Repository navigation
Releases
v0.2.2 — responsive capacity recovery
Faster capacity recovery for large legacy threads, with the final safety check preserved.
- Due retries take priority over background inspections.
- Legacy caching handles in-flight notification changes.
- Unsupported ephemeral threads are reported once per connection.
- Timing logs show where recovery time is spent.
Update:
uv tool install --reinstall git+https://github.com/ustas-eth/codex-retry.git
Restart only the retry runner. The app-server and workers can stay running.
Changes: #3
v0.2.1 — recover execution independently of goal state
Retries capacity-failed execution independently of goal state, preserving paused, completed, and limited goals. --no-resume-blocked-goals now preserves blocked goal status without suppressing execution recovery.
Improves legacy failure detection and recovery after transient reads or lost persistence. Queued failures remain tracked beyond the discovery window, with duplicate-start and uncertain-control safeguards preserved.
Restart only the retry runner. The app-server and Codex sessions can stay running.
Changes: #2
v0.2.0: recover goals and unloaded threads
Adds capacity recovery for recent saved threads, including unloaded work and failures surviving a server restart.
Blocked goals are reactivated with the same assignment, budget, and accumulated usage. Reactivation is best effort because Codex does not expose why a goal was blocked; --no-resume-blocked-goals preserves deliberate blocks. Also adds configurable lookback and read-only dry runs.
Install or update:
uv tool install --reinstall git+https://github.com/ustas-eth/codex-retry.git@v0.2.0Restart only the retry runner. The app-server and Codex sessions can stay running.