Maker here. I am a design student, and I also build sites for real clients. Every time I reached for an AI agent to move faster, I watched it produce the same generic page, the one with Inter and a purple gradient and a hero eyebrow chip. I would spend more time pulling it back toward something with a point of view than I would have spent building it by hand. So I started writing down what I actually do on real work: which palette commits to a brief, which easing an enter gets, how a counter should roll, the tells that mark a page as machine made. That turned into rules, then into a database the agent picks from, then into a detector that catches the slop deterministically instead of relying on taste it does not have. fluid is that, packaged as 6 skills any agent can read. Two honest notes. The rules are distilled from production and client work, but no client site runs on the tool, this is the craft pulled out of that work, not a case study. And it is early, made by one person. The detector is the part I am most proud of: 23 deterministic rules, no model, no API call, so you can run it in CI and it will tell you exactly which line looks like a template and why. Repo is https://github.com/Darcos-Loft/f... and the showcase is at https://fluid.plrstudio.fr. Install with npx fluid-skills install, or on Claude Code add it as a plugin marketplace. I would genuinely love feedback on the detector rules, what slop tell did I miss?

fluid

fluid, give your AI agent a design studio's taste

fluid, give your AI agent a design studio's taste