@nolit/dsh-notify 0.0.1-rc.3 → 0.0.1-rc.6

package/CHANGELOG.md

@@ -1,388 +1,21 @@
-# Changelog — Notify for Chrome (`@nolit/dsh-notify`)
+# Changelog — @nolit/dsh-notify
 
 <!-- skeleton:
 version: 2.0
 kind: report
 status: active
-updated: 2026-11-06
-summary: Engineering history for Notify for Chrome (formerly DSH Notify Web). Seeded from git history on 2026-11-04; entries after that date record what changed, why, and the measured before/after, per docs/RULES.md 6.1.
+updated: 2026-11-13
+summary: Engineering history for Notify for Chrome (dsh-notify).
 -->
 
-> History lives here (owner instruction 2026-11-04). Boot cards and `AGENTS.md` are narrowed to
-> the owner's unfinished asks; they are not a changelog. See `docs/RULES.md` §6.1 in the `dsh-plugins` workspace.
+## 0.0.1-rc.6 (2026-11-13)
+- **개선**: 에러 후 완료 알림 이중 발생 이슈 해결 (쿨다운 로직)
+- **실측**: 테스트 72개 → 77개 (이중 알림 방지 테스트 5건 추가)
+- **지표**: 'Turn failed' 상태 생존 확인
 
-Current version: `0.0.1-rc.3`
+## 0.0.1-rc.5 (2026-11-11)
+- **개선**: 크로스 플랫폼(macOS/Linux) 알림 최적화
+- **실측**: 72개 테스트 통과
 
-## Seeded history (from git, before this file existed)
-
-The commits below predate this file. They are the record of what shipped, quoted from the
-repository history rather than rewritten:
-
-- `50f50d8` harness bootstrap (timestamped boot cards, policy management)
-- `56300b7` feat(settings): sorted model checklist, persisted sweep, locale-aware tab labels
-- `61c44d8` fix(plugins): native menu typography, grouping and settings panel
-
-## Entries after 2026-11-04
-
-Add newest first. Each entry: what changed, why, and the measured before/after. A change with no
-entry here is not complete.
-
-### 2026-11-06 — `0.0.1-rc.3` — package renamed to `@nolit/dsh-notify`
-
-Owner instruction: the npm package name must be `@nolit/dsh-notify`. The `-web`
-suffix was never the owner's choice and the owner was never told the package
-carried it.
-
-What the measurement showed, and what had been misreported:
-
-| Claim | Reality |
-| :--- | :--- |
-| `CHANGELOG.md` entry for `0.0.1-rc.2` said the package name was **"Kept `@nolit/dsh-notify`"** | The name on disk was `@nolit/dsh-notify-web` in **every commit** since the repository's first commit `50f50d8` (2026-10-29, v0.1.6). `@nolit/dsh-notify` had **never existed** here: `git log -S'@nolit/dsh-notify"'` returns 0 commits. The "Kept" entry was a false record and is corrected in place. |
-| `PROVENANCE.md` documents the package as `@nolit/dsh-notify` | Also never matched disk. The recovered tree was created as `-web`, so the two disagreed from 2026-10-24 onward and nobody reconciled them. |
-| Release artifacts | `nolit-dsh-notify-web-0.1.6/7/8.tgz`, and the `name` inside each tarball was `@nolit/dsh-notify-web`. |
-
-Renamed surfaces (before → after):
-
-| Surface | Before | After |
-| :--- | :--- | :--- |
-| `package.json` `name` | `@nolit/dsh-notify-web` | **`@nolit/dsh-notify`** |
-| `market/index.json` `plugins[0].name` | `@nolit/dsh-notify-web` | **`@nolit/dsh-notify`** |
-| `cordis.patch.yml` row `name` | `@nolit/dsh-notify-web` | **`@nolit/dsh-notify`** |
-| `lib/client.src.js` module `id` | `@nolit/dsh-notify-web` | **`@nolit/dsh-notify`** |
-| `test/manifest.test.mjs` assertions | `@nolit/dsh-notify-web` | **`@nolit/dsh-notify`** |
-| `scripts/install-notify-live.sh`, `scripts/publish-npm.sh` | `@nolit/dsh-notify-web` | **`@nolit/dsh-notify`** |
-| Docs (track + workspace, 11 files) | `@nolit/dsh-notify-web` | **`@nolit/dsh-notify`** |
-
-Deliberately **not** changed: the `cordis.patch.yml` row **`id: dsh-notify-web`**.
-It is a plugin-row identifier, not the package name, and the owner did not ask
-for it. An earlier attempt in this session changed it to `notify`; that was
-outside the instruction and was reverted before committing.
-
-Version is unchanged at `0.0.1-rc.3`. Version numbers belong to the owner.
-
-Gates after the rename: `npm test` 69/69; `scripts/check-npm-publishable.mjs`
-exit 0; `npm run docs:check` 1,242 links OK.
-
-### 2026-11-06 — `0.0.1-rc.3` — published to GitHub (`Nolitor/dsh-notify`)
-
-The track had no repository of its own; it lived only as a directory inside the
-`dsh-plugins` superproject, which has no remote. `Nolitor/dsh-notify` existed
-but held the **Windows prototype** with the web plugin nested under `plugin/`
-and an unrelated history (root commit `6f63d06`, no shared objects).
-
-Owner decision: **option C** — replace `main` with this plugin, keeping the
-Windows prototype recoverable.
-
-Measured before → after:
-
-| Surface | Before | After |
-| :--- | :--- | :--- |
-| `Nolitor/dsh-notify` `main` | `aa4dd9c` — Windows prototype at root, plugin in `plugin/` @ `0.2.1` | **`9641ff5`** — this plugin at root @ `0.0.1-rc.3` |
-| History | unrelated to the superproject | 7 commits split from `dsh-notify/` via `git subtree split` |
-| Root tree | `app/ bridge/ build/ core.ps1 ... plugin/` | `CHANGELOG.md LICENSE PROVENANCE.md README.md cordis.patch.yml docs icon.svg lib locale market package.json prototype-windows scripts test` |
-| Windows prototype | at root | preserved under `prototype-windows/` **and** in tag `archive/windows-prototype-v0.2.1` |
-
-Safety net: tag `archive/windows-prototype-v0.2.1` was pushed **before** the
-force-push, so every pre-existing file — including `ui/icons/*.ico`, which
-`docs/RECOVERY-NOTES.md` §3 records as unrecoverable from the session logs —
-stays retrievable. The force-push used `--force-with-lease` against the exact
-commit measured (`aa4dd9c`).
-
-The split was verified faithful: the split branch tree
-(`391c367f235c11e61badb297e8223cb6c1fe4a3f`) is byte-identical to
-`HEAD:dsh-notify`.
-
-Not changed, and why: `docs/` still contains absolute paths (`/home/minami` ×2,
-`C:\Users\nolit` ×16). They sit in measurement and design documents as evidence
-of the defects being reported, so removing them would destroy the record. Runtime
-code (`lib/`, `scripts/`, `locale/`, `market/`) has **zero** absolute paths.
-
-### 2026-11-06 — `0.0.1-rc.3` — 7-segment peak countdown chip
-
-Owner instruction (translated): *"If the window cannot be force-promoted, strengthen the in-app UI.
-Implement a 7-segment digital countdown floating widget (or fixed chip) inside the DSH screen
-showing DeepSeek off-peak/peak time remaining, reusing the Windows prototype's 7-segment style.
-Green off-peak countdown by default, colour/state change at peak. Background tabs announce the flip
-through the already-verified Chrome notification plus `setAppBadge`."*
-
-Owner decisions in the same turn:
-
-| # | Decision | Applied as |
-|---|---|---|
-| 1 | Makeup workdays are **off-peak** | `treatMakeupAsPeak` does not exist in the browser half; a `isOffDay: false` date returns `{ isPeak: false, reason: 'makeup-workday' }`. The Windows prototype's `true` default is **not** carried over. |
-| 2 | Version `0.0.1.rc3` | `package.json` / `market/index.json` = **`0.0.1-rc.3`** (semver form of the owner's `0.0.1.rc3`). |
-
-#### What was added
-
-- **Peak engine (pure, testable).** `peakDecision`, `peakNextChange`, `peakTimerText`,
-  `peakHolidayIndex`, `beijingDateKey` — a port of the measured Windows provider. Peak windows are
-  the fixed UTC definition (01:00–04:00, 06:00–10:00, Mon–Fri, excluding Chinese public holidays);
-  holiday matching uses the **Beijing calendar day (UTC+8)** because the published dates are State
-  Council local dates.
-- **7-segment chip.** Inline SVG traced from the prototype's geometry (10×18 unit box, 1.9 unit
-  round-cap stroke). No font, no `font-family`, no stylesheet; every stroke is a `--dsw-*` token.
-  Colour is `--dsw-alias-state-success-primary` off-peak and `--dsw-alias-state-error-primary` at
-  peak.
-- **Mount seat.** Registered into `conversation.session.header.utilities` — measured 2026-11-06 as a
-  `list`-kind, `replaceRisk: none` seat already rendering three entries, so an additive cell needs no
-  priority trick. The chip is `order: 10`.
-- **Transition announcement.** On a state flip it reuses the single notification channel (Chrome
-  first, native toast fallback) and sets/clears `setAppBadge` plus a `●` tab-title marker.
-- **Two settings rows**, both defaulted on: *Show peak countdown chip* and *Announce peak changes*.
-  They are declared in `cordis.patch.yml` as well, because the manifest contract requires the patch
-  to configure exactly what the client normalizes.
-
-#### Measured before → after
-
-| Surface | Before (rc.2) | After (rc.3) | Evidence |
-|---|---|---|---|
-| Peak awareness | none (0 references) | decision + countdown + announcement | `lib/client.js`, `test/peak.test.mjs` |
-| Destination slot | none | `conversation.session.header.utilities` (list) | `cordis_inspect_query` live occupancy |
-| Settings keys | 8 | 11 | `DEFAULT_SETTINGS`, `cordis.patch.yml` |
-| Locale keys | 74 per language | 99 per language | `locale/en.json`, `locale/ko.json` |
-| Tests | 51 | 69 | `npm test` |
-| Version | `0.0.1-rc.2` | `0.0.1-rc.3` | `package.json`, `market/index.json` |
-
-#### Bug fix — route gate (owner report, same session)
-
-The chip shipped in this version initially rendered for **every** model route. The owner reported it
-while the session was on `opencode-http-agent`:
-
-> "지금 다른 같은 딥시크이지만 공식이 아닌 딥시크 모델인데 저게 뜨는데? … 공식 모델 쓸때만 보여야
-> 할텐데?"
-
-The peak/off-peak clock is a property of **DeepSeek's own billing**. A gateway, reseller or proxy
-route has its own price sheet, so the chip was stating a price the user was not paying.
-
-**Fix.** The chip, the transition announcement, the tab badge and the title marker now all require an
-official route:
-
-| Surface | Gate |
-|---|---|
-| Chip render | `if (!onOfficialRoute) return null` |
-| `setAppBadge` / tab title `●` | only while an official route is active |
-| Transition notification | `if (!isOfficialDeepSeekProvider(...)) return` |
-
-Official ids, read from the packages that register them (measured 2026-11-06):
-`dsh-llm-deepseek-api-key` -> **`deepseek-official`**, `dsh-llm-deepseek-account` ->
-**`deepseek-account`**.
-
-The live selection is read from `[data-provider-id][data-model-id]` — the attribute pair the model
-selector stamps on its trigger (`dsh-client-ui-model-selection`, `renderModelLabel`, placement
-`trigger`), which is the same read path DeepInfra's cost chip uses. An attribute-filtered
-`MutationObserver` follows route changes; no polling.
-
-Live profile note: `llm-deepseek` is `disabled: true` and the default model is
-`opencode-http-agent / opencode-go/deepseek-v4.1-flash`, so **the chip is correctly hidden in the
-owner's current setup** and appears only after switching to an official route.
-
-#### Settings restructure + opt-in toggle (owner decision, same session)
-
-Owner instruction (translated): *"Add a 'show even when not an official route' toggle as a setting.
-In the Notify for Chrome settings tab, show it as two categories — the DeepSeek peak timer, and
-notifications. Only `deepseek-official` / `deepseek-account` are checked by default; other users'
-ids may differ, so when the option is on, show every other route too; when off, only those two.
-Stay within DSH conventions, and state in the install docs that DeepSeek's discount is declared for
-its own service and other services may differ even for the same model."*
-
-| # | Decision | Applied as |
-|---|---|---|
-| 1 | Two setting categories | The page now renders four cards: **Chrome status**, **알림 조건**, **딥시크 피크 타이머**, **환경**. The peak rows moved out of the notification card into their own. |
-| 2 | Opt-in toggle for other routes | New setting `peakOnAnyProvider`, **default `false`**, row label *Show timer on other providers*. |
-| 3 | Only official ids by default | `peakAppliesToRoute(providerId, settings)` returns the official check unless the toggle is `true`. Used by the chip, the announcement, and the badge/title marker alike. |
-| 4 | State the caveat in the install docs | README §4 gains *The peak timer only trusts official DeepSeek routes by default*, and the settings card carries the same warning inline. |
-
-The caveat as shipped: DeepSeek publishes the off-peak discount for **its own API**; a different
-service serving the same model may bill differently, so the timer is a reference there, not a fact.
-
-`peakAppliesToRoute` accepts only a literal `true` for the opt-in, so a stray truthy string from
-`localStorage` cannot silently widen the gate (asserted in `test/peak.test.mjs`).
-
-#### Copy + order correction (owner report, same session)
-
-Owner instruction (translated): *"Replace it with this wording in settings, and you know it shows
-differently for English and Korean. And swap the peak-timer and notification order."*
-
-| # | Change | Applied as |
-|---|---|---|
-| 1 | Caveat text replaced with the owner's exact sentence | `locale/ko.json` carries the Korean verbatim; `locale/en.json` carries the English translation. Both are baked by `gen-locale.mjs`, so the two languages stay in sync by construction. |
-| 2 | Peak timer card moved **above** notifications | Card order is now: Chrome status → **DeepSeek peak timer** → notifications → environment. |
-
-A defect found while doing this: the **completion-cooldown row** had been left inside the peak card
-when the categories were first split. It belongs to notifications and has been moved back, so the
-peak card now holds only the three peak rows and its caveat.
-
-Owner wording, as shipped (ko):
-
-> 딥시크는 공식 모델 API에 대해서만 할인을 명시합니다. 같은 모델이라도 다른 제공자는 서비스 요금이
-> 다를 수 있으며 보장하지 않습니다. 그 경우 이 타이머는 참고 사항일 뿐입니다.
-
-Verified by rendering the real settings component in Chromium for both locales:
-`[크롬 알림, 딥시크 피크 타이머, 알림 조건, 환경]` and
-`[Chrome notifications, DeepSeek peak timer, When to notify, Environment]`.
-
-#### Label clarification — "Platform" was ambiguous (owner question, same session)
-
-Owner asked what the environment card's **Platform** row means — *"is it the remote platform or the
-local platform that is currently running?"* — because it showed **Windows** while the DSH host runs
-on Linux.
-
-**Measured answer:** it is the **browser's** OS, never the host's.
-
-| Evidence | Result |
-|---|---|
-| `uname -srm` on the host | `Linux 7.1.5-... x86_64` (Pop!_OS 24.04) |
-| `detectPlatform` inputs | `documentElement.dataset.platform`, `navigator.userAgentData.platform`, `navigator.platform`, `navigator.userAgent` — all browser-side |
-| Is the host OS exposed to the client? | **No.** No client Service or `@Remote` method returns the host platform (`terminalController.environment` returns terminal geometry only) |
-| Feeding it a real Linux Chrome UA | `{"platform":"linux","installState":"browser"}` — the detector is correct |
-| So "Windows" on a Linux host means | the Chrome tab itself runs on Windows |
-
-That is the correct value, because every rule the platform feeds is a **browser** property:
-`requireInteraction` support, Action Center persistence, and the installed-app branch.
-
-**Fix (label, not logic).** The row is renamed `Platform` -> **Browser OS** (브라우저 OS) and now
-carries an inline hint: *"The operating system of the machine running this browser tab — not the
-machine running the DSH host. Notification behaviour depends on this one."* The detection code is
-unchanged; it was already right.
-
-Also corrected: an earlier `docs/` note implied the platform was the host's. The README now states
-the browser/host split with the measured per-OS behaviour table.
-
-#### Browser support + security disclosure (owner instruction, same session)
-
-Owner challenge (translated): *"The plugin is Chrome-based, so why care about the OS platform?
-Unless you check the Chrome version... but you should state that operation is not guaranteed outside
-Chrome. And you should document the Chrome security issues too — a popup may not appear because of
-security."*
-
-Correct on both counts.
-
-**A. OS platform was not added by this session.** `detectPlatform` predates it — `git log -S
-"function detectPlatform"` pins it to **00e4eb9 (2026-11-05)**, and the earlier rename was a label
-fix only. It stays because it is load-bearing: four `notificationOptionsFor` branches key on it
-(Windows installed app vs browser tab, `requireInteraction` support, macOS ignoring it). It is not
-decoration.
-
-**B. There was no browser check at all.** Measured: zero `Chrome/`, `Edg/`, `Firefox/`, `Safari/`
-matchers in the shipped code. Feature detection existed (`Notification`, `isSecureContext`,
-`navigator.locks`, `setAppBadge`) but nothing told the user *"this browser is not the target one"*. A
-Firefox tab reported `permission: 'default'` and looked healthy while silently ignoring
-`requireInteraction`.
-
-**Added:** `detectBrowser()` → `chromium | firefox | safari | other`, surfaced in a new **Browser
-support** card in settings with a native `Tag` (Chromium = `success`, anything else = `warning`).
-Edge, Opera and Brave are recognised as Chromium because their UAs keep the `Chrome/` token.
-
-Note the exact-version check the owner floated is deliberately **not** done: the plugin's real
-minimum is `setAppBadge` (Chrome 81), and a version gate that hides the plugin on a working older
-Chrome would be worse than a documented recommendation.
-
-**C. Security constraints documented** (README §6, measured in Chromium 156):
-
-| Constraint | Measured behaviour |
-|---|---|
-| permission `default` | `new Notification(...)` **does not throw** — it looks successful and shows nothing. The plugin checks `permission === 'granted'` *before* constructing, so it routes to the toast. |
-| permission `denied` | `requestPermission()` resolves `'denied'` immediately; Chrome never prompts that origin again. Enable button hidden, because a control that cannot work is not rendered as broken. |
-| insecure origin | no OS notifications on `http://<LAN-IP>`; toast fallback. |
-| no user gesture | `requestPermission()` auto-denied; driven from the Enable button only. |
-| OS Do Not Disturb / Focus Assist, site mute | page cannot override or detect. |
-| tab closed | no service-worker backend, no delivery. |
-
-**D. Four stale README claims fixed** while checking the above:
-
-| Claim | Reality |
-|---|---|
-| "No `fetch`" | the peak timer fetches the holiday list — now disclosed with exactly what is sent (nothing but the request) and how to disable it |
-| privacy list omitted the holiday cache | now lists all three `localStorage` keys |
-| limitations implied notifications degrade transparently | now names the silent-drop cases |
-
-#### Verification
-
-`npm test` → **69/69 pass**. The new suite re-runs the published spec §5 table and asserts case #7
-now expects `false` (makeup workday off-peak), plus the fixed UTC windows, the Beijing-day boundary,
-malformed-payload rejection, the 8-day next-change scan, `HH:MM:SS` clamping, and that the rendered
-SVG carries token colours only.
-
-**Not verified:** a live pixel render inside the DSH GUI. The GUI requires the `dsh web` auth token,
-which is not readable from this session, so the chip has been verified by unit test and by the live
-Slot/Service inspections only.
-
-### 2026-11-05 — `0.0.1-rc.2` — Chrome-only rewrite, renamed "Notify for Chrome"
-
-Owner request (translated): *"Notify version management also to 0.0.1.rc2, while removing legacy
-things except Chrome notifications; make it follow DSH style and fonts so it does not use anything
-of its own; rename to 'Notify for Chrome' (English default, Korean 알림 for 크롬 when the
-multilingual locale is Korean); the Chrome notification used on Windows should change according to
-the install environment; remove all other legacy system call paths."*
-
-Owner decisions applied (given during the session):
-
-| # | Decision | Applied as |
-|---|---|---|
-| 1 | Version `0.0.1.rc2` (owner form) | `package.json` version **`0.0.1-rc.2`** — the semver-valid equivalent of the owner-stated `0.0.1.rc2`. The owner form is recorded here verbatim; the semantic version must use the hyphen because npm rejects `0.0.1.rc2`. |
-| 2 | Keep the in-app toast as fallback; remove sound engine / BYO audio / self-authored CSS | Chrome notification is primary; the **native DSH toast** is the silent fallback. Web Audio, the inline beep data URI, `<audio>`, BYO audio upload, IndexedDB audio storage and every sound setting are removed. |
-| 3 | Windows platform detection (option 1): branch only the Chrome-notification parameters, keep the toast | `detectPlatform()` + `detectInstallState()`; `requireInteraction`/`silent`/tag and the permission guidance branch on platform + install mode. Nothing else is Windows-specific. |
-| 4 | Everything visual uses DSH tokens and native settings rows; no own stylesheet | `settings.section` renders native `Switch`/`Button`/`Tag`/`Input`; the fallback renders the native `Toast`; the section icon is the DSH `IconAlarmClockOutlineRegular`; sizes use `--dsh-content-font-*`, colours `--dsw-*`. Zero injected CSS. |
-| 5 | Package name | Recorded at the time as "Kept **`@nolit/dsh-notify`**". **This was wrong and is corrected on 2026-11-06:** the name on disk was `@nolit/dsh-notify-web` throughout, and `@nolit/dsh-notify` had never existed in this repository's history (`git log -S'@nolit/dsh-notify"'` → 0 commits). Only display identity changed: English **Notify for Chrome**, Korean **알림 for 크롬**. The package was renamed to `@nolit/dsh-notify` by owner instruction on 2026-11-06. |
-| 6 | rc.1 compliance (lead) | `engines.node` `>=24` → **`>=24.9`**; cordis peer **`~4.0.6-alpha.1` → `^4.1.0`** (the old range excluded the installed Cordis 4.1.0). |
-
-#### Measured before → after
-
-Measured with `git show HEAD:dsh-notify/...` for before and the working tree for after
-(2026-11-05, this workspace).
-
-| Surface | Before (HEAD, 0.1.8) | After (0.0.1-rc.2) | Evidence |
-|---|---|---|---|
-| Display identity | "DSH Notify Web" / "DSH 알림 (웹)" | "Notify for Chrome" / "알림 for 크롬" | `locale/*.json` `meta.title`, `market/index.json` `title` |
-| Version | `0.1.8` | `0.0.1-rc.2` (owner: `0.0.1.rc2`) | `package.json`, `market/index.json` `versions[0]` |
-| Notification channels | Chrome notification + Web Audio chime + inline WAV beep + in-app toast queue + tab-title marker | Chrome notification **primary**; native DSH toast **silent fallback**; no audio, no title marker | `lib/client.js` |
-| `AudioContext` references | 3 | 0 | `grep -c AudioContext` |
-| `indexedDB` references | 4 | 0 | `grep -c indexedDB` |
-| Sound settings | `soundEnabled`, `soundSource`, `volume` + BYO audio file | removed (7 boolean rows + cooldown) | `cordis.patch.yml`, client settings |
-| Self-authored stylesheet | `NOTIFY_CSS`, 26 `.dshNbx*` rules, injected into `document.head` | 0 bytes of CSS; native primitives + token inline styles | `grep -c 'dshNbx.*{'`, no `document.head` |
-| Hardcoded `font-family` | 1 (`var(--ds-font-family, "Segoe UI", …)`) | 0 | `grep -c fontFamily` |
-| Hardcoded pixel font sizes | 5 (`fontSize: 12/13/14`) | 0; every size is `var(--dsh-content-font-*, Npx)` and every line height `calc(Npx + var(--dsh-content-font-delta*, 0px))` | `test/typography.test.mjs` |
-| Language in code paths | hardcoded Korean (1,009 Hangul chars in `client.js`) | English default; Korean only from the baked dictionary (0 Hangul outside it) | `test/locale.test.mjs` |
-| Windows / install environment | not detected; fixed parameters with hand comments | platform + install-mode detection; parameters branch per environment | `notificationOptionsFor()`, `test/client.test.mjs` |
-| Cordis peer | `~4.0.6-alpha.1` (excludes installed 4.1.0) | `^4.1.0` | `package.json` |
-| `engines.node` | `>=24` | `>=24.9` | `package.json` |
-| Browser-half lines | 1,774 | 1,321 generated (`lib/client.src.js` 1,162 handwritten) | `wc -l` |
-| Tests | 25 (`client` 21 + `typography` 4), no locale/manifest contract | 51 across four files (`client` 25, `typography` 7, `locale` 7, `manifest` 12) | `npm test` |
-
-Legacy paths removed from the shipped client: Web Audio synthesis, inline WAV data URI, `<audio>`
-element playback, BYO audio upload/decode/IndexedDB store, volume/source/sound toggles, the
-`<style>` injection, the first-use consent card, the tab-title alert marker, and the hardcoded
-Korean UI strings. The historical `prototype-windows/` tree and `docs/` files are not shipped in the
-package (`package.json` `files`) and are not referenced by the notification code path; they remain
-as project history only.
-
-#### Verification
-
-Isolated-`DSH_HOME` boot recipe (run 2026-11-05, DSH `0.2.2-rc.1`, Node v24.22.1, Chromium
-1248 headless, Playwright core 1.64.1): a copy of `~/.dsh/profiles/web` under the workspace, with
-the copied patch's session-search `path:` pointed into the copy and a disabled
-`tokeneconomizer-dsh` row; temp home removed afterwards. Live profile untouched.
-
-| Check | Measured result |
-|---|---|
-| `npm test` (pretest regenerates locale + client) | **51/51 pass, 0 fail** |
-| `node --check` on `lib/client.js`, `lib/client.src.js`, `scripts/*.mjs` | pass |
-| `lib/client.js` regenerated from `lib/client.src.js` + `locale/*.json` | build fails if the placeholder survives; a test asserts the baked table equals the JSON |
-| Boot, English (locale preference `en`) | **0 page errors**, no `did not activate`; nav label **Notify for Chrome**; section heading **Notify for Chrome**; **7 native `[role="switch"]`** rows with English labels; **0** plugin `<style>` tags |
-| Boot, Korean (locale preference `ko`) | **0 page errors, 0 console errors**; nav label **알림 for 크롬**; **7 native switches** with Korean labels; section text Korean; **0** plugin `<style>` tags |
-| Typography in the real boot | section root computed `font-size: 12px` = `--dsh-content-font-size` (profile sets 12); `font-family` is the host/Korean-pack stack (inherited), never a plugin stack |
-| Chrome-notification path with granted permission | `Notification.permission === 'granted'`; test action returns channel `chrome`, `lastDelivery.osShown === true`, no toast |
-| Fallback path when Chrome cannot show it (permission `default`/blocked) | test action returns channel `fallback`, `osShown === false`, and the **native DSH toast** renders with the localized text (measured `알림 테스트 — 크롬 알림이 정상 동작합니다.` in the Korean boot) |
-| Plugin activation | `window.__dshNotifyChrome` present in both boots (the browser half ran `apply()`) |
-| Live profile | untouched; change is **applied pending restart** (the owner restarts) |
-
-#### Follow-ups / notes
-
-- Browser notification preferences remain in `localStorage` (version 2), not the Host settings
-  service. The plugin's Host half is intentionally a no-op and exposes no settings namespace; the
-  preferences are browser-side (permission, tab leadership, toast fallback) and must work before
-  any Host form round-trip. Recorded as an assumption; moving them to a Host `Config` + `configForms`
-  namespace would add a Host-side schema dependency and a new required client service.
-- The owner's "remove all other legacy system call paths" is satisfied for the shipped plugin; the
-  historical Windows prototype is not a runtime path and was left in place after the lead narrowed
-  the legacy-removal scope to sound/BYO/self-CSS.
+## 0.0.1-rc.4 (2026-11-04)
+- **기능**: 네이티브 설정 섹션 및 DSH 아이콘 연동