Claude Code usage may drain fast when its to-do list updates

Claude Code’s 5-hour usage window can shrink very quickly in long coding sessions because of how its to-do tracking may interact with caching. When Claude Code uses TodoWrite, it adds a reminder block for the task list near the front of the conversation. Each time an item is marked done, added, or changed, that block can be written again.

Because the block sits in the part of the prompt that is normally cached, changing it can break the prompt cache. The next request may then process the whole long context again, sometimes hundreds of thousands of tokens, even when the visible answer is tiny. In a long session with many task updates, this can repeat many times and burn through usage quickly.

Related reactions describe sudden jumps in Claude Max limits, Claude desktop usage after an update, and Claude Code sessions where usage moved far faster than expected. The wider lesson also applies to Codex and other coding tools: large codebases and broad context can become expensive unless the tool reads only the files and lines it truly needs.