Haleclipse/codex rust-v0.109.0-cometix on GitHub

Release 0.109.0-cometix

New Features

Releases now ship one-step installers for macOS, Linux, and Windows, so Codex can be installed directly from published scripts instead of relying on npm-first setup. (openai#12740, openai#12741)
Plugins can now be enabled from config.toml, and plugin bundles can contribute skills, MCP config, and app manifests that load automatically at runtime. (openai#12864, openai#13401)
Memories gained more user control: Codex can forget or hard-clear stored memories, skip polluted captures, and optionally write memory files into the workspace. (openai#12900, openai#13008, openai#13085, openai#13467)
Multi-agent work is easier in the TUI: you can launch it from /agent, review cross-agent approvals, get clearer agent names, and receive plan-mode or user-input notifications. (openai#12995, openai#13218, openai#13246, openai#13495)
Realtime voice sessions now have an /audio picker for microphone and speaker selection, with device choices stored across sessions. (openai#12849, openai#12850, openai#13030)
Sandboxing is more flexible: Linux bubblewrap now supports restricted read-only scopes, and read-only policies can preserve network access when explicitly allowed. (openai#12369, openai#13409)

@ file search now follows Git's ignore boundaries, so a parent .gitignore outside the repo no longer hides valid project files. (openai#13250)
Diff and syntax highlighting now respect terminal and theme capabilities more reliably, fixing washed-out Windows Terminal diffs and broken ANSI-family syntax themes. (openai#13016, openai#13037, openai#13382)
/agent flows are more stable: pending messages, status display, Escape handling, profile interactions, and several race conditions were fixed. (openai#13130, openai#13131, openai#13235, openai#13240, openai#13248)
Command-line project trust overrides now correctly update project-local MCP transports instead of being dropped during config parsing. (openai#13090)

The deprecated app-server v1 surface was further removed in favor of v2, and a flat v2 schema bundle is now exported to simplify client code generation. (openai#13324, openai#13364, openai#13375)