The Clone Goes to a Separate Session, the Child Task Stays Local: How Claude Code v2.1.212 Split Parallel Work Between "/fork" and "/subtask"
In Claude Code v2.1.212, the meaning of /fork changed. It's now the command that clones the conversation and launches a separate background session, while the old "child task, right here" behavior moved to /subtask. Here's a look at the update that lets you choose between running in parallel and del
In Claude Code v2.1.212, shipped on July 17, the meaning of the /fork command was swapped out. Until now, /fork was the command for "spinning up a child agent (subagent) right here, partway through the current conversation." With this update, that role has moved to a separate command, /subtask, and /fork has become the command for "cloning the entire conversation and launching a separate background session." The name is the same, but because the change touches the steps a developer physically takes, let's walk through it in order.
Same "branching," but the place it runs has changed
The official changelog puts it this way: /fork now "copies the conversation into a new background session (which appears as its own line in claude agents), while you keep working. The in-session subagent that /fork used to launch has become /subtask." In other words, two behaviors that had been crammed into a single word are now split explicitly into two commands.
/fork — peel off into a separate session without stopping the main line
It clones the conversation state as-is and carves it out into an independent background session. The carved-out session lines up as its own row in the claude agents list, and your local conversation keeps going. When you want to "push the current thread forward, but also run a spun-off investigation or an alternative approach in parallel," you can grow a branch without leaving the main line.
/subtask — delegate to a child inside your local session
It inherits the old behavior of /fork: launching a child agent inside the same session. It suits "on-the-spot division of labor," where you carve out part of the current work, have a child solve it, and receive the result locally.
To lay out the difference between the two:
| Aspect | /fork | /subtask |
|---|---|---|
| Where it runs | A new background session | Inside the current session |
| How it appears in the list | Lines up as its own row in claude agents | Handled within the current session |
| What it's suited for | Running separate work in parallel while advancing the main line | Carving out part of the current work to a child |
Making "parallel" and "delegate" selectable at the level of the words themselves
Until now, when you wanted to branch, "do I want it running independently in a separate session?" and "do I want it running locally as an extension of the current work?" both lived inside a single command. With this split, developers can choose which mode they want before they start. Long-running explorations, or validations where you don't want failures to affect the main line, go to a separate session with /fork. If you want to strongly carry over the current context and hand off just a part, you keep it local with /subtask.
The same v2.1.212 also includes surrounding changes that make this parallel work easier to handle.
- A selection screen for /resume: typing
/resumein the agents view opens a list of past sessions, and you can pick from them including sessions you deleted from the list. The one you pick resumes as a background session. It's a safety-side change that lets you bring accidentally deleted work history back to your local view. - Automatic backgrounding of MCP calls: MCP tool calls that run longer than two minutes are automatically moved to the background, so the session no longer freezes up. The threshold can be tuned or disabled with
CLAUDE_CODE_MCP_AUTO_BACKGROUND_MS. The aim is to reduce the moments where a long wait on an external tool leaves you stuck. - auto-mode reset:
claude auto-mode resetreturns auto-mode settings to their defaults (with a confirmation prompt, skippable with--yes). It's a way back when you've fiddled with the settings so much that the behavior has become hard to read.
How it plays out in real operation
In actual development, it's increasingly common to advance a single main line while simultaneously juggling "an investigation on another branch," "an implementation of an alternative approach," and "keeping an eye on the tests." The split between /fork and /subtask distills that parallel work into a binary choice — "peel it off into a separate session" versus "delegate it to a child locally" — and lets you survey what's running in the claude agents list. Even in unattended, automated operation, there's now more room to detach spin-off tasks into a session separate from the main line. Combined with the automatic offloading of long MCP calls, the defining feature this time is that everything is aligned toward not stopping your local session.
Behind the convenience, a few things worth keeping in mind
The easier parallel work becomes, the harder it gets to grasp "how many sessions are running right now, and what each is doing." Work you've offloaded to the background easily slips out of sight, and if you leave something running and forget about it, that much token consumption quietly piles up. The fact that the same v2.1.212 also includes a change that sets a cap (default 200) on the number of subagent and web-search executions can be read as an intent to rein in this "runaway parallelism." It's safest to think of the tooling for parallel work and its brakes as a set — that kind of caution is worth holding onto.
That said, the core of the change is practical and unexaggerated. It's probably best understood as an update that lets you decide, up front and at the level of the command name, the one bit of extra effort of consciously choosing "separate session or local?" each time you branch.
References: anthropics/claude-code CHANGELOG (v2.1.212) / Claude Code changelog (official docs) / Releasebot — Claude Code Updates
