Path-scoped. Loaded only when relevant.
Open a matching file → the right rules attach. Zero token cost the rest of the time.
What this pillar does for you.
When a file approaches 1500 lines, the AI proposes an extraction plan before adding more.
New files land where they belong. No tangled spaghetti, no "utils.ts grew to 4000 lines".
Working code is left alone. Edits are scoped to the smallest blast radius - no rewrites disguised as refactors.
13 rules.
Each rule is a markdown file the AI loads when it edits a matching path. Click any to see related skills and where it's referenced.
Language-agnostic baseline: KISS/DRY/YAGNI, immutability, naming, ~200-400 line files, single-responsibility functions. Stack rules layer on top.
TypeScript strict baseline, monorepo layout, zod-at-the-boundary, Gitflow. The framework-agnostic floor every other rule sits on.
Behavioral rules that block silent assumptions, speculative architecture, unrelated edits, and invisible failure - think before coding, simplest first, surgical changes.
Required root scripts (format, lint, typecheck, test, build), repo-owned tooling only, lint-staged + Husky pre-commit gates - deterministic and hard to bypass.
WCAG 2.2 AA floor: semantic HTML, localized labels, keyboard reachability, visible focus, real-time region usage. EU Accessibility Act treated as launch risk.
Lighthouse mobile >=90 against production preview, Web Vitals budgets (LCP<=2.5s, INP<=200ms, CLS<=0.1), web-vitals wired to observability, regressions alerted per route.
Curly quotes, em/en dashes, ellipsis, multiplication signs - the punctuation LLMs get wrong. Applied silently across every UI artifact.
Central design gate before any UI change: purpose, primary action, 2-3 competitor references, state coverage, material honesty. No generic template-looking output.
Every page React.lazy + Suspense, dynamic-import heavy libs at usage site, narrow selectors, virtualize lists >100, derive expensive work in selectors not render.
44x44 touch targets, WebView-visible focus, Android-back closes topmost layer, prefers-reduced-motion respected, icon-only buttons have translated labels.
Lighthouse mobile still required against production preview, lazy-load page code and heavy plugins, real device smoke check before release, batch high-rate updates with rAF.
VoiceOver/TalkBack for critical journeys, accessibilityRole/Label/state, 44pt touch targets with hitSlop, dynamic type respected, focus managed after modals/redirects.
Hermes + New Architecture default, FlatList/SectionList with stable keys, flash-list only after measuring, lazy-load heavy modules at route boundaries, batch real-time at 30fps.
20 skills.
Skills are intent-triggered workflows. The AI router loads them by description; bodies on demand.
WCAG 2.2 Level AA implementation and audit. Maps component intent to semantic ARIA on web and accessibility traits on iOS/Android - focus order, target size, labels, hints.
Baseline cross-project coding conventions for naming, readability, immutability, and code-quality review - framework-agnostic floor below the language skills.
C++ Core Guidelines-based standards for modern, safe, idiomatic C++ when writing, reviewing, or refactoring.
Scaffold a typed React component with shadcn primitives, i18n keys, explicit loading/error/empty states, and tests - only after onboarding is resolved.
Scaffold a full React feature slice (pages, services, slices, hooks, tests) following the forced stack and folder layout.
Scaffold a typed RN component with NativeWind, RNR primitives, accessibility, and tests baked in - only after onboarding is resolved.
Full RN feature scaffold (screens, navigation, slices, RTK Query, types, tests) on the forced Expo + Expo Router stack.
Scaffold an Expo Router route/screen with typed params, safe-area boundaries, loading/error/empty states, and auth-gated layout when relevant.
Scaffold an RN service or RTK Query endpoint with zod validation, expo-secure-store token handling, and typed AppError mapping.
Scaffold a React route/page with lazy loading, SEO metadata, i18n keys, and a typed feature shell.
Scaffold an RTK Query endpoint or plain axios service with zod-parsed responses, typed AppError, and shared api-client wiring.
Structured visual UI/UX audit that produces a phased, implementation-ready plan and waits for approval before touching code. Purely visual - never edits logic or state.
Generate or audit a design system, check visual consistency, and review PRs that touch styling.
Surgical edits, simpler implementation, no overengineering, explicit assumptions, success criteria, fail-loud checkpoints, cost-aware reasoning/model routing.
Create distinctive, production-grade frontend interfaces with high design quality when visual direction matters as much as code quality - only after onboarding is resolved.
Audit and rewrite user-facing copy in React/RN to use translation keys. Extends the existing i18n module, never creates a parallel one. <Trans> for rich copy, t() for scalars.
Java standards for Spring Boot services: naming, immutability, Optional usage, streams, exceptions, generics, and project layout.
Refactor existing React code with surgical edits. Splits fat components along clean seams, keeps imports/tests green, never rewrites working code.
Audit, plan, and implement SEO across technical SEO, on-page optimization, structured data, Core Web Vitals, and content strategy.
Record polished UI demo videos using Playwright - WebM with visible cursor, natural pacing, and professional feel for tutorials and product walkthroughs.
One plugin, eight pillars.
Every rule and skill is part of a system. Jump across pillars to see how they reinforce each other.
Use the AI you already pay for.
Add superpowers.
One plugin. Senior-engineer rules, security, a backend, and one-command deploys — inside the editor you already use.