Released: 2026-08-26
When running multiple parallel tasks in the terminal, coordinating across workflows is now as natural as using team chat.
You can now use @ mentions to reference other Codex tasks directly in your prompt. You can also ask agents from the terminal to inspect existing tasks, spin up new ones, or dispatch messages to active tasks without manual copy-pasting.
When extracting a specific snippet or excerpt from a long reply, you no longer need to drag your mouse across terminal text. Running /copy now opens an interactive picker menu. You can explicitly choose between copying the full response, individual code blocks, or markdown blockquotes.
When juggling multiple background tasks in your terminal, identifying what each task is doing is now effortless.
Unnamed terminal tasks now automatically receive descriptive titles based on context. Furthermore, running /rename inspects your conversation history and suggests an editable title proposal, saving you from having to type names manually.
Reviewing responses that contain hyperlinks is now much cleaner in your terminal. On supported terminals, Markdown links render as tidy, clickable text labels. In environments without hyperlink support, the full visible URLs remain intact so you can still read and copy them easily.
You can bind cycling through permission modes to a custom shortcut. A permission mode determines how Codex handles confirmation before an action, and the bound key moves through the available modes in sequence.
When your work calls for a different confirmation approach, press the shortcut you configured to cycle to it.
When the same text change is needed in several places in Vim mode, pressing . now repeats your last edit. Vim mode is a keyboard-command way to edit text; . reruns the edit you just completed, so you do not need to type that same change again.
This is the editing-repeat feature for Vim mode.
When cancelling an active turn midway, automated cleanup and custom actions can now take over immediately.
Codex introduces Interrupt hooks. When an active top-level conversation turn is interrupted, these hooks can automatically run shell commands or trigger MCP handlers, allowing your environment to safely persist state or release held resources instead of abruptly terminating.
Opening repositories from unknown sources is now significantly safer. Untrusted projects are no longer allowed to supply project-level AGENTS.md instruction files, preventing malicious repositories from hijacking AI behavior with embedded prompts. Additionally, managed deny-read security rules remain strictly enforced even after interactive permission changes occur during a session.
Exporting diagnostic reports for troubleshooting is now much safer. Codex has expanded credential redaction across app-server diagnostics. Sensitive data is now automatically masked across provider details, authentication refresh flows, and hardware attestation fields. When you share debug output with maintainers, secret tokens and verification payloads will never leak into log files.
Connecting to remote tool extensions is now much more dependable.
Codex has resolved issues with remote MCP server connections. The system now reliably looks up bearer tokens and initializes required servers without failure, all while maintaining full backwards compatibility with older executor environments.
Running Codex on Windows with elevated permissions or special account names is now completely smooth. Previously, if a user profile folder contained non-ASCII Unicode characters (such as accents or non-Latin scripts), launch aliases would often fail. Elevated sandbox environments running with administrator privileges also encountered initialization errors. Both path handling and privilege escalation setups are now resolved.
Exiting Codex on Unix systems like macOS or Linux is now fast and reliable. Previously, shutting down could hang indefinitely if detached background processes still held terminal handles or if output buffers were completely full. The shutdown sequence now cleans up detached processes and flushes full buffers properly.
Running long conversations and multi-agent setups through cloud platforms is now much more reliable.
For developers using Amazon Bedrock models, Codex has fixed errors during conversation compaction when compressing long context windows. Multi-agent compatibility issues have also been resolved, allowing multiple sub-agents to collaborate smoothly.
Reference other Codex tasks with `@` mentions, and ask agents to read, create, or message tasks from the terminal. (#40308, #40315)`/copy` now offers a picker for full responses, individual code blocks, and blockquotes. (#39997)Unnamed terminal tasks receive descriptive titles automatically, and `/rename` suggests an editable title based on the conversation. (#40492, #40495)Markdown links appear as clickable labels in supported terminals, with visible URLs retained elsewhere. (#40471)Bind shortcuts to cycle permission modes, and use `.` in Vim mode to repeat your last edit. (#39873, #40521)Bind shortcuts to cycle permission modes, and use `.` in Vim mode to repeat your last edit. (#39873, #40521)New `Interrupt` hooks can run commands or MCP handlers when an active top-level turn is interrupted. (#40511)Untrusted projects no longer supply project-level `AGENTS.md` instructions, and managed deny-read rules remain enforced after permission changes. (#39837, #40004)Improved credential redaction in app-server diagnostics, including provider, authentication refresh, and attestation fields. (#39993)Fixed remote MCP bearer-token lookup and required-server startup while preserving compatibility with older executors. (#39926, #39952, #39979)Fixed elevated Windows sandbox setup and launch aliases under Unicode user paths. (#39971, #40570)Prevented Unix shutdown hangs caused by detached processes retaining a terminal or full terminal output buffers. (#40460)Fixed conversation compaction and multi-agent compatibility for Amazon Bedrock models. (#39804, #39825)