What is a “skill”? (beginner explanation)
Goal: Explain a skill with one narrow job (messy notes → weekly update): fixed I/O plus a human gate—not “do my job.”
Best for: You keep pasting the same long instructions every week.
A skill is a reusable instruction pack you can name, test, and version—not a vague “be helpful” vibe. Start here before you build a library.
Why beginners skip this
Calling a skill a “universal macro” feels explanatory—boundaries and fixtures get skipped. Say the one job first.
Step 1 — One sentence definition
A skill is a small playbook: written steps, examples, and rules for one recurring job (e.g. “turn meeting notes into decisions”). The AI loads that playbook when the job matches — so you do not retype the playbook every time.
Many products store a skill as a folder with a file often named SKILL.md. You do not need to memorize the file format on day one — start with the idea.
Step 2 — Skill vs prompt vs Project (no slang)
Use this table to place each thing; do not shove checklists into “help me work.”
| Thing | Use it when… | Lives where? |
|---|---|---|
| One prompt | Single ask, maybe never again | Inside today’s chat |
| Skill | Same process every week | Reusable library you (or your team) install |
| Project | Same files + ongoing context | A workspace with documents |
Step 3 — Spot a good first skill topic
- You did it at least three times in two weeks.
- The steps are stable (same checklist), only the input changes.
- Mistakes are annoying but not catastrophic if the AI slips — you still review.
Example
Filled prompt
Candidate skill: “Weekly status from bullet dump”
Input: messy bullets from Slack
Output: 5 wins / 5 risks / 3 asks for leadership
Human check: I delete anything I cannot defend in the meeting
What good output looks like
- Skill purpose is one job for “What is a “skill”?” — not a company brain
- Done + golden example present and tested
- Never-send / never-delete honored
- Pass/fail: fail-case run once (skills-what-is-a-skill #1)
Step 4 — What you will click later (preview)
Exact menus move. Look for words like Skills, Customize, Plugins, or Create. On Claude and ChatGPT, skills may sit under customize / settings / plugin directory. If you cannot find it yet, keep your playbook in a Project instruction — same habit, simpler surface.
Example
Filled prompt
Jobs: (1) apology email (2) status polish (3) one-off brainstorm.
Mark which deserve a skill vs one-off prompt. One-sentence purpose for the top skill only.
What good output looks like
- Matches the named files/tools/constraints in the fill
- Refuses invented facts or silent send/delete
- Pass/fail check you can mark in one minute
- Stops before irreversible steps without approval
Do not bookmark-only
Finish Practice now, then tick this guide on your path. Titles unread are not skills.
Close the loop
Write one line: next time I wave “universal macro” as the explanation, I refuse to start.
Check the unknown path
What should the model do when a field is missing—null, TBD, or ask you? Write it once in Boundaries.
Example
Filled prompt
Jobs this week: (1) apology emails (2) one-off brainstorm (3) status polish twice.
Mark skill vs one-off. One-sentence purpose for the top skill only.
What good output looks like
- Done + golden example present and tested
- Never-send / never-delete honored
- Pass/fail: fail-case run once (skills-what-is-a-skill #2)
- Skill purpose is one job for “What is a “skill”?” — not a company brain
Red flag
Calling everything a skill; cannot state purpose in one sentence—split before you build.
Save this card
Card: reusable job hat · purpose sentence · split if purpose blurs. Not the company brain.
Before you start
Open the skill card (or blank one-pager). Purpose must be one job for “Understand a skill as a reusable how-to the AI.” Golden example slot cannot stay empty. Guide: skills-what-is-a-skill. Keep the real job visible the whole time; a blank box means stop and fetch the artifact first.
How to use this week
This week explain one real skill to a teammate in three plain sentences: what it does, what it must not, how you test it.
Pitfalls
| Pitfall | Fix |
|---|---|
| Calling every prompt a skill | Skill = reusable pack with name/tests |
| One mega-skill for all jobs | One job per skill |
| No version or changelog | name-vN + one line |
Example
Filled prompt
Temptation: skip the habit for skills-what-is-a-skill to save five minutes.
Job still: Understand a skill as a reusable how-to the
Smallest correct pass: Done visible · checkable artifact · Unknowns [?].
Boundaries: draft only; no silent send; no invented metrics.
Return: artifact + refuse-line said aloud.
What good output looks like
- Names the shortcut refused (specific to skills-what-is-a-skill)
- Still delivers checkable output for “What is a “skill”?”
- Unknowns labeled; draft-only
- Pass/fail: refuse-line dated (skills-what-is-a-skill #4)
Practice now
Trigger test: say the trigger to a teammate; they must know when it fires (Asia/Shanghai).
Non-skill: write one thing that should NOT be a skill (Asia/Shanghai).
Skill-defn drill: one skill = trigger + instructions + boundaries—write all three (Asia/Shanghai).
Thin-spot drill: rewrite the Save card with real nouns from your work (repo, doc, inbox). Generic placeholders fail the drill.
Tonight explain one real skill to a teammate in three plain sentences: what triggers it, what it does, what it must never do.
Close the loop: date one refuse-line (Asia/Shanghai) after your first live pass today.
Buffer line: name the artifact path you used (Asia/Shanghai).
One sentence takeaway
Three plain sentences: does / must not / how you test.
Next
什么是「技能 / Skill」?(新手版)
目标: 用一个窄题目(比如杂乱要点 → 周报)说清 skill 是什么:固定输入输出,加上人把关,不是「帮我上班」。
适合: 听过 skill 概念、还没写出第一张候选人卡的人。
好的候选卡:题目窄、输出形状固定、写明人要删什么。吹嘘词帮不上忙。
为什么新手跳过
把 skill 当「万能宏」像省解释;边界与夹具会被跳过。先说清一件活。
步骤 1——一句话定义
Skill(技能)就是一份小剧本:针对一件反复出现的事(例如「会议记录变成决定清单」),写好步骤、例子和规则。任务对得上时,AI 会加载这份剧本,你就不用每次从头重打。
很多产品会把技能存成一个文件夹,里面常有名为 SKILL.md 的文件。第一天不必背格式,先抓住这个想法即可。
步骤 2——Skill、提示词、Project 有何不同
用这张对照表分清三样东西各放哪;别把清单塞进「帮我上班」。
| 东西 | 什么时候用 | 放在哪 |
|---|---|---|
| 单次提示词 | 就这一次,以后未必再用 | 今天的对话框里 |
| Skill | 每周同一套流程 | 可安装的技能库(个人或团队) |
| Project | 同一批文件 + 长期背景 | 带文档的工作区 |
步骤 3——什么样的事值得做成第一个技能
- 两周内至少做过三回。
- 步骤稳定(同一张清单),只是每次输入不同。
- 就算 AI 偶发失误,后果可接受:你仍会人工过目。
实例
填好的提示词
候选技能:「从杂乱要点生成周报」
输入:Slack 里粘来的凌乱要点
输出:5 件进展 / 5 个风险 / 给领导的 3 个请求
人工核对:会上讲不圆的,我一律删掉
边界:不编造;未知标 [?];只出草稿。
像样的输出大概长这样
- 题目很窄:「从杂乱要点生成周报」,不是「帮我上班」
- 输出形状固定:5 进展 / 5 风险 / 3 请求
- 写明人要怎么把关:会上讲不圆的一律删
- 输入输出边界清楚
- 像候选人卡,能决定值不值得做成 skill
- 一分钟内能对照清单判过/不过;缺项标红再继续
步骤 4——以后你会点哪里(先预习)
具体菜单会变,去找 Skills、Customize、Plugins、Create 这类词就行。在 Claude 和 ChatGPT 里,技能可能藏在自定义、设置或插件目录;如果暂时找不到,先把剧本写进 Project 说明:习惯一样练,入口更简单。
实例
填好的提示词
请用不超过 120 字向队友解释 skill 与一次性提示词的差别,以我们的「周报邮件」为例。不要用吹嘘词。
听完要能回答:这顶帽子下周还戴不戴。
边界:不编造;未知标 [?];发送前停下。
完成检查:一分钟内能判过/不过。
像样的输出大概长这样
- ≤120 字,以「周报邮件」为例讲清差别
- 不用吹嘘词
- 队友听完知道:skill=可复用的固定流程,不是一次聊天
- 可直接贴进团队介绍
- 短、准、不装
- 听完能回答:这顶帽子下周还戴不戴
- 未知标出;不编造负责人、日期或指标
- 一分钟内能对照清单判过/不过;缺项标红再继续
不要只收藏
先做完「现在就练一次」,再在学习路径上打勾。只收藏不练,标题变不成技能。
闭环
写下一行:下次用「万能宏」糊弄解释,我拒绝开讲。
检查未知路径
字段缺失时模型该怎么做:填 null、写 TBD,还是问你?在边界里写清楚一次。
红旗
把 skill 当「万能宏」,边界与夹具会被跳过;先说清一件活。
保存这张卡
记下:skill 名 + 版本号 + 红线夹具是否测过。标上日期。
开始前
信任该 skill 前,先确认已有完成标准 + 一个实例。
本周怎么用
本周用三句白话向队友解释一个真 skill:做什么、不做什么、怎么测。
常见坑
| 坑 | 改法 |
|---|---|
| 候选题目「帮我上班」 | 收窄到固定输入输出 |
| 无人把关说明 | 写明人要删什么 |
| 用吹嘘词介绍给队友 | ≤120 字、举周报例、不吹 |
现在就练一次
今晚就练:用三句白话向队友解释一个真 skill:做什么、不做什么、怎么测。
一句话带走
三句白话:做什么、不做什么、怎么测。
下一步
← Back to home · Learn · Tools · Compare · Start