claude-watch

Installation

1. Copy the scripts

cp fetch-usage.sh ~/.claude/fetch-usage.sh
cp statusline-command.sh ~/.claude/statusline-command.sh
chmod +x ~/.claude/fetch-usage.sh ~/.claude/statusline-command.sh

2. Merge settings.json into ~/.claude/settings.json

Add the statusLine and hooks blocks from settings.json into your existing ~/.claude/settings.json. If you don't have one yet, copy it directly:

cp settings.json ~/.claude/settings.json

3. Trigger an initial fetch (optional)

bash ~/.claude/fetch-usage.sh

The usage cache will otherwise populate automatically on the next tool call or Claude response.

What it shows

Everything on a single line:

Opus 4.6 (1M) | my-project • main | 5h: 3h 12m • 7d: 4d 22h | ctx 45% (430k/1000k)

Segment	Description
Model	Active model name (shortened, e.g. "Opus 4.6 (1M)")
Folder	Current working directory basename
Branch	Git branch (if in a repo)
5h / 7d	Time remaining until rate limit resets
ctx	Context window usage (percentage + token counts)

How it works

• statusline-command.sh — reads the JSON piped by Claude Code and renders a single-line status bar with model, folder, branch, rate limit reset countdowns, and context window usage.
• fetch-usage.sh — reads the OAuth token from the macOS keychain, caches it in /tmp/.claude_token_cache for 15 minutes, hits the /oauth/usage endpoint (3s timeout), and writes results to /tmp/.claude_usage_cache. On failure the stale cache is preserved.
• settings.json — wires up the statusline command and triggers fetch-usage.sh in the background on PreToolUse and Stop hooks.

Dependencies

• jq
• curl
• git (optional, for branch display)