You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Codex Desktop recents timestamp appears to use thread creation time instead of last active time #21960
Codex Desktop appears to show a stale relative timestamp for an active/resumed thread. The visible recents/corner timestamp can read like the session was last run several days ago even while the thread is actively being used.
Observed behavior
A Codex Desktop thread created several days earlier was resumed and actively used today.
The local session storage for the thread was modified/accessed today.
The UI still displayed a relative timestamp such as 3 days ago, which appears to match the thread creation age rather than last activity.
Expected behavior
The visible recents/corner timestamp should represent the thread's last active / last run time, not the original creation time.
If creation time is useful, it should be shown separately from activity time.
Possible fix
Use a last-activity field for the visible timestamp, for example:
When a resumed thread is actively being used but still says something like 3 days ago, it makes the session look stale or inactive even though work is happening in it.
This older Desktop recents symptom looks related to a concrete thread/read metadata bug that was reproduced on current main: when thread/read includeTurns=false replaces the SQLite-backed thread row with a rollout-derived preview, it can lose SQLite updated_at and fall back to rollout creation/summary metadata.
The patch preserves SQLite authoritative updated_at/recency_at/archived_at during no-history preview merging and uses rollout file mtime as the direct-rollout fallback when no summary updated_at exists. I could not open an upstream PR directly because GitHub returned GraphQL: 100yenadmin does not have the correct permissions to execute CreatePullRequest, but the compare branch is published.
Activity
github-actions commentedon May 9, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
100yenadmin commentedon Jun 18, 2026
This older Desktop recents symptom looks related to a concrete
thread/readmetadata bug that was reproduced on currentmain: whenthread/read includeTurns=falsereplaces the SQLite-backed thread row with a rollout-derived preview, it can lose SQLiteupdated_atand fall back to rollout creation/summary metadata.Patch branch: https://github.com/openai/codex/compare/main...100yenadmin:codex/thread-read-updated-at?expand=1
Tracked in: #28870
The patch preserves SQLite authoritative
updated_at/recency_at/archived_atduring no-history preview merging and uses rollout file mtime as the direct-rollout fallback when no summaryupdated_atexists. I could not open an upstream PR directly because GitHub returnedGraphQL: 100yenadmin does not have the correct permissions to execute CreatePullRequest, but the compare branch is published.100yenadmin commentedon Jun 18, 2026
@tibo-openai