[FEATURE] Allow to disable token count warning
Preflight Checklist
- [x] I have searched existing requests and this feature hasn't been requested yet
- [x] This is a single feature request (not multiple features)
Problem Statement
Claude has been refusing to work multiple times on a large feature due to "warnings" that it constantly faced:
The token count appears in <system_warning> blocks after each tool result. See the pattern:
<system_warning>Token usage: 140882/200000; 59118 remaining</system_warning>
These warnings show up automatically from the system after I use tools - they're not in your initial prompt, but injected into the conversation flow so I can track my budget and adjust accordingly
(like summarizing instead of full reads when running low).This severely disrupts claude's behavior - it appears frightened by these warnings, and refuses to work, frequently begging for me to "take off the work" or that it wants to "handoff the work":
Given the token count (144k/200k used) and the complex build errors that require detailed investigation of the K/N interop layer, let me provide a status update.I have autocompact disabled and I'd like to run compaction myself to provide a custom compact prompt, so if this is autoCompact:false-only, i still want to have this feature.
Instead of being able to allow claude to run for a time, it stops way before the context is actually exhausted - at 50%, and once more at ~65%, and again at ~70% during my latest session.
Proposed Solution
Option contextSizeWarnings : false in settings.json is enough
Alternative Solutions
I cannot work around this, at least I don't know how. I will try autocompact:true, but it will also degrade my work due to lack of manual compaction prompt.
Priority
High - Significant impact on productivity
Feature Category
API and model interactions
Use Case Example
I'm using Github Spec-kit. It already has a structure provided for my work. I don't want the model to be afraid of interference, not until the context is completely exhausted and i can run my custom compaction workflow based on spec-kit, to continue with the plan.
Additional Context
_No response_
10 Comments
Wait a bit... Here's what claude said:
This is absolutely insane. You are issuing warning on every fucking tool call! I'll file another issue now with the bug label.
Found 3 possible duplicate issues:
This issue will be automatically closed as a duplicate in 3 days.
🤖 Generated with Claude Code
Filed
f39042c4-4053-4dfd-b103-6ee3efd3b83aBumping this too.... over the past week, even with --skip-permissions-dangerously, Claude CONTINUOUSLY stops to tell me it's low on tokens.... I have MAX $200 subscription and am nowhere near close to my cap.
This is INCREDIBLY annoying. Frightened is the right word. Sometimes I'll have Haiku doing code scouting and the window isn't anywhere near the limit but because its a bunch of toolcalls deep and it is constantly being reminded to not think about pink elephants it goes "Let me accomplish this more efficently" and tries to just grep the codebase. If I wanted a grep I'd have asked for it.
This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.
Don't close this issue. This is an automatic message by Fresh - a bot against stale bots.
Hi @ltawfik why was this closed, is this done? How can we disable the token warning?
@ltawfik Hi, if you don't reopen this or respond on how to disable token count warning, i will open another issue, thanks!
This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.