by leeguooooo
Lightweight Claude Code statusLine: 5h/7d rate-limit usage, reset countdowns, model + context window, prompt-cache age — one line, 3 styles × 9 themes, daemon fast-mode
# Add to your Claude Code skills
git clone https://github.com/leeguooooo/claude-code-usage-barGuides for using cli tools skills like claude-code-usage-bar.
Last scanned: 5/30/2026
{
"issues": [],
"status": "PASSED",
"scannedAt": "2026-05-30T15:22:40.947Z",
"npmAuditRan": true,
"pipAuditRan": true
}claude-code-usage-bar is an open-source cli tools skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by leeguooooo. Lightweight Claude Code statusLine: 5h/7d rate-limit usage, reset countdowns, model + context window, prompt-cache age — one line, 3 styles × 9 themes, daemon fast-mode. It has 322 GitHub stars.
Yes. claude-code-usage-bar passed SkillsLLM's automated security scan — a dependency vulnerability audit plus prompt-injection heuristics — with no high-severity issues. You can read the full report in the Security Report section on this page.
Clone the repository with "git clone https://github.com/leeguooooo/claude-code-usage-bar" and add it to your Claude Code skills directory (see the Installation section above).
claude-code-usage-bar is primarily written in Python. It is open-source under leeguooooo on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other CLI Tools skills you can browse and compare side by side. Open the CLI Tools category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh claude-code-usage-bar against similar tools.
No comments yet. Be the first to share your thoughts!
Top skills in this category by stars
Lightweight status-line monitor for Claude Code, with a local AgentParty bridge for Codex workflows. In Claude Code it shows your 5h / 7d rate-limit usage, reset timers, current model, context window, prompt-cache freshness, and (optionally) session cost. In Codex + AgentParty workflows it can append the current AgentParty channel, identity, listener, unread count, and last-message preview from a local cache.
5h[ 27% ]⏰1h28m →42% | 7d[ 79% ]⏰11h28m →88% | Opus 4.8(350.0k/1.0M) | cache 4m23s
📖 Deep dive: Is that
cache 4m23sline actually accurate? — how the prompt-cache countdown is computed

3 styles × 9 themes, configurable in one command. Auto-updates from PyPI. For
Claude Code, run pip install claude-statusbar && cs --setup and restart
Claude Code.
cs doctor — self-diagnosticv3.29.12 (2026-07-15) — AgentParty status is session-correct end to end: sessions sharing one project now read channel, identity, unread count, message preview, and listener state from their own config-owned cache slot instead of mixing those fields with the workspace's last writer.
v3.29.11 (2026-07-11) — AgentParty identity is session-correct: two Claude Code sessions in the same project can use different AGENTPARTY_CONFIG files without both status bars collapsing to the last writer's agent name. Only config paths from actual shell tool calls count; paths quoted in chat do not override identity.
v3.29.6 (2026-07-10) — support docs cleanup: the README now states the support boundary explicitly. Claude Code is the full native statusLine integration; Codex support is the AgentParty local status bridge and does not include OpenAI quota/session accounting.
v3.29.5 (2026-07-09) — service daemon and AgentParty session fixes: cs daemon stop and upgrade drift-kill now recognize launchd/systemd-managed daemons, pidfile cleanup no longer deletes a newer daemon owner's pidfile, and the AgentParty block is gated on session transcript evidence so unrelated windows in the same repo do not inherit another agent's channel line.
v3.29.0 (2026-07-09) — AgentParty block redesign + daemon-restart fixes. The AgentParty line was unreadable (FAINT stacked on the dimmest grey) and always claimed watch down: the reader looked for heartbeat_at while the contract writes heartbeat_ts, so every live listener read as dead. Now it renders as two lines with monochrome glyphs that inherit the theme — a header that states the listening state outright, and the last message on its own line marked ● unread / ○ read and @ when it mentions you. Also fixes three daemon defects found while chasing "I upgraded but nothing changed": the code-drift tick burned its own 30s spawn debounce and left every session inline-rendering; _signal_outdated_daemon could SIGTERM a recycled PID belonging to an unrelated process; and the orphan-.tmp sweep + auto-update check were starved by sharing a timer seeded to daemon start (a daemon that restarts on drift never lived the 30 minutes needed to fire either).
v3.28.2 (2026-07-09) — uv-tool upgrade fix: cs upgrade now detects uv-tool Python symlinks correctly and selects uv tool install --upgrade claude-statusbar.
v3.28.1 (2026-07-09) — upgrade/version UX fix: cs upgrade upgrades the install channel that is actually running cs (uv tool, pipx, or plain pip), and cs -v, cs -V, cs -version all work like cs --version.
v3.28.0 (2026-07-09) — AgentParty / Codex bridge line (show_party, default on): when the same workspace has AgentParty local status, cs appends 🎈 #channel · 🤖/👤 name · 👂watch/serve · unread · last message under the project line. This is local-only (~/.agentparty/state/<workspaceId>/statusline.json), uses the same cwd-scoped workspace id fixtures as AgentParty, and marks stale/down listener state instead of pretending it is live.
v3.27.0 (2026-07-03) — IP-risk detection aligned with ip-check: China-cloud provider detection and ban-risk threshold handling now match the ip-check.leeguoo.com classifier.
v3.11.0 (2026-06-02) — rate-limit projections (show_projection, default on): after each ⏰<reset> timer the bar shows →NN%, your expected end-of-window usage. The 5h model blends recent pace, whole-window average, and a local baseline; the 7d model integrates learned coarse rhythm buckets (work hours, non-work hours, night, weekend) so a busy first day is not blindly extrapolated across the whole week. show_forecast remains the separate ⚠<eta> warning chip for imminent cap hits. Plus: context_window.used_percentage = null handled gracefully, and the daemon only renders windows active in the last 10s.
v3.10.0 (2026-06-02) — live-activity line (in-progress todo, active tool, completed-tool rollup), git ahead/behind + session duration/lines on the project line, running-subagent lines, an opt-in bar_shimmer starfield on the battery bars, and a self-hosted plugin marketplace. Plus: auto-update now actually runs in daemon mode (detached, non-blocking), and the cache countdown is per-session-correct. All new segments are opt-in except the todo line.
v3.6.0 (2026-05-08) — cs --setup now defaults to daemon (fast) mode: under 1% CPU continuously instead of ~3% inline. Pass --inline to opt back. Also: py3.9 compat fixes, GitHub Actions CI, animated hero GIF.
v3.5.1 — npx skills add install path, show_cache_age on by default.
v3.5.0 — consolidated claude-statusbar skill: say "switch theme to nord" / "余量颜色改成 #4ec85b" and Claude Code routes it to the right cs command.
v3.4 — per-segment color management (each metric colors itself by its own severity), classic style finally respects themes, two new themes (catppuccin-mocha, tokyo-night), per-severity color overrides via cs config set color_ok|warn|hot.
v3.2 — daemon fast-mode (now the default in v3.6.0) for ~5× lower CPU at refreshInterval=1.
📋 Full changelog: CHANGELOG.md · GitHub Releases — every version's changes, also linked from the PyPI page.
5h[ 27% ]⏰1h28m →42% | 7d[ 79% ]⏰11h28m →88% | Opus 4.8(350.0k/1.0M) | cache 4m23s | $ 1.42
⤷ claude-code-usage-bar ⎇ main● · +182 -47 · ⏱ 12m · v3.12.0
⚙ effort:high · think:on · fast:off · style:default
| Segment | Meaning |
|---|---|
5h[27%] |
5-hour rate-limit usage (rolling window from Anthropic API headers) |
⟳ 5h/7d stale·restart |
Shown (in place of the two bars) when the cached 5h/7d data has gone stale because cs stopped receiving fresh ticks — usually another tool displaced the statusLine, or the daemon died. Restart Claude Code to refresh; if it keeps happening, run cs --setup to reclaim the statusLine. cs doctor explains it in detail. |
⏰1h28m |
Time until the 5-hour window resets |
7d[79%] |
7-day rate-limit usage |
⏰11h28m |
Time until the 7-day window resets |
→42% / →88% |
Projected end-of-window usage at your current rhythm (show_projection, on by default). Muted < 80%, yellow ≥ 80%, red ≥ 100%. The 5h model blends recent pace + whole-window average + a local baseline; the 7d model integrates learned day/night/weekend buckets so a busy first day isn't extrapolated across the week. |
⚠~18m |
At-risk warning chip — only when a window is projected to hit 100% and the cap is imminent (≤ 1 h). Separate from the projection (show_forecast, on by default). |
Opus 4.8(350.0k/1.0M) |
Model name + current context window usage |
cache 4m23s / cache COLD |
Countdown to prompt-cache expiry — the TTL (5min vs 1h) is auto-detected from the transcript, so it's right on a subscription (1h) or an API key (5min). Green when comfortable, yellow under 1min, red on COLD. Cache hits consume ~10× less rate-limit quota — for subscribers, letting it go COLD eats your 5h / 7d windows ~10× faster. Enabled by default; disable with cs config set show_cache_age false |
$ 1.42 |
Session cost in USD as Claude Code reports it. For Pro/Max subscribers this is the API-equivalent value of your usage (i.e. what it would cost on the API), not money owed. Useful as an ROI signal. Opt-in: cs config set show_cost true |
bal[████ 52%] $26.00 |
**Relay account |