Claude Code Can Now Reach Remote Sessions by Name

Claude Code 2.1.225 added a small line with a useful consequence: SendMessage can start a conversation with a named Remote Control session on another machine. ListAgents now shows those remote sessions as name [ref].

That sounds like plumbing. In practice, it removes a very specific kind of friction.

The problem it solves

Imagine a long Claude Code task is still running on your desktop, but you have moved to a laptop. You do not need a new copy of the task. You need to reach the right live session and send one precise update.

The new behavior gives that update an address. Name the Remote Control session, find the reference, and send the delta to the session that already owns the work.

Useful deltas include:

• a newly discovered source
• a changed priority
• a correction to an assumption
• a request to stop before an irreversible action

What it does not solve

This is a messaging and addressing layer. It is not state synchronization.

It does not, by itself:

• move the runtime to the second machine
• copy files, credentials, tools, or browser state
• give the receiving session missing conversation history
• make two workspaces identical
• turn a short message into a durable handoff

That boundary matters. If a task must survive machines, models, or clients, the durable context still needs to live somewhere both sides can inspect. A repository note, task file, decision log, or other single source of truth is safer than assuming the chat transcript will carry everything.

A workflow I would trust

1. Name long-running sessions by outcome, not by device.
2. Keep the task goal, evidence, decisions, and next action in the workspace.
3. Use SendMessage for a short delta, not a complete hidden handoff.
4. Have the receiving session write important decisions back to the workspace.
5. Require an explicit receipt before any high-impact external action.

The same 2.1.225 release also adds more detail to gateway spend-limit warnings, a workspace trust prompt, and a batch of reliability fixes. The next release, 2.1.226, only lists bug fixes and reliability improvements, so I would not invent a feature list for it.

A portability lesson from Agent Modpack

I have been testing Agent Modpack, a file-based sample team, across Claude Code and Codex. The first lesson was simple: installable does not mean equivalent. Permissions, hook lifecycles, paths, login state, and session behavior still differ by host.

That is why I keep the portable layer in ordinary files and pair it with a capability matrix and install-time probes. At the time I checked today, Hugging Face showed 106 download sessions for the dataset. That metric is not 106 people or 106 complete installations; it is only a platform download-session count.

Named remote messaging fits the same pattern. It is valuable when the durable operating context is already outside the chat window.

What I would watch next

The feature becomes much more trustworthy if teams can answer four questions:

• Who is allowed to address a remote session?
• What happens when a name or reference is stale?
• Is delivery acknowledged and auditable?
• Can high-impact messages require a second confirmation?

The release removes one real obstacle. It does not remove the need for explicit state, access control, and receipts.

Official notes: Claude Code 2.1.225 and 2.1.226.

Disclosure: I used Codex to help research and draft this post, then checked the claims against Anthropic's release notes and edited the final text.