A VS Code extension that shows your Claude Code usage and rate limits directly in the status bar and sidebar, plus a real-time monitoring panel for background agents.
Features
Usage monitoring
Status bar — at-a-glance usage with countdown to reset
Hover tooltip — per-limit breakdown with progress bars and activity stats
Sidebar panel — full-detail view accessible from the Explorer, opens on status bar click
Auto-refresh — polls on a configurable interval; multiple VS Code windows share a cache to avoid hammering the API
Smart warnings — proactive notifications fired at startup, +1h, and +2h (max 3 per session):
Session hot — warns when session usage ≥ 80% with 1h+ until reset (you're burning through it fast)
Weekly hot — warns when weekly usage ≥ 80% with 24h+ until reset (may run out before it resets)
Use it or lose it — nudges when weekly usage < 60% with < 2 days until reset (unused capacity about to expire)
429 backoff — automatically backs off for 5 minutes if rate-limited
Background agent monitoring
Real-time panel for tracking Claude Code background agents (sidechain tasks).
Collapsible task cards — each agent is shown as a card with status, model, description, and block count; click to expand and see full output
Live updates — polls every 2s with incremental DOM updates (no page reloads), preserving scroll position, search input, and expand state
Session labels — each task shows which session it belongs to
Search — filter agents by description, ID, model, session name, or content (indexes tool calls, text blocks, and errors)
Status filter — toggle between All / Active / Done
Grouped view — when viewing multiple sessions, toggle between flat and grouped-by-session layout
Multi-session support — select one or more sessions to monitor simultaneously; only sessions with background tasks are shown
Workspace picker — switch between workspaces to monitor agents from different projects
Editor title icon — appears next to Claude Code's own icon when background tasks are active; opens the agent panel as a side editor
Markdown rendering — agent text output is rendered with headers, bold, italic, code blocks, lists, tables, and horizontal rules
Smart scrolling — auto-scrolls to bottom for running tasks, locks when you scroll up; Start/End navigation buttons appear on hover
Debug mode — toggle via command palette or claudeUsageBar.debugMode setting for development
What's shown
Metric
Description
Current Session
5-hour rolling window
Weekly · All Models
7-day all-models limit
Weekly · Sonnet
7-day Sonnet-specific limit
Extra Credits (Org)
Org-level add-on credits
Prompts today / this week
Counted from ~/.claude/history.jsonl
Requirements
Claude Code must be installed and signed in (credentials are read from the macOS Keychain or ~/.claude/.credentials.json)
macOS (Keychain credential reading), though the file fallback works cross-platform
Configuration
Setting
Default
Description
claudeUsageBar.refreshInterval
60
How often to refresh usage data (seconds)
claudeUsageBar.debugMode
false
Enable debug mode (always show agent icon, dev badge in toolbar)
How it works
Usage: Credentials are read directly from the Claude Code OAuth token stored locally — no additional sign-in required. Usage data is fetched from api.anthropic.com/api/oauth/usage and cached in ~/.claude/claude-usage-bar-cache.json so multiple VS Code windows don't each make independent API calls.
Background agents: Session data is read from ~/.claude/projects/ JSONL files. Agent output files are discovered at /tmp/claude-*/ task directories. The extension parses these incrementally using byte offsets to avoid re-reading completed content. Completion is detected via stop_reason=end_turn with a 30-second staleness heuristic.
