Impeccable Fights AI Coding Tools' Purple-Gradient Sameness

The open-source repository that made Paul Bakaus's design skill famous cannot actually be read to see how its 61 rules work: the detection engine ships as a compiled binary.

Impeccable, a skill that gives AI coding agents 61 rules against the purple-gradient sameness of AI-generated interfaces, has 71,200 stars on GitHub as of Friday, up 306 in the previous 24 hours alone, according to GitHub's trending page and its API.

Paul Bakaus, a former Google engineer credited as the creator of the Khroma color tool, published Impeccable as an open-source, Apache 2.0-licensed skill on Nov. 16, 2025, according to the repository's creation date. It reached 10,000 stars within four months and shipped version 1.5.1 on March 17, 2026, according to a report from the design publication Abduzeedo.

What the skill actually does

Impeccable installs into a coding agent's configuration folders and runs 24 commands, invoked as /impeccable audit or /impeccable polish, that check a project's interface against 61 deterministic detector rules, according to its GitHub README. The rules do not call any external API; they scan file edits directly and flag patterns like the Inter font, purple-to-blue gradients and nested cards that Bakaus says agents trained on similar SaaS templates repeat by default. Two configuration files, PRODUCT.md and DESIGN.md, store project-specific context, and the skill supports live browser iteration for testing visual variants. It works inside at least 16 named agent harnesses, including Cursor, Claude Code, GitHub Copilot, Codex CLI and Google's Antigravity, and its README credits Anthropic's own "frontend-design" skill as the project's starting point, part of the same Agent Skills ecosystem that Anthropic opened as a cross-platform standard on Dec. 18, 2025.

The deal that took it from 10,000 stars to 40,000

Bakaus launched a company called Renaissance Geek on June 22, 2026, backed by venture firm a16z in a round led by investor Anish Acharya, according to a post on a16z's site. The company struck a distribution partnership with GitHub the same day, and Impeccable began shipping pre-bundled with the GitHub Copilot app, the post said. By then the skill had passed 40,000 GitHub stars and more than 160,000 installs through the third-party registry skills.sh, according to the announcement. "AI raised the floor. Time to raise the ceiling," Bakaus wrote in the post, arguing that AI tools had made baseline competence free but that "clear intent, relentless iteration, judgment, and craft" still separated good design from generic output.

What Renaissance Geek has not published

The a16z announcement did not disclose the size of the funding round. Nor has Bakaus published any measurement of whether the 61 detector rules change the quality of a blind design test, or how many of the 160,000 recorded installs are still active rather than abandoned after a single run. The README itself notes a structural limit on outside scrutiny: the rule-detection engine ships as a self-contained binary, with Node.js needed only for the optional npx installer, so the logic behind the 61 rules cannot be read or audited from the open-source repository the way the rest of the project can.

Milestone	Date	Stars
v1.5.1 ships	March 17, 2026	10,000
Renaissance Geek launches, GitHub deal	June 22, 2026	40,000+
Current	Sept. 26, 2026	71,200

Bakaus is now hiring a chief of staff, design advocates and design engineers at Renaissance Geek, according to the June announcement, which framed the company around what it calls the "Renaissance Geek": a generalist with both taste and the ability to work directly in code. Whether that framing survives contact with a bigger team, or whether Impeccable's growth slows now that the GitHub bundling deal is months old rather than news, is not something either the repository or the funding announcement addresses. Other GitHub-trending skills and agent tools, like the sandboxed agent manager Paperclip, have kept climbing the trending page well after their own launch news faded, so a slowdown is not guaranteed either way.

Sources

1. pbakaus/impeccable (GitHub)
2. Impeccable: The Open-Source AI Design Skill for Better UI (Abduzeedo)
3. Impeccable by Design (a16z)