Claude Code Statusline Script

Overview

A Python script that generates a formatted status line for Claude Code, displaying the current model, working directory, and context usage information. The script provides real-time feedback on token consumption relative to the model's context window limit.

Project Structure

claude-code-status-line/
├── src/
│   └── claude_code_statusline/
│       ├── __init__.py          # Package initialization
│       ├── statusline.py        # Main statusline command
│       ├── config/              # Configuration system
│       │   ├── defaults.py      # Default widget configuration
│       │   ├── loader.py        # Config file loading
│       │   └── schema.py        # Pydantic schemas
│       ├── widgets/             # Widget system
│       │   ├── base.py          # Base widget class
│       │   ├── registry.py      # Widget registration
│       │   └── builtin/         # Built-in widgets
│       ├── parsers/             # Transcript parsing
│       │   ├── jsonl.py         # JSONL parser
│       │   └── tokens.py        # Real token counting
│       ├── cli/                 # CLI commands
│       │   └── commands.py      # install/uninstall/doctor
│       └── utils/               # Shared utilities
├── scripts/
│   ├── install.sh               # One-liner installation
│   └── uninstall.sh             # Clean removal
└── tests/                       # Test suite

Installation

Quick Install (Recommended)

Install with a single command:

curl -fsSL https://raw.githubusercontent.com/wpfleger96/claude-code-status-line/main/scripts/install.sh | bash

This will:

• Install claude-code-statusline using uv or pipx
• Configure Claude Code automatically
• Create a backup of existing configuration

Requirements: Either uv or pipx

Manual Installation

Option 1: Using uv tool

uv tool install claude-code-statusline
claude-statusline install

Option 2: Using pipx

pipx install claude-code-statusline
claude-statusline install

Upgrading

To upgrade to the latest version:

# If installed with uv
uv tool upgrade claude-code-statusline

# If installed with pipx
pipx upgrade claude-code-statusline

Uninstalling

To uninstall:

# Quick uninstall
curl -fsSL https://raw.githubusercontent.com/wpfleger96/claude-code-status-line/main/scripts/uninstall.sh | bash

# Or manually
claude-statusline uninstall
uv tool uninstall claude-code-statusline  # or: pipx uninstall claude-code-statusline

CLI Commands

claude-statusline              # Output statusline (reads JSON from stdin)
claude-statusline install      # Configure Claude Code integration
claude-statusline uninstall    # Remove Claude Code configuration
claude-statusline doctor       # Verify installation health
claude-statusline --version    # Show version

Health Check

Verify your installation is working correctly:

claude-statusline doctor

This checks:

• settings.json configuration
• Config file validity
• Statusline execution
• Claude Code directory

Configuration

Widget Customization

Customize your status line by creating ~/.config/claude-statusline/config.yaml:

version: 1
lines:
  - - type: model
      color: cyan
    - type: separator
    - type: directory
      color: blue
    - type: separator
    - type: git-branch
      color: magenta
    - type: separator
    - type: context-percentage
      color: auto  # auto-colors based on usage
    - type: separator
    - type: cost
      color: auto  # auto-colors based on amount
    - type: separator
    - type: session-clock
      color: white

Widget Options:

• type (required): Widget identifier (see Available Widgets list below)
• color (optional): Color name or "auto" for dynamic coloring
• bold (optional): Make text bold (default: false)
• metadata (optional): Widget-specific configuration

Color Options: white, black, red, green, yellow, blue, magenta, cyan, dim, auto, none

Auto-Generated Config: If no config file exists, one is automatically created with all widgets enabled. Delete ~/.config/claude-statusline/config.yaml to regenerate defaults.

Stale Config Warning: When new widgets are added, the script warns you:

Warning: Config is missing widgets from defaults: session-clock.
Delete ~/.config/claude-statusline/config.yaml to regenerate with new defaults.

Features

Customizable Widget System

• Configurable Layout: Customize which widgets appear and in what order via ~/.config/claude-statusline/config.yaml
• Built-in Widgets: 14 widgets including model, directory, git status, context, cost, session info
• Flexible Styling: Per-widget color customization and bold formatting
• Automatic Fallback: Missing config uses sensible defaults with all widgets enabled
• Stale Config Detection: Warns when config is missing newly-added widgets

Available Widgets:

• model - Claude model name (e.g., "Sonnet 4.5")
• subscription - Subscription type (Pro/Max) or API usage indicator
• directory - Current working directory
• git-branch - Active git branch name
• git-changes - Staged/unstaged changes count
• git-worktree - Git worktree name (if applicable)
• context-percentage - Context usage with progress bar
• context-tokens - Token count (current/limit)
• cost - Session cost in USD
• lines-added - Lines added in session
• lines-removed - Lines removed in session
• lines-changed - Combined added/removed
• session-id - Claude Code session UUID
• session-clock - Elapsed session time
• separator - Visual separator (default: "|")

Real Token Counting

• Primary Source: Uses Claude Code's context_window field from status payload for authoritative token counts
• Fallback Support: Falls back to transcript parsing for older Claude Code versions or when context_window is unavailable
• Actual Token Values: Reads real token counts instead of estimating (from context_window.current_usage or transcript message.usage fields)
• Compact-Aware: Correctly handles /compact boundaries, counting only active context
• Accurate Context Limit: Uses context_window.context_window_size directly from Claude Code instead of model lookups

Smart Context Tracking

• Priority-Based Sources: Prefers context_window payload data → transcript parsing → model lookups
• Session Metrics: Continues parsing transcripts for session duration and compact boundary detection
• Backwards Compatible: Gracefully handles missing or null context_window data from older Claude Code versions
• Calibration Tool: Includes calibrate_token_counting.py to validate and improve accuracy

Token Calculation

How It Works

Primary Method (Claude Code 2.0.70+):

• Uses context_window.current_usage from Claude Code's status payload
• Calculates current context: input_tokens + cache_creation_input_tokens + cache_read_input_tokens
• Uses context_window.context_window_size for accurate model limit
• No estimation or overhead calculation needed - values are authoritative from Claude Code

Fallback Method (Older Versions):

• Reads actual token counts from transcript message.usage fields (input_tokens + output_tokens)
• Falls back to estimation (character count ÷ 3.31) only when usage data is unavailable
• Retrieves model-specific context limits from cached API data, live fetches, or hardcoded fallbacks

Configuration

Token counting now uses Claude Code's authoritative context_window data when available. For older Claude Code versions using the fallback method, system overhead is automatically managed.

API Cache System

With context_window (Claude Code 2.0.70+):

• Context limits come directly from context_window.context_window_size
• No cache or API lookups needed - authoritative value from Claude Code

Without context_window (Older Versions): Context limits are retrieved from:

1. Local Cache: /tmp/claude_code_model_data_cache.json (1-week TTL)
2. Live API: LiteLLM Model Database
3. Fallback: Hardcoded model limits

Cache benefits: Faster rendering, reduced network dependency, automatic refresh.

Display Features

Progress Bar: 10-segment bar using filled (●) and empty (○) circles

Context Usage Colors:

Color	Usage	Condition
Grey	0%	No active transcript
Green	<50%	Low usage
Yellow	50-79%	Medium usage
Red	80%+	High usage

Cost Tracking Colors:

Color	Cost Range
Grey	$0.00 (exactly zero)
Green	$0.01 - $4.99
Yellow	$5.00 - $9.99
Red	$10.00+

Line Change Colors:

Metric	Zero (Grey)	Non-Zero
Lines Added	0 lines	Green (>0)
Lines Removed	0 lines	Red (>0)

Git Integration:

• git-branch: Shows current branch with fallback text when not in a repo
• git-changes: Displays staged/unstaged changes (e.g., "+5 ~3 -2")
• git-worktree: Shows worktree name if working in a linked worktree

Session Tracking:

• session-id: Complete Claude Code session UUID for file matching
• session-clock: Elapsed session time (e.g., "Elapsed: 2hr 15m")

Widget rendering is intelligent - widgets with no data hide automatically, and orphaned separators are removed.

---

Example Outputs

Default configuration with git integration:

Sonnet 4.5 | claude-code-status-line | main | ●●●●●●○○○○ 60% (120K/200K tokens) | Cost: $2.50 USD | +150 ~25 -10 | Session: 852d18cc | Elapsed: 2hr 15m

High context usage:

Sonnet 4.5 | my-project | feature/auth | ●●●●●●●●●○ 90% (180K/200K tokens) | Cost: $8.75 USD | +500 ~50 -100 | Session: a3f7b2cd | Elapsed: 45m

Low usage (widgets hide when zero):

Sonnet 4.5 | my-project | main | ○○○○○○○○○○ 0% (783/200K tokens) | Cost: $0.00 USD | Session: d6994160 | Elapsed: 5m

No active transcript:

Sonnet 4.5 | my-project | develop | No active transcript | Cost: $0.00 USD | Session: 7b89c123 | Elapsed: 1m

---

Debug Logging

For debugging and understanding how the script processes Claude Code sessions, you can enable detailed logging:

Enable Debug Mode

export CLAUDE_CODE_STATUSLINE_DEBUG=1

Log Files Location

Debug logs are written to per-session files in the logs/ directory:

• Session-specific logs: logs/statusline_debug_<session_id>.log
• Examples:
  • logs/statusline_debug_d6994160-5c39-4ecf-8922-65a36b984ec5.log (UUID-based sessions)
  • logs/statusline_debug_unknown.log (fallback for edge cases)

Debug Features

• Session metadata: Model ID, version, working directory logging
• Token analysis: Per-message-type breakdown with percentages and field contributions
• Compact detection: Precise /compact boundary locations with token reduction stats
• Per-session logs: Isolated debug files (logs/statusline_debug_<session_id>.log)
• Error handling: Comprehensive parsing error logging and fallback behavior

Example Debug Output

[2025-09-26 15:03:09] === SESSION METADATA ===
[2025-09-26 15:03:09] Model: claude-opus-4-1-20250805 (Opus 4.1) | Version: 0.8.5
[2025-09-26 15:03:09] Found compact boundary at line 148, using content from line 149 onwards
[2025-09-26 15:03:09] Token Breakdown: assistant: 17,816 tokens (57%) | user: 13,275 tokens (42%)
[2025-09-26 15:03:09] Total: Conversation=31,092, System=15,400, Reserved=45,000 → 91,492 tokens
[2025-09-26 15:03:09] Token reduction from compaction: 146,759 tokens saved

Note: Debug logs are automatically excluded from git via .gitignore to prevent accidental commits of session data.

---

Token Counting Calibration Tool

The project includes claude-calibrate, a command-line tool to verify and improve token counting accuracy against Claude's official measurements.

Purpose

• Compare script calculations against Claude's official /context command output
• Identify discrepancies and suggest calibration factors
• Validate token counting accuracy when new Claude Code versions are released

Usage

Semi-Automated Mode (Default)

uv run --no-config claude-calibrate session1.jsonl session2.jsonl --verbose

The tool provides precise instructions for resuming each session and prompts you to enter the official token counts.

Manual Override Mode

uv run --no-config claude-calibrate session1.jsonl session2.jsonl \
  --known-tokens 17.5k 68k --verbose

Provide known token counts to skip automatic session resumption (useful for sessions that can't be resumed).

Auto-Discovery Mode

uv run --no-config claude-calibrate --max-sessions 3 --verbose

Automatically finds recent session files from all Claude Code project directories and provides instructions for manual calibration.

Example Calibration Report

📊 Successfully calibrated 2 session(s)

INDIVIDUAL RESULTS:
❌ session1.jsonl: Script 46,520 vs Claude 68,000 tokens (+31.6% difference)
✅ session2.jsonl: Script 17,470 vs Claude 17,500 tokens (+0.2% difference)

SUMMARY:
Average discrepancy: +15.9% | Suggested calibration factor: 1.231
⚠️  Token counting has moderate accuracy - consider adjusting CHARS_PER_TOKEN ratio

Features

• Auto-discovery: Finds recent sessions from all Claude Code project directories
• Semi-automated: Provides precise resumption instructions for manual token collection
• Directory decoding: Handles hyphenated paths (-Users-name-Project → /Users/name/Project)
• Flexible input: Accepts raw numbers, K-suffixed values, or full /context output
• Accuracy validation: Compares against Claude's official measurements
• Calibration recommendations: Suggests specific improvements and correction factors
• uv compatibility: Uses uv run for portable execution across Python environments

---

Development

Setup

1. Clone and install dependencies:

   git clone https://github.com/wpfleger96/claude-code-status-line.git
   cd claude-code-status-line
   uv sync --no-config

2. Configure Claude Code to use local version:

   uv run --no-config claude-statusline install

Testing Changes

# Test statusline output
echo '{"workspace": {"current_dir": "/test"}, "transcript_path": "", "model": {"id": "test", "display_name": "Test"}, "cost": {}, "version": "test"}' | uv run --no-config claude-statusline

# Run test suite
uv run pytest

# Enable debug logging
export CLAUDE_CODE_STATUSLINE_DEBUG=1

Releases

This project uses semantic-release for automated versioning and releases.

Commit Message Format:

<type>(<scope>): <description>

[optional body]

[optional footer]

Types:

• feat: New feature (minor version bump)
• fix: Bug fix (patch version bump)
• docs: Documentation changes (patch version bump)
• chore: Maintenance tasks (patch version bump)
• refactor: Code refactoring (patch version bump)
• BREAKING CHANGE: in footer (major version bump)

Example:

git commit -m "feat: add support for new Claude models"
git commit -m "fix: correct token calculation for edge cases"

Release Process:

1. Push commits to main branch
2. GitHub Actions automatically:
   • Analyzes commit messages
   • Determines version bump
   • Updates version in files
   • Generates CHANGELOG.md
   • Creates GitHub release and tag
   • Commits changes back to repository

Contributing

Contributions are welcome! Please:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Use conventional commit messages
4. Push to your branch
5. Open a Pull Request

For bug reports or feature requests, please open an issue.