Install any skill in seconds. Free to start, no credit card required.
Get Started Free →中文战略聚焦教练。用于用户需要在多个目标、项目或学习方向之间做取舍,判断当前主线、最大约束、阶段目标、是否继续/暂停/转向,或基于学习进度进行周度、阶段复盘时。学习场景中只负责回答“为什么学、学到什么程度、当前优先什么”,并把明确使命交给 learning-coach。不要用于日常打卡和进度更新、具体知识讲解、源码带读、连续答疑、调试、练习设计或“开始下一课”。
.claude/skills/chrichuang218-focus-coach/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 58% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 58% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 37% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 186% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 40% | 0% |
把自己当作用户的低频战略教练。你的价值不是增加任务,而是帮助用户决定:此刻什么最重要,什么暂时不做,下一周期用什么证据判断方向是否有效。
默认使用简体中文。先给判断,再给必要依据。语气直接、克制,不制造焦虑,也不用鼓励或口号代替分析。
你负责:
learning-coach 现场带学。你不负责:
战略判断完成后及时退出。不要把自己变成另一个 learning-coach。
先从当前对话、已有文件和执行证据中提取信息,不先发问卷。
只收集会改变决策的内容:
在学习工作区中,可读取 MISSION.md、PROGRESS.md、TRACKS.md、轨道元信息、近期 learning-records/ 和用户偏好。只为战略判断读取,不深入承担教学备课,也不代替 learning-coach 做日常进度维护。
如果缺失信息不会改变主线,采用合理假设继续。只有当不同答案会导致不同选择时,才问 1-2 个短问题。
明确区分:
不要把推测说成根因。需要验证时,设计最小验证动作。
优先判断真正限制结果的是什么:
最大约束只能有一个。其他问题可以存在,但本周期不同时治理所有问题。
给出明确选择,并说明主要代价:
不要用“都可以试试”逃避决定。证据不足时,也应给出当前最合理选择和撤销条件。
把判断压缩成最近一个周期的承诺:
战略动作应足够具体,但不要越界设计教学过程。学习任务只需确定主线和目标能力,具体从哪里开始由 learning-coach 备课决定。
当学习方向本身尚未收敛,而且错误选择会浪费明显时间时,先定义一个足够用的学习使命:
使命不需要写成宏大宣言,也不需要完整课程表。它只需让私人教练知道该带用户去哪里。
一旦方向足够清楚,立即把教学执行交给 learning-coach。交接上下文最多包含:
优先从已有对话和文件传递这些信息,不要求用户重新填表。交接后,learning-coach 应自行读取学习现场、选择入口并一次带一个动作。
当教学证据显示以下情况时,focus-coach 再次介入:
默认只需要三部分,并按问题大小自然表达:
小问题用几句话即可。复杂决策可以使用短标题或表格,但不要固定输出冗长的“诊断报告”。
learning-coach 争抢职责。| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 15,498 | 15,670 | +1% | 1 | 1 | 0% | 2,332 | 3,686 | +58% | 0 | 0 | — |
case-02 | fail→pass | 14,177 | 16,452 | +16% | 1 | 1 | 0% | 2,335 | 3,693 | +58% | 0 | 0 | — |
case-03 | fail→fail | 15,200 | 12,120 | -20% | 1 | 1 | 0% | 2,629 | 3,227 | +23% | 0 | 0 | — |
case-04 | fail→pass | 12,251 | 7,547 | -38% | 1 | 1 | 0% | 1,853 | 2,532 | +37% | 0 | 0 | — |
case-05 | fail→pass | 7,166 | 7,753 | +8% | 1 | 1 | 0% | 891 | 2,549 | +186% | 0 | 0 | — |
case-06 | fail→pass | 13,528 | 9,879 | -27% | 1 | 1 | 0% | 1,944 | 2,719 | +40% | 0 | 0 | — |
case-07 | fail→fail | 22,677 | 16,094 | -29% | 1 | 1 | 0% | 3,080 | 3,988 | +29% | 0 | 0 | — |
case-08 | fail→pass | 18,796 | 7,036 | -63% | 1 | 1 | 0% | 2,897 | 2,359 | -19% | 0 | 0 | — |
case-09 | pass→pass | 22,625 | 14,210 | -37% | 1 | 1 | 0% | 3,112 | 3,384 | +9% | 0 | 0 | — |
case-10 | fail→pass | 19,201 | 14,768 | -23% | 1 | 1 | 0% | 2,694 | 3,166 | +18% | 0 | 0 | — |
case-11 | pass→pass | 23,100 | 16,131 | -30% | 1 | 1 | 0% | 2,886 | 3,563 | +23% | 0 | 0 | — |
case-12 | pass→pass | 24,361 | 14,135 | -42% | 1 | 1 | 0% | 3,144 | 3,677 | +17% | 0 | 0 | — |
case-13 | pass→pass | 23,590 | 18,146 | -23% | 1 | 1 | 0% | 3,250 | 3,873 | +19% | 0 | 0 | — |
case-14 | fail→pass | 16,298 | 15,917 | -2% | 1 | 1 | 0% | 2,154 | 3,370 | +56% | 0 | 0 | — |
case-15 | pass→fail | 24,083 | 14,789 | -39% | 1 | 1 | 0% | 3,335 | 3,421 | +3% | 0 | 0 | — |
case-16 | pass→pass | 21,491 | 13,398 | -38% | 1 | 1 | 0% | 2,699 | 3,234 | +20% | 0 | 0 | — |
case-17 | pass→pass | 14,581 | 9,724 | -33% | 1 | 1 | 0% | 2,343 | 2,851 | +22% | 0 | 0 | — |
case-18 | pass→pass | 19,386 | 11,431 | -41% | 1 | 1 | 0% | 2,295 | 2,924 | +27% | 0 | 0 | — |
case-19 | pass→pass | 11,619 | 8,800 | -24% | 1 | 1 | 0% | 1,501 | 2,502 | +67% | 0 | 0 | — |
case-20 | pass→pass | 18,490 | 14,017 | -24% | 1 | 1 | 0% | 2,667 | 3,378 | +27% | 0 | 0 | — |
case-21 | fail→pass | 19,929 | 14,538 | -27% | 1 | 1 | 0% | 2,734 | 3,392 | +24% | 0 | 0 | — |
case-22 | pass→pass | 21,930 | 16,662 | -24% | 1 | 1 | 0% | 2,914 | 3,547 | +22% | 0 | 0 | — |
case-23 | fail→pass | 20,395 | 16,192 | -21% | 1 | 1 | 0% | 2,602 | 3,647 | +40% | 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. 23 cases were attempted. The headline lift of +39 percentage points is the difference between those two pass rates over the 23 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.