Vue Gets Its Own AI Skills – Using a Large‑Model Elimination Contest to Vet Rules

After Vercel packaged a decade of React best practices into an AI Agent Skill, the Vue community quickly responded with the open‑source vue-skills project, which adopts a “scientific experiment” approach: large‑model competitions decide whether a rule survives.

The core of the project is a strict Multi‑Model Verification mechanism consisting of three stages:

Baseline Test : Run a powerful model (e.g., Claude 3.5 Sonnet) without any skill. If it solves the problem, the rule is deemed redundant and eliminated.

Skill Test : If the baseline fails, attach the proposed skill and test again. Success indicates the skill is useful.

Killing Zone : A weaker model (e.g., Claude 3 Haiku) attempts the same task without the skill. If the weaker model solves it, the skill is classified as low‑value information and removed.

This three‑step process ensures that every retained skill carries high entropy—information that AI truly lacks, tends to hallucinate, or is outdated due to rapid Vue ecosystem changes.

Surviving skills fall into two categories:

Capability (AI truly "doesn't know"):

Volar 3.0 Breaking Changes : IDE plugin configuration changes that AI may still write using old settings; the skill forces the correct update.

vue‑tsc strictTemplates : Template‑level TypeScript errors that AI cannot resolve without guidance.

@vue‑ignore : Explicit ignore directives unknown to AI, taught by the skill.

Efficiency (AI can write, but poorly):

SSR HMR : Server‑side rendering hot‑module‑replacement configurations where AI‑generated code runs but performs suboptimally; the skill supplies the optimal solution.

Pinia Store Mocking : Unit‑test mock data that AI often fabricates incorrectly; the skill provides a standardized mock pattern.

To use these hardened skills in editors such as Cursor or Claude, the simplest method is to install them directly, but managing an ever‑growing list of skills becomes cumbersome. The article therefore recommends pairing vue‑skills with the previously introduced Skills Hub , which imports the repository https://github.com/hyf0/vue-skills and automatically synchronizes validated rules to all supported AI editors.

In contrast to Vercel’s React Skill—described as the “official standard”— vue‑skills embodies the community’s “geek spirit,” relying on data‑driven tests rather than exhaustive documentation. This “Test‑Driven Prompting” approach, which maximizes context‑window efficiency, may become the dominant paradigm for AI‑assisted development.