claude-statusline

Minimal Claude Code status line. No pets, no leaderboards, no sync:

Opus 4.7 │ Ctx 318.5K (31.9% / u 39.8%) │ $26.45 (1h0m) │ +12 -4 │ 5h 66% (2h37m) · wk 93% (3d7h) │ main ±3

| Segment | Source | | --- | --- | | Model | stdin model.display_name (or shortened model.id) | | Context | tail-scan of transcript_path jsonl; % computed against a per-model window table — Opus 4.6/4.7 and Sonnet 4.6 = 1M, Haiku 4.5 = 200K, family fallback otherwise. u is against 80% of the window (compaction-relevant). | | Cost + duration | stdin cost.total_cost_usd and cost.total_duration_ms (Claude Code's own numbers, not estimated) | | Edits | stdin cost.total_lines_added / total_lines_removed | | Usage quota | stdin rate_limits; 5h = 5-hour rolling window, wk = 7-day window. Shows remaining % (100 − used_percentage) and time-to-reset in parens. Only present for Claude.ai Pro/Max, and only after the session's first API response — hidden otherwise. | | Git | branch via git rev-parse --abbrev-ref HEAD; ±N = dirty file count, ✓ = clean |

Color thresholds on the Ctx value: green < 50% (usable), cyan < 80%, yellow < 95%, red otherwise. On the quota %: green > 20% remaining, yellow ≤ 20%, red ≤ 5%.

Why this exists

ccpet hardcodes the context window at 200K (line 2e5 / 16e4 in its bundle), so any session past 200K reads Ctx: 100.0% even though Opus 4.7 has a 1M window. The repo has been stale since 2025-08-30 and the bug isn't tracked. This is a strip-down replacement.

Install

npm install -g cc-statusline-simple

Then point ~/.claude/settings.json at the cc-statusline command:

{
  "statusLine": {
    "type": "command",
    "command": "cc-statusline",
    "padding": 0,
    "refreshInterval": 60
  }
}

Restart Claude Code (or open a new session) for the status line to appear.

refreshInterval (seconds) re-runs the command on a timer in addition to the event-driven updates, so the usage-quota countdown stays current while the session is idle. Drop it if you don't show the quota segment.

Usage-quota persistence

Whenever Claude Code supplies rate_limits, each render also writes the quota to disk so you can analyze it outside the status line:

| File | Contents | | --- | --- | | ~/.claude/cc-statusline/usage.json | Global "current" snapshot. Per window the freshest non-stale value wins, so a long-idle session can't clobber an active one. Written atomically. | | ~/.claude/cc-statusline/usage-<session_id>.jsonl | Per-session change history — one line appended only when that session's numbers change. The durable time series. |

rate_limits is account-global, but Claude Code hands each session the value from its own last API response. A long-idle session therefore reports a window that has already reset. Two rules keep the data honest:

• Stale windows are dropped — any window whose resets_at_epoch is already in the past is neither shown nor persisted.
• History is per-session — keyed by session_id, so concurrent sessions never interleave or race on one file.

Each record carries updated_at, session_id, plus per window used_percentage, remaining_percentage, resets_at_epoch, resets_at_raw, and resets_in_seconds. For a combined analysis across sessions, cat ~/.claude/cc-statusline/usage-*.jsonl. Writes are best-effort and never block or crash the status line.

From source

git clone https://github.com/ihainan/cc-statusline-simple.git
cd cc-statusline-simple
npm install
npm run build
# then set "command": "node /absolute/path/to/cc-statusline-simple/dist/cli.js"

Verify

npm run smoke

Spawns the CLI four times with synthetic payloads (Opus 4.7 / Haiku 4.5 / unknown model / no transcript) and prints each rendered line.

Extending

The whole thing is one ~200-line file (src/index.ts). Common edits:

• New model windows — add to MODEL_WINDOW map.
• Drop git — delete the readGitState block; saves a few ms per render.
• More transcript stats (5h burn rate, cache hit %) — they're in readContextTokens's tail scan; aggregate while you're there.