一键导入
using-rune
Use when starting any conversation or after compact - establishes Rune iron laws, skill invocation rules, and red flags
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。
菜单
Use when starting any conversation or after compact - establishes Rune iron laws, skill invocation rules, and red flags
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
Use when starting any creative work — creating features, components, or functionality, modifying behavior, or solving ambiguous problems — before design or implementation. Covers product discovery, competitive research, feature analysis, technical design, and capability spec writing. Applies to every project; routes to writing-plans (non-UI) or design-workflow (UI).
Use when finishing an edit to source files — runs format, typecheck, and console.log detection before moving to the next task.
Use when implementation of a bug fix or feature task is complete — dispatches the appropriate language and global reviewers and returns a binary APPROVE/BLOCK verdict.
Use when about to git commit — validates commit message format, lints staged files, and scans staged content for secrets and debug artifacts.
Use when reaching the design-workflow V2-4 hard gate — adversarially reviews design artifacts (token coverage, contract completeness, artifact consistency, accessibility docs, responsive coverage, DESIGN.md compliance) against docs/designs/<feature>/ before development handoff.
Use when a task involves UI — new pages, components, visual interactions, or design tokens — including work in docs/designs/, *.pen files, or with Pencil MCP tools. Skip for backend-only, config-only, or refactoring without visual impact. Routes to writing-plans on completion.
| name | using-rune |
| description | Use when starting any conversation or after compact - establishes Rune iron laws, skill invocation rules, and red flags |
如果一个 skill 有 ≥1% 概率适用于当前任务,你必须 invoke 它。
这不是建议,不是可选项,你不能合理化绕过。
违反字面就是违反精神。任意一条被绕过即视为流程失败。
没有 L2 = "降级铁律"。 chore 类改动(修 typo、调 hook regex、改配置、改 README)不是绕过流程,而是由 brainstorm skill 的 Scale Gate 判定为 Truly Simple 后走简化路径——仍 invoke brainstorm 公告、仍跑 quality-gate、仍调 reviewer、仍走 commit 闸。模型不得自判"简单"直接跳到 edit,必须先 invoke brainstorm 让 Scale Gate 公开判定。
调用规则:用户消息进来 → 检查是否有 skill 适用 → 有 1% 可能 ⇒ invoke Skill 工具 → 公告"使用 [skill] 来 [目的]" → 严格按 skill 执行。Skill 调用先于阅读代码、先于 clarifying questions。
优先级:用户显式指令 > Rune skill > 默认系统提示
行为纪律:不要默默选一种解释就开始执行。有歧义时,将其显式化:
这些念头出现 = 流程偏离信号,立即停下。
| 借口 | 现实 |
|---|---|
| "这个简单,直接做不用 skill" | 简单事最容易踩未检验假设 |
| "我先看看代码再决定要不要 skill" | 看代码=已在脑补方案;skill 调用先于阅读 |
| "我记得这个 skill 怎么用" | skill 会演化,每次按当前版本读 |
| "用户没说要走流程" | 用户说 WHAT,规则定义 HOW |
| "这次只改一行,跳过 TDD" | 一行也能引入回归,30 秒写测试 |
| "几个任务一起审查更高效" | 跨任务合并审查不允许(L1) |
| "改动很小,不需要审查" | 一行就能引入 SQL 注入、原型污染 |
| "reviewer 误报了,跳过吧" | 误报需说明具体理由由用户确认 |
| "先 commit 再说" | 没审查的代码不许 commit(L3) |
| "我已经手动测过了" | 手动测试无记录、不可重放、不算覆盖率 |