菜单
Ask the agent to do agent-driven end-to-end validation for this Runelight repo. Use when the user mentions intelligence test, `.it.md`, agent-driven E2E, or wants confidence beyond unit/type/build checks.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Install, upgrade, or ensure Runelight preview/capture integration in a TypeScript React project or TypeScript Vue 3 project. Use when a project asks for "install Runelight", "set up Runelight", "upgrade Runelight", or "update Runelight"; this bootstrap skill detects the host, wires the integration, and verifies end-to-end.
Write, edit, or review Runelight React components in .g.tsx files. Use when working on React/TSX .g protocol components, GFrames, createGScopeHook, createGProvider, provider variants, JSX branch coverage, or React Runelight authoring diagnostics.
Write, edit, or review Runelight Vue SFC components in .g.vue files. Use when working on Vue .g protocol components, <g:frames>, template-first scope frames, Vite Vue preview, or Vue Runelight authoring diagnostics.
Polish Runelight-covered GUI through a required explicit sync before editing, then a red-team observe-edit-reobserve loop that works from product surface model down to details one confirmed point at a time. Use when the user asks to polish, refine, improve visual quality, clean up UI details, or make an existing `.g.tsx` / `.g.vue` interface feel better.
Refactor existing React TSX into production .g.tsx UI models without preview wrappers.
Refactor existing Vue 3 TypeScript SFCs into production .g.vue UI models without preview wrappers.
| name | intelligence-test |
| description | Ask the agent to do agent-driven end-to-end validation for this Runelight repo. Use when the user mentions intelligence test, `.it.md`, agent-driven E2E, or wants confidence beyond unit/type/build checks. |
Run tests like a smart human would: understand the change, choose the relevant projects/routes/commands, and verify real behavior.
If the user gives specific .it.md notes or scenarios, follow them; otherwise look under intelligence-tests/ and choose selectively based on impact area. Do not try to exhaust every local project by default.
Prefer real evidence over narration: dev server routes, HTTP responses, browser-visible results, screenshots/traces when useful, and deterministic commands such as tests/typecheck/build.
Clean up temporary files and stop servers. Report what was tested, what passed, and any important gap left untested.
