Claude Code 2.1.260: Uncapped Subagents and Safer Worktrees

Lachlan Reed

Picture this. You set off an autonomous Claude Code subagent on a massive background job, say, a ninety minute end to end test suite or a massive Docker build. You grab a coffee, come back an hour later... and boom. It was abruptly killed right at the sixty minute mark. Well, version 2.1.260 eliminates that invisible ceiling, brought to you by Jellypod AI.

James Turner

Man, that sixty minute cutoff was such a silent killer! You would have long running database migrations or web scrapers running in a subagent, and suddenly everything just dropped dead with zero warning.

Lachlan Reed

Oh, tell me about it mate! I, I, I actually had a multi repo test pipeline fail on me mid run last week because of that exact limit. I thought I was going bonkers! But yeah, according to the official release notes for Claude Code version 2.1.260, Anthropic removed the one hour time limit on background commands started by subagents.

James Turner

So now subagent background tasks just run indefinitely until they finish or until you manually stop them, which finally matches how the main session behaves.

Lachlan Reed

Spot on. And the practical workflow now is super clean, mate. You just dispatch your long background job through a subagent, and then you can monitor the whole process directly using commands like attach, logs, and stop. You do not have to worry about some arbitrary timer pulling the plug while you step away from the terminal.

James Turner

Okay, but, uh... wait. If subagents do not auto terminate background processes at sixty minutes anymore, what happens if a task gets stuck in an infinite loop?

Lachlan Reed

Ah, now you are asking the real questions .

James Turner

Seriously though! On build host machines or cloud instances, a hung or spinning subagent task will now literally run forever if you do not watch it. That is a quick way to burn through your cloud budget or spike host CPU to one hundred percent.

Lachlan Reed

Yeah, fair point. You definitely cannot just fire and forget without guardrails anymore.

James Turner

Exactly. You have to pair this uncapped execution with proper process monitoring, or wrap your commands in explicit timeout scripts, or even use Linux cgroups memory limits on your build runners so runaway processes get caught before they wreck your server.

Lachlan Reed

That is solid advice, mate. You gotta respect the power of an infinite runtime.

James Turner

Definitely. Were there any other noteworthy changes packed into 2.1.260?

Lachlan Reed

Yeah, a couple of really neat quality of life tweaks. For one, they added a text form of the advisor command, so you can run slash advisor, slash advisor followed by the model name, or slash advisor off. That is massive for headless setups, Remote Control, and Agent SDK workflows where you do not have an interactive UI picker.

James Turner

Oh nice! What about safety around git worktrees? Did anything change there?

Lachlan Reed

Glad you asked! Before, if you went to delete a background session worktree with unpushed work, it was pretty easy to accidentally wipe stuff out. Now, Claude Code gives you a clear safety prompt that spells out the exact branch name and the specific commit count before deleting anything. So if you see three unpushed commits on feature branch main, you get a chance to save your skin before double deleting!

James Turner

That alone is gonna save a lot of developers from a heart attack. Unlimited subagent runtimes plus safer worktree cleanup makes 2.1.260 a really solid release.

Lachlan Reed

No doubt about it, mate. Go update your CLI and let those builds run!