Repository navigation
Open
Labels
bugExisting behavior is brokenExisting behavior is brokenrecipe:remediationRoute: investigate/decompose before implementationRoute: investigate/decompose before implementation
Description
Problem
On the dev (branch) channel, a successful autoskillit update prints AutoSkillit updated successfully. and then exits 20 (failed-upgrade). After success, the command re-executes itself with its original arguments. The second invocation runs update again under the newly installed release, finds previous, observed and target commits all equal, and rejects the unchanged advance verdict.
Separately, running autoskillit update when already current also exits 20. Scripts and operators therefore see a failure for a successful update, and for a no-op update.
Reproduction (verified 2026-10-30, real packages in a scratch Docker home)
Environment:
- Old: pristine develop-track 0.10.1013 (
5c4924749b3b52fabc487233c3747e511ab35512). - Target: 0.10.1306 (
f84bae324c3497ec0851d5ac887770d0e895fbad), supplied through a scratch Git mirror.
- Install the old release. Run
autoskillit update.- Observed: the candidate is installed and published, and the entrypoint pivots.
AutoSkillit updated successfully.is printed.- Then:
Upgrade advance verdict unchanged on branch channel ...,AutoSkillit update did not complete (failed-upgrade)., exit 20.
- Run
autoskillit updateagain on the now-current candidate.- Observed: exit 20 in about 11 s, with previous, observed and target all equal to
f84bae324....
- Observed: exit 20 in about 11 s, with previous, observed and target all equal to
- Check the candidate.
- Observed:
autoskillit --versionandautoskillit doctorexit 0. The installeddirect_url.jsonrecords the target commit.
- Observed:
Evidence
src/autoskillit/cli/update/_update.py:64callsperform_restart()after success.src/autoskillit/cli/update/_restart.py:12runsos.execv(sys.executable, [sys.executable] + sys.argv), re-running the originalupdatearguments.src/autoskillit/cli/update/_transaction.py:165-170formats the unchanged advance verdict that the second run treats as failure._update.pyand_restart.pyare identical in 0.10.1013 and the currentdevelop. Upgrades from already-shipped releases hit this through the old caller plus the new callee.
Impact
- Every successful dev-channel update ends with a failure exit code and a contradictory message.
- An already-current
autoskillit updateis reported as a failed upgrade. - Automation that checks the exit status will treat working installs as broken.
Expected behavior / acceptance criteria
- A successful update exits 0. The post-update restart must not re-run the update transaction. Consider whether a restart is needed at all for
update. autoskillit updateon an already-current install reports "already up to date" and exits 0.- These outcomes hold when an older release's updater installs the new release. The new release's handling of a re-executed
updatemust cope with the old caller. - A regression test exercises the real restart and exec boundary. Existing updater tests mock
perform_restart.
Related
- Found in the 2026-10-30 release-readiness rehearsal (draft PR [DO NOT MERGE] Release readiness experiments for develop #5335,
docs/developer/release-readiness.md, "Reproduced upgrade failure").
Metadata
Metadata
Assignees
Labels
bugExisting behavior is brokenExisting behavior is brokenrecipe:remediationRoute: investigate/decompose before implementationRoute: investigate/decompose before implementation
Activity