Summary
Version 0.167.0 of the OpenAI Codex CLI (Rust build) adds an agent command center, cross-backend session switching, image previews in the composer, Google Cloud Vertex AI onboarding, experimental system-proxy support, and a set of reliability fixes. Under the hood, the release advances Code Mode output bounds, Guardian review of exec/command activity, elastic Guardian context budgeting, and MCP connection management.
Full diff: rust-v0.166.0...rust-v0.167.0 in the openai/codex repository.
Key Points
- Agent command center: opened with
codex agents --no-daemon; works in embedded sessions and marks unread results (#58238, #58239, #58263, #58264). Tasks are sorted by recency within status groups (#58714), and pending approvals are prioritized in the subagent picker (#58720). - Multi-backend sessions: the terminal session picker can switch between connected backends while preserving drafts (#58280, #58457); resume-picker pagination continues across backends (#58262).
- Onboarding & auth: API key and cloud provider options are grouped under “Sign in another way” (#58427);
gcp_setup_wizardconfigures Vertex AI via Application Default Credentials or a credential file (#57857); provider OAuth sign-in timeout extended to ten minutes (#57859). - TUI quality-of-life: local image thumbnails while composing and clickable image paths in the transcript (#58261, #58446); inline patch review during approval prompts (#58281);
Ctrl+Lclears the display while transcript history stays reachable viaCtrl+T(#58266); OSC 52 clipboard forwarding through GNU screen (#58716). - Experimental: opt-in macOS/Windows system proxy settings from
/experimental(#58157); experimental status-line UI tree renderer (#57970). - Reliability fixes: chats survive model-provider changes (transcripts readable, prompt editable if a chat can't resume, #58202); network approval prompts resolve when requests end and serialize per endpoint (#58350, #58351); realtime V3 transcripts persist to history (#58221); context compaction accounts for retained output from failed responses and responses missing token usage (#58404).
- Docs: app-server README documents managed network constraints under and prompt-user command rules requiring fresh human approval; the openai-docs skill recommends documented replacements when migrating off deprecated models.