Impeccable uses 61 model-free rules to spot generic AI pages

On 2026-11-03, a Juejin write-up introduced Impeccable, an open-source project from Paul Bakaus aimed at catching the templated look of landing pages produced by coding agents. Its core move is to take the problem out of the prompt: with the same model, even a long design prompt tends to collapse into the same familiar face, and negative instructions such as avoiding Inter or purple-to-blue gradients only govern the current conversation, not the next unattended edit. The repository README describes Impeccable as 1 skill, 24 commands, real-time iteration in the browser, and 61 deterministic detection rules, building on Anthropic's frontend-design skill. Both the command line and the browser extension run these rules without invoking a model or requiring an API key, which makes them usable in continuous integration. The README names the repeated tells it targets: Inter used everywhere, purple-to-blue gradients, cards stacked inside cards, grey text on colored backgrounds, and a rounded square icon above every heading. The article also notes two GitHub pages reporting different figures, and uses the repository page for totals while citing the daily trending page only for the same-day star increase.