[bernardopg]

Summary

The DankBar pill showed one quota window per provider, so a Claude user had to choose between the 5 hour and the 7 day number, and could not tell which one the bar showed without hovering.

DankBar usage window (barWindowOverrides) now accepts a set of windows:

• Primary + secondary (claude:primary+secondary) and All windows (claude:all) join the existing Primary / Secondary / Tertiary / Highest choices.
• Each shown window is tagged with its length and coloured on its own: Claude 5h 45% / 7d 80%. The vertical bar stacks the tag above each percentage. The tooltip names every window, with its reset when one provider is shown.
• top and auto pill modes rank such a provider by its highest window; with the default notifyWindowScope=displayed, every shown window raises its own alert.
• Absent windows are skipped (Claude's tertiary only exists with a model-scoped limit); when none of the chosen windows exists, the primary window is shown, so an override never blanks the bar.
• Single-window choices render exactly as before.

Slot parsing moves to BarWindows.js, shared by the widget and the settings body, which previously each carried a copy. Settings stores sets in canonical order (secondary+primary → primary+secondary, all three → all).

"Windows that raise alerts" now describes the displayed windows in the plural, and the bar-window description keeps its 5h 45% / 7d 80% example on one line in German and Spanish.

Related issues

Closes #86

Type of change

• New provider
• Bug fix
• Enhancement
• Documentation / i18n
• CI / tooling

Migration

Existing barWindowOverrides values keep their meaning. BarWindows.js is a new QML module: a hot reload leaves the plugin disabled with File name case mismatch, so the upgrade needs one dms restart (noted in CHANGELOG.md).

Testing

• tests/test-qml-smoke.sh: 14 checks on BarWindows.js (canonical slots, invalid/duplicate parts, alias-aware parsing, absent-window fallback, highest, maxPercent).
• tests/test-widget-runtime.sh: renders the real horizontal and vertical pills with claude-work:primary+secondary and checks the tags, per-window percentages, tooltip, notifyWindowsFor(), and the canonical value Settings stores.
• Both new assertions were mutated once to confirm they fail.
• Every tests/test-*.sh, check-metadata, check-changelog, check-dco, qmllint, shellcheck, actionlint, git diff --check and gitleaks pass locally.
• Manually on a live DMS 2.0 session (Hyprland): Primary + secondary and All windows in the bar and tooltip, back to Primary removing the pair; settings, dropdown, pill and popout in pt_BR, de_DE, zh_CN and es_ES.

Limitations

• Local kcov 43 reports 93.41% statement coverage because of its known multi-line SQL tracking bug; this PR changes no covered bash, and CI measures with the pinned kcov build.
• zizmor was not run locally; no workflow changed.

Checklist

• plugin.json .version and CHANGELOG.md are consistent (only if this is a release-bearing change; both version pills read the manifest at runtime)
• New/changed provider scripts are executable and emit valid JSON
• i18n/en.json keys have translations in every locale (or the strings are user-neutral)
• Every commit is signed off (git commit -s) under the DCO
• CI is green