OpenAI Codex release notes: app 26.608 makes Claude migration easier

The June 9 OpenAI Codex release notes are not another big model headline. The practical change in Codex app 26.608 is that migration, plugins, and settings become easier to find and review before a coding agent gets more responsibility. A coding agent is an AI assistant that can read code, change files, and run tools inside a bounded workspace.

OpenAI Codex release notes: app 26.608 is about migration and control

Codex app 26.608 adds Migrate to Codex flows for importing supported setup from Claude Code and Claude Cowork, including during onboarding. The same release revamps the plugins screen with separate tabs, marketplace and category filters, keyboard navigation, and clearer install actions. Settings search also finds more options, including Git and pets.

Source: OpenAI Codex changelog, Codex app 26.608

For Nordic teams, the point is not to “switch agents” in one sweep. The point is visibility. If you already have routines in Claude Code or Claude Cowork, Codex can now help bring over supported setup, but you still need to check what actually moved: plugins, skills, environment variables, permissions, test commands, and review steps.

An approval gate is a point where a human must approve file changes, commands, or access before the agent continues. In practice, migration should go together with scoped API keys, secret managers, least privilege, redacted logs, and clear run receipts.

Codex Mobile gained more work control the same day

The same OpenAI Codex release notes page lists ChatGPT for iOS 1.2026.153. Codex Mobile gained support for choosing a branch, creating a worktree, and running an environment setup script for new threads. It also gained a Codex profile with usage stats and token activity, /goal support for creating and managing goals from mobile, inline review comments, side chat from selected transcript text, and editing the latest sent prompt.

Source: OpenAI Codex changelog, ChatGPT for iOS 1.2026.153

That matters for teams treating Codex as a work queue, not just a chat box. A worktree is a separate working copy of a Git project where the agent can work without immediately disturbing the main folder. Goal mode is Codex's long-running work mode: OpenAI documents /goal <objective> to set a goal, /goal to check it, and /goal pause, /goal resume, /goal clear to control the run.

Source: OpenAI Codex use case: Follow a goal

What Hammer readers should test now

Start with a migration inventory, not a large production workflow. If your organization already has prompts, local skills, plugins, or MCP connections in other agent tools, Codex 26.608 is a good moment to create a receipt: what can move, what needs human control, and what should be rebuilt as a governed routine.

For Tool Forge, this shifts the question away from “which agent is best?” and toward how the work is packaged: the right plugin, the right permission, the right log, the right stop point. That is where agent work becomes reliable enough for real workflows.

Short example: use the new Codex feature

Human step: open Codex app 26.608 and use the Migrate to Codex flow if it is offered for your setup. Then let the agent review the result, rather than asking it to click around inside the app for you.

Review the setup that was migrated into this Codex project. Compare plugins, skills, environment assumptions, and approval rules. Produce a migration receipt: what is ready, what needs a human decision, and which secrets or scoped permissions must be checked before Codex runs tasks. Do not edit files.

Good output should:

• list the plugins, skills, and settings that actually exist, without guessing
• separate safe next steps from decisions a human must make
• call out env vars, secret manager, scoped permissions, output redaction, and logging where integrations are involved
• suggest one small Codex task to run after the review