You just hit a Claude Code usage limit mid-session
The reply cut off halfway through a diff. The status line went red. You had one small follow-up left, and now it says the 5-hour window is exhausted and the next reset is in fifty-two minutes. You'd have paced yourself if you'd known.
Claude Code enforces two rolling limits on the Pro and Max plans: a 5-hour rolling window and a weekly budget. Both reset on their own schedules, both are invisible from the shell you're actually working in, and both can end a session earlier than you expected. When you're SSH'd into a remote box and running claude there, the friction is worse — the client that would show you the limits is on a webpage, in a different window, on a different machine, and you're a session away from it.
Most people just eat the interruption. A few work around it.
How people usually deal with this
The mental clock
You start a session and try to remember when. Two hours later you're deep in a refactor and the clock is the last thing on your mind. This works about as well as remembering to drink water.
A browser tab on claude.ai/settings/usage
Open in a spare tab. Refresh occasionally. Fine until the browser goes to sleep, or the tab gets closed, or you're in a full-screen terminal for two hours and never see it.
claude --usage at the start of each prompt
Explicit, always accurate. Adds a token check and a small delay to every command, which is what you were trying to avoid by using a shell in the first place.
Running a background poller
A small script that pings the usage endpoint every few minutes and echoes to a file. Works, but now you're maintaining a script for something that ought to be visible in the terminal itself.
What's actually useful
What would help is a usage meter that lives in the same window as the SSH session — not in a browser tab, not in a separate poll. Somewhere your eye lands anyway when you're waiting for a response. Ideally with an alert as you approach the limit, so you can wrap up the current thought instead of getting cut off inside it.
Ctrl-V Terminal puts this in the tab bar. When you're using Claude Code (locally or over SSH), the meter shows how much of your 5-hour window is left and how much of your weekly budget is left. It updates on its own, doesn't cost you a token per check, and gives you an unobtrusive nudge when either limit is inside the danger zone — 15%, 5%, exhausted — before it becomes a mid-diff surprise.
The rest of the SSH-Claude workflow
Usage awareness is only half of what makes Claude Code over SSH pleasant. The other half is not fighting the terminal itself:
- Screenshots as arguments.
Ctrl+Vin a session drops your clipboard image onto the remote via SFTP and pastes the file path at your cursor.claude "look at $(pwd)/screenshot.png"just works. More on that here. - Tabbed sessions. Keep the box you're pairing on in tab one, its logs in tab two, and a scratch shell in tab three. A busy tab indicator pulses while its terminal is producing output, so you notice when a long Claude run finishes.
- Session state that survives. If the app crashes or updates, tabs, names, colours and window layout come back the next launch. You don't lose the mental model of where you were.
- Master password (Pro) + screensaver lock. Saved session passwords stay in a vault; step away and the lock kicks in so a shoulder-surfer doesn't inherit your Claude session.
Try it
Ctrl-V Terminal is a free download for Windows. The usage meter itself is a free-plan feature — you don't need to pay to see how close you are to a Claude limit. The Pro plan adds unlimited saved sessions and near-limit alerts.