OpenAI Codex release notes: Appshots and Goal mode in Codex 26.519

Adam Olofsson HammareAdam Olofsson Hammare
OpenAI Codex release notes: Appshots and Goal mode in Codex 26.519

OpenAI Codex got a very practical update on May 21. This is not just another pile of small UI fixes. OpenAI is making Codex easier to use as a managed workbench: send context from another app, give Codex a longer goal, follow the work and keep access under control.

OpenAI Codex is OpenAI's coding agent for reading, writing, testing and proposing code changes. Codex CLI is the terminal version of that agent. A sandbox is a bounded environment where the agent can work without free access to the whole computer, and an approval gate is a step where a human must approve sensitive actions.

OpenAI Codex release notes and changelog for May 21

The new app version, Codex 26.519, adds Appshots to the Codex app on macOS. Press both Command keys to send the frontmost app window to Codex with a screenshot and available text. That means you no longer have to describe a UI bug from memory. Codex can see what you see.

The same changelog says Goal mode is no longer experimental. It is now available in the Codex app, IDE extension and CLI. OpenAI documents it as a slash command: set the goal with /goal <objective>, check status with /goal, and use /goal pause, /goal resume or /goal clear when you need to control the run.

Source: OpenAI Codex changelog, Codex 26.519

Source: OpenAI Codex use case: Follow a goal

Codex CLI 0.133.0 makes the same shift easier to control

For terminal users, Codex CLI 0.133.0 is the important line. Install it with:

npm install -g @openai/[email protected]

The release notes say goals are now enabled by default, with dedicated storage and progress tracking across active turns. codex remote-control behaves more like a normal foreground command: it waits for readiness, reports machine status and keeps explicit start and stop commands. Permission profiles gained list APIs, inheritance, managed requirements.toml, runtime refresh and stronger Windows sandbox support.

That sounds dry. But for a team that wants a coding agent to work longer than one question, these are the details that decide whether the work can be followed, stopped and reviewed.

Source: OpenAI Codex changelog, Codex CLI 0.133.0

Source: GitHub release rust-v0.133.0

What this means for Nordic teams

If you build automations, internal tools or customer portals, the point is simple: Codex is becoming less like an answer box and more like a work queue. That fits when a developer, product owner or technical project lead wants to give the agent a goal, let it work through several steps and then review diffs, tests and decisions.

Start with real but bounded work: a broken onboarding screen, a stronger test suite or an internal report that needs more reliable data loading. When Codex needs tools, environment variables or external systems, use scoped API keys, a secret manager, least-privilege access, approval gates and logs. That lets the agent do useful work without dropping passwords or production access into a chat thread.

For Hammer readers, this belongs in Tool Forge: build the workflow around the agent, not only the prompt.

Short example: use the new Codex feature

Paste this as a /goal command in Codex after updating to CLI 0.133.0, or when running Codex 26.519. Do not type a normal prompt that says "use Goal mode"; OpenAI's docs show that you set the goal with /goal <objective>.

/goal Get our customer portal ready for a demo. Read the README and test commands first. Make a short plan, change things in small steps, pause and ask for approval before touching auth, payments or environment variables. Stop when the demo flow works, tests have run and you have summarized changed files and remaining risks.

Good output should include:

  • A plan before larger changes.
  • Small diffs that can be reviewed one by one.
  • Clear pauses when Codex needs permissions or secrets.
  • A final note with tests, files and open questions.

Watch the next Codex changelog

This is worth tracking because OpenAI is now tying the app, IDE and CLI around longer agent work. The next thing to watch is whether Goal mode, Appshots and permission profiles get more team administration: policies, audit trails and better reporting. That is where Codex becomes easier to introduce in real organizations, not just in one developer's terminal.

The Forge newsletter

Get new articles in your inbox

Pick the topics you care about. No noise, at most one email a week.

Get new articles in your inbox

We follow GDPR. Unsubscribe anytime.