Anthropic released Claude Code version 2.1.224 on Friday with two new tools that let active coding sessions locate one another and send short text messages. The change removes the need for developers to copy context by hand between separate Terminal windows running on the same machine.
The news
The update adds ListAgents and SendMessage to the Claude Code tool set. ListAgents returns a list of other active sessions on the same Mac. SendMessage delivers a plain-text payload to one of those sessions. A receiving session displays the incoming text as a labeled card that includes a link back to the sender session.
Anthropic’s documentation states that only the message text itself travels between sessions. Conversation history, attached files, and permission grants do not move. The feature is available on both macOS and Linux. No equivalent capability has been announced for Windows.
Context
Before this update, each Claude Code session operated in isolation. Engineers working on related tasks had to paste summaries or error logs manually when one session needed information held by another. The new feature keeps that isolation intact while adding a narrow channel for coordination.
Developers can use the channel to alert another session when a change breaks a shared dependency or to forward an answer that the second session is waiting for. The documentation also notes that an incoming message cannot trigger permission approvals on the receiving side. This preserves the existing security model in which each session must request its own access rights.
Details
The implementation is deliberately minimal. ListAgents only reports sessions that are currently running on the local machine. SendMessage accepts a target session identifier and a text string, with no support for attachments or structured data. The receiving side renders the message in a card format that links back to the originating session for quick reference.
Because the channel carries only plain text, it cannot be used to transfer code files, environment variables, or tool permissions. Anthropic explicitly documents that an incoming message will never cause the receiving session to grant file-system access or execute commands on its own. This design choice limits the feature to signaling and short coordination notes rather than full context sharing.
The release is limited to macOS and Linux terminals. Windows users continue to rely on manual copy-and-paste or external note-taking tools when running multiple sessions.
Why it matters
The update addresses a practical friction point for engineers who keep multiple Claude Code sessions open on the same machine. By giving sessions a lightweight way to signal one another, Anthropic reduces the manual bookkeeping that previously interrupted focused work. At the same time, the strict limits on what can pass between sessions preserve the existing security boundary that each session must request its own permissions.
The result is a modest but concrete improvement in how the tool handles parallel tasks without widening the attack surface. Teams that already run several Claude Code instances for different branches or services now have an in-tool method to pass brief alerts. The restriction to text-only messages and the explicit block on permission propagation keep the change narrow enough that it does not require new threat-model reviews for most users.
Over time, this kind of minimal coordination primitive may become table stakes for any AI coding agent that runs in multiple terminal windows. For now, the change simply removes one small source of context switching for developers who already work that way.
---
Sources:
{"word_count": 612, "sources_used": 1, "expanded_sections": ["context", "details", "why_it_matters"]}
No comments yet