[BUG] Missing session closure option when reaching a limit
Preflight Checklist
- [x] I have searched existing issues and this hasn't been reported yet
- [x] This is a single bug report (please file separate reports for different bugs)
- [x] I am using the latest version of Claude Code
What's Wrong?
In case of "You've reached your Fable 5 limit. Run /usage-credits to continue or switch models with /model.", the session cannot be closed in a way that allows you to continue with a cheaper model in a new session. Due to the forced model switch, the AI agent has to re-read the entire session, wasting a lot of tokens unnecessarily. It is not clear how much of Fable's recent work is lost, but it seems to me that it is a lot. I ask for a fix for this error and I would like to point out that I am not happy about my wasted tokens.
Clean session closure offer is also missing when the 5-hour and weekly limits are exceeded. This should be available even though there are no tokens left to be used at that time.
What Should Happen?
When the limit of a 5 hours or a weekly token or model usage is reached, Claude Code should offer to close the current session cleanly, allowing it to continue in a clean session with a different model or even manually.
Such session closing should not cost more tokens.
Error Messages/Logs
You've reached your Fable 5 limit. Run /usage-credits to continue or switch models with /model.Steps to Reproduce
Just reach any limit in a long session.
Claude Model
Other
Is this a regression?
No, this never worked
Last Working Version
_No response_
Claude Code Version
2.1.207
Platform
Anthropic API
Operating System
macOS
Terminal/Shell
iTerm2
Additional Information
_No response_
Showing cached comments. Read the full discussion on GitHub ↗
3 Comments
This has burned me multiple times - session hits limit mid-task, work is lost, no clean way to save state.
What I started doing preventatively:
Would love a proper "session running low, want to save state?" prompt from the UI though. Hitting the wall unexpectedly is the worst UX.
Yes — this is exactly what I've been documenting this weekend (full write-up: #76987). Concrete numbers from my own case:
Hit the Fable limit mid-session. There was no clean closure offered — I had to give it explicit instructions to switch models myself, it did not happen automatically. On the next resume of that same window, the cost was real and measurable: two screenshots of the usage panel taken 2h21m apart (same session, confirmed by the countdown timer landing on the identical boundary both times) showed weekly Fable-only usage going from 8% to 20% — nearly tripled — while zero actual work happened in that window. That lines up with your point about not knowing how much work is lost: I can't tell either, only that the meter moved a lot and the output didn't.
+1 on the ask: a clean "you're about to hit the limit, want to checkpoint and continue in a new session" prompt, at zero token cost, would have prevented this outright.
Switching with
/modelkeeps your whole conversation — the new model continues from the same context, so nothing needs to be re-read and no work is lost. If you'd rather stop and come back later, just exit: the session is saved automatically andclaude --continue(or/resume) reopens it, optionally on a different model withclaude --continue --model <name>. Neither step spends extra tokens.Docs: https://code.claude.com/docs/en/model-config
🤖 Generated with Claude Code