Seven bugs in the scripts, plugin and CI, before and after

Each pair is a real run: the old code (before) and the fixed code (after), on this Mac. Each fix also has a test that failed before it.

1The build script builds the target it is given

bun scripts/build.ts linux-x64 without --out dropped its first argument and built this Mac's target. Now it builds the one named, and a mistyped option stops it.

Before
$ bun scripts/build.ts linux-x64
dist/obeya-0.13.0-darwin-arm64  (80 MB)
built in 0.3 s
After
$ bun scripts/build.ts linux-x64
dist/obeya-0.13.0-linux-x64  (101 MB)
built in 2.2 s

$ file dist/obeya-0.13.0-linux-x64/obeya
dist/obeya-0.13.0-linux-x64/obeya: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), dynamically linked, interpreter /lib64/ld-linux-x86-64.so.2, for GNU/Linux 3.2.0, BuildID[sha1]=a9ea3707c32f242dc5ec7fee3c5fdd9bc99893b3, not stripped

$ bun scripts/build.ts --al
build: Unknown option '--al'. …

2The app check starts the app, not the screenshot file

check-app.ts took the first word without -- as the program, so --screenshot out.png <program> tried to start out.png. Now it takes the program, and says so at once when the program is not there. (No app is installed on this Mac, so the check itself did not run.)

Before
$ bun scripts/check-app.ts --screenshot out.png /Applications/Obeya.app/Contents/MacOS/obeya
ENOENT: no such file or directory, posix_spawn '/private/tmp/w3-before/out.png'
    path: "/private/tmp/w3-before/out.png",
 syscall: "posix_spawn",
After
$ bun scripts/check-app.ts --screenshot out.png /Applications/Obeya.app/Contents/MacOS/obeya
check-app: no program at /Applications/Obeya.app/Contents/MacOS/obeya
usage: bun scripts/check-app.ts <the app's program> [--fake-mic] [--screenshot <png>] [--keep]

3A scratch Obeya's --say and --artifact keep every word

Both dropped each word of the text that starts with --. The same line sent once from the old script and once from the new, to a card of a scratch Obeya: the old one in the card's history, the new one on top.

A card whose conversation shows the line from the new script with both --dry, and below it in the history the old script's line without them

4The site republishes its video player when any of its sources change

site/player.js is built from ten files; the Pages workflow listed only four of them, so a change to the player's words (demo-page.ts) or the caption language choice (locale.ts) did not reach obeya.si. Now it lists them all, and a test fails when a new source is missing.

File player.js is built fromBeforeAfter
scripts/site-player.ts✓✓
src/core/player.css✓✓
src/server/demo-page.ts✗✓
plugin/skills/demo/lib/subtitles.ts✗✓
plugin/skills/demo/lib/locale.ts✗✓
plugin/skills/demo/lib/timeline.ts✗✓
src/core/caption-pages.ts✓✓
src/core/frame.ts✗✓
src/core/video-player.ts✓✓
plugin/skills/demo/lib/word-times.ts✗✓

5A demo render without Playwright says what is missing

Without Playwright, node demo.ts and node revoice.ts crashed at import. Now the render gets to its setup check, which lists Playwright with everything else that is missing and how to install it.

Before
$ node demo.ts      # on a machine without Playwright
node:internal/modules/package_json_reader:343
  throw new ERR_MODULE_NOT_FOUND(packageName, fileURLToPath(base), null);
        ^

   … 12 lines of traceback …
}

Node.js v26.10.0
After
$ node demo.ts      # on a machine without Playwright
Error: cannot render yet:
Demo setup on macOS (arm64):
  ✓ node         v26.10.0
  ✗ playwright   not found
      bun install --cwd "…"
  ✓ browser      /Applications/Google Chrome.app/Contents/MacOS/Google Chrome
  ✓ ffmpeg       9.0.2
  ✓ uv           0.12.23
  ✗ voice        kokoro: Kokoro — not installed
      node "…/plugin/skills/demo/lib/voices.ts" install
      or install it in Obeya's settings (Configuration → Demos)
Missing: playwright, voice. A render stops until they are installed.

6A voice sample on a spent quota gives the message, not a traceback

When a hosted voice's daily quota was used up, a sample (the Sample button in Configuration → Demos) ended in two tracebacks; only a render said what to do. Now both say it in one line. Captured against a stand-in server that answers as Gemini does then.

Before
$ python tts.py --sample voice.json en "Hello." sample.wav      # the daily quota is used up
Traceback (most recent call last):
   … 37 lines of two tracebacks …
  File "plugin/skills/demo/lib/tts.py", line 435, in _post
    raise QuotaExhausted(f"the daily {self._service} TTS quota is used up")
QuotaExhausted: the daily http TTS quota is used up
After
$ python tts.py --sample voice.json en "Hello." sample.wav      # the daily quota is used up
the daily http TTS quota is used up; choose another voice (or DEMO_GEMINI_MODEL) or wait a day

7The browser skill's input gives up after 30 s

An input step against a page that never answers waited for good, where every other command stops after 30 s. Now it stops after 30 s too.

Before
$ node browser.ts input 'eval:for (;;) {}'      # a page that never answers
(no answer, still waiting after 75.0 s: the capture stopped it)
After
$ node browser.ts input 'eval:for (;;) {}'      # a page that never answers
Timeout waiting for Runtime.evaluate
(ended after 30.2 s, exit code 1)

+Also in this change