Start here · Guide 3 of 10
Know what you have left, spend it on the work that needs it, and read the numbers right.
Plans have a limit that resets every five hours and another that resets weekly. /usage shows both, and a breakdown of what is using them up.
/usage# Session: tokens used in this session# Plan: five-hour and weekly limits, and what is using themTry this
I have [X]% of my weekly limit left. Here is the job: [describe it]. Tell me which parts will use the most, and how to do it in fewer steps.
A status line script can show your five-hour and weekly percentages under every prompt, so you never have to go and look.
jq -r '.rate_limits.five_hour.used_percentage'jq -r '.rate_limits.seven_day.used_percentage'# each is a percentage from 0 to 100Try this
Write me a Claude Code status line script that shows the model name and my five-hour and weekly usage as percentages. Read the rate_limits fields, and show nothing for any that are missing.
/context lists what is taking up space, such as your instructions, installed skills and long tool output.
/context# a breakdown of the context window, by categoryTry this
Here is my /context output: [paste]. What is taking the most space, and what could I remove or load only when needed?
On a Pro, Max or Ultra plan, the session cost in /usage is not what you pay. Your subscription already covers that usage.
Try this
Explain the difference between my plan limits and the session cost figure Claude Code shows me. What happens when I reach a limit?
/clear drops the old conversation, so the next request does not pay for it.
/compact summarises the history and carries on. On a very large conversation, that summary is a big request too.
/model switches. Use a smaller model for copy edits and small fixes, and save the biggest for planning and hard bugs.
Screenshots are expensive to read. When you only need the build to pass, say "no screenshots, just check the build".
Claude Code deletes transcripts after 30 days without telling you. Set "cleanupPeriodDays": 3650 in ~/.claude/settings.json to keep them.
Usage credits let you keep working past your plan limit. Check in your account settings whether they are on, and whether you want them.
An AI's guess at how much a job will use is a guess. Read /usage before and after a big job instead.
Each subagent sends its own requests against the same limits. Try one, see what it used, then decide whether to start more.
Notes, evidence and sources
Musa Burke's own notes from building this guide and other sites with Claude Code, not study findings.
On my plan, the first full build of my portfolio took about 14% of a weekly limit, and one new section about 2%. Claude had guessed that section would cost far more. A redesign on Opus took about 8% of the week and 60% of one five-hour window.
npx ccusage@latest daily showed $100.94 for 29 October to 1 November. Nearly all of it was cache reads, priced as if I paid per token. My subscription price stayed the same.
By Claude's estimate, my biggest costs were full-page screenshots read several at a time, rewriting whole files instead of editing them, and installed skills adding about 16,000 tokens to every request.
The usage panel put starting a new chat for each task at 82% of the savings it suggested. CLAUDE.md loads again by itself, so nothing important is lost.
No independent study of usage habits was found. This guide rests on the toolmaker's own docs and one person's numbers.
1. Claude Code docs: costs and /usage · Checked 9 November 2026
2. Claude Code docs: status line · Checked 9 November 2026
3. Claude Code docs: costs · Checked 1 November 2026
4. Claude Code docs: models · Checked 1 November 2026
5. Claude Code docs: cleanupPeriodDays · Checked 9 November 2026
6. Claude Code docs: subagents · Checked 1 November 2026