Usage diagnostics

Claude Code Usage Checklist

Run through these checks before assuming your plan is too small. Many usage problems come from context bloat, model choice, or unintended API billing.

Before a long session

  1. Run /clear if you are starting a new task.
  2. Use Sonnet for execution unless Opus is clearly needed.
  3. Plan the file scope before letting the agent edit widely.
  4. Check whether ANTHROPIC_API_KEY is set.
  5. Use smaller prompts instead of dumping unrelated context.

When you hit a limit

  1. Read the reset time in the Claude Code message.
  2. Switch to a lighter model if available.
  3. Use /compact if the context window is the problem.
  4. Use /cost if running through API mode.
  5. Upgrade only if limits interrupt normal work repeatedly.

Five common causes

Long conversation

Old context keeps getting carried forward. Clear or compact when the task changes.

Large repo context

Reading many files and producing many diffs increases context pressure.

Opus-heavy use

Use Opus where reasoning matters, then switch down for execution.

Tool-heavy sessions

Search, file context, and repeated tool calls can increase usage.

API key mode

API key billing behaves differently from plan-limited subscription usage.

Weekly caps

Some paid plans have weekly usage limits in addition to session windows.

Sources