We just published research on how AI agent skills perform across model tiers. Ke

We just published research on how AI agent skills perform across model tiers.

Key finding: Knowledge skills are a bigger win on cheaper models. The correctness lift roughly triples from frontier to smallest model.

But there's a nuance: the "taste" (design system, anti-pattern guardrails) transfers down-tier, but the verification loop (render, critique, fix) doesn't — it needs a browser and enough model capability.

We added an automated quality gate to address this.

Full article: https://splatdev.com/blog/do-ai-agent-skills-help-weaker-models-more-we-measured-it/