The statusline

tokenmaxxing renders Claude Code's statusline natively (while teeing the usage feed into the session's account file, usage/<uuid8>.json, when the session has a seat). The design principle: quota is carried in color, not numbers. One line, for example:

tm-wt-demo Fable 5 (high) +156/-23  ◆ 𝒇 2h 1d  ◇ 19h  ◇ 𝒇 2d  ◇ full  ✗ ?

• Each window token is its time-to-reset, painted on a continuous green-to-yellow-to-red ramp by used percent (truecolor when COLORTERM says the terminal supports it, stepped onto the 256-color cube otherwise). The ramp anchors at pure green for 0, yellow at 75, and red at 95 and above, so the old severity bands survive as ramp anchors.
• The model name is painted by context-window fill on the same ramp.
• A green ◆ leads the windows of the session's own account, the seat the supervisor gave it; every other pooled account gets its own cyan ◇ (red ✗ if it needs reauth): healthy accounts first, sorted by earliest upcoming reset, then the accounts needing reauth. A session started outside the supervisor has no seat: its own windows still render behind the ◆, and every pooled account renders as ◇.
• An account walled by an enforced limit (a server refusal the StopFailure hook recorded) paints its marker full: a walled parked account renders one full-painted token carrying the wall's time-to-reset in place of its cached windows, and a walled active seat keeps its measured windows behind a full-painted ◆ and appends the same wall token (3d100 without color).
• The seat's block shows every per-model row the sample reported as a bare painted initial (Fable renders 𝒇; other families use their uppercased first letter, always family-matched), then the session window, then the weekly window. Every other account's block shows its weekly window, preceded by a per-model initial only when that cap is more used than the weekly aggregate.
• An untouched parked account renders a green full. Measured usage with an unknown reset renders a painted ?, an unmeasured per-model cap on a gated model renders an unpainted 𝒇?, and a seat with no rate-limit data yet or a pooled account never sampled renders an unpainted ? - unmeasured must never look safe.
• Reset countdowns show the largest unit only, never "now" or "0m"; a window with no upcoming reset renders a green 0 (empty again).
• A linked git worktree's basename leads the line; a main checkout renders nothing there. The detection reads the filesystem for a .git file and never spawns a git subprocess.
• With NO_COLOR set, window tokens fall back to a glued numeric format (2h5 means resets in 2 hours, 5% used; a per-model token is its initial plus the percent, F62; ctx 42 for context fill) and the markers and the words full and ? print unchanged - without color, a drained window must not look fresh.

Buffered usage quota

Buffered usage quota: every usage percent the statusline paints or prints, for the seat and for every parked account, is measured against that account's switch threshold, not against the provider's full limit. The threshold is the account's own entry in thresholds.accounts, else the global threshold, minus policy.projectionMargin for the session window, and it is lifted near a reset the way switching lifts it (policy.accountReleaseMs). 100 means the account reached its switch threshold, where tokenmaxxing stops placing sessions on it unless its thresholds.accounts entry sets credits, and usage past that point stays at 100. The figure therefore reads higher than the usage Claude Code or Codex reports: an account with a threshold of 75 at 75 percent provider usage reads 100, and one with a threshold of 90 at 45 percent reads 50. A session started outside the supervisor has no seat account, so its own session and weekly windows keep the provider figure. The context-window fill of the model name and of the subagent rows is not usage quota and is unchanged.

Subagent tasks get their own panel rows via claude's subagentStatusLine surface: one row per active task with the family name painted by that task's context fill, its effort in parentheses, and the task's label.