Codex Can Import Workflows From Other AI Agents

OpenAI is giving ChatGPT Work and Codex a way to bring over an existing agent workflow instead of starting from scratch. The ChatGPT desktop app can import supported projects, chats, skills, and plugins, review import history, and opt into automatic updates from the Settings surface. (OpenAIDevs on X)

The durable ChatGPT import documentation makes the scope broader than a file copy. Depending on the source and destination, the importer can carry over instruction files, settings, skills, plugins, project folders, recent chats, MCP configuration, hooks, slash commands, and subagents. Existing agent setup remains unchanged.

Key Takeaways

• The ChatGPT desktop app imports from Claude Code, Claude Cowork, and Cursor.
• Codex CLI imports from Claude Code and Cursor.
• The import surface includes work and configuration: projects, recent chats, instruction files, settings, skills, plugins, MCP configuration, hooks, slash commands, and subagents.
• Desktop users can enable automatic updates to keep imported work in sync with the original agent.
• Import history is reviewable from Settings > Import.
• Importing is non-destructive. The existing agent setup is not changed or deleted, although imported plugins and connections may need authorization.

What Is Portable

OpenAI's documentation separates the items that move into their corresponding ChatGPT or Codex surfaces:

The importer also detects setup that needs a follow-up authorization step. After import, the desktop app can show a status card identifying plugins or connections that still need setup. (ChatGPT Learn, “Import from another agent”)

This is useful because an agent workflow is rarely just a prompt file. Team conventions may live in repository instructions, skills, hooks, MCP settings, and project memories. OpenAI's import model tries to preserve those pieces while mapping them to the destination's equivalent configuration.

Desktop App Versus Codex CLI

The source-agent list differs by destination:

In the desktop app, the setup flow is Settings > Import. If the Import section is not visible yet, the documentation says to open General and find Import other agent setup. Choose the source agents, select the items to bring over, and continue. (ChatGPT Learn, “Import from another agent”)

Codex CLI uses the /import command. The documentation says to choose Claude Code or Cursor, select the supported setup, project files, and recent chats, then review the imported configuration before continuing. The command is unavailable during a running task, in a remote session, or while connected to a local app-server daemon. (ChatGPT Learn, “Import from another agent”)

Automatic Updates and Import History

The desktop app can keep imported work connected to its source. Open Settings > Import and turn on automatic updates; OpenAI says this keeps imported work in sync with the original agent. The same section provides import history so you can review what has been brought over and how the sync has progressed. (OpenAIDevs on X; ChatGPT Learn, “Import from another agent”)

Automatic updates are most useful when the source workflow is still changing. They can reduce the cost of re-importing a repository's skills or project setup, but they do not remove the need to review permissions, MCP credentials, hooks, or prompt templates that behave differently after migration.

Limits and Review Checklist

The documentation says Codex CLI imports up to 50 chats from the last 30 days. Desktop and CLI support different source agents, and project-level setup comes from the repositories and folders selected for import. User-level setup comes from files on the local machine. (ChatGPT Learn, “Import from another agent”)

Before relying on an imported workflow, review:

• Tool restrictions and permissions in imported skills and agents.
• MCP server settings that use custom authentication, headers, environment variables, or custom transports.
• Sign-in requirements for MCP servers.
• Hooks whose behavior may differ after import.
• Plugins or marketplaces that require manual setup.
• Prompt templates or slash commands that depend on arguments, shell interpolation, or file-path placeholders.

Those checks are not busywork. A successful import means the configuration was mapped; it does not mean every permission, credential, hook, or integration has been validated for the new runtime. (ChatGPT Learn, “Import from another agent”)

Bottom Line

The new import flow makes switching into ChatGPT Work or Codex more practical for people who already have an established Claude Code, Claude Cowork, or Cursor workflow. Start with the desktop app if you want the broadest source support and automatic updates; use /import in Codex CLI when you want a local, terminal-first migration from Claude Code or Cursor. Review the imported permissions and integrations before turning the workflow loose.

Sources

• OpenAI Developers, “You can now keep your work from other agents in sync with ChatGPT Work and Codex” (August 11, 2026): https://x.com/OpenAIDevs/status/2087242829076791392
• ChatGPT Learn, “Import from another agent”: https://learn.chatgpt.com/docs/import