What Is Codex Record & Replay (Recording to Create Skills)
How Record & Replay works (from recording to replay)
Record & Replay is a feature added to Codex, the coding-assistant service developed by OpenAI, that records a sequence of actions on your Mac and turns it into a reusable skill. It was released on June 18, 2026, in version 26.616 of the Codex app. Let us first walk through what it is, how it works, how it differs from traditional automation, and what you need to use it.
"Added Record & Replay, a macOS feature that turns a demonstrated workflow into a reusable skill." — from the changelog entry (June 18, 2026 / Codex app 26.616)
What Record & Replay Is (Turning One Demo into a Reusable Skill)
Record & Replay lets you perform your usual task once on the Mac screen, and Codex learns the steps so it can carry them out for you next time. Its biggest feature is that you can teach a task by showing the actual operation once, instead of writing a long set of instructions.
"Record & Replay lets you demonstrate a workflow on your Mac and turn it into a reusable skill." — from the opening of the Record & Replay documentation
The "skill" produced here is not a recording that bakes in the raw operations. It is closer to an editable how-to document, where Codex reads the recorded content and writes out, in words, when to use the task, what inputs it needs, what steps to follow, and how to verify the result. OpenAI itself describes this output as a skill you can inspect and edit.
"Codex turns that demo into an inspectable, editable skill. You control when recording starts and stops." — from the Introducing Record & Replay announcement
The Difference from Traditional RPA (Keeping Intent, Not Coordinates)
Until now, the typical way to automate routine tasks was RPA (Robotic Process Automation). RPA records operations such as "click this coordinate" or "press this button" in order and reproduces the same movements. Because RPA reproduces the recorded coordinates and steps as-is, it tends to break the moment the screen layout changes even slightly.
Record & Replay, by contrast, keeps the intent of the task as text rather than the operations themselves. Because a goal such as "download the expense CSV, sum the amounts, and enter them into the application form" is described in words, Codex is reported to be able to judge and proceed even when the file name, the date, or screen details change. Where RPA traces fixed movements, Record & Replay runs flexibly with an understanding of the intent—that is the fundamental difference between the two.
How It Relates to Codex and Computer Use
Record & Replay is not a standalone product; it is provided as one feature of the Codex app. To run it, you need Computer Use enabled in the Codex app settings. Computer Use is the foundation that lets Codex operate the actual screen, such as a browser or an app, and Record & Replay is built on top of that mechanism.
"Record & Replay is available on macOS. Initial availability excludes the European Economic Area, the United Kingdom, and Switzerland. Computer Use must also be available and enabled." — from the requirements section
Codex was originally OpenAI's service for assisting with writing and editing code. With Computer Use and Record & Replay added, it expanded beyond helping you write code toward taking over routine on-screen tasks themselves. In the same context of delegating work to AI, it also helps to review what ChatGPT (GPT-5) can do and how to use it to see the full map of OpenAI's tools.
How to Use Codex Record & Replay (From Recording to Replay)
Steps to record a skill
The actual operation requires no difficult setup. You start recording from the Codex app screen, work as usual, and stop. Here we look at the recording steps, how the skill is created afterward, and how to replay and edit the skill you made.
Steps to Record a Skill (Plugins → Record a skill)
You start recording from "Plugins" in the Codex app. Choosing "Record a skill" from the "+" menu makes Codex suggest a prompt to confirm what the task is. After you review it, add any needed context, and approve recording, all that is left is to demonstrate your usual task on screen.
You control when recording starts and stops yourself. You can stop from the menu bar or overlay, or by telling Codex you are done. Because Codex observes the screen operations and window contents while recording, starting once your task is fully prepared keeps things tidy.
"Open Plugins in the Codex app." / "When you are done, stop recording from the menu bar, overlay, or tell Codex that you are done." — from the Start a recording steps
How the Skill Is Created Automatically After Recording
When you stop recording, Codex reads the series of recorded operations and drafts the skill. The skill it creates documents, in writing, when to use the task, the inputs it needs, the steps to follow, and how to verify the result.
"The skill explains when to use the workflow, what inputs it needs, what steps to follow, and how to verify the result." — from the Record & Replay documentation
Because of this description, the skill works as a procedure you can reason about, not just a macro that repeats the same movements. Even if the date or target file changes in a monthly task, the intent of the steps remains, so Codex can run it with the context in mind.
Replaying and Editing the Recorded Skill
You can call the finished skill from a new thread and replay (run) it. Because the recorded skill is inspectable and editable, you can review the steps or add explanations to improve its accuracy.
When something does not run well, being able to edit the skill's description directly is the big difference from an opaque macro. By filling in steps whose explanation was vague, or adding a verification step, you make later replays more stable. Rather than aiming for perfection in one go, refining the description as you run it a few times is the realistic approach.
Use Cases and Suitability for Record & Replay
Tasks Record & Replay suits and does not suit
Tasks it suits
- Monthly expense reports and time-off requests
- Downloading recurring reports
- Creating issues in a fixed format
- Publishing and uploading videos
- Routine bookings such as reserving a parking space
Tasks it does not suit
- Tasks whose steps change every time
- Sites whose layout changes frequently
- Tasks done only once
- Tasks with vague success criteria
- Tasks that unavoidably require sensitive input
Record & Replay is not an all-purpose automation tool; its strengths are clear. OpenAI explains that it suits tasks that are repetitive, depend on your preferences, and are easier to show than to describe. Here we organize the concrete uses, the tasks it does not suit, and the cautions before you use it.
Routine Tasks It Suits (Expense Reports, Report Downloads, and More)
The examples OpenAI gives are filing expenses, booking a parking space, creating a configured issue, publishing a video, and downloading a recurring report. The more stable the steps and the clearer the criteria for whether it succeeded, the more Record & Replay pays off on repetitive tasks.
"Use it when the workflow is repetitive, depends on your preferences, or is easier to show than to describe in a prompt." — from the Record & Replay documentation
Such tasks tend to get long when you try to explain the steps accurately in words. Showing them once gets the point across, so you save the very effort of crafting a prompt. If you have routine clerical work that comes up every month, once you finish the first recording, later runs are just a matter of calling the skill.
Tasks It Does Not Suit (One-off Tasks, Frequently Changing Screens)
Conversely, there are tasks where the effect is hard to see. For tasks whose steps change every time, sites whose screens change frequently, or tasks done only once, the effort of recording and turning it into a skill does not pay off. Tasks with vague success criteria are also less stable, because Codex has trouble verifying the result.
The official guidance, too, says it shows its strength most when the steps are stable and the success criteria are clear. In other words, tasks that do not meet those conditions may be faster done by a person rather than forced into automation. When considering adoption, the shortcut is to judge first whether "the task follows the same steps each time and you can clearly tell whether it finished."
Cautions and Drawbacks to Know Before Using It
It is a useful feature, but its use comes with some prerequisites and constraints. It runs on macOS only, requires Computer Use to be enabled, and during recording you must take care not to show passwords or other sensitive data on screen.
"Use realistic inputs, but avoid secrets and sensitive data." — from the guidance on recording
There are also regional limits. The European Economic Area (EEA), the UK, and Switzerland are outside the initial scope, while Japan is among the regions that can use it. It also does not currently support Windows. These constraints are worth checking before deciding whether to adopt it at work. If you share and operate recorded tasks within a company, it is reassuring to also confirm your workplace's rules on handling sensitive information.
Requirements and Summary for Record & Replay
Where it fits by type of user
Finally, we organize the requirements worth knowing when adopting Record & Replay and the overall picture of its usage conditions. The table below lists the supported OS, the required setting, the available regions, and the release timing.
A List of Supported OS, Required Setting, and Regions
The conditions to check before adopting it are easier to grasp in a table. You can use it only when macOS, an enabled Computer Use, and an available region all line up.
| Item | Details |
|---|---|
| Supported OS | macOS only (Windows not supported) |
| Required setting | Enable Computer Use in the Codex app |
| Available regions | EEA, UK, and Switzerland excluded at launch (Japan can use it) |
| Release timing | June 18, 2026 (Codex app 26.616) |
For enterprise use, an arrangement where an administrator enables Computer Use through a configuration file is also envisaged. For personal use, enabling it yourself from the Codex app settings is all the preparation you need.
The Difference Between Three Automation Approaches
Record & Replay is not the only way to delegate routine tasks. Instructing with a prompt each time, traditional RPA, and Record & Replay each shine in different situations. Knowing the differences makes it easier to judge which fits your task.
| Aspect | Prompt instructions | Traditional RPA | Record & Replay |
|---|---|---|---|
| How you teach it | Describe steps in text | Record operations | Demonstrate and record the task |
| Resilience to screen changes | Adapts via context | Weak to layout changes | Adapts, as it keeps intent |
| Inspect and edit | Edit the prompt each time | Needs setting adjustments | Inspect and edit the skill |
| Tasks it suits | Tasks that change each time | Fully fixed-procedure tasks | Stable, repetitive tasks |
When you want AI to organize materials or manuals you have on hand, formatting them into Markdown in advance keeps the heading and table structure intact and improves how accurately they are read. When you want to format a web page directly, the following tool helps.
The text and files you enter are processed within your environment and are not sent to sakutto's servers, so you can format work materials with peace of mind. To widen the range of tasks you delegate to AI, reading the practical guide to local LLMs and Claude (Anthropic's generative AI) explained alongside this helps you tell the tools apart.
Summary and Official Sources for Record & Replay
Record & Replay is a feature that lets Codex learn a repetitive Mac task by showing it once, so you can hand it off from then on. Keeping the actual operation, intent and all, rather than relying on a long set of instructions, is what sets it apart from traditional automation. If you carry a lot of stable clerical or recurring work, recording that first task is worth a try. On the other hand, it does not suit tasks whose steps change every time or one-off tasks, so judging the use and using it selectively is the realistic approach.
This article is organized based on the following primary sources (official documentation and announcement). Specifications, supported environments, and regions may change, so always check the latest official information before use.