-status-chip.test.tsx with behavior changes.
• GitHub PR associations: retain terminal/merged siblings for tabs/unlink; derive openPRs only for aggregate status/automation and test desktop/mobile terminal unlink plus two-to-one collapse/focus.
• Task repository labels and projections: user-facing task/card repo chips should display a stable repo slug or name (owner/repo when known, otherwise the repo name), not a local filesystem path; local paths belong in metadata. Tasks with no repository should not render a repo chip. When a UI projection adds normalized collection data for grouping, preserve scalar filter fields such as repositoryPath and test both grouping and filtered views.

Internationalization (i18n)

Externalization is complete. A hardcoded user-facing literal is a regression, not leftover migration work. New copy goes through t() / <Trans> wherever you write it.

Add English keys to src/locales/en/<namespace>.json; use useTranslation() in components and module-level t only inside plain helper calls. <Trans> is only for markup, and a t() child corrupts its tag indices. Do not translate domain data, identifiers, test IDs, discriminants, comparison tokens, or map keys; split display copy from logic first. Keep lib/i18n/provider.tsx module initialization; removing it blanks the app. Use _one/_other plural keys, never English suffixes. Never capture t() in a module-level constant; it freezes the boot locale. No Unicode em dash (U+2014) in copy or locale values.

pnpm lint fails on hardcoded UI strings: i18next/no-literal-string is an error on every .ts/.tsx file (tests, *.test-helpers.*, *.test-utils.* and e2e/** excluded). It was scoped to i18nGuardFiles during the migration; measured at zero violations across all 2560 source files, it was widened. i18nGuardFiles remains the migration record and the lint:i18n <path> preview scope: append when you externalize a path, never delete an entry (check-guard-allowlist.mjs rejects that). i18n:ratchet guards new/changed lines independently. The rule sees only JSX literals; SCREAMING_CASE tables, plain .ts helpers, parameter defaults and toast/setter arguments are gated by scripts/check-nonjsx-copy.mjs, which scans the whole tree by exclusion. Silence a legitimate one with // i18n-exempt: <reason> (required) as a // LINE comment — the detector's pattern is line-anchored, so a marker inside a /** */ block is silently ignored.

Real-locale catalogs gate. pt-pt, zh-cn, zh-hk, zh-tw, ja are complete; check-i18n-keys.mjs fails on a missing/extra key, a dropped {{placeholder}} or <n> tag, an empty value, or a value identical to English. Untranslatable values are handled in two tiers: those looksLikeCopy rejects as non-copy need no declaration; prose that reads the same in the target language goes in src/locales/<locale>/_verbatim.json (or the shared _verbatim.json) with a mandatory reason — reasonless or stale entries are errors. For zh-tw/zh-hk run pnpm run i18n:zh-hant, do not hand-translate.

i18n:check also gates key/catalog drift, <Trans> indices, inline plurals, module-scope t(), em dashes, and the pseudo-locale check. Needs Node 24. Guide: docs/i18n.md; spec docs/specs/platform/requirements/i18n.md.

Markdown safety

Any renderer that enables embedded raw HTML must pair rehype-raw immediately with rehype-sanitize, and enable that combination only on the intended surface. Do not broaden raw-HTML support to chat, comments, or other renderers without a separate security decision. Add regression coverage for permitted README markup and for stripping executable HTML and unsafe URLs.

Code-quality limits

Enforced by apps/web/eslint.config.mjs (warnings, will become errors):

• Files: ≤600 lines · Functions: ≤100 lines · Absolute-positioned virtualizer rows do not collapse CSS margins; do not layer row padding/gaps over an item's existing margin. Preserve the old rendered gap with a focused bounding-box regression.
• Cyclomatic complexity: ≤15 · Cognitive complexity: ≤20
• Nesting depth: ≤4 · Parameters: ≤5
• No duplicated strings (≥4 occurrences) · No identical functions · No unused imports
• No nested ternaries

When you hit a limit, extract a helper function, custom hook, or sub-component. Prefer composition over growing a single function.

Plugin system

The public frontend contract is apps/packages/plugin-sdk; docs/plans/plugins/PLUGIN-API.md and lib/plugins/types.ts are its detailed host implementation — all three must change together. lib/plugins/registry.ts is the reactive singleton PluginRegistry; every register* call needs matching cleanup in unregisterPlugin and totalCount(), or a disabled/uninstalled plugin leaks a stale registration.

• Task panels (registerTaskPanel): one generic dockview component, "plugin-panel", shared by every plugin — identity lives in params: { pluginId, panelKey } (id helpers in lib/state/layout-manager/plugin-panels.ts). renderPanel in dockview-shared.tsx and dockview-panel-content.tsx each get exactly one "plugin-panel" case (lookup tables, not switches). PluginTaskPanel (components/task/) resolves the registration behind a PluginErrorBoundary; mobileEnabled: true also renders it via the phone bottom nav (session-mobile-bottom-nav.tsx) with presentation: "mobile".
• Task contributions: registerTaskMenuAction({ group: "edit", ... }) adds card-only actions to the Edit submenu. Group "primary" adds top-level actions to card and desktop/mobile task-row menus; declaring items(context) renders an action of either group as a submenu of its children instead, with run kept as the flat fallback. Card indicator/tag slots stay card-specific; task-row-metadata is generic for sidebar and /tasks rows.
• Sidebar workspace actions: registerComponent("sidebar-workspace-actions", ...) renders after Quick Terminal/Quick Chat in the desktop sidebar's New Task row and in the shared phone navigation sheet, forwarding SidebarWorkspaceActionsSlotProps with presentation: "desktop" | "mobile"; mobile plugin controls own a 44px touch target and accessible name.
• host.storage: authenticated per-user key/value storage (lib/plugins/host-api.ts) backed by /api/plugins/{id}/user-state/... (docs/decisions/2026-08-01-per-user-plugin-storage.md); subscribe (lib/plugins/user-state-sync.ts) wraps registerWsHandler with own-plugin filtering and own-tab echo suppression via a per-tab writerId.
• host.ui.RichTextEditor/RichTextReadOnly (components/editors/tiptap/rich-text-editor.tsx): narrow Plan-panel-tiptap wrappers; update PLUGIN-API.md before widening props beyond { taskId, value, onChange, placeholder, className, testId } / { value, className, testId }.

Sidebar task views

sidebarViewsByWorkspace stores personal view state by workspace ID. Use selectSidebarViews, preserve workspace identity through async saves and rollback, and keep sidebarViews only for legacy wire/hydration compatibility. The backend owns migration/defaults; writes use scoped sidebar_view_state, never legacy global fields.

Testing notes

• vitest.config.ts pins process.env.NODE_ENV = "test", and that line is load-bearing. React exports act() only from its development build, so under NODE_ENV=production — which the runtime image sets (Dockerfile) and every container/agent shell inherits, while CI's image does not — @testing-library/react falls back to react-dom/test-utils and every render()/renderHook() throws TypeError: React.act is not a function before any assertion, with CI still green. vitest-environment.test.tsx and the vitest.setup.ts preflight fail by name if the pin goes, and the Run tests step in .github/workflows/frontend-tests.yml exports NODE_ENV=production on purpose so those guards fire in CI too rather than only in a container.
• E2E waits name the cause, they do not budget for the effect. The suite's dominant flake is await expect(x).toBeEnabled({ timeout: 30_000 }) — an assertion on a UI shadow of a backend event whose hand-picked budget expires under load. e2e/helpers/causal-waits.ts has one primitive per transport (waitForHttp, watchWs().waitForEvent, watchWs().waitForResponse): arm it before the action, await it after, then assert the UI with its default timeout. "The backend reached state X" needs no primitive — expect.poll against e2e/helpers/api-client.ts already reads the backend, not the DOM. watchWs(page) only sees sockets opened after it is called, so it goes before the first page.goto(). Confirm a causal chain by probing a live run with a throwaway page.on("response") logger; reading the components predicts it wrong often enough to matter. The only sanctioned wall-clock wait is dwell(page, ms, category, reason), or dwell(ms, category, reason) where no Page exists (fixtures, api-client retries) — category is the closed DwellCategory union (negative-assertion is permanent, unverified is debt to drive to zero), reason says why no event exists, no options; a delay inside a page.route() handler is injectLatency(ms, reason) instead, since the delay is the stimulus rather than a wait; raw page.waitForTimeout and promise sleeps are not sanctioned. Guide and worked examples: e2e/README.md. The sleep ban is enforced in two layers, the same shape as the i18n guards, and the conversion is complete so both cover the whole tree. eslint-rules/no-unsanctioned-sleep.mjs is an AST rule (e2e/ has ~700 test.setTimeout() calls, Playwright's per-test timeout setter, which a regex cannot separate from a sleep — nor a sleep from a Promise.race guard); it is an error across all of e2e/ via e2eSleepGuardFiles, and it also rejects a dwell/injectLatency not imported from e2e/helpers/causal-waits, since a failed import otherwise lints clean and throws on exactly the loaded shard the wait existed to survive. pnpm run e2e:sleep-ratchet (CI + pre-commit) is the second layer and judges the change, not the file. The guard only ever widens — never narrow it to make a build pass; e2e-sleep-wiring.test.ts asserts coverage via ESLint's own config resolution and fails if the CI step or hook disappears. Details: e2e/README.md ("How this is enforced").
• jsdom secure cookies need cookie-setter interception; Radix Tooltip tests use keyboard focus, while Playwright covers pointer hover with locator.hover(). Scope terminal selectors to the active panel/container; mobile and dockview may mount multiple instances, so shared helpers must not use global selectors. Local full Vitest suites use the configured 20-percent worker budget; do not pass all-worker overrides or overlap suites.