Codex as a subagent of Claude Code, and the other way round

In Shuriken 5.3 you can use Codex as a subagent of Claude Code by asking for it. The agent in any chat can start background agents and new chats on any provider you have, and Codex or Gemini CLI can do the same with Claude Code. Each agent it starts is a Shuriken chat on one of your accounts, with hot swapping like any chat you open.

What was missing before 5.3#

Before 5.3, the agent in a chat could open a new chat for you, on another provider too, once you confirmed a card. It could not start work in the background, hear when that work was done, or read what came back. Claude Code's own subagents run inside its session, on Claude, so handing a part of a task to Codex meant opening a second chat and pasting the context into it yourself.

5.3 gives every agent in a Shuriken chat list_providers, start_agents, read_result, continue_chat and adopt_result, next to the new_chat it already had. The feature page has a table of what each one does.

What you can use it for#

We don't prescribe a workflow. These are requests the tools can carry out:

• "Have Codex write the tests for this module while you finish the UI."
• "Ask Gemini to review this diff."
• "Run the data migration on Codex in the background and tell me when it's done."
• "Claude is at its weekly limit. Carry this chat on in Codex."
• "Try this on three providers, each on its own branch, and merge the one whose tests pass."

The agent first asks Shuriken what this computer can run: which agents are installed, their models, and how much of each of your accounts' 5-hour and weekly limits is used. It can start up to 8 agents in one call, listed under the chat that asked. They work in that chat's folder, like its own subagents, unless it gives each a branch of its own. When one finishes, the chat hears about it and can read its last answer.

Handing a chat to another provider#

An agent can also hand the chat it is in to another provider. It writes a handover (the goal, what was decided and done, what is left), and a new chat on, say, Codex starts from it, linked back to the Claude Code chat it came from. The old chat stays as it was. Claude Code and Codex in one project covers doing that switch by hand.

The limits we put on it#

Only a chat you are in can start agents, and one chat starts at most 24 chats an hour. The agent never sees a sign-in, a key or an e-mail address, only the names you gave your accounts and how much each has left. Everything it starts is a normal chat on your own plans: hot swapping moves it to another of your accounts when one nears its limit, and when a provider has no account with room it waits for the reset.

Try Shuriken