v1.2.0

Two proactive review reflexes, validated with an eval-driven loop. The skill's leverage on a strong base model is consistency and proactivity, not new knowledge — so these are behavioral forcing functions.

Added

• The clutter audit — a countable method for density judgment (border-nesting depth, signals-per-state, always-on markers, chrome-vs-data ratio, the removal test). "Feels busy" becomes named, specific cuts.
• Responsive design — breakpoints and the floor — the breakpoint ladder (wide / standard / narrow / too-small) and mechanics (relative units, load-bearing priority, SIGWINCH, 80×24 minimum + too-small message).
• SKILL.md applies both reflexes to any layout review even when the user only asked about something else — the most-missed behavior in practice. Review checklist + decision flow reference them so they fire from every entry point.
• evals/evals.json — the design-review eval set used to validate the change.

See CHANGELOG.md.

v1.1.0

Accuracy/freshness pass on the ecosystem references. No structural changes; routing, universal principles, and the trigger description are unchanged.

Highlights

• Bubble Tea v2 is stable — Cursed Renderer, pure Lipgloss, progressive keyboard enhancements, new charm.land/bubbletea/v2 import path, + a v1 → v2 migration note.
• Ratatui 0.30 named — modular workspace crates, no_std, and the actual crossterm_0_28 / crossterm_0_29 compatibility flags.
• Removed hardcoded GitHub star/download counts that go stale.
• Added CHANGELOG.md; manifests bumped to 1.1.0.

See CHANGELOG.md for details.

v1.0.0

First release of the tui-design Claude Skill. See README for install instructions.