AI-generated design tends to fall into predictable visual tropes: vibrant purple gradients, nested cards within cards, oversized headlines, purposeless whitespace. The problem—known as “AI slop”—isn’t merely aesthetic. It affects accessibility, performance, and credibility. This reality has sparked a new category of tools designed to catch these patterns early, during the generation process itself.
Impeccable is an open-source toolkit created by Paul Bakaus under the Apache 2.0 license. It positions itself explicitly as a design toolkit for AI coding agents. On Claude Code, it installs as a skill through npx impeccable install followed by /impeccable init. It also works with Codex CLI, Cursor, Gemini CLI, Copilot, and OpenCode—offering the same tools across any agent.
The toolkit includes five core commands. Audit produces a technical report with assessments for accessibility, performance, theming, and responsive design. Critique delivers a user experience evaluation with a numerical score. The other three—polish, bolder, quieter—allow precise, targeted aesthetic adjustments. Each command operates on real code and provides verifiable, concrete feedback.
At its core is a detection function built on 62 deterministic rules based on heuristics, not language models. You can invoke it directly via npx impeccable detect <path or URL>, analyzing both local files and live web pages. The detection requires no API keys, cloud credentials, or external calls—it runs locally.
The toolkit also offers a live mode for iterative work. A helper starts on your running application, injects a selector into the page, and lets the agent explore and modify source code in real time. You can generate variants for each element and observe changes applied instantly.
Impeccable doesn’t reinvent the design process. It doesn’t change the model and it doesn’t add new generative capabilities. Its role is to add verification and context to the existing workflow. Final quality still depends on the model you use and the instructions you give.
The promotional material for Impeccable concludes with a recommendation for DevSwarm, a platform that orchestrates multiple agents in parallel, each in an isolated Git worktree with its own terminal and IDE. Worth noting: DevSwarm is presented explicitly as a commercial offering with a free trial at the Pro tier—not a permanently free plan.
Impeccable is a pragmatic approach to a real, visible problem. It won’t transform AI-generated design into world-class work, but it does offer concrete technological guardrails and verifiability. For teams wanting more control over agent design outputs, these deterministic tools are worth investigating.
