# Claude Code 2.1.224: AI Sessions Can Now Message Each Other > Source: > Published: 2026-08-22 13:11:08+00:00 Claude Code v2.1.224, released August 7, 2026, broke something fundamental to how the tool has worked since day one: sessions are isolated from each other. Two new tools — **ListAgents** and **SendMessage** — let separate Claude Code sessions communicate directly over a local socket, without touching Anthropic’s servers. It is on by default. You configure nothing. And it is the most significant shift to multi-session development workflows since nested sub-agents landed in [version 2.1.172](https://byteiota.com/claude-code-v2-1-172-sub-agents-can-now-spawn-sub-agents/). ## How Cross-Session Messaging Works Claude discovers other active sessions using the **ListAgents** tool and sends text to a target session via **SendMessage**. You don’t call these tools yourself — Claude does, either when you ask it to or when it determines on its own that another session needs to know something. Messages travel over a local socket on the same machine. Nothing goes through Anthropic’s infrastructure. Both sessions stay entirely independent — no conversation history is shared, no files pass between them, and no permissions transfer. A message is exactly what it sounds like: a piece of text one Claude writes for another. Requirements: macOS or Linux (including Linux inside WSL 2), Claude Code v2.1.224 or later, both sessions running. That’s it. ## When Claude Sends a Message Without You Asking There are two conditions where Claude initiates cross-session communication on its own: - A change in one session affects what another session is working on - One session settles a question that another session is currently blocked on In practice, this covers the two most common coordination failures in parallel development: the surprise breaking change, and the duplicate problem-solving effort. ## Three Scenarios That Show Why This Matters **Parallel refactor and test.** Session A is refactoring your user API — the endpoint shape is changing. Session B is actively running tests against that endpoint. With 2.1.224, Session A notifies Session B before the breaking change lands. Instead of discovering test failures after the fact and manually explaining what changed, Session B adapts in real time. **Unblocking without context switching.** Session A is stuck on a dependency conflict. Session B, working on a different module in the same repo, just resolved that exact conflict. Without cross-session messaging, you only realize this if you happen to check Session B’s output. With 2.1.224, Session B forwards the resolution when it recognizes the dependency is shared. Session A unblocks without you lifting a finger. **Reviewer plus verifiers.** A reviewer Claude fans out one verifier sub-agent per finding. Each verifier runs its check and reports back. The multi-step exchange stays invisible; only the final verdict surfaces in your main session. As the [DevelopersIO team found in their hands-on](https://dev.classmethod.jp/en/articles/20260807-cc-updates-v2-1-224/): the back-and-forth stays buried, with only one clean result landing in the main window. ## The Limitations Are Real — Know Them **Text only.** You cannot pass files, code blocks, conversation history, or session state between sessions. A message is a brief handoff: a change summary, a test result, a question, an answer. If you need to move large context between sessions, you still need a shared file or a CLAUDE.md update. **Same machine.** Cross-session messaging is local-socket-based, which means sessions on different machines cannot communicate. If your team runs sessions on separate workstations or cloud instances, this feature does not bridge them — that is a different problem. **macOS and Linux only.** Windows users outside of WSL 2 are not supported in this release. ## What Else Ships in 2.1.224 Three other changes in this release are worth noting. The 200-sub-agent-per-session cap is gone — long-running autonomous sessions that previously hit the ceiling and stalled will no longer do so. A sandbox path-traversal bug is fixed: rules using a trailing slash (such as `denyRead: "~/.aws/"` ) were silently bypassed on Linux and macOS; they are now enforced correctly. And Claude now receives a clear explanation of which file or network access was denied when a sandbox violation triggers, rather than seeing nothing at all. The self-hosted runner (`claude self-hosted-runner` ) also shipped in this release, available on Team and Enterprise plans. The [official self-hosted environments documentation](https://code.claude.com/docs/en/self-hosted-environments) covers the setup in full. ## How to Update ``` # Check your current version claude --version # Update to latest npm install -g @anthropic-ai/claude-code@latest ``` Once on v2.1.224 or later, cross-session messaging is active with no additional configuration. Open two terminal sessions, run Claude Code in each, and they can find and message each other. The [official cross-session messaging documentation](https://code.claude.com/docs/en/cross-session-messaging) covers edge cases and platform-specific behavior. Sub-agents gave Claude Code vertical depth — one session spawning workers below it. Cross-session messaging adds horizontal reach — peer sessions coordinating as equals. The “one Claude, one session” mental model no longer describes how this tool works.