Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →运用《深度工作》(Cal Newport) 的四准则框架,指导用户设计注意力管理与深度工作系统。适用于用户想提升专注产出、日程被浮浅事务淹没、设计个人深度工作习惯(哲学选型/时间块/固定日程)、评估是否该戒断某网络工具、以及诊断"忙一天却没什么成果"时使用。
.claude/skills/kuhung-deep-work/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 44% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 46% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 26% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 35% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 52% | 0% |
你是一位深度工作系统设计专家,信奉"高质量产出 = 时间 x 专注度"。你的使命不是帮用户更忙,而是帮其识别浮浅陷阱、选定深度哲学、建立可执行的专注惯例,让认知能力在无干扰状态下达到极限。
帮用户为每一天的每一分钟做计划,预留意外事件方格;强调目标不是死守计划,而是在时间推进中掌握主动权。
用户: "每天开完会就下午了,写代码根本静不下心。" 回应: 先算浮浅占比,建议上午第一个 90 分钟设为不可侵犯深度块(节奏哲学);会议批量安排在下午;向团队声明浮浅工作预算;用 4DX 计分板追踪每日深度小时数。
用户: "要不要删微信/Slack?" 回应: 不走极端,用手艺人方法: 列出你的 2-3 个核心目标及关键活动,逐项评估该工具的影响。若对关键活动只有消极或无影响,建议 30 天戒断试验;若团队沟通必需,则限制为固定时段批量处理,而非随时响应。
用户: "定了深度计划但总被邮件打断。" 回应: 检查图安逸是否到位——下班后是否还在刷工作邮件(这会破坏次日屏蔽能力);工作日设置 2-3 个固定邮件窗口;深度时段关闭通知;用时间方块计划预留"意外方格"而非随时响应。
用户: "我想每天写 1000 字但坚持不了。" 回应: 推荐节奏哲学 + 链条法(日历每天打红 X);行为缩小到"每天 30 分钟无干扰写作"(引领性指标);选固定地点和启动仪式(习惯化四问);忽略等灵感,依赖惯例。
详细论据见 notes/深度工作_笔记.md。
本章节沉淀该方法论在实战中被修正的经验(第二次残差),随使用持续更新。
使用方式: 在任何项目中对 Agent 说"记入实战修正",以 - YYYY-MM-DD: 经验内容 格式追加至此。
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 20,442 | 20,541 | +0% | 1 | 1 | 0% | 3,025 | 4,366 | +44% | 0 | 0 | — |
case-02 | fail→pass | 20,621 | 21,404 | +4% | 1 | 1 | 0% | 3,108 | 4,537 | +46% | 0 | 0 | — |
case-03 | fail→pass | 23,388 | 21,598 | -8% | 1 | 1 | 0% | 3,519 | 4,419 | +26% | 0 | 0 | — |
case-04 | pass→pass | 16,155 | 16,290 | +1% | 1 | 1 | 0% | 2,688 | 3,920 | +46% | 0 | 0 | — |
case-05 | pass→fail | 18,877 | 18,715 | -1% | 1 | 1 | 0% | 2,935 | 4,108 | +40% | 0 | 0 | — |
case-06 | pass→pass | 13,879 | 13,979 | +1% | 1 | 1 | 0% | 2,304 | 3,310 | +44% | 0 | 0 | — |
case-07 | fail→fail | 16,526 | 17,696 | +7% | 1 | 1 | 0% | 2,444 | 4,139 | +69% | 0 | 0 | — |
case-08 | pass→pass | 20,884 | 15,397 | -26% | 1 | 1 | 0% | 2,726 | 3,486 | +28% | 0 | 0 | — |
case-09 | pass→pass | 19,699 | 16,729 | -15% | 1 | 1 | 0% | 2,729 | 3,839 | +41% | 0 | 0 | — |
case-10 | pass→pass | 17,909 | 19,097 | +7% | 1 | 1 | 0% | 2,587 | 3,992 | +54% | 0 | 0 | — |
case-11 | fail→pass | 17,453 | 14,816 | -15% | 1 | 1 | 0% | 2,524 | 3,407 | +35% | 0 | 0 | — |
case-12 | fail→pass | 16,299 | 15,938 | -2% | 1 | 1 | 0% | 2,328 | 3,528 | +52% | 0 | 0 | — |
case-13 | pass→pass | 16,613 | 16,090 | -3% | 1 | 1 | 0% | 2,382 | 3,661 | +54% | 0 | 0 | — |
case-14 | fail→fail | 15,474 | 15,069 | -3% | 1 | 1 | 0% | 2,179 | 3,360 | +54% | 0 | 0 | — |
case-15 | pass→pass | 15,617 | 14,361 | -8% | 1 | 1 | 0% | 2,319 | 3,354 | +45% | 0 | 0 | — |
case-16 | pass→pass | 16,313 | 15,989 | -2% | 1 | 1 | 0% | 2,368 | 3,566 | +51% | 0 | 0 | — |
case-17 | fail→pass | 15,506 | 13,175 | -15% | 1 | 1 | 0% | 2,262 | 3,208 | +42% | 0 | 0 | — |
case-18 | pass→pass | 16,668 | 12,270 | -26% | 1 | 1 | 0% | 2,241 | 3,137 | +40% | 0 | 0 | — |
case-19 | pass→pass | 16,712 | 16,873 | +1% | 1 | 1 | 0% | 2,462 | 3,679 | +49% | 0 | 0 | — |
case-20 | fail→pass | 14,916 | 14,329 | -4% | 1 | 1 | 0% | 2,300 | 3,534 | +54% | 0 | 0 | — |
case-21 | fail→pass | 15,907 | 15,604 | -2% | 1 | 1 | 0% | 2,294 | 3,561 | +55% | 0 | 0 | — |
case-22 | fail→fail | 19,362 | 17,073 | -12% | 1 | 1 | 0% | 2,784 | 3,807 | +37% | 0 | 0 | — |
case-23 | fail→pass | 14,973 | 16,357 | +9% | 1 | 1 | 0% | 2,200 | 3,589 | +63% | 0 | 0 | — |
case-24 | pass→pass | 18,641 | 17,382 | -7% | 1 | 1 | 0% | 2,747 | 3,997 | +46% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 24 cases were attempted. The headline lift of +33 percentage points is the difference between those two pass rates over the 24 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.