Codex CLI 0.131.0: Unified Mentions & codex doctor

Unified @ Mentions Across Files, Plugins & Skills

Previous versions of Codex CLI separated the mention flow by context type — files in one place, plugins in another. In 0.131.0, the @ mention system is fully unified: typing @ in the TUI composer opens a single picker that simultaneously searches files, directories, plugins, and skills, with results backed by app-server plugin metadata for accurate discoverability.

This removes the friction of knowing in advance which category to search. Rather than switching mental modes mid-prompt, developers can type @ and filter across everything at once. The implementation also resolves edge cases where explicit mentions lost context or plugins were filtered unexpectedly, making the feature more predictable across different session configurations.

codex doctor: Built-In Diagnostics

Debugging Codex CLI setup problems — authentication failures, network proxy issues, sandbox misconfiguration, incompatible terminal emulators — previously required digging through logs or manually testing each layer. The new codex doctor command performs a systematic sweep across six diagnostic domains:

• Runtime — verifies the Codex binary and its runtime dependencies are intact
• Authentication — checks API credentials and current session state
• Terminal — tests terminal compatibility and capability detection
• Network — probes connectivity and identifies proxy or firewall issues
• Configuration — validates config.toml syntax and resolves profile conflicts
• Local state — inspects the SQLite state database for corruption or version mismatches

The output is structured for easy sharing with support channels or bug reports, making codex doctor the recommended first step whenever Codex behaves unexpectedly. Running the command gives an instant, organized picture of what is working and what is not.

Richer TUI Status Line

The status line at the bottom of the Codex TUI now surfaces more context about the current session without requiring any extra commands:

• Blended token usage — a combined count across all active model calls, not just the current turn
• Permission and approval mode — the active permission profile (e.g., :workspace, :read-only, :danger-no-sandbox) and approval level are displayed inline
• Effective workspace roots — shows which directories are actually in scope, useful when profiles are inherited or overridden by config

Additionally, Markdown tables rendered in model responses are now displayed responsively within the TUI rather than as raw text, making structured output from models significantly more readable directly in the terminal.

Bug Fixes

Several rendering issues are resolved in this stable release. URL-adjacent prose no longer wraps incorrectly, light-mode selection contrast is improved, Shift+Enter handling in tmux csi-u panes is fixed, Esc no longer dismisses /side conversations, and network approval history now renders correctly by target. On Windows, the sandbox hardening adds deny-read rule parity with macOS/Linux, scopes write-root capability SIDs correctly, and avoids PowerShell stop-parsing forms that caused edge-case failures.