CODEX 0.122 HARDENS SANDBOXING AND PLUGINS, SMOOTHS TUI; WATCH FOR UPGRADE SNAGS
OpenAI’s Codex app 0.122 ships stronger sandboxing, revamped plugin workflows, and TUI upgrades, with a few upgrade issues reported by users. The new release o...
OpenAI’s Codex app 0.122 ships stronger sandboxing, revamped plugin workflows, and TUI upgrades, with a few upgrade issues reported by users.
The new release of the Codex app adds deny‑read glob policies, managed deny‑read requirements, and platform sandbox enforcement, plus isolated exec that ignores user config. Plugin workflows get tabbed browsing, inline enable/disable toggles, and support for remote, cross‑repo, or local marketplace sources.
TUI gains quick “/side” conversations and queued input now supports slash commands and bang shell prompts. Tool discovery and image generation are on by default, with better MCP and image metadata handling. Security flows tighten token handling and Windows sandbox permissions.
Around the edges, the OpenAI Agents SDK exposes a UI message stream helper, and langchain-openai 1.1.15 fixes streaming dict items and Azure chat profile inference. Some users report the latest Codex update wiped data and confusion around a new limit system; one post flags a wipe.
Stronger sandbox and filesystem policies cut accidental secret exposure when running toolchains or agent exec.
Plugin and TUI improvements make multi-tool, multi-repo workflows faster to stand up and operate.
-
terminal
Create deny-read glob rules (e.g., /.ssh/, */.env) and confirm Codex exec and Plan Mode cannot read them on macOS/Windows/Linux.
-
terminal
Host an internal plugin in a separate repo and load it via the new remote marketplace source; verify inline enable/disable and TUI /side flows.
Legacy codebase integration strategies...
- 01.
Back up Codex workspaces before upgrading; a few users reported data loss after update; validate migration on a throwaway profile first.
- 02.
Audit and codify filesystem and exec policies; require trusted workspaces; regression-test agents built with LangChain streaming and Agents SDK UI streaming.
Fresh architecture paradigms...
- 01.
Adopt a sandbox-first posture: enable MCP tool discovery by default, isolate exec, and define deny-read globs from day one.
- 02.
Use createAiSdkUiMessageStream to stream tokens into your app UI while Codex or agents run background work.
Get daily OPENAI + SDLC updates.
- Practical tactics you can ship tomorrow
- Tooling, workflows, and architecture notes
- One short email each weekday