buninu-linux 0.2.0

package/initramfs/buninu/CHANGELOG.md

@@ -0,0 +1,580 @@
+# Changelog
+
+## 0.4.12 - 2026-09-20
+
+### Fixed
+
+- Make Buninu's shell scripts parse and run under bunmsh as well as a POSIX
+  shell. The jsmdcui and musl-la launchers now use the `for name; do` form
+  accepted by bunmsh, and jsmdcui avoids bracket expressions in parameter
+  trimming while preserving case-insensitive Markdown suffix detection
+- Make `make-bunBin.sh` bunmsh-compatible by replacing its `case`, shell-option,
+  command-scoped `CDPATH`, `cd --` and `trap` usage with portable conditionals,
+  explicit failure handling and temporary-directory cleanup
+- Use explicit `1>&2` stderr redirection in `bun.sh` and `make-bunBin.sh`.
+  bunmsh interpreted the abbreviated `>&2` form as an output file named `&2`
+
+## 0.4.11 - 2026-09-20
+
+- Emergency fix: multicall.sh on Buninu Linux
+  * /proc/$$/cmdline: /bin/bun /bin/sh ...
+
+## 0.4.10 - 2026-09-20
+
+### Fixed
+
+- Sync bunmsh to 0.3.7. Its fallback `tr` builtin now decodes NUL, octal and
+  common backslash escapes in character sets, so commands such as
+  `tr '\0' '\n' < /proc/$$/cmdline` correctly split NUL-delimited process
+  arguments. Previously the NUL bytes passed through unchanged; the Buninu
+  multicall launcher could then misidentify a command invoked with arguments
+  and exit with status 127 before reaching its shell wrapper, for example
+  `jmi hlw.js` even though plain `jmi` worked.
+
+## 0.4.9 - 2026-09-16
+
+### Changed
+
+- Sync bunmsh to 0.3.6
+  * VS Code compatibility: accepts leading &
+
+## 0.4.8 - 2026-09-16
+
+### Documentation
+
+- Add an Android-only optional-tools walkthrough for installing `bunproot`,
+  cloning `js-udocker`, and starting an Alpine container. Both tools remain
+  user-downloaded rather than bundled with Buninu; the README records
+  `bunproot` as GPL-2.0-or-later and `js-udocker` as Apache-2.0
+
+## 0.4.7 - 2026-09-14
+
+### Added
+
+- `--readme-tui` and `--readme-wui` open README.md through jsmdcui instead of
+  printing it, so its headings and its table of contents become links to
+  follow -- the first as a terminal UI, the second served at a URL for a
+  browser. jsmdcui writes five generated files beside any Markdown it opens,
+  so README.md is copied into a directory under `TMPDIR` and run from there:
+  an installation stays the user's, and nothing generated reaches an
+  `--export` or survives an update
+- `--bunmsh` joins `--jsgotty`, `--jsmdcui` and `--musl-la` as a first
+  argument that spawns that app directly, forwarding every remaining argument
+  and exiting with its exit code
+
+### Changed
+
+- `package.json` now carries `repository`, `homepage`, `bugs`, `keywords` and
+  `author`, so the npm page links back to the source and the package can be
+  found by search
+- `apps/musl-la/NOTICE` records where the bundled musl loader came from:
+  Alpine's `musl` 1.2.5-r10 for aarch64, with the SHA-256 of the binary. That
+  pkgrel is no longer on Alpine's CDN, which keeps only the current one, so
+  the note points at an archived copy of the same build
+- Seven licence texts in `apps/musl-la/LICENSES/` covered components this
+  package does not ship -- ggml, libssh2, curl, zlib, LLVM, a bare Apache-2.0
+  and LGPL-3.0 -- and have been removed. The three binaries there are covered
+  by `LICENSE_musl.txt`, `GPL-3.0.txt` and `gcc_runtime_exception.txt`, which
+  stay
+- README.md reorganised: a shorter opening that reaches `Install Bun` and
+  `Start` sooner, the command reference moved ahead of the sections about
+  keeping and updating an installation, grouped command listings, and the
+  naming convention for platform-specific binaries (`la`, `lx`, `aa`, `wx`,
+  `ma`) written down
+
+### Fixed
+
+- `--local` was missing from `--help` entirely, and now leads the option list
+- `showimg` in a browser terminal
+  * Bind the scroll and render listeners before decoding the image rather than
+    after it. The server sends the placement ahead of the synthetic newlines
+    that scroll room for the image, so that scroll could land while the decode
+    was still running and the very first image of a session had no listener to
+    correct it -- it came out blank while every later image was fine
+  * Re-run the layout once the image element has actually loaded
+  * Lay placements out against xterm's own geometry until the client's first
+    resize message arrives, instead of against the 80x24 placeholder
+  * Let the terminal size settle for up to 400ms before emitting an image. A
+    window that was just opened, restored, resized or rotated reports a couple
+    of columns for a few hundred milliseconds, and an image sized against that
+    was emitted as a one-cell speck -- indistinguishable from the command
+    having printed nothing -- while the next run in the same session was correct
+  * Walk the prototype chain for the native `WebSocket.onmessage` descriptor.
+    Safari/WebKit defines it on a parent prototype rather than on
+    `WebSocket.prototype`, so the single-level lookup found nothing and image
+    messages were never processed; fall back to `addEventListener` when no
+    native setter is available
+  * Thanks @hcyuser for the contribution! (#3)
+
+## 0.4.6 - 2026-09-14
+
+### Fixed
+
+- Experimental macOS support
+  * Fall back to `$0` when `/proc` is unavailable
+  * Rename the shell wrapper's `status` variable to `_buninu_status`; `status`
+    is read-only in zsh, so the web terminal died the moment it connected on a
+    default macOS shell
+  * Thanks @hcyuser for the contribution! (#1, #2)
+
+## 0.4.5 - 2026-09-10
+
+0.4.4 is intentionally skipped 不吉利
+
+### Changed
+
+- Sync bunmsh to 0.3.5
+- Sync jsgotty to 1.1.11
+  * Security fix CSWSH
+- Sync jsmdcui to 0.19.1
+  * Security fix CSWSH for CDP server
+
+## 0.4.3 - 2026-09-05
+
+- Don't check exit status for bunx
+
+## 0.4.2 - 2026-09-03
+
+### Changed
+
+- Update the bundled bunmsh from 0.2.0 to 0.3.3, which is what a `--local`
+  session runs. It gains a `curl` PATH-fallback builtin written on Bun's own
+  `fetch`, so a device that ships no `curl` binary can still run the download
+  and API scripts that expect one; `pspa` and `pspac`, the process listing this
+  tree's `.bashrc` aliases used to be the only way to get; and a `kill` that
+  works on Windows. Its JavaScript mode binds `$` to the shell's own variable
+  table, so `$.HOME` reads a shell variable and `$.TAG = "v1"` writes one that
+  later commands see, and completion learned shell variable names, command
+  names inside `$(`, and JavaScript lines. A real `curl`, `pspa` or `pspac` in
+  `PATH` still wins over any of the builtins; `builtin curl ...` selects
+  bunmsh's. Its own [CHANGELOG](apps/bunmsh/CHANGELOG.md) has the rest.
+
+### Fixed
+
+- The process-list helpers section, left behind by the bunmsh update above,
+  still described `--local` as a session with neither the bundled `.bashrc` nor
+  `$HOME/.bashrc`, and sent you to source the bundled file to get the aliases
+  back — where `pspac`, being two commands joined by `;`, would fail to define
+  as a bunmsh alias. bunmsh now has `pspa` and `pspac` as PATH-fallback
+  builtins, and colour aliases for `ls`, `grep` and `diff` of its own, so a
+  `--local` session already answers all five of those names and has no reason
+  to source anything. The section says so now, and says where the builtins
+  differ from the aliases they stand in for: `pspac` colours the table inline
+  rather than writing `$HOME/.pspidargs.sh` and paging it through `glow`, and
+  `pspa` answers on Windows as well, by querying `Win32_Process` through
+  PowerShell, where `ps -eo pid,args` has no counterpart — so the paragraph
+  ruling Windows out is now about the aliases rather than about both. Nothing
+  changed for the other routes: Android still loads the bundled file through
+  `ENV` for the shell jsgotty starts, and a remote Bash or mksh session on
+  Linux or macOS still sources it by hand.
+
+## 0.4.1 - 2026-08-30
+
+### Fixed
+
+- A freshly installed tree no longer reports files as locally changed when
+  only their file mode differs. Publishing normalises modes, so a file can come
+  back from the registry with a different executable bit and byte-identical
+  contents, which `bun pm diff` counts as a modification; an update was
+  listing those and asking about files nobody had touched. They are now
+  recognised by the mode having changed while the contents did not, and left
+  out of the list, with a line saying how many were ignored so the count still
+  adds up against the command below.
+
+### Changed
+
+- An update prints the `bun pm diff` command it is about to run, on a line of
+  its own before the result. Run it by hand without `--json` to read the list
+  directly, or without `--name-only` to see the changes themselves rather than
+  which files hold them. It goes to stdout rather than stderr with the rest of
+  the output, so `2>/dev/null` leaves just the command.
+
+## 0.4.0 - 2026-08-30
+
+### Added
+
+- `--install [dir]` (`-i`) and `--strip-install [dir]` (`-si`) copy the running
+  installation into a directory that outlives the `npx` cache. `--install`
+  creates `<dir>/buninu` as the new `BUNINU_HOME`, `--strip-install` makes
+  `<dir>` itself that root, and both default to the current directory. The
+  implementation is `bin/install.js`, which runs on its own as well as being
+  imported by `bin/init.js`; run directly with no arguments it installs.
+  Relative symbolic links are recreated exactly as stored rather than followed,
+  so `bin/androidNativeLibs -> ../..` stays a link instead of dragging the
+  parent tree into the copy. `cp -a` is used when `cp` is on `PATH`, and an
+  equivalent `node:fs` walk otherwise; the two were verified to agree on
+  content, link targets, permissions and whole-second timestamps across every
+  entry of the tree.
+- Installing over an existing Buninu updates it in place. The copy only ever
+  adds and overwrites, so `apps/<name>/` commands, `bin/*.sh` overrides and the
+  Bun binaries `bin/bun.sh` extracted are left alone, and the three files that
+  belong to the package and to the user at once are merged: the local `buninu`
+  section of `package.json` is kept, command names added to `apps/cmdlist` are
+  kept, and `.bashrc` is kept whenever it only adds lines to the shipped one,
+  including lines inserted in the middle. A `.bashrc` that cannot be merged
+  that way is left exactly as it is, with the shipped version written beside it
+  as `.bashrc.dist`.
+- Before replacing the package's own files, an update runs `bun pm diff`
+  against the version the installation reports and lists the files that no
+  longer match it, then asks before continuing. Nothing is recorded inside the
+  installation to make this work — the published package is the reference — so
+  the check needs the network, and when it cannot run, or does not finish
+  within 60 seconds, it asks rather than assuming there was nothing to lose.
+  `--yes` answers ahead of time, which a run without a terminal needs;
+  `--force` replaces everything without merging or checking. An existing
+  installation is always named by absolute path before anything is written.
+
+- `buninu.xdgDataHome` points `XDG_DATA_HOME` at a directory of your choosing:
+  `true` puts it inside the installation at `$BUNINU_HOME/.local/share`, and a
+  string names one outright, resolved from the `package.json` directory when it
+  is relative. The directory is created at startup. What this moves in practice
+  is bunmsh's command history, which lives at `$XDG_DATA_HOME/bunmsh/history`,
+  so `true` is what makes a shell's history travel with a copied installation
+  instead of staying on the machine it was typed on. `HOME` is deliberately not
+  touched by it, so SSH keys and Git configuration in the user's own home keep
+  working and bunmsh can still import an existing `~/.bash_history`.
+
+- Refusing to install into an occupied directory now says which check the
+  directory failed — no readable `package.json`, or one naming a different
+  package — instead of reporting every case as "is not a buninu installation".
+  Whether a directory is an installation to update is decided by that name
+  alone: a Buninu missing `bin/init.js` is a damaged installation rather than
+  somebody else's directory, so an update repairs it, merging the local
+  configuration back in, where before `--force` was the only way through and
+  took that configuration with it. Every refusal names the directory by
+  absolute path.
+
+### Fixed
+
+- `HOME` no longer falls back to the package directory when the environment
+  does not set it. Windows does not set `HOME`, so a Windows session had its
+  home pointed at the disposable `npx` cache, taking anything written there —
+  bunmsh's command history included — with it when the cache was cleared. It
+  now falls back to the operating system's own home directory for the user.
+
+## 0.3.2 - 2026-08-29
+
+- Added bunmsh to bin and cmdlist
+
+## 0.3.1 - 2026-08-29
+
+- Added bunmsh to buninu
+
+## 0.3.0 - 2026-08-21
+
+### Added
+
+- Add `openWebView`, `evalWebView`, `showWebView` and `currWebView` to
+  `apps/native-bridge` (short spellings `openwv`, `evalwv`, `showwv`,
+  `currwv`, both accepted over the wire and listed in the host's `_discover`,
+  the same arrangement `getcb`/`setcb` already had), plus the
+  `WEBVIEW_CURRENT`/`WEBVIEW_CONSOLE`/`WEBVIEW_APP` constants. The host app
+  has exactly two WebViews, alive from startup and never created or closed at
+  runtime: `0` is the console showing the jsgotty terminal, `1` is the app
+  WebView, which starts blank and behind. An id of `-1` means whichever is in
+  front. `openWebView` loads without bringing its target forward, so loading
+  the app WebView while the user keeps looking at the terminal is one call;
+  `showWebView` is the only thing that changes what is on screen, and
+  `showWebView(-1)` switches to the *next* WebView rather than being a no-op.
+  `evalWebView` resolves to the value the expression produced rather than a
+  string containing it, with `undefined`, functions and thrown exceptions all
+  arriving as `null` because WebView itself does not distinguish them.
+- `xdg-open` gained `MINAPK_WEBVIEW`, which names the WebView to open a URL
+  in (and brings it to the front) instead of handing the URL to the
+  platform's default handler. Read here rather than by the host app on
+  purpose: it is an environment variable of this process tree, so the host
+  could not see a value exported in the shell a moment ago, which is the
+  whole point of having it. Only URLs are redirected -- a filesystem path
+  still goes to the host, whose `content://` provider can actually serve it,
+  unlike a `file://` URL in a WebView. A value that is not a plain integer is
+  reported and ignored rather than guessed at, and a WebView the host does
+  not have falls back to the default handler after saying so.
+- Add `demo-server.js`: serves one fixed page on an OS-assigned port and
+  opens it in the app WebView through `MINAPK_WEBVIEW=1 xdg-open`, as the
+  smallest end-to-end exercise of the above.
+- Add `buninu.backToConsole` (default `true`), read by the minapk host app,
+  not by Buninu itself: it decides whether the Android back key, pressed on
+  the app WebView with no page of its own to go back to, switches to the
+  console WebView or takes the leave-the-app path. The host treats a missing,
+  unreadable or non-boolean value as `true`, so it cannot fail in a way that
+  leaves the back key broken. minapk's `--no-back-to-console` sets it for a
+  single build.
+
+## 0.2.5 - 2026-08-19
+
+- Fixed fish shell startup breaking buninu
+- Start fish after entering the shell please
+
+## 0.2.4 - 2026-08-19
+
+### Added
+
+- Add the `xdg-open` command (`apps/xdg-open`), which opens a file or URL
+  with whatever the platform considers its default handler: native-bridge's
+  `xdgOpen` on Android (the host resolves a path under Buninu's home and
+  hands it to another app through a read-only `content://` provider, since
+  that directory is otherwise private to the APK), `termux-open` when
+  running under plain Termux with no native bridge, `open` on macOS, `start`
+  on Windows, and the real system `xdg-open` on Linux. That last one skips
+  Buninu's own `bin` while searching `PATH`, since this script is itself
+  registered as `xdg-open` and would otherwise find and run itself.
+- Add `showimg`, a `bin/`-only shorthand for `jsgotty --viu` (no `apps/`
+  subfolder, since it has no logic of its own beyond forwarding args).
+- Add `rz`/`sz`, `bin/`-only companion scripts wrapping the existing
+  `apps/jsgotty/rz.js`/`sz.js` -- `rz` uploads a file over ZMODEM, `sz`
+  downloads one -- over the same terminal connection jsgotty renders in a
+  browser or WebView.
+
+- Add `apps/native-bridge`, a Bun module (`toast`, `clipboardRead`/`getcb`,
+  `clipboardWrite`/`setcb`, and the raw `call(func, args, envp)` it and the
+  CLI both sit on top of) that reaches minapk's Android native bridge over
+  `PKG_BRIDGE_SOCK` -- a Bun unix socket, filesystem-path or Linux
+  abstract-namespace depending on how that env var is encoded -- using
+  jsmdcui's `rpc.mjs` `switchBackend`. Only meaningful inside an APK built by
+  minapk with the native bridge wired in; elsewhere `available()` returns
+  `false` and calls throw a clear error instead of doing nothing silently.
+  Every call times out after 5s by default, so a stuck or unresponsive host
+  can never hang the caller. `bun apps/native-bridge/native-bridge.js [func]
+  [args...]` runs it directly; a bare invocation defaults `func` to
+  `_discover` and lists what the host implements.
+- Add the `xclip` command (`apps/xclip`), ported from the DroidScript-era
+  `../tmpk/bin/xclip.sh` to use native-bridge instead of `dsapi.pipe`.
+  `-selection primary` (the default) stays local-file-only, matching real X11
+  semantics where the primary selection is never the same thing as the
+  clipboard; `-selection clipboard`/`-clip` bridges to the real Android
+  clipboard through native-bridge. jsmdcui's own clipboard backend detection
+  already shells out to `xclip` on Linux-like platforms (`isLinuxLike()`
+  counts `process.platform === "android"`), so jsmdcui's middle-click paste,
+  selection auto-sync, and `PastePrimary` command all work inside an APK with
+  no further wiring once `xclip` is on `PATH`.
+- Add `speak`/`ttsStatus`/`tts` to native-bridge and the `tts` command
+  (`apps/tts`) built on them. Android's TextToSpeech completion signal is an
+  asynchronous callback with no way to push it to the far side over this
+  protocol, so `speak(text, speed, pitch, flush)` never blocks -- it returns
+  a handle immediately, and `ttsStatus(handle)`/`tts(handle)` (`"speaking"`,
+  `"done"`, `"error"`, or `"unknown"` once a terminal state has already been
+  consumed) is polled to find out when it finishes. `pitch`/`speed` both
+  default to `1.0` (normal), the same convention as `termux-tts-speak`'s
+  `-p`/`-r`; native-bridge passes them straight to `TextToSpeech.setPitch()`/
+  `setSpeechRate()` with no unit conversion. `tts <text>` wraps the polling
+  loop for the common case, blocking the calling process until speech
+  finishes -- the same as any other CLI tool that waits for the thing it
+  started, and, like real TTS tooling (espeak-ng has no timeout concept at
+  all; Windows SAPI's `WaitUntilDone` documents `-1`/infinite as the default),
+  waits as long as it takes with no wall-clock cap by default. Pass
+  `--timeout <ms>` to opt into a bounded wait instead, or `-a`/`--async` to
+  not wait at all. `--pitch`/`--speed` override the environment; without
+  them, `tts` reads `$TTS_PITCH`/`$TTS_SPEED` itself (falling back to `1` for
+  either that is unset), so jsmdcui's own `TTS_PITCH`/`TTS_SPEED` convention
+  (which it sets on `Bun.env` before spawning a TTS command, inherited here
+  like any other child process env var) is honored automatically -- jsmdcui's
+  own `detectTtsCmd()` already falls back to a bare `Bun.which("tts")` after
+  termux-tts-speak/espeak-ng/espeak, which this `tts` now satisfies, but that
+  particular fallback branch passes no `-p`/`-r`-equivalent flags at all, so
+  reading the env vars directly here is the only way `tts`'s pitch/speed
+  actually reaches Android's TTS engine through that path.
+- Give `xclip` and `tts` a real fallback on non-Android platforms too, since
+  Buninu is cross-OS, not Android-only. `xclip`'s `-selection clipboard`
+  register now also tries jsmdcui's own `ClipboardManager`
+  (`src/platform/clipboard.js`, imported directly rather than reimplemented)
+  on win32/darwin, then `wl-copy`/`wl-paste` on Linux; `tts` gains a
+  `detectTtsCmd()` copied verbatim from jsmdcui's `src/index.js` (same
+  pitch/speed/lang math and command construction, including the win32
+  PowerShell/SAPI SSML branch, copied rather than hand-retyped since that
+  quoting is easy to silently break and this had no Windows machine to
+  verify it on) covering termux-tts-speak/say/PowerShell/espeak-ng/espeak.
+  Both deliberately skip the one path that would reintroduce the exact
+  self-reference risk they otherwise avoid: `ClipboardManager`'s Linux-like
+  branch searches PATH for a binary named `xclip`, and jsmdcui's own
+  `detectTtsCmd()` ends with a `Bun.which("tts")` fallback -- since these two
+  commands are themselves registered as Buninu's `xclip`/`tts`, ahead of
+  anything else of the same name on PATH, using either the way jsmdcui does
+  from inside xclip.js/tts.js would just be each script invoking itself.
+- Add `bin/*.bat` launchers for the new commands (`native-bridge`, `xclip`,
+  `tts`, `showimg`, `rz`, `sz`, `xdg-open`). POSIX platforms get their
+  `bin/<name>` entries as symlinks rebuilt by `bin/init.js` on startup,
+  which Windows cannot use, so each command needs its own `.bat` there.
+
+### Changed
+
+- Update the bundled jsmdcui to 0.18.1, which adds `switchBackend()` to
+  `src/cui/rpc.mjs` so an RPC caller can point at something other than the
+  default `rpc` endpoint -- specifically a unix socket, which is what
+  `native-bridge` dials. Also picks up 0.18.1's fix for decoding a
+  percent-encoded socket path, needed because `new URL("unix:" + sock)`
+  turns a leading NUL (a Linux abstract-namespace socket) into `%00`.
+- `native-bridge`'s CLI now prints `_discover` one function per line instead
+  of as a single line of JSON, which makes the list greppable. The value
+  itself is untouched -- rpc.mjs dispatches every call against that object,
+  so it stays exactly what the host returned; only the CLI's own display of
+  it changed.
+
+## 0.2.3 - 2026-08-18
+
+### Added
+
+- Add `buninu.exitAfterCmd` package.json setting (default `false`). When
+  `true`, the shell/PTY exits with `buninu.command`'s status instead of
+  falling back to an interactive shell once the command finishes; `false`
+  keeps the existing fall-back-to-shell behavior.
+- Add `--export-config [output.json]`, which writes this entire `package.json`
+  to `buninu.json` in the current directory by default (like `--export`, the
+  output path can be overridden), instead of exporting the whole installation
+  like `--export` does. If the output path already exists, it now asks for
+  confirmation (`[y/N]`) before overwriting.
+
+## 0.2.2 - 2026-08-17
+
+### Added
+
+- Add a "Commands inside the shell" section to README.md, split out of
+  Command-line information (which is only `bin/init.js`'s own flags, resolved
+  before Buninu starts). Lists everything in `apps/cmdlist` and moves the
+  `buninu-help`/`bunx` write-ups there, including a note that `bunx` can't
+  install anything on Android until
+  [oven-sh/bun#39084](https://github.com/oven-sh/bun/pull/39084) merges
+  upstream (Android's seccomp policy kills `bun i -g` with SIGSYS during bin
+  linking until then).
+
+## 0.2.1 - 2026-08-17
+
+### Added
+
+- Add `bunx`, a POSIX and Windows multicall command that ensures a package is
+  installed via `bun i -g` and then runs its matching binary, forwarding the
+  remaining arguments. Reinstall timing is adapted from real bunx's own rules
+  (an explicit dist-tag like `@latest` always reinstalls; otherwise a cached
+  binary is reused until it's older than 24h, or stat fails, or bun's own
+  `install/global/package.json` shows a different pinned version than the one
+  requested — this script has one shared install location instead of bunx's
+  per-version cache, so this last check substitutes for that). The global bin
+  directory is resolved via `bun pm bin -g`, falling back to
+  `$BUN_INSTALL`/`$HOME/.bun` on the very first install before that project
+  exists (which `bun pm bin -g` requires).
+  - Runs the target through `bun <target>` instead of exec'ing it directly:
+    on Android the resolved binary typically sits under storage mounted
+    noexec, so a direct exec is refused even though the file is readable and
+    executable-bit set. Trade-off: a package whose bin is a real native
+    executable rather than a JS/bun script won't run this way.
+  - On Android (detected the same way as `bin/init.js`, via
+    `/system/bin/linker64`), installs with `--backend=copyfile` since
+    hardlinks/symlinks frequently fail across Android's storage, and sets
+    `$PREFIX=/data/data/com.termux/files/usr` for the target process (not the
+    install step) so CLIs that check for a Termux environment still find one,
+    unless `$PREFIX` is already set.
+
+## 0.2.0 - 2026-08-17
+
+### Added
+
+- Add `--jsgotty`, `--jsmdcui`, and `--musl-la` as the first argument to
+  `bin/init.js` to spawn that app directly, forwarding every remaining
+  argument to it and exiting with its exit code. This bypasses the shell and
+  startup-command flow entirely, and lets `--jsgotty --help` (and the same for
+  the other two) show the app's own option reference instead of Buninu's.
+- Add `buninu-help`, a POSIX and Windows multicall command that renders
+  README.md with jsmdcui's `--cat` mode and then displays `icon.png` with
+  jsgotty's `--viu`. The default startup greeting now points to it instead of
+  separately mentioning `glow`/`README.md`.
+
+### Security
+
+- Bind jsgotty to `127.0.0.1` by default instead of `0.0.0.0` in
+  `scripts.start`. Previously the terminal server accepted connections from
+  any device able to reach the host on its network, relying only on the
+  random port and `--random-url` path segment for protection. Pass
+  `--address <value>` through to `bin/init.js` (forwarded arguments override
+  the flags baked into `scripts.start`) to opt back into listening on other
+  interfaces.
+
+## 0.1.9 - 2026-08-17
+
+### Fixed
+
+- Correct `apps/musl-la/NOTICE`: it listed 11 third-party components
+  (llama.cpp, nghttp2/nghttp3, OpenSSL, libssh2, curl, zlib, libidn2,
+  libunistring, brotli, c-ares, libpsl) with no corresponding files anywhere
+  in the package — leftover from an unrelated bundle. The only files actually
+  shipped in that directory are `ld-musl-aarch64.so.1`, `libgcc_s.so.1`, and
+  `libstdc++.so.6`; the notice now covers only those, identifying the exact
+  build (GCC 14.2.0, packaged as Alpine Linux 3.22's `libgcc`/`libstdc++`
+  14.2.0-r6, aarch64) with matching SHA-256 hashes and a link to Alpine's
+  build recipe pinned to the commit for that package revision, satisfying the
+  GCC Runtime Library Exception's corresponding-source requirement.
+
+## 0.1.8 - 2026-08-17
+
+### Changed
+
+- Run an optional `libmain.so` from Android's native-library PATH at startup,
+  falling back to the normal Android welcome message when it is unavailable.
+
+## 0.1.7 - 2026-08-17
+
+### Added
+
+- Add `buninu --export [output.tgz]` to create a gzip-compressed tar archive of
+  the current Buninu installation. The archive retains exactly one top-level
+  directory and defaults to `./buninu.tgz`. An existing output is backed up and
+  restored if archive creation or replacement fails.
+- Add `apps/cmdlist` as the validated source of POSIX multicall commands.
+  Normal startup now recreates every listed command as a `shloader` symlink,
+  including after npm installs that omit package symlinks.
+- Add the Buninu-only `# syntax: markdown` first-line marker for Markdown
+  highlighting in extensionless files such as `apps/cmdlist`.
+
+### Changed
+
+- Update the default and Android startup greetings to point users to
+  `glow apps/cmdlist`.
+- Update `musl-la` to prefer Android's native `libld-musl.so`, fall back to the
+  bundled loader, add the target ELF directory and existing `LD_LIBRARY_PATH`
+  to its library search path, and support `-e` for the environment-variable
+  launch mode.
+- Document how to add POSIX multicall commands and their separate Windows
+  batch launchers.
+
+### Fixed
+
+- Keep the POSIX startup wrapper and Android greeting command on one line so
+  process listings do not expose embedded LF (`Ctrl+J`) characters.
+
+## 0.1.5 - 2026-08-16
+
+### Fixed
+
+- Enable jsmdcui's mdcui encoding for `--cdp-maze`, which selects the bundled
+  Markdown maze without passing a `.md` filename or a `--demo*` argument.
+
+### Changed
+
+- Prefer PowerShell (`pwsh.exe`, then `powershell.exe`) as the default Windows
+  shell, falling back to `%COMSPEC%` or `cmd.exe` when PowerShell is unavailable.
+
+## 0.1.4 - 2026-08-16
+
+### Changed
+
+- Enable jsmdcui's mdcui encoding when any launcher argument begins with
+  `--demo`, in addition to the existing `.md` file detection.
+- Run jsmdcui's `src/index.js` entry point directly from the Unix and Windows
+  `jsmdcui`, `jmi`, and `glow` launchers instead of going through its `tui`
+  wrapper. This gives demos such as `--demo-img-change` the expected `Bun.main`
+  path so their bundled image assets can be resolved correctly.
+
+## 0.1.3 - 2026-08-16
+
+### Fixed
+
+- Preserve a single canonical child-process path variable: `Path` on Windows
+  and `PATH` on other platforms. This ensures the Buninu `bin` directory is
+  available inside the shell started by jsgotty on Windows.
+- Keep the jsgotty reconnect overlay clickable in WebGL mode by placing it
+  above the canvas and preventing Kitty image pointer handling from intercepting
+  its mouse and touch events.
+
+### Changed
+
+- Enable jsmdcui's mdcui encoding from the Unix and Windows launchers only when
+  at least one command-line argument is a `.md` file. Launching without a
+  Markdown file now retains the editor-first behavior.

package/initramfs/buninu/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Dr. John (醫者小智)
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.