Calm Precision UX Guide
Grounds every UI, writing, and deck preference a coding agent makes in cited research, not taste — one doctrine spanning web, iOS, and macOS.
The Problem
Design guidance for AI coding agents is usually either too vague to act on (“make it clean”) or trapped in a single medium — a web style guide that says nothing about a deck, a doc, or a data table. An agent building a dashboard, then a README, then a slide needs one coherent doctrine that adapts across all of them, not three disconnected rulebooks.
What I Built
A Claude Code plugin that packages design principles, platform conventions, and owned-app evidence into 21 trigger-shaped skills any agent can discover and load on demand. The spine is a cross-medium doctrine — six universal jobs (orient, focus, connect, progress, recover, resolve) and a set of non-negotiables — with medium adapters for writing, images, UI, decks, docs, data, and ops.
What’s Inside
- Cross-platform topic skills — hierarchy, navigation, responsive, feedback, motion, states, components, data-viz, tokens, and an evidence policy that separates source-of-truth reads from live capture.
- Platform skills — iOS HIG, macOS (window chrome, toolbars, menu-bar extras, materials, density), and web (semantic HTML, ARIA decision tree, WCAG 2.1 AA).
- Pattern ingestion — capture a live URL or image, map it to a 7-section schema, tag it, and audit new work against the captured library.
The latest release distributes an external expert UI design guide (47 cited sources, a five-dimension MECE taxonomy) across the library, with the full guide kept as a canonical reference. Every preference claim traces back to design-evidence files rather than taste.
The guide ties each preference claim to design evidence
This is the design layer the build loop loads before touching a component, a page, or a flow. It turns “make it look good” into a discoverable, evidence-backed set of moves that hold whether the agent is shipping a screen, a slide, or a sentence.