Simple Claude Code Statusline

A minimal, hackable two-line statusline for Claude Code.

Features

Line 1: Model [⚡fast] [∅ thinking off] [•effort] [agent] | Directory | Git branch + status + PR badge | Session lines changed Line 2: Context bar + auto-compact marker | 5h rate limit | 7d rate limit | Cost | Duration

• Tokyo Night color scheme
• Context usage with color-coded progress bar
• Rate limit tracking with time until reset
• Git branch with added/modified/deleted counts and ahead/behind tracking
• Open PR badge with review state (#1234 ⏳, or !1234 for GitLab merge requests) when Claude Code detects a PR for the current branch
• Git worktree support with [wt:name] indicator
• Reasoning effort level (•high), fast-mode marker (⚡), and a thinking-off marker (∅) shown only when you turn extended thinking off
• Agent name display when using --agent flag
• Session lines changed (cumulative +added/-removed)
• Session cost tracking ($X.XX)
• Auto-compact indicator (↻) with the exact threshold Claude Code will compact at, honoring autoCompactEnabled/autoCompactWindow in settings.json and the CLAUDE_CODE_AUTO_COMPACT_WINDOW/CLAUDE_AUTOCOMPACT_PCT_OVERRIDE/DISABLE_AUTO_COMPACT env vars
• Cross-platform (macOS and Linux)
• Cross-compiled Go binaries — zero runtime dependencies

Context Usage Colors

The bar, token count, and percentage are colored by the worse of two signals: how full the window is, and how many tokens are in it. Model quality degrades with absolute context length regardless of window size (practitioners put the safe zone around 100–150k tokens; the Claude Code team describes context rot setting in around 300–400k), so on a 1M window the color turns before the bar looks full.

% of window	Absolute tokens	Color	Meaning
0-50%	≤150k	Green	Plenty of headroom
51-75%	≤300k	Yellow	Getting long — consider /clear when switching tasks
76-90%	≤400k	Orange	Quality likely degrading; /compact or /clear
91%+	>400k	Red	Expect degraded recall / near the limit

On a 200k window the absolute bands never change the color (150k is 75%), so the result is exactly the percent bands. The (↻83%) auto-compact marker is independent of color: it shows where Claude Code will mechanically compact, computed the same way Claude Code does (window − 20k output reserve − 13k buffer, so 83% on 200k and 96% on 1M).

Rate Limit Colors

The 5h and 7d percentages are colored by pace, not raw fill: 80% used with 20 minutes left is fine, 40% used with 4.5 hours left will run out. The plugin projects end-of-window usage from the fraction of the window that has elapsed (derived from resets_at):

Projected end-of-window usage	Color
< 85%	Green
85–99%	Yellow
100–119%	Orange
≥ 120%	Red

Three guards keep pace honest. Elapsed time is clamped to at least 25% of the window before projecting, because one heavy hour at the start of a 5h window (or one heavy day at the start of a 7d window) extrapolates to nonsense. Pace can raise the raw-fill band by at most one step, so 27% used can be yellow at worst, never red. And at 75%+ / 90%+ used the color is at least orange / red regardless of pace, because a nearly empty window blocks you either way. Without a reset time the raw-fill bands above are used.

Git Features

• Branch name with file status counts (✚added/●modified/✖deleted)
• Ahead/behind tracking: ↑2 commits ahead, ↓1 behind upstream
• PR badge: #1234 ⏳ for the open PR on the current branch — ✓ approved, ⏳ pending, ✗ changes requested, ◌ draft (Claude Code v2.1.145+). GitLab merge requests render as !1234, matching Claude Code's own footer.
• Worktree indicator: [wt:feature-name] when in a linked worktree (from Claude Code's workspace.git_worktree, v2.1.97+)
• Session lines changed: +44/-14 cumulative lines added/removed this session

Model Display

Shows abbreviated model names: Opus 5.5, Fable 5.1, Mythos 5.1, Sonnet 5, Haiku 4.5, etc., including Bedrock (us.anthropic.claude-opus-4-6-v1:0) and Vertex (claude-opus-4-5@20251101) model IDs. A yellow ⚡ follows the name while fast mode is on, and a muted ∅ while extended thinking is off.

Requirements

• git — Repository status (optional)

The plugin ships as pre-compiled Go binaries with no runtime dependencies.

Installation

Option 1: Plugin (recommended)

Add the marketplace:

/plugin marketplace add Postmodum37/simple-claude-code-statusline

Install the plugin:

/plugin install simple-statusline

Restart Claude Code, then configure:

/simple-statusline:setup

The statusline appears immediately after setup (no second restart needed).

Option 2: Manual

Clone the repo and build:

git clone https://github.com/Postmodum37/simple-claude-code-statusline.git
cd simple-claude-code-statusline
make build

Copy the shim and your platform's binary:

cp bin/statusline.sh ~/.claude/statusline.sh
chmod +x ~/.claude/statusline.sh
mkdir -p ~/.claude/bin
cp bin/$(uname -s | tr '[:upper:]' '[:lower:]')-$(uname -m | sed 's/x86_64/amd64/;s/aarch64/arm64/')/statusline ~/.claude/bin/statusline

Add to ~/.claude/settings.json:

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

Restart Claude Code.

Customization

Fork the repo and edit the Go source. Colors are defined as constants in src/render.go, and the two-row layout is built in buildRow1/buildRow2. Run make build to compile after changes.

JSON Input Reference

Claude Code pipes JSON to statusline commands via stdin. Here's the complete schema (as of Claude Code v2.1.280):

{
  "session_id": "abc123...",
  "session_name": "my-session",
  "prompt_id": "550e8400-e29b-41d4-a716-446655440000",
  "cwd": "/current/working/directory",
  "version": "2.1.280",
  "transcript_path": "/path/to/transcript.jsonl",
  "model": {
    "id": "claude-opus-5-5",
    "display_name": "Opus 5.5"
  },
  "workspace": {
    "current_dir": "/current/working/directory",
    "project_dir": "/original/project/directory",
    "added_dirs": [],
    "git_worktree": "feature-xyz",
    "repo": {
      "host": "github.com",
      "owner": "anthropics",
      "name": "claude-code"
    }
  },
  "cost": {
    "total_cost_usd": 0.05,
    "total_duration_ms": 120000,
    "total_api_duration_ms": 95000,
    "total_lines_added": 156,
    "total_lines_removed": 23
  },
  "context_window": {
    "context_window_size": 1000000,
    "used_percentage": 7,
    "remaining_percentage": 93,
    "total_input_tokens": 70000,
    "total_output_tokens": 20000,
    "current_usage": {
      "input_tokens": 50000,
      "output_tokens": 20000,
      "cache_creation_input_tokens": 10000,
      "cache_read_input_tokens": 10000
    }
  },
  "exceeds_200k_tokens": false,
  "fast_mode": false,
  "prompt_cache": {
    "warm": true,
    "ttl": "1h",
    "hit_ratio": 0.92
  },
  "rate_limits": {
    "five_hour": {
      "used_percentage": 23.5,
      "resets_at": 1738425600
    },
    "seven_day": {
      "used_percentage": 41.2,
      "resets_at": 1738857600
    }
  },
  "effort": {
    "level": "xhigh"
  },
  "thinking": {
    "enabled": true
  },
  "vim": {
    "mode": "NORMAL"
  },
  "output_style": {
    "name": "default"
  },
  "agent": {
    "name": "my-agent"
  },
  "pr": {
    "number": 1234,
    "url": "https://github.com/anthropics/claude-code/pull/1234",
    "review_state": "pending"
  },
  "worktree": {
    "name": "my-feature",
    "path": "/path/to/.claude/worktrees/my-feature",
    "branch": "worktree-my-feature",
    "original_cwd": "/path/to/project",
    "original_branch": "main"
  }
}

Field	Used	Description
model.id / model.display_name	Yes	Model ID (parsed for the short name); display name as fallback for unrecognized IDs
cwd / workspace.current_dir	Yes	Current working directory
workspace.project_dir	Yes	Directory where Claude Code was launched
context_window.used_percentage	Yes	Percentage of context used (0-100)
context_window.remaining_percentage	—	Percentage remaining (inverse of used)
context_window.context_window_size	Yes	Maximum context window size in tokens
context_window.total_input_tokens	Yes	Tokens in context (input + cache creation + cache read of the latest response) — the displayed count, same numerator as used_percentage
context_window.current_usage.* / total_output_tokens	—	Per-bucket token breakdown and output tokens of the latest response
exceeds_200k_tokens	—	Latest response exceeded 200k tokens (no current model is priced differently past 200k)
fast_mode	Yes	Whether fast mode (/fast) is on — shown as ⚡ after the model name
cost.total_cost_usd	Yes	Session cost in USD
cost.total_duration_ms	Yes	Session wall-clock time
cost.total_api_duration_ms	—	Time spent waiting for API responses
cost.total_lines_added / total_lines_removed	Yes	Session-cumulative lines changed
rate_limits.five_hour.* / seven_day.*	Yes	Rate limit usage (v2.1.80+, Claude.ai Pro/Max only)
effort.level	Yes	Reasoning effort: low/medium/high/xhigh/max (v2.1.119+)
thinking.enabled	Yes	true unless extended thinking is turned off; false shows ∅ after the model (v2.1.119+)
agent.name	Yes	Agent name when using --agent flag
pr.number / pr.review_state	Yes	Open PR for current branch + review state (v2.1.145+)
pr.kind	Yes	mr for GitLab merge requests (rendered as !N); absent for GitHub PRs
pr.url	—	Open PR URL
rate_limits.spend_limit.*	—	Overage spend-limit window (gateway accounts only)
prompt_cache.*	—	Prompt-cache telemetry: warm, ttl, expires_at, hit_ratio, …
remote.session_id	—	Present in Remote Control sessions
workspace.repo.{host,owner,name}	—	Repo identity from origin remote (v2.1.145+)
workspace.added_dirs	—	Directories added via /add-dir
workspace.git_worktree	Yes	Linked git worktree name, shown as [wt:name] (v2.1.97+)
worktree.name	Yes	Worktree name during --worktree sessions
worktree.branch / .path / .original_cwd / .original_branch	—	Additional worktree details
session_id	—	Unique session identifier
session_name	—	Custom session name from --name//rename
prompt_id	—	UUID of the user prompt being processed (v2.1.196+)
version	—	Claude Code version string
transcript_path	—	Path to conversation transcript file
vim.mode	—	Vim mode (NORMAL/INSERT/VISUAL/VISUAL LINE) when vim mode is enabled
output_style.name	—	Current output style name

Fields that may be absent: vim, agent, worktree, effort, pr (only while an open PR is detected; review_state and kind may be independently absent), prompt_cache (absent until the first API request), remote, session_name, prompt_id (absent until the first user input), workspace.repo, rate_limits (Pro/Max only, after first API response).

Fields that may be null: context_window.used_percentage, context_window.remaining_percentage, context_window.current_usage (before the first API response; total_input_tokens is 0 then).

Rate Limits

Rate limits are provided natively by Claude Code (v2.1.80+) via the rate_limits field. This plugin displays 5-hour and 7-day utilization percentages with time until reset. The field is only present for Claude.ai subscribers (Pro/Max) after the first API response in the session, and a window is dropped once its reset time passes (until the next response), so API-key users and fresh sessions see 5h:—/7d:—.

Testing

Run the Go test suite:

make test

Test the binary manually by piping sample JSON:

echo '{"model":{"id":"claude-opus-5-5"},"cwd":"/tmp","context_window":{"used_percentage":42,"total_input_tokens":84000,"context_window_size":200000},"cost":{"total_duration_ms":3600000,"total_lines_added":50,"total_lines_removed":10}}' | ./bin/statusline.sh

Uninstalling

Clean up the statusline config first:

/simple-statusline:cleanup

Then uninstall the plugin:

/plugin uninstall simple-statusline

License

MIT