Jellypod, Inc.

The Claude Code Changelog

TechnologyNews

Listen

All Episodes

Claude Code v2.1.179 Fixes Drops, Globs, and WSL2

This episode covers Claude Code v2.1.179, including better handling of dropped socket connections, smarter sandbox glob processing, and reduced prompt bloat for large repositories.

We also dig into quality-of-life fixes for WSL2 scrolling and faster remote plugin startup over SSH.

This show was created with Jellypod, the AI Podcast Studio. Create your own podcast with Jellypod today.

Is this your podcast and want to remove this banner? Click here.


Chapter 1

Claude Code v2.1.179: Surviving Connection Drops and Taming Heavy Sandbox Globs

Lachlan Reed

Ever had Claude deep in a twenty-step refactor only for your Wi-Fi to drop, taking the entire session with it? [exhales sharply] Lachlan here, along with James, and this update is brought to you by Jellypod AI.

James Turner

That twenty-step refactor pain is so real. [chuckles] There is nothing worse than watching your terminal hang indefinitely on a stale connection. But with Claude Code v2.1.179, Anthropic is finally resolving this by preserving partial tool responses when socket connections drop, instead of raising a raw, session-killing terminal error.

Lachlan Reed

Yeah, it’s a massive win for CLI stability. Previously, if your connection blinked while Claude was halfway through executing a command, the agent state would completely desync, leaving you with a frozen "running tool" spinner. Now, the CLI handles the TCP timeout gracefully, keeps the output received up to the disconnect point, and unlocks the prompt so you can resume without losing your entire agent execution history.

James Turner

And keeping those partial tool responses is crucial because it saves you from re-running heavy build or test commands from scratch. But honestly, the change that's going to save the most compute is how they've optimized the Linux sandbox permission globs. Specifically, the way broad allowRead and denyRead policies are handled.

Lachlan Reed

Oh, those glob patterns were absolute context hogs! [laughs] If you specified a broad allowRead pattern across a massive project tree to secure your sandbox, the runtime would serialize those expanded paths directly into the Bash tool system prompt. You’d end up blowing your context window on thousands of lines of raw system metadata before the model even started writing code.

James Turner

Right, you were essentially paying in tokens to tell the model what it *couldn't* see. [dryly] In v2.1.179, they’ve refactored how the sandbox limits are passed to the agent. Instead of serializing the flat, expanded directory list into the Bash tool definition, the CLI now evaluates those glob rules dynamically, keeping the system prompt lean and preventing session bloat on massive repositories.

Lachlan Reed

Which makes a huge difference in response latency, especially when you're working in monster monorepos. And speaking of development environments, they also snuck in some crucial quality-of-life fixes for WSL2 and remote setups. If you were running WSL2 inside VS Code or Windows Terminal, the mouse-wheel scroll behavior was completely broken in the terminal emulator, which is now fully fixed.

James Turner

That WSL2 scrolling issue was a massive headache when trying to read long stack traces. [sighs] They also parallelized the initialization phase for remote plugins, which previously blocked the CLI startup sequence on sequential network roundtrips over high-latency SSH connections.