Detailed issue report: Context compaction on long threads instantly consumes 5-hour Pro quota

Hello everyone, I wanted to share a specific context-window issue I ran into with Claude Pro (€22/month tier) during a multi-day coding session, in hopes of finding workarounds or providing constructive feedback on how limits interact with long threads. I’ve been working on a web application codebase using Sonnet 5 in High thinking mode. As the project grew, the conversation history naturally expanded. I noticed that the longer the thread became, the faster the 5-hour usage limit was reached, which makes sense given how input tokens accumulate over extended contexts. Today, a specific behavior effectively locked me out without allowing any progress: 1. 12:00 PM: I submitted a list of bugs to fix. Claude started processing, but hit the usage limit 10 minutes in. 2. 5:00 PM (Exact 5-hour reset): I returned to the thread right at the reset time and simply replied "Continue". 3. The Loop: Claude responded that it needed to clear/compact older parts of the conversation history due to length. It spent ~60 seconds performing this auto-compaction. 4. Immediate Limit: Right after finishing the compaction output, the system immediately informed me that my usage limit was reached again for the next 5 hours. Because the context window was at its maximum limit, the single prompt required to summarize/compact the thread consumed 100% of the token quota allocated for that 5-hour session in less than two minutes, without generating a single line of actual code or fixing any bug. Is there a way to prevent Claude from attempting auto-compaction when close to context limits? For those working on long-term codebases, what is your workflow to avoid this thread-bloat? Do you manually summarize progress and open a fresh chat every few hours, or use Claude Projects with specific context files instead of long chats?