Agent Skills

UI audits, typography, docs, PR review, and releases, loaded on demand by your coding agent

24 skills for the parts of shipping that code review never covers: the loading states, the type scale, and whether half the diff is AI slop.

Install

npx skills add mblode/agent-skills -g --agent claude-code,codex -y

Cursor and OpenCode work too: swap the --agent list. The bundle is listed at skills.sh.

The judgement these skills encode is trainable. Taste Training is 319 exercises for the eye behind the rules. The first unit is free.

codebase-architecture: Design a structure, deepen an existing one, or harden it with guardrails.
scaffold-nextjs: Next.js turborepo with Blode UI, Ultracite, GitHub, and Vercel.
scaffold-cli: TypeScript CLI package: ESM, tsdown, vitest, changesets, CI.
multi-tenant-architecture: Tenant isolation, routing, and custom domains on Cloudflare or Vercel.

product-design: What the interface should do, before anyone builds it.
ui-design: Visual direction, Tailwind builds, screenshot to markup, dark mode, responsive.
ui-animation: Springs, gestures, easing, and curves pulled from a screen recording.
presentation-creator: Decks with a story spine, speaker notes, and a QA pass. Ships as markdown or a web app.

copywriting: Product and marketing copy, persuasion frameworks, brand voice charts, AI-ism removal.
docs-writing: Diataxis doc types, 48 rules, audit and writing modes.
readme-creator: Writes or rewrites a README as a public-facing shop window.

Personal voice training, evaluation, and long-form blog writing live in ghostwriter, where posts are drafted from the blog platform profile.

ui-audit: React and Next.js UX audit: 82 rules, 12 playbooks, ship verdict.
ax-audit: Agentic experience audit: 23 rules, ship verdict.
dx-audit: Libraries, CLIs, and SDKs: root-cause findings, agent-friendly checks.
typography-audit: 78 rules: punctuation, fonts, sizing, spacing, hierarchy, pairing.
optimise-seo: Next.js SEO: sitemaps, metadata, structured data, hreflang, CWV.

planning: Builds the plan, then scores it to 5/5 before any code.
pr-reviewer: Read-only diff review: bugs, structure, AI slop, security.
tidy: Hunts complexity across five angles and applies the simplifications.
pr-creator: PRs with short human descriptions, Linear IDs, tidied commits.
pr-babysitter: Watches an open PR: conflicts, CI, comments. Fixes what it can.
autoship: npm releases with changesets: fix loop, CI watch, publish verify.

MIT