Codex CLI 0.96.0: Thread Compaction & Rate Limits
Codex CLI 0.96.0 brings async thread compaction, websocket rate limit events, and unified execution on non-Windows platforms. Plus fixes for TUI notes, thread listing, and tool injection atomicity.
TL;DR
- Async thread compaction API lets you trigger cleanup without blocking
- Websocket rate limit signaling now available via new
codex.rate_limitsevent unified_execenabled on all non-Windows platforms for better execution consistency
New
- Async thread compaction —
thread/compactadded to v2 app-server API as an async RPC trigger, so you can start compaction and track completion separately without blocking your session. - Websocket rate limit signaling — New
codex.rate_limitsevent provides rate limit metadata over websockets with parity for ETag and reasoning metadata handling. - Unified execution on non-Windows —
unified_execnow enabled across all non-Windows platforms for consistent command execution behavior. - Source provenance in requirements — Constrained requirement values now include source information, enabling source-aware debugging in UI flows like
/debug-config.
Fixed
- Esc key handling in TUI notes — When notes are open in
request_user_input, pressing Esc now exits notes mode instead of interrupting the entire session. - Thread listing resilience — Thread listing now queries the state DB first (including archived threads) and falls back to filesystem traversal only when needed, improving correctness and resilience.
- Thread path lookup validation — Fixed thread path resolution to require that the resolved file actually exists, preventing invalid thread-id resolutions.
- Dynamic tool injection atomicity — Tool injection now runs in a single transaction to avoid partial state updates.
- Request rule guidance — Refined
request_ruleguidance used in approval-policy prompting to correct rule behavior.
Under the Hood
- Migrated state DB helpers to versioned SQLite filename scheme and cleaned up legacy state files during initialization.
- Expanded runtime telemetry with websocket timing metrics and simplified internal metadata flow in core client plumbing.
Update via: npm install -g @openai/codex-cli@0.96.0 or check the full release on GitHub.
Source: Codex