Codex CLI: Instant Interrupt Steering & Compact Welcome Screen
CodexCodex CLI 0.159.0 added an opt-in instant_interrupt setting that lets new input steer Codex immediately during model responses or long-running code-mode calls. The release also introduced a compact welcome screen, richer native Mermaid diagram rendering and transcript scrolling while reviewing plans. Fixes cover stray Windows console windows, Markdown-preserving copy and drafts kept when switching tasks.
Key Takeaways
- Instant interrupt is opt-in, letting new input steer Codex right away during model responses or long-running code-mode calls.
- New sessions show a compact welcome screen with consistent headers and occasional tips during and after turns.
- Native Mermaid rendering now handles more flowchart edges, labels and node groups directly in the terminal.
- Developers can scroll the transcript while deciding whether to implement a plan, which helps when reviewing long outputs.
- Windows console flashing was fixed for MCP servers, code-mode hosts and piped commands, with 0.159.2 covering background processes.
- Copied transcript selections keep Markdown tables and whitespace, and blank sessions retain drafts when switching tasks.
Sources & Mentions
1 external resource covering this update
Steering Codex Mid-Response
Codex CLI 0.159.0, released on September 29, 2026, added an opt-in instant_interrupt setting that changes how new input reaches the agent while it is working. With it enabled, a message typed during a model response or a long-running code-mode call steers Codex immediately instead of waiting for the current step to finish. For developers who watch the agent head in the wrong direction, this shortens the gap between noticing a problem and correcting it.
The feature builds on the existing steering behavior in the CLI, where pressing Enter during a running task injects new instructions into the turn. Because instant interrupt is opt-in, the default behavior is unchanged and developers can try it when they want faster course corrections.
A Cleaner Start and Better Terminal Reading
New sessions now open with a compact welcome screen and consistent headers, with occasional tips shown during and after turns. The warnings viewer dismisses reviewed warnings when it is closed, and pressing k keeps one for later. Developers can also scroll the transcript while deciding whether to implement a proposed plan, so they can re-read earlier output before approving.
Native Mermaid rendering in the terminal supports more flowchart edges, labels and node groups, which makes diagrams in Codex responses more likely to display inline instead of falling back to source text.
Fixes Worth Knowing About
- Windows launches no longer flash stray console windows for MCP servers, code-mode hosts and piped commands. Version 0.159.2 suppressed the remaining flashes when Codex starts background processes.
- Copying transcript selections now preserves Markdown tables, formatting and significant whitespace.
- Blank sessions keep drafts when switching tasks, and threads can be archived and listed before their first turn.
- Local ChatGPT sign-in opens the browser reliably, and onboarding offers a shortcut to copy the login link.
- Approved commands retain explicit filesystem denials, and
.awsdirectories are protected by default under writable roots. - macOS TLS access is fixed in network-enabled sandboxes and in remote environments that require proxy access.
Removals
Codex removed automatic follow-up prompt suggestions along with the tui.prompt_suggestions setting, and dropped the bundled plugin-creator skill. App-server clients also gained the ability to paginate thread history from a specific item.