Codex Desktop keeps running Git turn-diff snapshot tasks on stale repos/worktrees and repeatedly fails, likely contributing to window instability
# Codex Desktop keeps running Git turn-diff snapshot tasks on stale repos/worktrees and repeatedly fails, likely contributing to window instability
Summary
On my Windows machine, Codex Desktop continues to run background Git turn-diff / checkpoint tasks against stale repositories, old worktrees, and historical directories that are not the current active workspace.
These tasks repeatedly fail with errors like:
Failed to retain a turn diff tree snapshotFailed to complete turn diff analytics capture
This appears to be a Codex Desktop background workspace/repo-watcher issue rather than a problem in the active project itself. The repeated failures may be contributing to window instability / apparent "flash crash" behavior after the app stays open for a few minutes.
Environment
- Platform: Windows
- Date observed: July 4, 2026
- Active workspace at time of testing:
C:\Users\<user>\Documents\<active-workspace>
Important context
I already identified and fixed a separate local root cause earlier the same day:
- there was an accidental
C:\.git - which caused
C:/to be treated as a repository root - that issue has been removed
After fixing that, Codex no longer treats C:/ as the active repo root, but the turn-diff snapshot failures still continue.
So this report is not about the earlier C:/-as-repo problem.
This is about Codex Desktop continuing to watch and process stale historical repos/worktrees in the background.
Actual behavior
Codex Desktop continues to run background Git analytics / checkpoint operations against directories that are not the current active workspace.
Observed failure pattern includes background commands like:
git update-ref refs/codex/turn-diffs/checkpoints/...- returning
exitCode=128
Observed error messages include:
Failed to retain a turn diff tree snapshotFailed to complete turn diff analytics capture
The failures appear on historical directories such as:
C:/Users/<user>/Documents/<historical-projects>/...C:/Users/<user>/Documents/<historical-projects>/<old-worktree>C:/Users/<user>/.codex/skills/...
At the same time, logs show repeated background watcher/origin scans like:
Starting git repo watcherworker-complete dirCount=15worker-complete dirCount=14worker-complete dirCount=29
This makes it look like Codex is still holding onto a broad set of historical repo contexts and trying to process them even though they are not the active workspace.
Expected behavior
Codex Desktop should:
- scope Git turn-diff / snapshot work to the currently relevant active workspace(s)
- avoid repeatedly processing stale historical repos or old worktrees unless explicitly reopened
- gracefully skip incompatible or non-standard repos/worktrees
- avoid repeated noisy failures when internal checkpoint refs cannot be updated
- ensure these background failures do not affect renderer/window stability
Why this seems like a Codex Desktop issue
This does not appear to be a normal user Git workflow failure.
Reasons:
- The failing operations are Codex internal background tasks, not user-invoked Git commands.
- The failures occur mostly on stale repos / old worktrees / historical directories, not the active workspace.
- Even if a historical repo is in an unusual state, Codex should degrade gracefully instead of repeatedly failing in the background.
So the stale repo state may be a trigger, but the robustness issue appears to be in Codex Desktop's background repo-watching / turn-diff handling.
Possible areas to investigate
- source of repo watcher directory enumeration
- why stale historical repos/worktrees remain eligible for turn-diff analytics
- whether old worktrees and non-active repos should be automatically excluded
- retry / backoff / cleanup behavior after
update-refcheckpoint failures - whether failed turn-diff tasks can accumulate and impact renderer stability over time
Impact
- persistent background log noise
- unnecessary Git activity on non-active directories
- possible memory / task buildup over time
- possible contribution to the app window disappearing / becoming unstable after several minutes
Notes
There is also a separate remote SSH auto-reconnect issue on this machine, but that is a different problem and should be tracked separately.
This issue is specifically about background Git turn-diff snapshot failures on stale repos/worktrees.