Guides

Running several Claude Code sessions without losing track

Give every session one task and its own git worktree, give it a name, and watch all of them from one place: agent view for background sessions, or the sidebar in Claude Desktop. Get an alert only when a session needs you, with the project’s name in it, and review finished sessions in batches. With two or three independent sessions most people keep up; past that, the limit is your attention, not the agents.

By Dmitry Mishunin, who builds Meowdoro · October 27, 2026 · checked against the Claude Code documentation on October 26, 2026

• The short answer
• One worktree each
• Names
• One place to see them
• Alerts that say which one
• The decision queue
• How many is too many
• Questions

The short answer

• Isolate: claude --worktree feature-auth starts a session in its own checkout, so two sessions never edit the same files.
• Name: /rename each session after its task, and come back to it with claude --resume.
• Watch from one place: claude agents opens agent view, which groups sessions into needs input, working and completed. Claude Desktop does the same in its sidebar.
• Alert on what blocks: a permission prompt or a question, with the project in the title. Finished sessions can wait.
• Batch the review: look at finished sessions together, not one by one as they land.

One task and one worktree per session

Two sessions in the same folder will sooner or later edit the same file. A git worktree gives each session its own checkout of the repository on its own branch. Pass --worktree (or -w) with a name and Claude Code creates it under .claude/worktrees/<name>/ on a branch called worktree-<name>, then starts the session there (worktrees):

claude --worktree feature-auth

When you exit a worktree that has work in it, Claude Code asks whether to keep it and prints the command to return to it. Sessions you dispatch from agent view move into a worktree of their own before they edit anything, and Claude Desktop runs parallel sessions with git isolation as well.

Keep one task per session. A session that fixes a bug, then reviews a pull request, then explores a new library carries all three in its context, and the third task pays for the first two.

Name every session

“Which one was the auth fix?” is the question that eats time with four tabs open. Name each session with /rename as soon as it has a task, and treat sessions like branches: one per line of work. claude --resume then shows the names, and claude --continue picks up the last one. Some terminals also show each session’s state in its tab, working, waiting on you or done, through the Program Status Protocol (OSC 7501), which Claude Code 2.1.295 and later reports on its own.

One place to see them all

• Agent view. claude agents opens one screen for your background sessions, a research preview at the time of writing: each row shows a session’s name, its latest status or question and a time, grouped into needs input, working and completed. You can dispatch new tasks from it, peek and reply, or attach to the full conversation (agent view).
• Claude Desktop. The Code tab lists sessions in a sidebar and notifies you when one finishes while you’re looking at another.
• Inside one session. /tasks lists background commands and subagents, running or finished.

Alerts that say which session needs you

A generic “Claude Code needs your attention” doesn’t help with four sessions open. A Notification hook receives the session’s folder, so a few lines of shell put the project name into the banner. The script is in our guide to Claude Code notifications. While agent view is open, two more notification types fire for background sessions: one when a session starts waiting on your input, one when it finishes or fails.

The decision queue

With several agents, your job turns into answering them. One developer running six sessions described the result as “processing a decision queue you accidentally built”. Treat it like a queue on purpose:

• Blocked first. A session waiting on a permission or a question does nothing until you answer. Those come before reviewing finished work.
• Read before you approve. Approving on reflex is the failure mode of many sessions. In a widely discussed experiment with 40,000 game runs, people approving AI agent commands missed about one threat in three. If you approve everything anyway, an allowlist or auto mode is more honest than clicking yes.
• Review in batches, at the end of a focus block, with fresh eyes. A fresh session or a subagent is also a good reviewer, because it isn’t attached to the code it just wrote.

How many sessions is too many?

There’s no fixed number, but the warning signs are clear. You reread a diff twice because you forgot what you asked. You approve without reading. You can’t say what each session is doing without looking. In developer threads, people running nine to twelve projects at once describe being exhausted by the evening, and several say they went back to two or three, or to one at a time. Parallel sessions also multiply token usage, so they reach usage limits sooner. When the tiredness doesn’t go away, read AI coding burnout: why agent work wears you out.

If a task is only a side job, such as researching a library or reading logs, a subagent inside your current session is often better than another session: it works in its own context and returns a summary.

All sessions at a glance in the notch

We make Meowdoro, so this is a disclosure as much as a tip. It’s a focus timer around the MacBook notch that lists every Claude Code and Codex session in a small black island: working, waiting for you, finished or failed, grouped by agent, each with its name or what it needs. A click opens the session in Claude Desktop or ChatGPT, or brings forward the app where it runs. Sessions on other machines show up too. Sessions you’ve already seen dim and drop off. During a break the list disappears until the next focus (agents in the island).

Questions

Can I run multiple Claude Code sessions at the same time?

Yes, in separate terminals, in agent view or in Claude Desktop. On the same repository, give each its own worktree so their edits don’t collide.

Can two Claude Code sessions talk to each other?

Yes. With cross-session messaging, Claude can list and message your other Claude Code sessions on the same machine, on another machine or in the cloud (cross-session messaging). For a coordinated team with a shared task list, there are agent teams, an experimental feature that is off by default.

Do parallel sessions share my usage limit?

Yes. They all draw on the same plan, so several busy sessions use it up proportionally faster.

Should I use subagents or separate sessions?

Subagents for side tasks that would flood your conversation with search results or logs. Separate sessions for independent pieces of work you want to steer and review yourself.

Sources: the Claude Code documentation on running agents in parallel, worktrees, agent view, sessions, hooks and best practices, read on October 26, 2026; the Reddit and Hacker News discussions linked above.