Switching between Claude Code and Codex mid-task? /import brings chats into Codex, but only setup comes back

#What changed

A post making the rounds on X on Oct 17 lays out how to hop between Claude Code and Codex in the middle of a task. We checked the claims against both tools' docs. The short version is that the trip from Claude Code into Codex carries your recent chats, but the trip back only carries setup, so you need your own handoff note to keep the thread of the work.

Going into Codex, start a local Codex CLI session and type /import, then choose Claude Code. You can pick setup, project files, and recent chats. Codex CLI brings over up to 50 chats from the last 30 days, and it also maps instruction files to AGENTS.md, settings.json to config.toml, slash commands to skills, plus MCP servers, hooks, subagents, and Claude Code project memories. The ChatGPT desktop app has the same flow under Settings > Import.

Going back, Claude Code also has an /import command (v2.1.213 or later). Its docs say it brings over another agent's configuration: instruction files, MCP servers, commands, subagents, and skills. They don't list chat history, so your Codex conversation stays in Codex.

#Who is affected

Developers who use both Claude Code and Codex on the same repo, especially anyone who switches when one tool hits a usage limit or gets stuck partway through a change.

#What to do now

• Before switching to Codex, run /import from a local Codex CLI session with no task running. It doesn't work in a remote session or while connected to a local app-server daemon.
• Pick only what you need. If your Codex setup is already in sync, import just the recent chat you're continuing.
• After importing, review MCP servers that use custom auth, headers, or environment variables (you may have to sign in again), hooks, and tool permissions in imported skills and agents. OpenAI's docs call these out.
• Before switching back to Claude Code, ask Codex to write a short handoff file: the goal, what changed, what's left, and how to run the tests. Start the Claude Code session by pointing it at that file.
• Keep one AGENTS.md as the shared rulebook. Note that Claude Code's /import appends a one-time copy of AGENTS.md into CLAUDE.md, and that copy won't follow later edits. If you want one file to stay the source of truth, put @AGENTS.md in CLAUDE.md instead of relying on the copy.

#What is not confirmed

DotsFeed checked the command behavior against OpenAI's and Anthropic's docs but didn't run the round trip itself. How well an imported chat continues the work in Codex will depend on the task. The original post's handoff template sits behind an X article link and wasn't reviewed.

#Sources

• @Cuiszt_Cui on X: https://x.com/Cuiszt_Cui/status/2111639751925395666
• ChatGPT Learn, "Import from another agent": https://learn.chatgpt.com/docs/import
• Claude Code Docs, "How Claude remembers your project" (/import and AGENTS.md): https://code.claude.com/docs/en/memory