10.0.301: HTML launch (direct=1) fails with Error #204 on cold start, 10.0.286 works

Summary
Since the Web Store started serving 10.0.301 (first seen 2026-10-03 ~21:50 UTC), launching a macro by opening its exported HTML file with ?direct=1 fails on a cold browser start. The page shows “Error #204: It seems UI.Vision RPA is not installed yet - or you need to turn on Allow access to file URLs”. No macro runs and no log is saved. Allow access to file URLs is on (newAllowFileAccess: true). The same profile and HTML file on 10.0.286 work.

Environment

  • Chromium 152.0.7977.82 (Ubuntu snap, arm64) and Chromium 154.0.8037.57 (Ubuntu snap, x86_64), both affected.
  • Launch: chromium --profile-directory=… “file:///…/macro.html?direct=1&closeBrowser=1&closeKantu=1&savelog=…”
  • 10.0.301 CRX sha256 ef6699aaef8c24aeaa26440649a50d628264a9f21371362961aa69693abbe407

Reproduction

  1. Take a profile with ui.vision 10.0.301 installed and file URL access on.
  2. Start the browser cold with a macro HTML file and ?direct=1.
  3. The HTML page polls for data-kantu on , never sees it, and shows Error #204 after its retries.

Same machine and macro on 10.0.286: the macro plays (2 of 2 cold launches). On 10.0.301: Error #204 (4 of 4).

What we observed

  • Chromium’s stderr on the failing launches shows DidStartWorkerFail gcbalfbdmfieckjlnblleoemohcganoc: 5.
  • Toggling the extension off and on in chrome://extensions makes launches work in that browser session only. The next cold start fails again.
  • Making the HTML page retry for 30 s instead of ~2 s does not help.

Likely cause (read from the code, not confirmed in a console)

  • In 10.0.286, content_script.js sets data-kantu on document.documentElement as soon as it runs, and only afterwards reads config.
  • In 10.0.301, data-kantu is set only inside jt().then(…), i.e. after get(“config”) resolves (around switch(Rt(“file:”===window.location.protocol?!!t.allowRunFromFileSchema:…).
  • If the background service worker has not started (see the stderr line), that promise never resolves, so the page is never marked and the launcher reports #204.

Request
Could you restore the 10.0.286 behaviour of setting data-kantu immediately (or not make it depend on the service worker), and let us know which version carries the fix? We run unattended command-line launches, so an automatic update that breaks the launch path stops every job at once. Is there a supported way to pin the extension version?

Thanks. The 10.0.281 issue was fixed very quickly last time and we appreciate it.

Thanks for the report, we are looking into it asap.

10.0.301 launches fine with the HTML launcher a current UI.Vision export generates. Our failing macros were exported by an older version, and their launcher block loses a timing race against 10.0.301.

Same profiles on 10.0.301, same browser session:

  • A macro carrying the newer launcher block (setTimeout(main, 500) before the first check, parseInt on the reload counter, noImport: false in the kantuSaveAndRunMacro detail) runs normally.
  • A macro carrying the older block (main() called immediately, no parseInt) shows Error #204 every time.
  • Swapping the newer block into the failing macro file, with the macro table untouched, makes it run.

Two observations that may be useful to you:

  1. In the older launcher, var count = 1 + (url.searchParams.get(‘reload’) || 0) concatenates strings. After the first reload count becomes “11”, which is greater than MAX_TRY, so it gives up after one reload, about one second, instead of three.
  2. 10.0.301 seems to set data-kantu later than 10.0.286 did. In 10.0.286 it was set as soon as the content script ran; in 10.0.301 it’s set after get(“config”) resolves. The older launcher’s one-second window is enough for 286 but not for 301.

Since many users have macros exported by older versions, setting data-kantu immediately again, as 10.0.286 did, would keep old exports working. Failing that, a release note saying to re-export, or replace the launcher block, would help.

Sorry for the false lead in the first post, and thanks again.

Thanks for the additional info!

Correct! We are looking into this.