What it shows you
- Context you can feel. A gauge, the token count and the percentage — so you know you are getting close before Claude tells you.
- Both rate-limit windows. The 5-hour and the 7-day, each with the time until it resets — and the 7-day per model, on a seat where one model has its own weekly limit.
- Cost as you go. The session’s spend, and on a seat billed against a monthly budget rather than the rolling windows, that budget beside it.
- Your subagents. A second panel, a row per subagent, with its name, model, what it is working on, tokens, elapsed time and a status glyph.
- Git, from the filesystem. Branch, worktree and dirty markers, and the
repository’s own name from its
origin, resolved by reading the repository rather than shelling out for everything.
It is one Rust binary. No runtime, no Node, nothing to keep up to date — Claude
Code invokes it directly, and it starts in about a millisecond. Most of a render
is spent asking git about your working tree, so inside a repository the whole
thing costs tens of milliseconds rather than a couple; outside one it is barely
more than the process starting.
It can also stop a session running past the line
A PostToolUse hook watches your usage and, when you cross a cap, tells Claude
to wrap up and hand off rather than letting the session run on. The shipped caps
are 65% of the context window, 95% of the 5-hour window, 98% of the 7-day one
and 90% of a monthly budget — all four are yours to change, and a repository you
cloned cannot raise them past what you set. See Configure.
Getting it
Apple Silicon macOS, with Homebrew:
brew trust --formula virajp/tap/claude-status
brew tap virajp/tap
brew install --formula virajp/tap/claude-status
claude-status --configureOr with mise:
mise use --global "github:virajp/claude-status@latest"
claude-status --configurebrew trust is not optional — Homebrew 6 will not load a third-party formula it
has not been told to trust, so the install fails without it. You do it once.
brew tap is not optional either: the qualified name carries brew trust, but
brew install cannot find the formula without the tap.
--configure is the real last step, not a formality: it is what wires Claude
Code to the binary. Install covers both routes in full.
Where to go next
| Page | For |
|---|---|
| Install | getting the binary, and wiring Claude Code to it |
| Configure | the three config layers, and what you can change |
| Per-repo | naming one repository in the bar — the only per-repo setting |
| Segments | every segment, what it draws, and when it sits out |
| Diagnosing | --doctor, and the questions it answers |
Wondering why something behaves the way it does? The decision record carries the reasoning behind each choice — including the ones that were later reversed. It is written for implementers rather than users; these pages are the user-facing documentation.