OpenAI Codex release notes: alpha 0.142 makes waiting states easier to control

OpenAI Codex release notes got a small but useful CLI signal overnight: 0.142.0-alpha.8 lets some slash commands work while Codex already shows a running task or waits for MCP startup. Narrow? Yes. But for teams using Codex with several tools, remote environments, or slow MCP startup, this is the kind of friction that decides whether agent work feels controllable or stuck.

OpenAI Codex release notes: what 0.142.0-alpha.8 changes

OpenAI published the GitHub pre-release rust-v0.142.0-alpha.8 on June 21, 2026. The release page is short, but the tagged commit explains the change: the Codex TUI should no longer block every settings command just because the interface treats an active task or MCP startup as a running task.

Codex CLI is OpenAI Codex's local coding agent for the terminal. TUI means the terminal interface you work in when you run codex. MCP, Model Context Protocol, is a way to connect an agent to extra tools and data sources without hard-coding every integration.

The verified new behavior is:

• /resume is allowed while a task is running, but still blocked while a real agent turn is active.
• /model, /permissions, /personality, and service-tier commands such as /fast can be used while task mode is active.
• Model selection, reasoning effort, and other turn settings are captured when a turn starts. If you change them during an active turn, they apply to the next turn, not the one already running.
• The status bar may update immediately, even while the active turn still uses its earlier settings.

Source: OpenAI Codex GitHub release rust-v0.142.0-alpha.8, commit d667082 on resume and settings during MCP startup, OpenAI Codex CLI documentation.

Why it matters for Nordic automation teams

This is not a big product launch. It is a control improvement. When Codex waits for MCP servers, remote environments, or slow tool startup, you want to adjust the next step without tearing down the whole session.

For practical AI workflows, it means three things:

• You can switch model or permission level for the next turn when the task needs different accuracy or more tools.
• You can use /resume in more waiting states instead of restarting and losing context.
• You get a clearer boundary between the current work and the next instruction: the change affects the next turn, not the one already running.

For Tool Forge-style work, the point is simple: give the agent useful tools, but make access visible. Use scoped permissions, environment variables or secret managers for sensitive keys, redact sensitive output from logs, and route larger changes through human approval before they merge.

Source: the commit behavior note on next-turn settings and status bar updates.

Alpha means: test it, but do not build the routine on it yet

0.142.0-alpha.8 is a GitHub pre-release. npm still has 0.141.0 as latest, while 0.142.0-alpha.8 is on the alpha channel. That is a useful signal, but not the same as telling the whole team to move its stable Codex routine today.

The best test now: run it in a separate development environment, start a Codex session with MCP or remote tools, and check whether you can adjust model, permissions, or service tier while the task is still initializing. Write down what happened in the run log.

Source: @openai/codex npm registry data, GitHub pre-release page.

Human step first

Open an existing Codex CLI session or start a separate test session. If the interface is stuck in task mode or MCP startup, try the verified slash commands in the TUI: /resume, /model, /permissions, /personality, or a service-tier command such as /fast. Do not put those commands in the agent prompt; they are human TUI commands.

Source: commit d667082 lists the allowed commands and the active-agent-turn limit.

Short example: use the new Codex feature

After you have done the human step above, the next agent prompt can stay short:

Treat this as the next Codex turn after I changed model or permissions.
Inspect the current repo state and summarize:
- what task state we are continuing from
- which files you need to inspect before editing
- any MCP/tool startup or permission issue that should be logged
- the smallest safe next diff to propose

Good output should:

• show whether the agent is continuing existing work or needs a new plan,
• name the files or tools it wants to check before writing code,
• separate MCP or permission issues from the code task itself,
• propose a small next change that a human can review.

What we watch next

The next important signal is whether the same behavior lands in a stable Codex CLI release after the alpha channel. Then the question is not only "can I change a setting while something starts?" but how teams should document model choice, permissions, and run logs as a normal part of agent operations.