OpenAI Codex release notes: app 26.609 improves browser debugging

OpenAI Codex received a genuinely useful app update on June 11. This is not a new model release, and it is not mainly a Codex CLI story. The useful part of the Codex app 26.609 release notes is that browser debugging, computer control, and project instructions become easier to govern for teams that want a coding agent in day-to-day work.

• Source: OpenAI Codex changelog, Codex app 26.609

OpenAI Codex release notes: what changed in app 26.609

OpenAI Codex is OpenAI's coding agent for local projects, cloud threads, and app-based development work. A coding agent can read code, suggest changes, and sometimes run tools, but every integration still needs a clear sandbox and an approval gate. A sandbox limits what the agent can touch; an approval gate is the point where a human approves sensitive actions.

In 26.609, three changes are worth testing first:

• Developer mode for Browser use in Chrome and the Codex in-app browser. OpenAI says this gives Codex controlled Chrome DevTools Protocol access for performance profiling and deeper debugging of network traffic, console output, runtime errors, and page state.
• The /init command is now available in the app composer for creating project instructions with the same initialization workflow as Codex CLI.
• Computer Use expands: Enterprise users get Computer Use outside the European Economic Area, the United Kingdom, and Switzerland, and Windows gets support for per-app access controls.

OpenAI also lists rate-limit reset banking for Plus and Pro, faster Browser use, clearer activity summaries, improved plugin management, better usage-limit errors, and the Cmd + Enter and Ctrl + Enter shortcuts for submitting custom approval feedback.

• Source: OpenAI Codex changelog, June 11, 2026

Why Developer mode matters for web teams

Developer mode for Browser use is the part I would test first. OpenAI says Codex now gets controlled CDP access for deeper debugging. That is concrete: the agent does not just say a page "feels slow". It can inspect network traffic, console output, runtime errors, and page state while working with a local or public page.

OpenAI's in-app browser documentation says the browser is designed for local development servers, file-backed previews, and public pages that do not require sign-in. Browser use requires the Browser plugin to be installed and enabled. The same documentation shows that you can reference the browser with @Browser, including a prompt that asks Codex to capture a performance trace and inspect network traffic.

• Source: OpenAI In-app browser documentation

For Nordic teams, the practical use case is simple: use Codex when a human and an agent need to inspect the same rendered page. Let the agent gather technical observations, but keep changes to authentication, payments, environment variables, and external integrations behind human approval. That fits a Tool Forge setup where env vars, secret managers, scoped API keys, redacted logs, and audit trails are in place before the agent gets more responsibility.

Short example: use the new Codex feature

Human step: open the project in Codex app 26.609, make sure the Browser plugin is installed and enabled, and open the local page in the in-app browser. If you only want to create or refresh project instructions, use /init in the app composer, because 26.609 adds that command there.

• Source: OpenAI Codex changelog, 26.609 and OpenAI In-app browser documentation

Then copy a short agent instruction:

Use @Browser to capture a performance trace for the local page that is already open. Inspect network traffic, console output, runtime errors, and page state. Name the likely bottleneck, propose the smallest safe patch, and ask for approval before changing auth, payment, environment variables, or external integrations.

Good output should include:

• A clear observation from Browser use, not just a guess from the code.
• Exact files or components that should change.
• A small proposed patch or plan, not a broad rewrite.
• A separate approval question if the change touches permissions, secrets, or external systems.

What to check before scaling it up

This is a useful release for teams already testing Codex in web work. Start with an internal interface, a public test page, or a local preview. Run one clear debugging task. Save the agent's observation log. Compare it with what a developer would have checked manually.

I would not start by giving Codex broad access to every app. Use the per-app controls on Windows where available, keep Computer Use scoped, and document which websites, apps, and keys are allowed. That makes Codex less like another loose chat tab and more like a workbench where agent work can be reviewed.