Problem
When the registration config edit rejects, WorkspaceService.create, createMultiProject, fork and rename return Err, but they do not roll back what already exists:
create / createMultiProject (src/node/services/workspaceService.ts ~5667 / ~6108): the worktree(s) or container directory, the session, the init state and the abort controller stay behind.fork (~11456, via config.addWorkspace): the same leftovers.rename (~8282): the runtime has already renamed the directory and branch on disk. The rename is not reverted, so disk and config disagree.
This already happens for every existing editConfig rejection source (the corrupt-file gate, a lost registration lock, a throwing transform). With #4444, a failed save rejects too, so disk-full or unwritable-root cases hit these paths.
Suggested direction
Give each flow one cleanup path for "registration edit rejected" that reuses the existing failure cleanup (createScratch already does this at ~5372). Keep it to one design that covers success, failure, cancellation and opt-outs; see #4440 for why per-path fixes did not converge.
Refs #4444 (follow-up F1 from its plan).
Generated with xum • Model: anthropic:claude-opus-5-5 • Thinking: high • Cost: $34.57
Problem
When the registration config edit rejects,
WorkspaceService.create,createMultiProject,forkandrenamereturnErr, but they do not roll back what already exists:create/createMultiProject(src/node/services/workspaceService.ts~5667 / ~6108): the worktree(s) or container directory, the session, the init state and the abort controller stay behind.fork(~11456, viaconfig.addWorkspace): the same leftovers.rename(~8282): the runtime has already renamed the directory and branch on disk. The rename is not reverted, so disk and config disagree.This already happens for every existing
editConfigrejection source (the corrupt-file gate, a lost registration lock, a throwing transform). With #4444, a failed save rejects too, so disk-full or unwritable-root cases hit these paths.Suggested direction
Give each flow one cleanup path for "registration edit rejected" that reuses the existing failure cleanup (
createScratchalready does this at ~5372). Keep it to one design that covers success, failure, cancellation and opt-outs; see #4440 for why per-path fixes did not converge.Refs #4444 (follow-up F1 from its plan).
Generated with
xum• Model:anthropic:claude-opus-5-5• Thinking:high• Cost:$34.57