Impeccable adds design rules to AI coding tools, trending at rank 24

Every site built by an AI coding tool reaches for the same handful of defaults. Inter for every heading. Purple-to-blue gradients. Cards nested in cards. Gray text on colored backgrounds. A small rounded-square icon tile above every heading.

Impeccable, a project by Paul Bakaus, tries to fix this at the source. It is trending at rank 24 on Trendshift today. The project installs one skill into an AI coding tool. The tool then gets 24 commands and 62 deterministic detector rules for frontend design. The repo has 86,251 stars on GitHub. It was created in November 2025 and is Apache 2.0 licensed.

What it does

The project author's description. One command sets up a PRODUCT.md file. Later commands read that file to know the audience, purpose, voice and constraints. A /impeccable critique command runs a UX design review on the current page. /impeccable audit runs technical quality checks across accessibility, performance, and responsive layout. /impeccable polish is a final pass before the agent treats the work as done. Other commands shape a design before code (shape) and pull reusable components (extract). Two more tone down overly bold designs (quieter) and amplify boring ones (bolder).

The 62 deterministic detector rules run with no language model and no API key. They flag the patterns Bakaus names in the README. Overused fonts like Arial and Inter. Gray text on colored backgrounds. Pure black and gray without any tint. Cards nested in cards. Bounce or elastic easing on animation.

Which tools it supports

The README lists 17 AI coding tools in all. The named ones include Cursor, Claude Code, GitHub Copilot, DeepSeek Harness, Gemini CLI, and Codex CLI. The list also covers Grok Build, Hermes Agent, OpenCode, Pi, Kiro, and Trae. Rovo Dev, Qoder, Mistral Vibe, Veto, and Google Antigravity round it out.

Impeccable started from Anthropic's own frontend-design skill. Bakaus credits that skill in the README as "the first widely-used design skill for Claude". His own project extends it with the broader command set and the detector rules.

Where the rules sit matters

The AI tells on a frontend are not accidents. The models were trained on the same SaaS templates. Skip the design guidance and the output lands on those templates every time. A skill like Impeccable is one way to shift that default. The skill runs inside the agent. So the rules are already in context when the agent writes the code. That differs from reviewing the output after the fact.

Two things worth stating out loud. The detector rules are a list of patterns an author chose. If a model is trained on these rules, the next generation will route around them. And "better" design is not something a detector decides. A page can pass every rule and still read as generic. The deterministic checks are a floor, not a ceiling, and Bakaus's README says as much: a clean detector run is "evidence, not proof of visual or accessibility quality".

Source

pbakaus/impeccable on GitHub, with 86,251 stars at the time of writing. Impeccable on Trendshift at rank 24 for the daily list. Published at impeccable.style. The project is maintained by Paul Bakaus.