--- category: skills topic: overview content_type: reference status: latest retrieved: 2026-05-16 source_url: https://github.com/pbakaus/impeccable license: Apache 2.0 tags: [design-quality, ai-agents, anti-patterns, typography, color, motion, accessibility, synthesis] --- # Impeccable — Design Skill Overview ## What it is Impeccable is a Claude Code / cursor / Codex skill for frontend design quality by Philip Bakaus. It ships as a system prompt with 7 domain reference files and 23 named commands. Its core thesis: LLMs trained on the same SaaS templates produce the same handful of tells on every project unless given explicit design guidance. Impeccable provides that guidance. **Origin:** Built on top of and extending Anthropic's official `frontend-design` skill. --- ## The AI slop problem Without explicit guidance, AI agents consistently default to: - Inter for everything - Purple-to-blue gradients - Cards nested in cards - Gray text on colored backgrounds - Rounded-square icon tiles above every heading - Medium blue primaries - Near-white `#FAFAFA` surfaces - 6–8px uniform border radius across all components These are not random errors — they are structural biases from training on a corpus dominated by generic SaaS templates. The problem is invisibility: the output is functionally correct and aesthetically unremarkable. It passes code review but fails design review. --- ## 7 domain reference files The skill loads these reference files alongside every design command: | Reference | Covers | |---|---| | typography | Type systems, font pairing, modular scales, OpenType features | | color-and-contrast | OKLCH, tinted neutrals, dark mode, accessibility | | spatial-design | Spacing systems, grids, visual hierarchy | | motion-design | Easing curves, staggering, reduced motion | | interaction-design | Forms, focus states, loading patterns | | responsive-design | Mobile-first, fluid design, container queries | | ux-writing | Button labels, error messages, empty states | --- ## 23 commands The skill provides a shared design vocabulary via `/impeccable [command]`: **Planning:** `craft`, `teach`, `document`, `extract`, `shape` **Review:** `critique`, `audit`, `polish` **Adjustment:** `bolder`, `quieter`, `distill`, `harden`, `onboard` **Specialized:** `animate`, `colorize`, `typeset`, `layout`, `delight`, `overdrive`, `clarify`, `adapt`, `optimize`, `live` --- ## Anti-pattern framework Impeccable includes two layers of anti-pattern detection: **27 deterministic rules** — Run without an LLM, catching specific measurable failures (use of a banned font, pure black as a surface token, opacity-derived container colors, etc.). Each rule is tied to specific design guidance. **12-rule LLM critique pass** — Applied by the AI after generation, checking for second-order pattern failures (e.g., landing in a predictable aesthetic pocket even while technically avoiding first-order reflexes). --- ## Key anti-patterns flagged - Overused fonts: Arial, Inter, Roboto, system defaults - Gray text on colored backgrounds - Pure black or pure gray without tint - Wrapping everything in cards or nesting cards inside cards - Bounce/elastic easing (dated feel) - Glassmorphism as a default rather than a deliberate choice - Color-only state signaling (always pair with a second signal) - Identical padding/radius/shadow across all card types --- ## Register concept Impeccable distinguishes two design registers: - **Brand register** — Marketing, landing pages, campaigns. Design IS the product. Maximum expressive latitude. - **Product register** — App UI, dashboards, tools. Design SERVES the product. Restraint and consistency over expression. The active register is identified from the task context and determines which reference guidelines apply most strictly. --- ## Relationship to Anthropic frontend-design Impeccable explicitly extends Anthropic's `frontend-design` skill (adding 7 reference domains, 23 commands, and deterministic anti-pattern rules). Both share the core philosophy of anti-generic design and deliberate aesthetic direction, but Impeccable provides more scaffolding for ongoing projects via PRODUCT.md and DESIGN.md context files.