Press ctrl+enter — or ctrl+x followed by ctrl+s — in Claude Code version 2.1.275 or later and the new send-now key makes Claude Code send queued messages immediately, interrupting the current turn instead of making you wait for it to finish. Until this release, anything you typed while the model was working sat in a queue until the turn ended; now you decide when the queue flushes, and both sent and queued messages are shown in grey until the model actually receives them, so you always know what has landed and what is still waiting.
📺 Watch: Building AI Agent Operating Systems with Claude
🔥 Get the Agent OS as a free bonus: AI Profit Boardroom members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls. → Get inside · Want AI SEO help 1-on-1? Book a free SEO strategy session →
The change arrived in Claude Code version 2.1.275, released on 17 September 2026, per the official Claude Code changelog published by Anthropic. The changelog entry reads: "Added a send-now key (ctrl+enter, or ctrl+x ctrl+s) that interrupts the current turn and sends all queued messages at once; sent and queued messages show in gray until the model receives them." Small line, big quality-of-life difference — and this page covers exactly how the mechanism works, when to use it, and where it fits alongside the rest of the September 2026 Claude Code updates.
Claude Code Send Queued Messages: How The Send-Now Key Works
Claude Code has always let you keep typing while the model is mid-turn. Those messages queue up and get delivered when the current turn completes. That design is sensible as a default — the model finishes its thought, then reads your follow-ups in order — but it has an obvious failure mode: you spot the agent heading down the wrong path thirty seconds into a five-minute turn, type the correction, and then sit there watching it burn time on work you already know you will throw away.
The send-now key is the manual override. Press ctrl+enter and three things happen, per the changelog description: the current turn is interrupted, every message in the queue is sent at once, and the display marks messages in grey until the model receives them. The alternative chord, ctrl+x followed by ctrl+s, does the same thing — useful if your terminal setup already binds ctrl+enter to something else.
The grey styling matters more than it sounds. Under the old behaviour it was easy to lose track of whether a message you typed mid-turn had been delivered or was still pending. Making delivery state visible removes the guesswork: grey means in flight, normal means received. If terminal rendering quirks are a sore spot for you generally, the Claude Code no-flicker mode guide covers the other big display improvement worth knowing about.
When To Interrupt And When To Let The Queue Wait
Having the send-now key does not mean every message should jump the queue. The distinction that matters is whether your message changes what the model should be doing right now.
- Use send-now for course corrections. Wrong file, wrong approach, a requirement the model missed — every second the turn continues is wasted work. Interrupt, deliver the correction, let it restart on the right track.
- Use send-now for new urgent context. If you have just noticed something that changes the task — an error in another window, a stakeholder message, a better idea — get it in front of the model before it commits to more output.
- Let the queue wait for additive follow-ups. "After this, also update the tests" does not need to interrupt anything. Delivered at the end of the turn, it becomes the natural next instruction with full context of the finished work.
If you want to go beyond keyboard shortcuts and build agent workflows that actually pay for themselves, check out the AI Profit Boardroom → see what members are building here. Want a second pair of eyes on your own setup first? Book a free SEO strategy session and get a concrete plan.
This is the same judgement call agent operators make constantly: steering early is cheap, steering late is expensive. Workflows that lean on long-running agent turns — like the AI SEO systems in the Claude Code AI SEO guide, the closest companion article to this one — benefit most, because turns there routinely run for minutes and a bad direction compounds fast.
Queued Messages In Cloud And Resumed Sessions
The send-now key is the visible half of a broader clean-up of queued-message behaviour in this release. The same 2.1.275 changelog also lists a fix for "resumed cloud sessions leaving unanswered questions open after queued messages superseded them" — the situation where the model asked you something, your queued messages made the question irrelevant, and the resumed session still treated it as pending. If you run Claude Code sessions in the cloud and steer them from your phone or another machine, that fix removes a genuinely confusing state.
Queue control matters more the more asynchronously you work. On a local machine you can always just watch the turn; in remote and cloud sessions, messages pile up precisely because you are not watching in real time. Deciding whether those messages should interrupt or wait is now an explicit choice rather than something the timing of the turn decides for you. If you are weighing up local versus hosted setups generally, the Claude Code local guide covers the trade-offs.
📺 Watch: Claude AI SEO: 0 to 403 Clicks a Day!
What Else Shipped Alongside The Send-Now Key
Version 2.1.275 was a dense release. Beyond queued messages, the 17 September 2026 changelog highlights three more changes worth knowing:
- Account-level skills and plugins sync. Skills and plugins enabled on your claude.ai account now sync to terminal sessions signed in with the same account, with opt-out settings (syncClaudeAiSkills: false and syncClaudeAiPlugins: false) if you want the environments separate.
- Gateway sign-in confirmation. When a Claude apps gateway names the signed-in account, you confirm it before the credential is saved, and /status shows the account in use.
- A credentials hygiene fix. Plugin and marketplace messages and logs no longer display passwords or tokens embedded in URLs.
Two adjacent releases round out the picture. Version 2.1.274, also dated 17 September 2026, added a visible warning when memory usage turns critical, with steps to free memory or restart. And version 2.1.276, released 18 September 2026, fixed a 2.1.275 regression where every request failed with a 400 error for setups pointing ANTHROPIC_BASE_URL at a proxy or gateway — so if you use a gateway, update straight past 2.1.275 to the newest build.
Getting Set Up To Use The Feature
If ctrl+enter does nothing in your terminal, work through this checklist:
- Check your version. Run claude --version — the send-now key needs 2.1.275 or later, and given the 2.1.276 regression fix, later is better.
- Try the alternative chord. Some terminal emulators capture ctrl+enter before applications see it. The ctrl+x ctrl+s chord exists for exactly this reason.
- Confirm there is actually a queue. The key sends queued messages; if nothing is queued and no turn is running, there is nothing for it to do.
If you are new to the tool entirely, start with the learn Claude Code guide and the Claude Code free access guide — the send-now key is a refinement you will appreciate more once long agent turns are part of your daily workflow. And once they are, structure matters more than shortcuts: the Agent OS guide covers turning ad-hoc agent use into a repeatable operating system, the agentic OS with Claude Code guide shows the Claude Code specific build, and the Goldie Bench write-up covers how the underlying models compare in hands-on tests when you are choosing what to run those systems on.
Why Small Control Features Like This Matter
A send-now key looks like a minor convenience next to model releases and benchmark jumps, but control features are what make agents usable for real work. The gap between "the agent does what I meant" and "the agent does what I said twenty minutes ago" is exactly where time and money leak out of AI workflows. Interruptibility, visible delivery state, and predictable queue behaviour all shrink that gap — which is why Anthropic keeps shipping them, and why the operators getting the most out of Claude Code in 2026 are the ones who learn each control the week it lands rather than months later.
If you want daily tutorials on updates like this — plus the prompt libraries and agent skills to go with them — check out the AI Profit Boardroom → get inside here. Rather map out your AI strategy 1-on-1? Book a free SEO strategy session — no cost, no catch.
