Claude Code Gets a Send Now Key

James Turner

You are three minutes into watching Claude Code completely refactor your database models... except you realize ten seconds in that it is using the old schema. The legacy one from two sprints ago.

Lachlan Reed

Oh mate, the absolute worst feeling in the world! Because up until now, your options were... well, pretty painful, right? You hit control C, but that completely kills the execution context, dumps your progress, and you have to start from scratch.

James Turner

Or you type out a correction, hit enter, and then just sit there watching it run ten more wrong tool calls because your message is stuck in the queue until the current turn finishes.

Lachlan Reed

Trapped in line! Like standing at the post office behind someone mailing fifty packages. But look, that is exactly what Anthropic just tackled in version 2.1.275. They added what they are calling a send now key.

James Turner

Yeah, this is huge. Now when you see the agent drifting off track, you can hit control enter... or if your terminal eats that shortcut, control x control s... and it immediately interrupts the active turn right mid tool call.

Lachlan Reed

So it drops what it is doing right then and there, takes whatever text you had queued up in your buffer, and forces Claude to ingest all of it into context at once. No lost state, no waiting for a five minute task to wrap up.

James Turner

And the visual feedback in the terminal UI is super clean now too. Any queued or sent messages sit in gray text until the model officially acknowledges them. So you know exactly what Claude has seen and what it has not.

Lachlan Reed

That is fair dinkum brilliant. I mean, steering autonomous agents feels a lot less like throwing a bottle into the ocean and hope it lands on the right beach. You actually have a steering wheel now.

James Turner

Exactly. And speaking of keeping things in sync, 2.1.275 also introduced account wide skill and plugin syncing. So if you have custom skills or plugins enabled on your claude.ai account, your terminal sessions automatically pull them down when you log in.

Lachlan Reed

Wait, so if I tweak a custom prompt skill on the web app, my local terminal environment just gets it? No manual copying or repo pulls needed?

James Turner

Yep, it just pulls it in! Though if you want to keep your CLI isolated for specific projects, you can opt out in your settings file by setting syncClaudeAiSkills or syncClaudeAiPlugins to false.

Lachlan Reed

Nice. They also smoothed out marketplace installs too, right? With that new marketplace flag on plugin install?

James Turner

Right! If you try to install a plugin from a marketplace your system does not know about yet, passing slash plugin install with the marketplace flag will offer to add the marketplace source automatically right before installing.

Lachlan Reed

Saves you that extra setup step. Now, we should mention, right on the heels of 2.1.275, Anthropic pushed out a quick follow up patch, version 2.1.276. What was that hotfix for?

James Turner

That fixed a regression from 2.1.275 where anyone pointing ANTHROPIC BASE URL at an enterprise gateway or custom proxy was getting hit with a bad request 400 error on every single turn, mentioning an input tag advisor error.

Lachlan Reed

Ouch! Imagine getting locked out of your terminal proxy right in the middle of a release cycle! Glad they patched that within hours.

James Turner

Yeah, 2.1.276 cleans that up completely. So the big takeaway here is update your CLI, get used to smashing control enter when your agent strays, and keep active control of your prompt execution.

Lachlan Reed

Dead set. Give control enter a spin on your next build, and stop watching wrong code run! Catch you all next time.