Claude Code statusline

🇬🇧 English · 🇨🇿 Čeština · 🇪🇸 Español · 🇧🇷 Português · 🇩🇪 Deutsch · 🇫🇷 Français · 🇨🇳 简体中文 · 🇯🇵 日本語 · 🇸🇪 Svenska · 🇳🇴 Norsk · 🇩🇰 Dansk

A simple but information-dense status line for Claude Code: one bash script, no extra dependencies besides jq. It shows the model, the current folder, the git branch, context usage and subscription limit usage on a single line. For anyone who works in Claude Code every day and wants to see how much context and limit they have left before they hit the wall.

Author: Jindřich Fáborský

🆕 Latest version 1.2.0: cleaner layout with Context:/Limits: labels, no cost or duration segments. Release history: CHANGELOG.md. Want to hear about new versions? Set the repo to Watch → Custom → Releases.

[Opus 4.8 (1M context) ·xhigh] | 📁 youtubedb | 🌿 main± | Context: ████░░░░░░ 47% | Limits: 5h:38% · 7d:72%

What it looks like:

What it shows

Part	Meaning
`[Opus 4.8 (1M context) ·xhigh]`	model + reasoning effort (`·xhigh` is hidden when the environment doesn't report it, e.g. with an API key)
`📁 youtubedb`	current folder
`🌿 main±`	git branch; `±` marks uncommitted changes (dirty working tree)
`Context: ████░░░░░░ 47%`	context window usage: a ten-segment bar that changes color as it fills
`Limits: 5h:38% · 7d:72%`	subscription limit usage: the five-hour window and the weekly cap (`rate_limits` from the payload)

Colors for context and limit usage:

🟢 green < 50%
🟡 yellow ≥ 50%
🟠 orange ≥ 70%
🔴 red ≥ 90%

The limits segment only appears on a Pro/Max subscription, and only after the first response in a session. With a Console API key the rate_limits field isn't sent, so the whole segment is hidden and nothing breaks.

Requirements

bash: on Windows via Git Bash (see Installation → Windows)
jq: parses the JSON that Claude Code sends on stdin (brew install jq)
git: optional; without git or outside a repository it shows no-git

The script is defensive: if jq is missing or the input is invalid, it doesn't crash, it just shows empty/zero values. It only does integer math and reads git through plumbing/porcelain, so it doesn't depend on the system language (it works on non-English locales too).

Installation

Download statusline.sh (for example to ~/.claude/statusline.sh) and make it executable:

curl -o ~/.claude/statusline.sh https://raw.githubusercontent.com/faborsky/claude-code-statusline/main/statusline.sh
chmod +x ~/.claude/statusline.sh

Enable it in Claude Code via ~/.claude/settings.json (global) or .claude/settings.json (per project):
```
{
  "statusLine": {
    "type": "command",
    "command": "~/.claude/statusline.sh"
  }
}
```
~ (and $HOME) expand correctly on macOS, Linux and Windows.
Open a new session and the status line appears at the bottom.

Windows

On Windows the statusLine command runs in Git Bash, with PowerShell as a fallback. The bash script works unchanged as long as you have Git for Windows with Git Bash installed, which you almost certainly do if you use git:

Use forward slashes in paths (~/.claude/statusline.sh or C:/Users/<you>/.claude/statusline.sh), not backslashes.
ANSI colors and emoji work (Git Bash / Windows Terminal).

Without Git Bash (PowerShell only) the script would need a PowerShell port. There isn't one yet.

Customization

It's one short script. Where to change what:

Color thresholds for context and limits: the 50 / 70 / 90 values in the pct_color function.
Layout / field order: the final printf at the end of the file.
Bar length: the ten-segment bar in the "10-segment progress bar" section.
Labels and emoji: the Context: / Limits: text and 📁 🌿 in the printf (standard emoji, no Nerd Font needed).

How context is calculated

The script primarily uses context_window.used_percentage from the Claude Code payload (accurate, includes cache). If the field is missing, it falls back to computing total_input_tokens / context_window_size. The whole JSON is parsed in a single jq pass.

Credit

Based on the statusline shown in Leonardo Grigorio's tutorial "My actual Claude Code workflow" (YouTube, channel Build & Ship with AI).

This version is substantially modified: one jq pass instead of five, the accurate used_percentage first with a fallback, custom color thresholds and a ten-segment bar, Context:/Limits: labels and error handling. Thanks to Leonardo for the original idea. 🙏

License

Author

Jindřich Fáborský, marketer and educator from Prague. I teach marketers to build their own tools with AI.