Claude Code Statusline

A labeled-row status bar for the Claude Code TUI. Displays location, weather,
environment info, context window usage, git status, and session metrics.
Works on macOS and Linux.

─── | CC STATUSLINE | ────────────────────────────────────────────────────
LOC: Ithaca | 5:02p | Sun Mar 01 | ⛅  25°F · 10mph · 56%
ENV: CC: v2.1.63 | OAuth | anthropic.claude-4.6-sonnet
● CONTEXT: ●●●●●○○○○○○○○○○○○○○○○○○○ 23% used | In:3.4M  Out:21.0k
◆ GIT: ai-cli-workspace | Branch: main | ↑1 ↓0 | clean
+ SESSION: +30 -5 lines | 37m54s | #476c2e1 | 🔋 30% | ~$10.52 est

Requirements

• macOS or Linux (including WSL)
• Claude Code v2.x+
• jq, curl, bc (brew install jq bc on macOS, sudo apt install -y jq curl bc on Debian/Ubuntu)

The setup.sh script runs a preflight check and tells you exactly which commands to run if anything is missing.

Install

Option 1: Let Claude Code install it

Just point a Claude Code session at the repo:

Clone https://github.com/pete-builds/claude-code-statusline and run its setup.sh

Claude handles cloning, dependency checks, and the settings.json merge.

Option 2: Setup script

git clone https://github.com/pete-builds/claude-code-statusline.git
cd claude-code-statusline
./setup.sh

Windows/WSL note: clone from inside WSL (not from Windows Git or git-bash) so the shell scripts get LF line endings. The repo pins eol=lf via .gitattributes, but some tools ignore attributes. If you see bad interpreter: /bin/bash^M, run dos2unix statusline.sh setup.sh.

Option 3: Manual install

1. Copy statusline.sh to your Claude config directory:

cp statusline.sh ~/.claude/statusline.sh
chmod +x ~/.claude/statusline.sh

1. Add the statusline config to ~/.claude/settings.json. If the file already
   exists, merge this into your existing settings:

{
  "statusLine": {
    "type": "command",
    "command": "~/.claude/statusline.sh"
  }
}

Start a new Claude Code session — the statusline appears automatically.

What each row shows

Auth display

Labeling gateway keys

If you run more than one key against the same gateway (e.g. a personal key and
a work key that both go through litellm.example.com), set
ANTHROPIC_KEY_LABEL before launching Claude Code and the label appears after
the gateway host on the ENV row: GW:example·work. Capped at 16 characters.

The usual pattern is a shell alias or wrapper that exports the label alongside
the base URL and key it belongs to.

Bash / zsh (Linux, macOS, WSL, Git Bash):

alias claude-work='ANTHROPIC_BASE_URL=https://litellm.example.com \
                   ANTHROPIC_AUTH_TOKEN=sk-... \
                   ANTHROPIC_KEY_LABEL=work claude'
alias claude-personal='ANTHROPIC_BASE_URL=https://litellm.example.com \
                       ANTHROPIC_AUTH_TOKEN=sk-... \
                       ANTHROPIC_KEY_LABEL=personal claude'

PowerShell (Windows): aliases can't set env vars inline, so use a function:

function claude-work {
  $env:ANTHROPIC_BASE_URL   = "https://litellm.example.com"
  $env:ANTHROPIC_AUTH_TOKEN = "sk-..."
  $env:ANTHROPIC_KEY_LABEL  = "work"
  claude @args
}

Windows cmd: use a .bat wrapper that sets the three variables and then
calls claude %*.

Nothing is validated server-side; the label is a local display hint. If
ANTHROPIC_KEY_LABEL is unset, the ENV row shows GW:host on its own as
before.

Notes for Cornell AI Gateway users

If your ANTHROPIC_BASE_URL points at https://api.ai.it.cornell.edu/, the
ENV row shows GW:cornell (the parser takes the second-to-last dotted
segment of the host).

The gateway rate table in statusline.sh is preloaded with Cornell's flat
$/1M-token pricing and only fires when GW_HOST == "cornell", so no other
gateway is affected. When rates change, edit the case "$MODEL" block —
values were last verified against the Confluence pricing page (541787315,
v187) on 2026-07-20.

ANTHROPIC_KEY_LABEL is handy alongside the bridge-keychain flow when you
run more than one Cornell key (e.g. personal vs a shared project key)
against the same base URL — set it in whichever alias exports the key and
the bar shows GW:cornell·<label>.

Adapting the rate table for your own gateway

Non-Cornell users can copy the pattern. Replace "cornell" in the guard
with the second-to-last hostname segment of your gateway (whatever appears
after GW: in the bar), then edit the case "$MODEL" block to match your
gateway's model IDs and rates.

Active skill marker

The ENV row can show which Skill (plugin skill or slash command) was most
recently invoked as 🔌 <name> for a short window after it runs. Useful when
you want the bar to reflect that a plugin like codex:rescue or code-review
is what's driving the current turn.

The Skill tool doesn't propagate into the statusline JSON payload, so a
PostToolUse hook stamps a marker file that the statusline reads on the next
render. The marker is treated as fresh for 60 seconds; tune MARKER_TTL_SECS
in statusline.sh if you want it shorter or longer.

Enable it:

1. Copy the hook script:

mkdir -p ~/.claude/hooks
cp hooks/skill-marker.sh ~/.claude/hooks/skill-marker.sh
chmod +x ~/.claude/hooks/skill-marker.sh

1. Add a PostToolUse hook for the Skill tool to ~/.claude/settings.json.
   If your hooks block already exists, merge this entry into it:

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Skill",
        "hooks": [
          {
            "type": "command",
            "command": "bash \"$HOME/.claude/hooks/skill-marker.sh\""
          }
        ]
      }
    ]
  }
}

The hook is opt-in — the statusline silently no-ops when the marker file is
absent or stale, so leaving it unwired costs nothing.

Context bar colors

The bar fills left to right as your context window fills up. Color indicates
how close you are to the context limit:

• Green → normal
• Yellow → approaching limit, start wrapping up your current task
• Red → near limit, time to manage context

Context management options:

• /compact — summarizes conversation history in place. Convenient but can lose context or misrepresent what was discussed. Use with caution on complex tasks.
• Safer pattern: ask Claude to write a summary of the current state to a markdown file, then run /clear, and open the new session by reading that file. You get a clean context with reliable continuity.

Data sources

• Location: ipapi.co — free, HTTPS, no API key, cached 1 hour
• Weather: Open-Meteo — free, no API key, cached 10 minutes
• Both services must be reachable. If blocked by a Pi-hole or firewall, whitelist
  ipapi.co and api.open-meteo.com

Privacy & trust

Worth knowing before you install.

Third-party network calls. Every status refresh sends cached requests (not every keystroke) to three providers:

• api.ipify.org returns your public IP
• ipapi.co geolocates your IP to city and lat/lon
• api.open-meteo.com returns current weather for those coordinates

None of these calls are authenticated. Nothing is sent to me or to any server I control. But those three providers can log your IP and approximate location. If that's not acceptable for your setup (corporate VPN, privacy-conscious workflow, etc.), either edit the script to remove the location and weather blocks or run it behind a filtering proxy.

Supply chain. The script runs with your shell's privileges every time Claude Code refreshes the statusline. If this repository or my GitHub account is compromised, anyone who pulls updates or re-runs setup.sh gets the attacker's code executed automatically. The repo has no commit signing enforcement. To freeze what you're running, pin to a specific commit SHA:

git checkout <commit-sha>
./setup.sh

Local writes. Setup installs to ~/.claude/statusline.sh and merges a statusLine key into ~/.claude/settings.json, preserving other keys. Runtime writes go to ~/.cache/claude/statusline/: location, weather, and git caches, plus context_window_debug.log, a rotating log capped at 200 lines containing short session IDs, token counts, and model names. Nothing leaves your machine through this log.

Non-risks. Session JSON from Claude Code is shell-quoted via jq @sh before use, so there's no command injection path. API keys, when present, are only rendered as their last 4 characters. The script doesn't call any network endpoint beyond the three listed above.