Context limits / session interruptions are destroying productivity — significant regression vs 2 months ago
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?
I'm a paying Claude Code user and over the last few weeks something has changed dramatically in how sessions are managed. Two months ago I could work through complex, multi-file engineering tasks in a single session without interruption. Now:
Sessions get cut mid-task due to "token limits" with waits of up to 3 hours before I can continue
I hit weekly reset limits that force me to ration my prompts — I'm thinking about tokens instead of thinking about my work
Context compression kicks in aggressively, losing critical context and causing regressions in the work already done
When a session is cut, the next session starts cold and makes mistakes the previous one had already resolved — consuming even more tokens to fix
The net effect: I now spend more mental energy managing Claude's limitations than solving actual problems. That's the opposite of what an AI assistant should do.
I understand infrastructure has costs. But this feels like a deliberate squeeze to push users toward the $100/month plan — a 5x increase. If the product genuinely can't sustain my usage at the current plan, say so clearly. What's not acceptable is a silent regression that degrades the experience without explanation.
I'm not asking for unlimited usage. I'm asking for consistency and transparency. Tell me what the actual limits are, document them, and don't change them without notice.
This is the feedback channel you pointed me to. I hope someone actually reads it.
What Should Happen?
Sessions should continue uninterrupted through complex multi-file tasks, or at minimum provide clear advance warnings before hitting limits. Usage limits (daily, weekly, context window) should be explicitly documented and consistent — not silently changed. If a session must end, the next one should resume with enough context to avoid regressions that waste additional tokens fixing mistakes the previous session had already resolved.
Error Messages/Logs
Steps to Reproduce
Start a Claude Code session on a complex, multi-file engineering task (e.g. PHP/CodeIgniter app with controllers, models, views, routes — 10+ files touched in one session)
Work continuously for 1-2 hours with back-and-forth iterations, debugging, and applying fixes
Session gets cut due to token/context limits with no advance warning
Wait up to 3 hours for reset
Resume session — Claude starts cold, loses prior context, repeats mistakes already fixed, consuming additional tokens to recover
Hit weekly reset limit on a heavy-use day, forced to ration prompts for the rest of the week
Frequency: Multiple times per week. Regression started approximately June–July 2026. Was not happening 2 months ago on the same plan.
Claude Model
Sonnet (default)
Is this a regression?
Yes, this worked in a previous version
Last Working Version
_No response_
Claude Code Version
2.1.215 (Claude Code)
Platform
Anthropic API
Operating System
macOS
Terminal/Shell
VS Code integrated terminal
Additional Information
_No response_