Skip to content

Skill

GPT Taste

leonxlnx

The same push away from templated pages, aimed at Codex and GPT because they drift differently: it forces a fresh pick of hero shape, type stack and scroll choreography each build instead of reaching for last time's, and leans hard on scroll-driven motion. What separates it from the general version is the model it corrects, not the work it covers.

A situation it fits

The team's frontend pipeline runs on Codex end to end, and every landing page it emits arrives with the same hero shape and the same type stack as the last one. They want the pressure against templated output applied where the repetition is actually coming from, and switching models is not on the table.
GPT Taste. This is the same anti-templating job aimed specifically at Codex and GPT, on the argument that they drift differently, and it forces a fresh pick of hero shape, type stack and scroll choreography on every build rather than reaching for last time's. What separates it from its sibling is the model it corrects rather than the work it covers, and this brief turns on precisely that. The most tempting wrong answer is design-taste-frontend, the general version of the same push, which would be right if the model were not the variable named in the problem. image-to-code changes the order the work happens in, not what the pipeline keeps repeating. redesign-existing-projects improves pages that exist; here every page is new and arrives identical.

Skills it gets confused with

These share a family with GPT Taste, which is another way of saying they are the ones you might reach for by mistake.

Every comparison with GPT Taste

Knowing what GPT Taste does is the easy half. Telling it apart from the others under time pressure is the game.

Today's session

GPT Taste is part of leonxlnx/taste-skill. Licence: MIT. The description above was written for this game, not taken from the skill.