What is Impeccable? The design skill for AI coding agents, and what its install changes

Impeccable is Paul Bakaus's open-source design skill for AI coding agents: one `/impeccable` skill with 24 commands, plus a command-line detector that flags 59 common "AI-made" design patterns without calling a model. The part most install guides skip is the hook. On Claude Code, Cursor, Codex, GitHub Copilot and Grok Build, npx impeccable install also writes a project hook that runs the detector whenever the agent edits a UI file, and in Claude Code that hook runs even when you deny the agent's own launcher command.

The repository is pbakaus/impeccable, Apache-2.0. We read it on 8 October 2026: the latest skill release was 4.5.0 (2 October), the engine v0.1.12 (published late on 7 October UTC), and the repo showed 78,339 stars. Stars say it is popular, not that it is safe; this guide is about what it changes in your project.

Why people are searching for it

We checked Google Trends on 8 October 2026, about 09:00 Asia/Riyadh, worldwide, past 7 days, for the seed query claude skills:

Trends values are relative, not search counts. The same list also showed ui/ux pro max (+80%) and superpowers github (+60%); both are already on Skill Rank, so this guide covers the one that was not.

On X, posts containing pbakaus/impeccable, npx impeccable or impeccable.style (reposts excluded) numbered 134 in the seven days to 06:00 UTC on 8 October, highest on 3 October with 29. Several of those posts, between 3 and 5 October, described the repo as trending on GitHub. The questions in those threads were mostly about what it installs, which agents it supports and whether it needs an API key.

Impeccable at a glance

The README says Impeccable "started from" Anthropic's frontend-design skill and adds three things: one setup flow, a shared command vocabulary and the detector rules.

What the 24 commands do

Everything runs through one skill: /impeccable <command> <target>. The README groups them loosely like this:

• Setup and documentation: init asks about your product and writes PRODUCT.md; document writes a DESIGN.md from existing code; extract pulls tokens and components into a design system.
• Review: critique (hierarchy, clarity), audit (accessibility, performance, responsive checks), polish (final pass before shipping).
• Direction: bolder, quieter, distill, colorize, typeset, layout, animate, delight, overdrive.
• Robustness: harden (errors, i18n, overflow), onboard (empty states, first run), clarify (UX copy), adapt (devices), optimize (performance).
• Building: shape (plan before code), craft (shape then build), live and generate (variants in the browser).

You can pin a favourite as its own shortcut: /impeccable pin audit gives you /audit.

The skill file itself is opinionated. Its anti-pattern list tells the agent not to use "overused fonts (Arial, Inter, system defaults)", gray text on coloured backgrounds, nested cards or bounce easing. Its SKILL.md also says "The brief wins": if your brief pins a font or palette, the agent should keep it even when it trips a warning.

What npx impeccable install changes in your project

This is the part to read before you run it on a work repository. According to the README on 8 October 2026:

1. It asks where to install. It lists the agent folders it detects (for example ~/.claude, ~/.codex or a project .cursor), then asks for project or global scope. Scripts can pass --providers= and --scope=.
2. It installs a hook on five agents. The hook runs the detector on direct UI file edits:

1. It may download a binary. The skill's launcher runs a self-contained engine that either ships next to it or is "downloaded once on first run into ~/.impeccable/bin/". No Node is needed except for npx itself.
2. It creates a working folder. Commands write screenshots, caches and session state under .impeccable/. The README gives a .gitignore block, and lists the files that should be committed: config.json, live/config.json, design.json, surfaces/*.md and critique/*.md.

Two sentences in the README deserve a second read:

"In Claude Code, installed command hooks run independently of model-tool approval. The first edit or Stop event can therefore download and cache the engine even if the session denies the model's launcher command."

"Applying copy edits automatically runs package.json's optional scripts["impeccable:manual-edit-validate"] command in a shell, with your user permissions."

Neither is hidden. Both mean that saying "no" to the agent in chat is not the same as turning Impeccable off. The README's own switches are --no-hooks at install time, hook.enabled in .impeccable/config.json, and, for Claude Code, launching with --settings '{"disableAllHooks": true}'.

A safe first run

If you want to try it without committing to the hooks:

1. Run the detector alone on a branch: npx impeccable detect src/. Exit code 0 means no primary findings, 2 means findings, 1 means a target could not be scanned.
2. Read the findings. The README is careful here: a clean run "is evidence, not proof of visual or accessibility quality".
3. If you like the vocabulary, install with npx impeccable install --no-hooks, run /impeccable init, and try one review command such as /impeccable critique on a single page.
4. Add the hook later, once you know what the detector flags in your codebase. Add the .gitignore block at the same time.

Our general checklist for reading a skill before installing it applies here too: open the SKILL.md, look at what scripts it runs, and check which files it writes.

Impeccable or another design skill?

• Want a small, plain skill file and no hooks? Anthropic's frontend-design is the base Impeccable started from.
• Want a shared command vocabulary, a written product brief and a detector that runs on every UI edit? That is what Impeccable adds.
• Already compared taste-style design skills? We covered taste-skill, Emil Kowalski's skill and awesome-design-md on 6 October.

FAQ

Is Impeccable free? The repository is Apache-2.0 and the detector needs no API key. The commands run inside your coding agent, so they use whatever model and plan your agent uses.

Does Impeccable work in Cursor? Yes. The README lists Cursor, and on Cursor the hook checks proposed writes before they land. The README's manual-copy note for Cursor mentions enabling Agent Skills in settings.

How do I update it? npx impeccable update. On Codex, approve the hook again in /hooks if it changed.

How many detector rules are there? The README said 59 on 8 October 2026. Some posts on X quote 61; check the README for the current number.

Limits of this guide

• Based on the repository README, the compiled Claude Code SKILL.md, GitHub release data and the npm registry, read on 8 October 2026. We did not install Impeccable or run its hooks.
• Google Trends figures are relative. X counts depend on the query; posts mentioning only "Impeccable" without the repo name, install command or site are not included.
• Stars, trends and post counts measure attention, not quality or safety.