curl -fsSL https://raw.githubusercontent.com/thevibeworks/claude-code-statusline/main/install.sh | bash
-
- curl -fsSL https://raw.githubusercontent.com/thevibeworks/claude-code-statusline/main/install.sh | bash
+
+ Anatomy of the line
-| +131/-16 6m $7.53 | -Session activity. Lines added/removed, API time, session cost — cost flashes its increment (+.37) for a minute after each jump. |
| fabl5[1m]xh[█░░░░░12%] | -Model + context. Abbreviated model, 1M-window tag, reasoning-effort badge, context pressure bar. On 1M models the bar also carries the premium-pricing band. After /compact the bar snaps to [░░░░░░0%] and flashes the drop. |
| fb[67%] | -Model-scoped weekly quota. When the API carries a per-model weekly limit for the model you're running (Fable here), it hugs the model block. Other models' limits stay hidden — only the one constraining this session is signal. |
| 5h[38%@14:00] | -5-hour window. Wall-clock reset time, deliberately not a countdown: the line only re-renders on activity, so @1h38m would decay into a lie during idle gaps. @14:00 stays true in a frozen frame. |
| 7d[55%@2d] | -7-day window, forecast-colored. A learned per-weekday burn profile projects whether quota outlasts the window — your heavy Tuesday counts more than a generic average. Red means "at your pace, dry before the reset". |
| 5h ▂▅█▃▮▯▯▯▯▯ 0.8x @14:00 | -5h ledger. This sitting as ten half hours, height = the 5h points each cell cost; ▮ now, ▯ ahead, ˍ ran-but-negligible, ░ no sample. Ends with its pace (used ÷ elapsed) and the reset its right edge stands for — axis labels, not badges restated. |
| 7d ▅▁▂ ▃▅ˍ▃▅ … ▆▆▮▯▯▯▯ 0.7x @Wed 09:00 | -7d ledger. The week as its 5h windows (34 cells, oldest left, a gap at each local midnight in history only), height = the 7d points each burned; × where the pool runs dry at your pace. Reconstructed from your own usage log — the same history ccpace reads. |
| + fb 91% vs 7d 55% · go op | -The pin. Left of the ledgers, compacted to the room line 1 leaves (cut at the rightmost joint, so it gives up as little as possible). It stays while the condition holds. The number you act on is bold; the hue is the stance — cyan opportunity, yellow/red pressure, dim budget. |
| + fb weekly 91% against 7d 55% · the model caps first… | -The flash. Row 3: the same notice in full, for ~90 s after the condition first appears in this session — then it disappears and leaves the pin. What it knows: which limit binds first (fb 91% vs 7d 55% = the model caps before the account), an out-of-band quota re-base (7d rebased 53%→12%), the last 5h window of the week, expiring surplus checked against your own burn ratio. --notice off keeps row 3 quiet. |
| ≡!419k ≡~ !net | -Truth badges. ≡!419k — prompt-cache break on resume (bold red past 200k); ≡~ — rebuilding; stale-data markers say why a fetch failed. Silent while healthy. |
Read the meter
+Every reading answers twice: the left column states the number, the right column says what to do about it. The № cites the register frame above where it appears.
+@14:00 stays true in a frozen frame where @1h38m would decay into a lie.▮ now, ▯ ahead, ˍ ran-but-negligible, ░ no sample. Ends with its pace: above 1x you cap before the reset....▯5hx14 answers "14 slots to the reset", and × red answers "the tail runs dry".≡!419k — a full-prefix rewrite at 20× the read rate), a rebuild in progress, a fetch that failed and why. Silent while healthy: zero width spent on a warm cache.Built like it matters
+Built like a meter should be
An advisor, not a dashboard
-A second row that interprets the badges: ! 5h caps ~05:18, 42m before reset when you'll hit a wall, + 7d resets @07:00, 56% unused · spend it when paid capacity is about to expire unused. Quiet means no row at all.
Row 2 interprets the badges: ! 5h caps ~05:18, 42m before reset when you will hit a wall, + 7d resets @07:00, 56% unused · spend it when paid capacity is about to expire. Quiet means no row at all.
Numbers that don't lie
-Wall-clock resets instead of decaying countdowns. A visible [░░░░░░0%] instead of a hidden bar. Stale data is always badged, never passed off as fresh.
Cache break detection
-Resume after an idle gap longer than the TTL and ≡!419k fires — the full-prefix rewrite you can't see coming, sized and colored by severity. Silent while healthy; no width wasted on a warm cache.
Polite to the API
-N concurrent Claude Code instances share one cached fetch behind an atomic lock — at most one usage request per account per TTL window. Escalating backoff honors Retry-After.
Works behind TLS inspection
-Corporate mitm proxy or a local tracer: it builds a combined CA bundle from NODE_EXTRA_CA_CERTS automatically, so fetches survive where plain curl dies.
Documented wire contract
-The observed /api/oauth/usage schema is versioned in-repo and re-synced against CLI captures, so parser changes check against ground truth.
384 tests on one bash file
-bats suite covering quota math, lock contention (20-contender single-winner), forecast pacing, ANSI rendering, and the installer. CI on every push.
Wall-clock resets instead of decaying countdowns. A visible [░░░░░░0%] after /compact instead of a hidden bar. Stale data is badged, never passed off as fresh. Burn is a monotone envelope, never a re-earned dip.
Nothing running in the background
-No daemon, no telemetry, no npm. It runs when Claude Code renders the prompt, reads stdin + one cached API call, prints a line, exits.
No daemon, no telemetry, no npm. It runs when Claude Code renders the prompt, reads stdin plus one shared cached API call, prints its rows, exits. 417 bats tests on the one file.