miniSCP/docs/gotchas.md

# Gotchas — landmines this codebase has already stepped on

Every one of these caused a real bug or debugging spiral. Read before
touching the adjacent code.

## Angular (v22, zoneless, signals, OnPush)

- **Never consume push streams through signal+effect.** An effect that reads
  a signal and (directly or transitively) writes signals re-triggers itself —
  this froze the entire app on the first transfer. Push messages
  (transferUpdate etc.) go through plain callback registries in RpcService.
  Signal coalescing would also silently drop intermediate progress snapshots.
- **No signal writes during template render** (NG0600). Anything that loads
  panels from code paths reachable during render defers with
  `queueMicrotask` (see `PanelsService.newWorkspace`).
- **`[value]` on `<select>` is applied before `@for` options render** and
  silently selects nothing. Use per-option `[selected]` bindings (connection
  manager stack picker).
- Template bindings with special characters: `[class.bg-primary/15]` works;
  `[class.shadow-[...]]` (brackets) does NOT parse — use `[style.box-shadow]`.
  Static class attributes may use arbitrary values freely.
- Functions called from templates ARE reactive: signals read inside them
  register on the view's consumer (call-stack-based tracking). That's why a
  template calling `i18n.t()` re-renders on a language switch — `t()` reads
  the `locale` signal.
- Dedicated tools > shell: IDE diagnostics between paired template/class
  edits are often stale mid-sequence — trust `bun run build:ui` (full AOT
  type-check), not the first red squiggle.
- Tailwind v4 canonical class names: `wrap-break-word` not `break-words`,
  `accent-primary` not `accent-[var(--primary)]`, `z-60` not `z-[60]`.
  Every color used in a template must resolve through `@theme inline`.
- Stylesheet `url()` must resolve at BUILD time: assets referenced from
  `styles.css` live under `ui/src/` (esbuild emits them hashed into `media/`
  and rewrites the url), NOT `ui/public/` — a public path fails the build with
  "Could not resolve".
- **Two `class` attributes on one element: the second is silently dropped.**
  No warning, no build error — the layout classes simply never apply. This
  collapsed the tab strip and two window headers at once (`fe028a9`), because
  a scripted edit appended `class="…"` to elements that already had one.
  When removing a conditional binding, check what the branch *contained*: the
  bindings there were genuinely dead, but the layout classes inside them were
  not, and nothing replaced them.
- **A component `template:` is a TS template literal, so no backticks inside
  it** — not even in an HTML comment. A backtick in a comment terminates the
  literal and the errors point at the prose, not at the quoting.
- **Reach `core/` through the `@core/*` alias, never a relative path.**
  Components sit at varying depths and get moved; the alias is what let 67
  imports survive two relocations untouched. A `../../core/...` path is a
  future merge conflict.
- **The build DOES type-check components nothing imports.** ngtsc compiles the
  whole program, so an unreferenced component still fails `bun run build:ui` on
  a type or template error (verified by injecting an error into an
  unreferenced file).
- **`overflow-x: auto` also clips vertically.** CSS refuses to pair `visible`
  with a non-visible axis, so the computed `overflow-y` becomes `auto` too:
  anything poking out of a horizontally scrolling strip (a raised tab, an
  underline at `-bottom-px`) is silently clipped. Reserve padding instead.

## Theming (`ui/src/styles.css`)

- **A new surface token needs an AMOLED line, or AMOLED stops being
  all-black.** The two `[data-dark-tone='darker']` blocks re-declare every
  surface (`--chrome`, `--toolbar`, `--background`, `--row-alt`, …) as `#000`;
  a token added only to the dark blocks shows through in AMOLED. Tokens derived
  in `:root` by `color-mix` (`--toolbar-2/-3`) follow on their own.
- **Hover/selected color is a translucent overlay (`--accent`).** An opaque
  hover vanishes on a surface of its own lightness — light's old 0.967 hover
  on the 0.965 header strip did exactly that. Keep `--accent` alpha-based when
  retoning surfaces.
- **A native control ignores `border-radius` and background until
  `appearance: none`.** WebKit draws checkboxes, radios and selects itself;
  the square restyle in the `@layer base` block depends on it. Put `pl-*`,
  never `px-*`, on a `<select>`: utilities outrank the base layer, so a `px`
  utility would erase the padding-right the drawn caret lives in.
- **A custom `::-webkit-scrollbar` turns off macOS overlay scrollbars.** The
  moment the pseudo-element is styled, WKWebView draws classic always-visible
  scrollbars that take layout width. Deliberate here (square design); don't
  reach for it to "tint" scrollbars without accepting that.

## The interface scale (root font-size lever)

- **electrobun's native page zoom is a silent no-op on Linux — do not reach for
  it.** `webviewSetPageZoom` in the shipped `libNativeWrapper.so` (v1.18.1, and
  the `_cef.so`, and the copy in our own build output) is a compiled
  empty-body stub: the disassembly has no `call` instruction, and
  `webkit_web_view_set_zoom_level` is not even in the library's imports.
  `webviewGetPageZoom` returns a hardcoded `1.0` from `.rodata`, so the failure
  is unobservable through the API — set 2.0, read back "1.0 = success". The
  JSDoc's "(WebKit only)" means macOS WKWebView. This is why the scale is CSS.
  (Disassembled on 1.18.1; 2.0.2's zoom fixes are CEF-only — re-check the
  native Linux wrapper before trusting it.)
- **Everything the UI sizes with must be rem.** The scale works by setting the
  root font-size, so a px value freezes while everything around it grows. The
  px reservoirs were the panel-chrome tokens (`--row-h` etc.) and 93
  `text-[11px]` arbitraries. 1px hairlines and focus rects stay px deliberately
  (crisp at any scale).
- **px constants in TS that mirror rem-sized CSS are latent scale bugs.** The
  context menu's `MENU_W = 224` was a hardcoded mirror of `w-56`
  (14rem): box grew, math didn't, flip fired at the wrong edge. Convert through
  `@core/metrics` `remPx()` at use time — it reads the LIVE root font-size,
  and per call, since the scale changes mid-session. The editor's canvas-based
  `charWidth` is cached keyed on the computed font string for the same reason.

## Electrobun

- **The `electrobun` npm package (2.x) is a bootstrap whose every export is a
  throwing stub** ("Electrobun 2.x APIs come from the Hutch devkit, not
  node_modules"). Every `electrobun/*` import must resolve through the
  projected `.hutch/devkit/`: the root tsconfig extends
  `.hutch/devkit/tsconfig.json`, and `ui/tsconfig.json` maps `electrobun/view`
  to `../.hutch/devkit/api/browser/index.ts` (the exact export-map target). Let
  an import fall through to `node_modules/electrobun` and the webview throws at
  load. tsc and the UI build fail on unresolved `electrobun/*` imports until
  `bunx electrobun prepare` has run (postinstall and every dev/build script do
  it). Never point anything into `~/.hutch` — that store is Hutch-owned.
- **`build.mainProcess` defaults to Cottontail in 2.x**, and Cottontail ignores
  the whole `build.bun` block — the ssh2 externals included, silently (verified
  with `bunx electrobun config --env=stable`). `mainProcess: "bun"` in
  `electrobun.config.ts` is load-bearing.
- **The browser SDK is raw TS compiled inside the Angular program**, so the UI's
  strictness flags apply to upstream code: 2.0.2's `createTransport().send`
  returns on one path only, which is why `ui/tsconfig.json` has
  `noImplicitReturns: false`. Its `import "./global.d.ts"` also prints "File
  '…/global.d.ts' not found in TypeScript compilation" on every build — benign,
  and adding the file to the program does not silence it.
- **Window `close` is not cancellable, but 2.x added `will-close`.** The `close`
  event's response type is `{}`; `will-close` (fired by the titlebar/traffic-light
  close and `win.requestClose()`) honors `event.response = {allow:false}`, and
  `win.close()` skips it. The editor close safety predates this: all
  unsaved-state guards still intercept app-controlled close paths (⌘W, buttons)
  or survive the close (editor draft mirroring) — a `will-close` prompt is a
  possible follow-up.
- **`Updater.applyUpdate()` asks the `before-quit` handlers first (2.x)**, so the
  transfers/unsaved guard can veto an update restart. Its "Quit Anyway" must
  re-run `Updater.applyUpdate()` (`events.applyUpdate` flags the quit as
  update-initiated) — a plain `Utils.quit()` there quits without installing.
  `Utils.quit()` runs the same approval, which is why the guard checks
  `quitConfirmed` first.
- **macOS ⌘-keys go to the PAGE first; only unclaimed ones reach the app
  menu.** WKWebView's `performKeyEquivalent` hands key equivalents to the page
  while the webview is first responder (WebKit `WebViewImpl.mm`: "lets
  webpages have a crack at intercepting key-modified keypresses"); a key the
  page `preventDefault`s never reaches the menu. Old comments here claimed the
  opposite ("the menu intercepts ⌘W first") — what was really happening is that
  windows which didn't claim ⌘W let it fall through to the menu's Close Tab.
  Two consequences: a handled shortcut MUST `preventDefault` (`claim()` does),
  or the menu runs it a second time; and a ⌘W a secondary window doesn't claim
  still arrives as the MAIN window's `close-tab` action — bun checks the
  focused Settings window, then editors, first and closes THAT, so a new
  window kind must either claim close-window or join that check.
- **One rpc instance per window** — BrowserView binding steals the rpc's
  transport; a shared instance silently sends to the wrong webview.
- **RPC before socket open was silently lost on 1.18.1.** 2.x Electroview queues
  sends while its `hostSocket` (v1: `bunSocket`) is CONNECTING and falls back to
  the native bridge; RpcService still gates everything on the socket-open
  promise (`ready`) as belt-and-braces.
- The emitter has no `off()`. Window event closures must not capture records
  you want garbage collected.
- `electrobun build` defaults to the **dev** channel; release builds need
  `--env=stable`. The stable app name differs from the dev one
  (`miniSCP` vs `miniSCP-dev`), including the app-data dir implications when
  testing.
- Tray: never call `setImage` after construction on macOS — the template
  treatment is lost and the icon renders black on dark menu bars.
- `ApplicationMenu` is macOS-only in our usage; accelerators on Windows are
  single-character only.
- **`ContextMenu.showContextMenu` is a silent no-op on Linux** (electrobun's own
  `.hutch/devkit/api/sdks/main/proc/linux.md` says so, still in 2.0.2; the app
  also prints "Application menus are not supported on Linux" at boot). Because
  the UI suppresses the webview's own right-click menu, that left EVERY panel
  context menu dead on Linux — no error, nothing happens. The menu is plain
  HTML now (`ui/src/app/skins/shadlike/context-menu.ts`) on every platform;
  don't reintroduce the native one for "nativeness" on mac/win without keeping
  a Linux path.
- **Window-open flash prevention is three-layer** — remove any one and the
  white/inverted blink returns: (1) windows are created `hidden: true` and
  revealed by `showWhenReady` on the webview's first `dom-ready` (800 ms
  fallback), (2) the bun preload bakes the persisted `data-theme`/tone
  attributes and the interface scale before first paint (`buildPreload`),
  (3) `ui/src/index.html` carries an inline `<style>` with the html background
  per theme — its oklch values MUST stay in sync with the `--background` tokens
  in styles.css. (The win7 branch once replaced (3) with a transparent rule and
  the modern skin silently lost it — the other two layers do not cover it.)
- Dev vs prod URL: `getMainViewUrl()` resolves the ng-serve dev server only
  when `DEV_SERVER_URL` is set AND the updater channel is `dev` AND the
  server answers a HEAD request; secondary windows must reuse the resolved
  URL, never hardcode `views://`.
- Upstream Bun panic (`FFI_Callback_threadsafe_call` → `web_worker.spin`):
  rare segfault under long sessions; upstream electrobun×Bun issue, restart
  recovers, nothing app-side to fix. Seen on Bun 1.3.13 (Electrobun 1.18.1);
  2.0.2 ships Bun 1.4.0 — unconfirmed whether that fixed it.
- **Preloads and frames are construction-time only.** The preload is a
  persistent `WebKitUserScript` at DOCUMENT_START, frozen at webview
  construction — `webview.loadURL()` re-runs the OLD one
  (`updatePreloadScriptToWebView` is only reachable from electrobun's internal
  `<electrobun-webview>` handler). `titleBarStyle`/`transparent` have no
  setters. Anything `buildPreload` bakes is a first-paint snapshot the UI must
  correct live over RPC. Recreating the main window is not an escape hatch:
  `exitOnLastWindowClosed`, push closures and the tray capture it, and there
  is no `off()`.
- **`trafficLightOffset` changed meaning in 2.x.** 1.18.1 nudged the buttons
  by the offset but clamped them inside the stock titlebar, so `{12, 13}` really
  meant "as low as the titlebar allows" (left 21 / top 18 on macOS 26). 2.x adds
  the full offset to macOS's default position (9/9 in the 32pt Tahoe titlebar)
  and grows the titlebar to fit — the same numbers dropped the lights 4pt.
  `chromeOptions()` now passes `{12, 4}`: left 21 / top 13, the 14pt buttons
  vertically centred in the 40px `h-10` header (1.18.1's clamped spot, top 18,
  was 5pt low). The default shifts with macOS's titlebar metrics (28pt / 7
  before macOS 26), so re-check after an OS release. `win.setWindowButtonPosition(x, top)` is the absolute
  alternative (re-applied on resize/fullscreen) if per-version drift matters.
- **Drag regions ignore `preventDefault`.** The preload starts a window move
  from a bubble-phase `mousedown` on `document` and never reads
  `defaultPrevented`; a `pointerdown` handler doesn't stop it either (the
  compat `mousedown` still fires). `isAppRegionDragTarget` lets a `no-drag`
  region win — the `.electrobun-webkit-app-region-no-drag` class (or, since
  2.x, `app-region: no-drag` CSS) on the element is the only reliable
  suppressor (`.hutch/devkit/api/preload/dragRegions.ts`; re-checked on 2.0.2).
- **No resize, maximize-event or double-click-maximize primitives (1.18.1, still
  absent from 2.0.2's `BrowserWindow`).**
  `titleBarStyle: "hidden"` drops the whole frame (Cocoa `Titled` — the only
  way to remove the traffic lights; `gtk_window_set_decorated(false)` on
  Linux) and with it the WM resize borders; nothing restores them (no
  `startWindowResize`, no `gtk_window_begin_resize_drag`, no
  `setResizable`/min-size options — only `setFrame`/`getFrame`). No maximize
  event exists (derive it from `resize` + `isMaximized()`), and drag regions
  don't double-click-maximize. A custom-chrome window has to build all three.

## Linux

- **`ERR_DLOPEN_FAILED` on `bun run start`** means a missing system library, not
  broken app code. `libNativeWrapper.so` links `libayatana-appindicator3.so.1`
  *hard* (for the tray), so the app cannot start at all without it — even though
  the tray is a minor feature. Diagnose with
  `ldd build/dev-linux-x64/*/bin/libNativeWrapper.so | grep 'not found'`; the
  `libasar.so => not found` line in that output is a red herring (it ships beside
  the wrapper and resolves at load). Install list per distro is in
  [platform-notes.md](platform-notes.md).
- **Black window on NVIDIA + `Failed to create GBM buffer of size WxH: Invalid
  argument`** (usually with `X11 Error: GLXBadWindow (code 170)`). The app boots,
  logs "miniSCP started", reports no error — and paints nothing. WebKitGTK's
  DMABUF renderer can't allocate buffers on the NVIDIA proprietary driver;
  `WEBKIT_DISABLE_DMABUF_RENDERER=1` is the upstream workaround.
  - **The app cannot apply this to itself.** WebKit reads the variable when it
    forks its web process, and **Bun's `process.env` writes do not propagate to
    child processes** (verified: a spawned child sees it unset). Setting it in
    `src/bun/index.ts` — even in a module imported before `electrobun/main` —
    does nothing. It must be in the environment before the app process starts.
  - Only the dev launch paths inject it: `scripts/linux-gpu.ts` (`launchEnv()`)
    for `bun run dev` / `bun run start`. **A new dev launch path must call
    `launchEnv()` or it will render black on NVIDIA.** The release is the
    official 2.x Setup installer now (the v1 portable tarball and its
    `miniSCP.sh` wrapper are gone), so an installed build on NVIDIA needs the
    user to launch it with `WEBKIT_DISABLE_DMABUF_RENDERER=1` (e.g. in the
    desktop entry's `Exec`).
  - Gated on `/proc/driver/nvidia` so Intel/AMD keep hardware compositing, and
    an explicit `WEBKIT_DISABLE_DMABUF_RENDERER` from the user always wins.
  - A "does it boot?" log check does NOT catch this — grep for the GBM line, or
    look at the window.
- **`bun run start` leaves the old instance running, and
  `import -window miniSCP` grabs the OLDEST match.** Instances pile up across
  rebuilds, so every screenshot silently comes from a stale build and the fix
  you just made looks like it did nothing. Cost three wrong readings in one
  session, including a "the CSS is in the bundle but does not apply" dead end.
  Kill first (`pkill -f dev-linux-x64`), confirm with
  `pgrep -cf 'dev-linux-x64/miniSCP-dev/bin/launcher'`, then launch and capture.
  Note `pkill -f 'electrobun dev'` also matches the shell running it and kills
  your own command (exit 144) — match on the build path instead. That is only
  half the trap: `pkill -f dev-linux-x64` does the same thing, because the
  pattern appears in the killing command's OWN argv. Break the literal with a
  character class — `pkill -f 'dev-linux-x[6]4'` — the way `ps | grep [x]yz`
  has always worked.
- **Launching the app steals focus — ask first.** `bun run start` opens a real
  window that takes focus, which interrupts whatever the user is doing; there
  is no Xvfb/xvfb-run on this machine and Wayland offers no
  launch-without-focus. Batch every live check into a single launch.
- Benign startup noise on Linux — non-fatal, the app runs normally:
  `libayatana-appindicator is deprecated. Please use libayatana-appindicator-glib`
  (upstream electrobun's choice, not ours).

## Keyboard (`src/shared/keymap.ts`, `ui/src/app/core/shortcuts.service.ts`)

- **WKWebView skips buttons, checkboxes and radios on Tab** unless the user
  turned on System Settings → Keyboard → Keyboard navigation (or presses ⌥Tab).
  Dialogs, Settings and the manager were mouse-only on a default Mac. Tab is
  walked by hand: `cycleFocus(root, event)` (core/focus-cycle.ts) — use it in
  any new dialog or window.
- **WebKit doesn't focus a clicked button — focus drops to `<body>`** (Windows
  parks it on the button, where Space clicks it again). Every list key died
  until the list was clicked. `App` `(mousedown)` keeps focus on chrome-button
  clicks and forwards `<body>` keys to the active list; keep both.
- **Match modifiers exactly.** The old panel switch caught every Tab, so
  Ctrl+Tab flipped the active panel AND switched tabs; `isMod = metaKey ||
  ctrlKey` made ⌃F (Cocoa's forward-one-character) open the editor's find.
  `matchesEvent` compares all four modifiers, platform-exact.
- **AltGr is Ctrl+Alt on Windows/Linux.** A Ctrl+Alt+letter binding steals
  "@", "€", "ł"… — never a default there, never fired from a text field, and
  an event with `getModifierState('AltGraph')` matches nothing.
- **Type-to-filter takes every printable key.** A plain-letter binding (vim
  j/k) must resolve BEFORE it — and then that letter no longer starts a query
  (Settings warns). While a query is being typed, printable keys (Space too)
  always extend it.
- **A dialog's Esc must `stopPropagation`.** The connection manager closes on
  a window-level Escape; the dialog cleared its own signal first, so the
  window handler saw "no dialog open" and closed the whole window.
- **An unclaimed key behind a dialog reaches the macOS menu** — ⌘T opened a
  tab behind the rename prompt. The main window swallows window commands and
  the menu bar's panel commands while a dialog or menu is up.
- **WebView2 reloads the page on unclaimed F5 / Ctrl+R** (default browser
  accelerators), dropping every session and the editor's unsaved text. Every
  window root ends with `blockReloadKeys`; a blanket `stopPropagation` (the old
  path bar) hides keys from it — don't add one.
- **Hold-to-repeat** fired ⌘W once per repeat — every tab closed. `claim()`
  repeats only movement-like commands (`isRepeatable`).
- **Electrobun's native menu accelerator turns a combo with no modifier token
  into ⌘+key** ("f7" shows and fires ⌘F7), its `backspace` / `delete` tokens
  are characters AppKit matches unreliably, and AppKit matches Shift+punctuation
  by the shifted character ("}" for ⇧]). `toAccelerator` returns null for all of
  those; the menu item then shows no key and the page still handles it.
- **Recording a key must take the keydown before every window handler** — and
  must NOT take reserved ones: preventDefault on ⌘Q while recording would make
  the app unquittable from the keyboard.
- **Focus after a state change needs `afterNextRender({ injector })`.** Under
  zoneless CD `queueMicrotask` runs before the render that creates an `@if`
  element (the editor's find box and textarea), so `.focus()` hit nothing.

## Domain logic

- **sessionId ≠ filesystem identity.** Two panels on the same server are two
  SSH sessions. Every same-file/nested-path guard compares `fsKey`
  (`resolvedIP:port` / `"local"`); comparing sessionId caused a
  truncate-to-zero data-loss bug. When fsKeys differ but paths collide, the
  probe-file check unmasks aliased hosts. Never "simplify" this back.
- `createWriteStream` truncates BEFORE the source is read — that's why
  self-copy is destructive, and why guards must run before any stream opens.
- The synthetic `'..'` row: type `'dir'`, must be excluded from selection,
  targets(), filtering match-count, kind/rights/owner display, and drag
  sources. Any new row-level feature must decide what `'..'` does.
- `Semaphore.wake()` must reserve slots synchronously; waking waiters that
  increment asynchronously loses budget under concurrent release.
- Settings: the backend spread-merges (`{...DEFAULT_SETTINGS, ...data}`) so
  new keys auto-default, but the UI SettingsService reads field-by-field —
  a new setting silently stays at default until wired in all five spots.
- **Saves are whole-object: a window must not save before its first
  `getSettings` landed.** A ⌘+ pressed while an editor opened wrote
  `DEFAULT_SETTINGS` (no keyboard remaps) over the user's file.
  `SettingsService.persist()` is a no-op until `applyAll` ran once.
- **Every window's RPC must serve `saveSettings`.** The scale shortcuts
  shipped in editor and manager windows while only the main RPC served it:
  the request went unanswered, the change applied locally, and main's next
  broadcast snapped it back. All three factories now share
  `saveSettingsHandler`, and `settingsSaved()` in index.ts rebroadcasts to
  every window but the sender — an echo would drag a held Cmd+ backwards.
- The editor draft mirror sends full content per tick (2 MB cap upstream);
  don't shrink the 400 ms debounce.
- `panel.touched` gates localDir auto-navigation; `connectToken` guards
  against a slow connect resolving into a panel the user already reused.

## i18n

- **Translations live in `src/shared/i18n/*.ts` (TS, not JSON)** — deliberate:
  it sidesteps `resolveJsonModule` differences between the Angular esbuild and
  Bun builds and gives full typing. `en.ts` is the source of truth; other
  locales are partial and fall back to English then the raw key.
- **`i18n.t()` is reactive because it reads the `locale` signal** — that's what
  makes live language switching work with no pipe. If you cache a translated
  string in a plain field it WON'T update on switch; call `t()` in the template
  (or in a signal-reading getter) instead. Same for `i18n.date()/size()`.
- Dictionary values are TS string literals: **French/other apostrophes break
  single-quoted entries** (`'…c'est…'`). Use double quotes for any value
  containing an apostrophe. Plurals are `{ one, other }` maps (not ICU strings);
  zh/ja supply only `other`.
- **Native menus can't use the Angular service** — `menu.ts` goes through backend
  `i18n.ts`, and the app menu must be rebuilt when the locale changes (wired in
  index.ts `settingsChanged`), or it keeps the old language. The context menu no
  longer has this problem: it is Angular now and reads `I18nService` directly.
- `eu`/`zh`/`ja` are AI drafts pending native review (see the i18n README).
- **Silent auto-refresh must snapshot sessionId+cwd and discard a mismatched
  result.** A normal `refresh()` is masked by `busy` (input blocked), but the
  silent variant (poll/watch) runs while the user can navigate — without the
  guard, a poll that started in dir A resolves and stamps A's listing over the
  dir B the user just opened. It also never calls `scrollCursorIntoView`
  (scroll must stay put) and returns early if `busy()` or already refreshing.

## Build / tooling

- Angular CLI rejects Bun's Node-compat version — `scripts/node-version-shim.ts`
  is preloaded via `bunfig.toml` at the root AND in `ui/`. Removing either
  breaks `ng`.
- ssh2's optional native deps must stay `external: ["cpu-features", "*.node"]`
  in `electrobun.config.ts` — bundling them breaks the build; ssh2 falls back
  to pure JS at runtime. (Inert without `mainProcess: "bun"` — see Electrobun.)
- **2.x installers dropped the `stable-` prefix; the updater feed kept it.**
  `macos-arm64-miniSCP.dmg` / `win-x64-miniSCP-Setup.zip` /
  `linux-x64-miniSCP-Setup.tar.gz` next to `stable-<os>-<arch>-update.json`,
  `.tar.zst` and `.patch`. Anything that globs `artifacts/stable-*` misses the
  installers — release.ts collects by `<os>-<arch>-`, deploy.sh publishes
  `artifacts/*` verbatim (release.ts empties `artifacts/` first), and the site
  links the new names.
- `bun run --cwd ui build` (flag order matters; `bun --cwd ui run build` does
  not do what you want).
- `version.gen.ts` is gitignored and regenerated by postinstall + every build
  script; a fresh checkout fails `tsc` until `bun install` (or any script)
  has run once. Never commit it — it would bump the commit count it records.
- Icon pipeline is `scripts/gen-icons.ts` (resvg) — regenerate after touching
  `assets/logo.svg` or `assets/tray/tray-glyph.svg`; the .ico and tray PNGs
  are build inputs, not build outputs.