OpenAI Codex CLI · Release notes
OpenAI Codex CLI Updates: Instant Interrupts, Guardian Reviews, and Plugin Caching
The latest OpenAI Codex CLI releases introduce instant interrupt steering, opt-in Guardian reviews for agent handoffs, and performance improvements through plugin manifest caching. Here is what matters for developers writing and using AI coding skills.
OpenAI released a rapid sequence of Codex CLI updates between September 28 and October 1, 2026, spanning versions rust-v0.158.0 through rust-v0.160.0. These updates focus on improving interactive control, enhancing agent context management, and optimizing performance for developers who rely on the CLI for daily coding tasks.
For developers writing or using skills, the most significant changes involve the new instant_interrupt feature, the expansion of Guardian review capabilities, and performance improvements in plugin loading. These updates affect how agents handle user input, manage context during handoffs, and execute complex workflows.
Instant Interrupts for Real-Time Steering
Version rust-v0.159.0 introduced the opt-in instant_interrupt feature, which allows new user input to steer Codex during model responses or long-running code-mode calls. This change addresses a common pain point where developers had to wait for a model to finish generating a response before they could correct its direction or provide additional context.
The feature works by preempting model responses when new user input arrives, ensuring that the agent can immediately adjust its behavior based on the latest instructions. This is particularly useful for complex coding tasks where the initial prompt might need refinement as the agent progresses.
- Enable
instant_interruptto allow real-time steering of model responses - Prevents the need to wait for long-running code-mode calls to complete
- Improves the interactive experience for developers who prefer to guide agents step-by-step
Guardian Reviews and Agent Handoffs
Version rust-v0.160.0 added opt-in Guardian review capabilities that can retrieve earlier user instructions and include context from agent handoffs. This enhancement allows agents to maintain a more coherent understanding of the project's history and user intent, even when tasks are handed off between different agents or sessions.
The Guardian review system now preserves encrypted agent messages and includes handoff-aware root context, ensuring that sensitive information is handled securely while still providing the necessary context for subsequent agents. This is crucial for multi-agent workflows where context continuity is essential for maintaining quality and consistency.
- Opt-in Guardian reviews retrieve earlier user instructions
- Includes context from agent handoffs for better continuity
- Preserves encrypted agent messages for security
Performance Improvements in Plugin Loading
Version rust-v0.160.0 also included performance improvements that reduce repeated plugin-loading work by caching parsed manifests and reusing HTTP connections for remote plugin requests. These changes help speed up the startup time of Codex CLI, especially when working with a large number of plugins.
Additionally, background reclamation of unused log database space was added to reduce disk usage, ensuring that the CLI remains efficient over time. These optimizations are particularly beneficial for developers who use Codex CLI in CI/CD pipelines or other automated environments where performance and resource usage are critical.
- Caches parsed plugin manifests to reduce loading time
- Reuses HTTP connections for remote plugin requests
- Reclaims unused log database space in the background
Windows and macOS Sandbox Fixes
Several bug fixes were included in the rust-v0.159.0 and rust-v0.160.0 releases to address issues with Windows and macOS sandboxes. These fixes include suppressing unwanted console windows from background helpers, repairing long-path permissions on Windows, and fixing macOS TLS access in network-enabled sandboxes.
These changes ensure that Codex CLI works reliably across different operating systems, which is important for developers who work in heterogeneous environments. The fixes also improve the stability of the CLI when running complex workflows that involve multiple agents and tools.
- Suppresses unwanted console windows on Windows
- Repairs long-path permissions on Windows
- Fixes macOS TLS access in network-enabled sandboxes
What to Do Next
Developers who use Codex CLI should update to the latest version to take advantage of these improvements. The instant_interrupt feature and Guardian review capabilities are opt-in, so users should review the documentation to understand how to enable and configure these features.
For those writing skills, the performance improvements in plugin loading and the enhanced context management in Guardian reviews provide a better foundation for building complex, multi-agent workflows. Developers should also review the bug fixes to ensure that their workflows are not affected by any known issues.
- Update to the latest Codex CLI version
- Review documentation for
instant_interruptand Guardian reviews - Test workflows to ensure compatibility with the latest changes
Skills to try
- codex-review: Run an independent code review using the OpenAI Codex CLI in headless mode. Gets a second opinion from a different model family (the current Codex models) on recent changes, a PR, …
- handshake: Persistent, branch-aware context handoffs for coding agents. Handshake distills long, messy sessions into structured project state, letting Claude Code, Codex, and other agents sta…
- autoreview: Structured Codex, Claude, Amp, Pi, or Kimi code review when explicitly requested.
- code-review: Thorough code review — checks correctness, security, performance, readability, and test coverage. Gives actionable feedback ranked by severity.
Sources
- OpenAI Codex CLI rust-v0.160.0
- OpenAI Codex CLI rust-v0.159.3
- OpenAI Codex CLI rust-v0.159.2
- OpenAI Codex CLI rust-v0.159.1
- OpenAI Codex CLI rust-v0.159.0
- OpenAI Codex CLI rust-v0.158.0
Written by
Skills-Hub Team
Skills-Hub is the open registry for AI coding skills, with SKILL.md files synced daily from Anthropic, Google, Microsoft, and 90+ official sources. Free + MIT.