28 Ways to Stop Hitting Claude Code Usage Limits

Usage limits are almost never a plan problem. They are a context problem.

Claude re-reads your entire conversation on every single message. Not a summary of it, the whole thing. So a chat that has been running for two hours is paying for two hours of history on every reply, whether or not any of it is still relevant. The cost of a message is set by how much you are dragging behind you, not by how hard the question is.

Almost everything below is a way of dragging less.

01 Run /compact before the chat gets long, not after

/compact replaces the conversation so far with a tight summary and keeps going. You lose the verbatim history and keep the conclusions, the decisions, and the current state of the work.

The mistake is waiting until you are already struggling. By then you have paid full price for every message in between. Compact at natural boundaries: when a feature is finished, when a bug is fixed, when you are about to change subject.

You can pass an instruction to /compact telling it what matters. Use it. A summary that keeps the spec, the decision and the open bug is worth far more than a generic one.

02 Switch to /model opusplan

This puts Opus on planning and Sonnet on execution, automatically, in the same session. You get the stronger model where judgment actually matters and the cheaper, faster one for the part that is mostly typing.

Most coding work is not hard reasoning. It is applying a decision you already made across a lot of files. Paying top-tier rates for that is where sessions quietly burn out.