OpenAI Codex release notes: app 26.616 records workflows as skills

OpenAI Codex got a very practical app update on June 18: Codex app 26.616 adds Record & Replay, a macOS feature that turns a demonstrated workflow into a reusable skill. For Swedish and Nordic teams, the catch matters: initial availability excludes the European Economic Area, the United Kingdom, and Switzerland. The release note is still worth reading now because it shows where Codex workflows are heading.

OpenAI Codex is OpenAI's coding agent for work in code, apps, and connected tools. A skill is a reusable instruction with steps, inputs, and verification. Computer Use means Codex can see and operate graphical apps under separate permissions.

Source: Codex app 26.616 in the OpenAI Codex changelog

OpenAI Codex release notes: what changed in app 26.616

The main update is Record & Replay for macOS. OpenAI describes it as a way to turn a demonstrated workflow into a reusable skill. The requirements are explicit: Record & Replay requires Computer Use to be available and enabled, and the first rollout does not include the EEA, the UK, or Switzerland.

The same release adds bulk actions to automation run history. You can mark every run as read or archive eligible runs. For teams already letting agents run recurring jobs, this is a small, useful cleanup feature: less noise in the history makes it easier to see which runs actually need human attention.

OpenAI also mentions new deep links for managing SSH connections and improved Browser Use, where visible-tab routing and annotations persist when a draft browser session moves to the server.

Source: Codex app 26.616 in the OpenAI Codex changelog

Why Record & Replay matters for practical automation

Many workflows are hard to describe in a prompt but easy to show once: creating a correctly configured issue, downloading a recurring report, publishing material in the right order, or filling an internal form with the right defaults. Record & Replay is built for that kind of habit. You demonstrate the workflow, Codex observes what it needs, and Codex can then draft a skill with when to use it, which inputs vary, which steps to follow, and how to verify the result.

For Nordic organizations, the availability note is as important as the feature itself. If your team is in Sweden, Norway, Denmark, Finland, or Iceland, this is probably something to prepare for rather than enable for everyone today. The preparation is still concrete: choose one recurring workflow, write down what done means, mark where human approval belongs, and keep secrets in environment variables or a secret manager. That is Tool Forge work: making AI work repeatable, reviewable, and safe enough to connect to real systems.

Source: Record & Replay in OpenAI Codex docs

Short example: use the new Codex feature

Human step first: in an environment where the feature is available, open the Codex app, go to Plugins, open the plus menu, and choose Record a skill. OpenAI's documentation also says to give Codex helpful context, approve recording when you are ready, perform the workflow on your Mac, and then stop recording from the menu bar, the overlay, or by telling Codex you are done.

Then paste a short instruction to Codex after the recording is finished:

Review the workflow I just demonstrated and turn it into a reusable Codex skill for customer follow-up. Write when it should be used, which inputs vary, which steps require human approval, and how the result should be verified. Ask for approval before using other apps, external systems, or saving anything.

Good output should include:

• A skill that explains when the workflow should and should not be used.
• Clear variables, such as customer name, date, channel, and responsible person.
• Approval points before Codex touches external systems or saves changes.
• A short verification note a human can read before the routine runs again.

Source: Record & Replay: start recording and reuse a workflow

What you can test now

If Record & Replay is not available in your region yet, you can still do the prep work. Pick one workflow that repeats every week and write down three things: what the human does, which fields or decisions vary, and what counts as a good result. When the feature becomes available, the recording will be better. And if you later build a plugin or MCP integration, you will already know the permission, logging, and human-control boundaries.

Automation history is also worth checking if you already use Codex Automations. The bulk actions in app 26.616 are not glamorous, but they help teams keep runs readable. That is often where useful agent work gets stuck: not in the first demo, but in the everyday queue where someone needs to see what ran, what is waiting, and what needs review.