An OpenClaw migration stops every Omarchy update on 4.0.5, and the fixes are heading for 4.0.6
Machines with the openclaw package can’t get past migration 1790397381 since 4.0.5 shipped. Here is why it happens, the one-line workaround from the Omarchy team, and the string of fixes merged on October 18 for the next release.
Image: omacom/omarchy PR #15809 (GitHub)
If you have OpenClaw installed and ran omarchy update after Omarchy 4.0.5 landed on October 18, 2026, you probably hit a wall: “Running migration (1790397381)”, then “Something went wrong during the update!”. The bug was reported the same morning in issue #15804, and three more users confirmed it on different machines during the day, all on the stable channel with omarchy 4.0.5-1 and openclaw 2026.10.4-1.
The cause is laid out in PR #15809 by @emirb. 4.0.5 shipped a migration that moves OpenClaw from its Arch package to a self-updating copy under ~/.openclaw (the “Make Hermes and OpenClaw agents self-updating” line in the release notes). That move needs a new OpenClaw “seed” package, and at release time that package hadn’t shipped: every channel still served the runtime package, so the migration exited 1 on every machine with openclaw installed. Because the migration runner stops at the first failure, the five 4.0.5 migrations after it never ran either, including the one that puts Omarchy’s package repository ahead of Arch’s. According to the same PR, about 4,800 OpenClaw packages were downloaded on stable in the week before the release, so this is not a corner case.
There is a simple workaround while you wait. In his reply on the issue on October 18, @emirb suggested marking that one migration as done and running the update again:
touch ~/.local/state/omarchy/migrations/1790397381.sh then omarchy update. OpenClaw keeps working from its package, and the remaining migrations run normally. Users on the issue report that this got them through, with one caveat: the migration that raises [omarchy] above the Arch repos needs sudo, so run the interactive omarchy update from a terminal rather than a non-interactive omarchy-migrate.
The fix itself took a few turns on October 18, all on the quattro branch. #15809 retired the migration at 08:17 UTC so it would finish without changing anything; @ryanrhughes reverted it a few hours later (#15854). Then @spencerbull merged #14212 at 16:34 UTC, which takes a cleaner route: the old migration is retired for good, and the OpenClaw move comes back as a brand new migration, 1792334754. The new number matters, because 1790397381 is already marked done on every machine that used the workaround, and Omarchy never re-runs a migration marked done. Two minutes later the seed package PR, omarchy-pkgs #797, was merged too. Per its description, OpenClaw leaves the “fast ring”, so the seed builds into edge first and only reaches rc and stable as the release train advances.
#14212 originally came with a general “deferral” mechanism for migrations waiting on another repository’s package. That part was pulled back out the same evening in #15864: the new migration now simply completes without moving anything when the seed isn’t there, and the launcher sets OpenClaw up when it is launched from Omarchy later. A final piece, #15875 by @emirb (merged at 21:46 UTC, “Fixes #15804”), covers people who installed their own OpenClaw with npm i -g: instead of failing every update, the migration prints the installer’s reason, leaves that copy alone and finishes, and the move completes the next time OpenClaw is launched from Omarchy once the user’s copy is gone.
How did this get past review? #14212 is candid about it: neither end-to-end review of the original change ran the migration through omarchy update. It was run from a terminal, whose PATH includes ~/.local/bin, while omarchy update runs migrations on a fixed PATH without it. The test helper had the same blind spot. The suite now has a run_update helper that uses the update’s PATH, and it fails against the original installer.
All of this is merged but not released. #15875 mentions testing in “a real VM with 4.0.6 as Spencer built it” and says the change “cherry-picks with #14212 and #15864” for 4.0.6, and @emirb told the issue reporter “We’re fixing it in the next release.” As of this morning, v4.0.5 is still the latest release on GitHub, so treat 4.0.6 as in preparation rather than announced. One loose end noted in #15809: Install > AI > OpenClaw on 4.0.5 also fails with “cannot set up a self-updating OpenClaw”, because the installer needs the same seed; that was flagged as needing a fallback in its own PR.
