Top AI Product

AI coding agents have a tell: the frontends they generate all look the same — templated, safe, forgettable. Taste-Skill, an open-source project trending on GitHub, is a set of skill files that tries to fix that by giving agents like Claude Code, Cursor, and Codex “good taste.”

## How it works

It’s not a model — it’s installable skill files that change how the agent approaches UI. A recent v2 rewrite has the agent read the brief, infer the right design direction, and ship interfaces that don’t look templated, with stricter variants tuned for GPT and Codex that push more layout variance and motion. Install is a single command.

## Why it’s interesting

“Slop” — generic, mediocre AI output — is becoming its own problem as more of the web gets agent-generated. Taste-Skill is part of a small wave of tools aimed not at capability but at quality and aesthetics: making the agent’s default output less obviously machine-made. Whether “taste” can really be packaged as a skill file is the bet, but the demand for it is clearly real.