How to make your Claude Code limit last longer: the tools that help

On a Pro, Max, Ultra, Team or Enterprise plan, /usage shows a breakdown of what counts against your plan limits: recent usage split by skills, subagents, plugins and MCP servers, with a suggested change for anything behind ten per cent or more. When the footer shows a hint such as f to fix 2 of 3, pressing f has Claude Code apply the suggestions it can, showing you each change first. It needs Claude Code 2.1.305 or later.

What it does not do: it covers this machine only, for the last day or the last week, and leaves out usage from other devices and claude.ai.

/context draws the current context as a coloured grid and suggests what to trim: tools that take up a lot of context, a memory file that has grown large, and how close you are to the limit. Because every request re-sends this context, anything it shows you trimming is saved on every turn after.

What it does not do: it shows this session as it is now, not what used your window earlier.

Claude Code passes your status line the share of the 5-hour and 7-day limits you have used, and when each resets, so the percentages can sit at the bottom of the terminal as you work. People who rarely get cut off mid-task use it to wrap up near the end of a window. How to check your Claude Code usage shows how to set it up with /statusline. In Codex, /status shows the same windows.

What it does not do: it shows how full the window is, not what filled it, and Anthropic documents the limit fields for Pro, Max and Ultra plans.

Caveman is a free, open-source add-on for Claude Code, Codex and many other coding agents. Its best-known part is a skill that makes the agent answer in short, plain sentences while leaving code, commands and error messages exactly as they are. The full install adds more: on your own machine it shrinks logs, JSON, CSV and test output before the agent reads them, and, if you sign in to it, it can pick the model and effort for each request.

Its README is frank about the numbers. On its own test, telling the model "Answer concisely" gets most of the saving on replies, and the caveman voice cuts 3 per cent more at the median, its strongest mode 35 per cent more. It reports a JetBrains test of 86 coding tasks that found 8.5 per cent fewer output tokens and no measurable loss of quality. The part that shrinks what the agent reads did more in its own runs: 33.2 per cent fewer input tokens across six test sessions, though one file type came out 9.9 per cent worse.

What to know: the command-line tool sends usage statistics by default (caveman telemetry off stops it), and the model-picking part, once you sign in to it, sends your latest requests to Caveman's own service. The skill on its own sends nothing to Caveman.

ccusage is a free command-line tool that reads the usage data coding agents keep on your machine, for Claude Code, Codex, OpenCode and others, and turns it into daily, weekly, monthly and per-session reports. ccusage blocks shows Claude Code's 5-hour windows, and it can also feed a status line. Run it with npx ccusage@latest.

tokscale is a newer free tool of the same kind that reads dozens of coding tools. It can also show your current Claude Code and Codex 5-hour and weekly percentages, by asking the vendor with the sign-in already on your machine, and it has an optional public leaderboard for people who choose to submit their totals.

What they do not do: they report and change nothing, so they help you find what to cut rather than cutting it, and they read only the machine they run on. ccusage, tokscale and the other local trackers compared with OpenReason Intel goes further.

Session Forensics is a free page from OpenReason that reads your Claude Code and Codex session files in your browser, with no install and no sign-up, and your files are never uploaded. It lists the limits you hit in the last seven days and the turns that used each one up, with a plain reason such as a long chat re-read at every step. That tells you which of the habits on this page would have saved the most.

What it does not do: it looks back at one computer, it changes nothing, and for Claude Code its 5-hour figure is an estimate, because Claude Code's files do not record the meter.

Not a product but the habit people credit most: before you stop, ask the agent to write a short note of where the task stands, then start a fresh session and point it at the note. A long session re-read after a break, or compacted, is one of the costliest things you can do to a window, and starting fresh with /clear costs nothing. The same note works to hand the task from Claude Code to Codex or back. Several community skills write the note for you; a plain request to the agent works as well.

What it does not do: the new session knows only what the note says, so a note that skips a decision means explaining it again.

OpenReason Intel, a dashboard for Claude Code and Codex usage, keeps a record of your 5-hour and weekly windows over time from every machine you use, with each turn's estimated share of the window and what it cost once the window was full. That shows which of the causes above is yours, and whether a change, such as trimming MCP servers or adding Caveman, made your windows last longer.

It shows you what happened; it does not stop a session as you work, it does not shorten anything itself, and it cannot change your vendor's limits. For the habits that go with these tools, see what people do when they run out.