[90% Don't Know] "20 Skills to Give AI Expertise" Saved by 20,000 People

"AI, how many times do I have to explain the same thing...😩"

"I taught you so much yesterday, why are we starting from zero again..."

This isn't the AI's fault; it's your own "way of teaching AI."

Have you ever experienced this while using AI tools?

• Every time you open a new conversation, you have to re-explain the same premises from scratch. Yesterday's work isn't carried over to today's AI.
• Even though you spent time writing a prompt, the quality of the output varies every time.
• You've heard that "Skills" are becoming popular lately, but you don't know what the right answer is.
• You have no sense of "building up" work with AI. Every day feels like a reset.

Many of you probably relate to this.

By reading this article to the end, you will gain the method to switch from a "life of writing prompts every time" to a "life of giving AI expertise."

An article published by m0h (@exploraX_), who regularly posts viral articles in the overseas AI community, is currently going viral with 5.3 million views and 20,000 bookmarks.

This time, I will break down that content for Japanese businesspeople and go even further to explain the "thinking method for turning your own tasks into skills," which isn't in the original article.

Before we start, I have just two requests.

1. Save this article and set aside 20 minutes this week.
2. If you have colleagues using AI, please share this article with them.

Here is the original post.

https://x.com/exploraX_/status/2039269234253934811

Prompts Disappear, but Skills Remain Forever

The reason 20,000 people bookmarked this article isn't just the list of 20 skills itself.

What this article conveyed was the fact that "the era of writing prompts every time is over." It was a great article including that insight.

Think about it. Writing a prompt every time is the same as teaching the AI from scratch every time. The "way to write a proposal" you spent an hour teaching yesterday disappears without a trace today.

How does this change with Skills?

You write instructions for the AI in a text file and save it. Once written, the AI will read that file from then on and reproduce the same quality output as many times as you want.

To be more specific, the way we teach AI has evolved significantly over the past few years.

Prompts disappear after one use. You have to rewrite them every time.

CLAUDE.md is a configuration file where you tell the AI, "Operate according to these rules." It's just text, but its presence alone stabilizes the AI's behavior.

Skills are a systematic organization of CLAUDE.md into folders. You can increase the AI's areas of expertise for each task, such as "How to write a proposal" or "How to do competitive analysis."

A Personal OS is the state where the AI runs your entire business as a "dedicated OS for you" once enough Skills have been accumulated.

Prompts disappear with every conversation.

But Skills remain forever as files. Saving Skills as files means giving the AI a memory. Because it has memory, it can reproduce the same quality, and as memory accumulates, the AI's capabilities build up.

In the past few months, the overseas AI community has seen a succession of viral hits: Karpathy's CLAUDE.md explanation, a 33-page Skills guide, and now these 20 Skills.

This is no coincidence; it's happening within the major shift from the "era of prompts" to the "era of Skills."

Personally, I think this is the most important change right now. Instead of "writing every time" as a prompt, you "write once and use forever" as a Skill.

Whether you can make this switch is the turning point between those who use AI as a tool and those who give AI expertise.

If You Can Structure and Explain Your Work, You Can Use Skills

When you hear "Skills," it might sound like you need programming knowledge.

But in reality, the content is just a text file. It's in .md file format, but it's no different from text you'd write in Notepad or Notion.

The original article title says 20 skills, but including two meta-skills, all 22 skills are built with the same structure:

• name: Name of the skill
• description: What the skill does
• features: What it can do
• output format: What format to output in
• instructions: Specific steps
• constraints: What not to do

In short, this is a "business manual for AI."

When you teach a new employee, you tell them, "In this situation," "in this order," and "output in this format," right? Skills are just writing that same thing in text.

To put it more simply, it's like a "Technical Machine (TM)" in Pokémon.

If you use one TM, that Pokémon learns a new move. Once learned, it can be used in every battle. Skills work exactly the same way; if you set one .md file, the AI learns a new area of expertise and demonstrates it every time.

I explained Claude Code Skills using the TM analogy in the article below, so check it out if you're interested 👇

UT ClaudeCode Lab

@ClaudeCode_UT

·

Apr 25

Article

[Complete Preservation Edition] Thorough Guide to Making AI Remember How to Work with Claude Code Skills

"I'm worried if I can use AI because I have zero programming knowledge..."

"I'm scared of looking at the black terminal screen and don't feel like it."

"But I'm anxious about being left behind by the AI trend !😱"

Even in that state, Claude Code...

1

88

827

589K

That's why I can say this:

Anyone who can write CLAUDE.md—that is, anyone who can structure and explain their own work—can use Skills.

If you can write a proposal, you can write Skills.

Of course, it's not a cure-all.

Skills are incredibly effective for tasks that can be standardized, but for creative tasks that require completely different judgments every time, they remain merely an "aid."

Just knowing this boundary should eliminate the misconception that "I can't use Skills."

8 of the 22 Skills are Effective for Non-Engineers

The original article introduces 22 skills flatly across five categories: Writing, Visual, Research, Video, and Coding.

In this article, I will reorganize them for Japanese users and present them in a way that is closer to actual business practice. Specifically, I've selected 8 that directly link to daily tasks in planning, marketing, corporate planning, and business development, organized from a different angle than the original article.

Below, let's compare the world with and without Skills for four business challenges.

• These are sample skills, so please apply them as a base.

■ The 2 hours spent worrying about proposal structure disappears

Skills used: Structured Writing + Hook Generation

Monday morning, your manager says, "Submit a new business proposal by Friday." When you sit at your computer, the most time-consuming part isn't writing the content. It's the structural work of deciding "what to write in what order," and before you know it, 2 hours have passed. That feeling of not making progress in front of a blank screen is something many people recognize.

By adding the structured writing skill called the SCQA Writing Framework in the original article, this structural work becomes almost zero. SCQA stands for Situation → Complication → Question → Answer, a framework used by consulting firms like McKinsey.

Just by giving it the theme "Internal proposal for new business," a skeleton structured into these four parts will come back in 10 minutes.

All you have to do is focus on the judgment of "what to advocate."

The other one, Hook Generator, is a skill to strengthen the beginning of proposals or presentations. If you ask it to "write the first 3 lines of this proposal," it will return multiple patterns of opening sentences that grab the reader's attention.

The task of "thinking about the opening phrasing for 30 minutes" turns into a judgment of "which angle to pull them in with."

I've created a skill tailored to the Japanese business scene to show what kind of file you should actually make. You can copy and paste this as is.

1---
2name: Proposal SCQA Structure
3description: Used when generating the skeleton of internal proposals/plans with SCQA structure
4---
5
6# Proposal SCQA Structure Skill
7
8## Input
9Receive the proposal theme in one line
10
11## Output Format
12Return a skeleton composed of the following 4 parts
13
14### S (Situation)
15- 2-3 lines of facts that stakeholders will nod along to
16- Use premises or numbers shared within the company if available
17
18### C (Complication)
19- Identify specific problems arising from the current situation
20- 1 line on the impact of "leaving it as is"
21
22### Q (Question)
23- One sentence for the question naturally derived from the complication
24
25### A (Answer)
26- Answer the question conclusion-first
27- Execution steps in 3 or fewer
28
29## Constraints
30- Keep each part within 3 lines
31- Do not use technical jargon. Write in words a manager can understand in 5 seconds
32- Do not use ambiguous expressions like "will consider" or "will promote"

■ Summarizing meeting minutes ends in 2 minutes

Once you can create materials that communicate well, the next step is to efficiently collect and organize the information that forms the basis of those materials.

Skills used: Long-form Summarization + Information Structuring

A one-hour meeting ends. Spending 30 minutes manually summarizing while re-listening to the recording. Spending 20 minutes looking for where last month's minutes were. Spending another 15 minutes searching when asked, "What did we decide on that matter last time?"

In information-handling jobs, this "organizing work" slowly eats up time.

If you pass meeting transcript text to the Long-Form to Summary Compressor skill mentioned in the original article, a structured summary of 5 key points, action items, and decisions will come back in 2 minutes.

Furthermore, combining it with the Knowledge Structuring Skill will, for example, structure "10 customer feedbacks from this month" by category.

The important thing here is how to chain these two together.

Compress with Summary, systematize with Structuring, and accumulate. Creating this flow structurally resolves the problem of "information being scattered and unfindable." Instead of just hoarding, by structuring and hoarding, it becomes knowledge that can be used later.

Information organization that took 30 minutes shrinks to 7 minutes. And you can focus the saved time on judgments like "is this information worth keeping?" or "how should I organize it so it's usable in six months?"

Here is the skill that chains this together.

1---
2name: Knowledge Assetization from Meeting Minutes
3description: Used when accumulating knowledge by chaining summarization and structuring from meeting recordings or notes
4---
5
6# Meeting Minutes → Knowledge Assetization Skill
7
8## Input
9Meeting recording text, rough notes, or chat logs
10
11## Processing (Chained in 2 steps)
12
13■ Step 1: Summarization (Compression)
14- Compress each speaker's points into 1 line
15- Separate decisions and pending items
16- Extract action items by owner + deadline
17- Output: 3-part structure of "Key Points," "Decisions," "TODO"
18
19■ Step 2: Structuring (Systematization)
20- Categorize Step 1 output by project/theme
21- Show connections if related to past meeting notes
22- Automatically extract "items to confirm at the next meeting"
23
24## Output Format
25Summarize "Current Summary" + "Thematic Structuring" + "Next Confirmation Items" in one file
26
27## Constraints
28- Do not change the speaker's intent
29- "To be considered" is pending. Do not put in decisions
30- One file per meeting. Do not mix multiple meetings

■ Half a day of competitive research becomes 30 minutes

Once information is organized, let's deepen research and analysis based on that information.

Skills used: Research Report Generation + Competitive Analysis

In corporate planning and business development, quarterly competitive reports are unavoidable. But if you spend half a day "opening competitor sites one by one, picking up news, and summarizing in Excel," you won't have time left for the crucial task of "how to respond to these trends."

If you give a theme like "Latest trends in the SaaS industry" to the Deep Research Synthesizer in the original article, it will return a research report that extracts patterns from a large amount of information. It includes citations, so you can track where the information came from.

The Competitive Intelligence Skill is a skill that automatically generates a SWOT analysis just by giving it the names of three competitors.

Work that took half a day to collect data ends in 30 minutes. And you can reclaim the time you should originally spend on strategic judgments like "how will our company differentiate against these competitor trends?"

Here is a skill that integrates these two.

1---
2name: Competitive Analysis Report
3description: Used when researching competitor trends and creating SWOT analysis reports
4---
5
6# Competitive Analysis Report Skill
7
8## Input
9- Competitor company names to analyze (1-3 companies)
10- Our company's business area in one line
11
12## Output Format
13
14■ Overview of each company
15- Business content / Recent moves / Points of interest within 3 lines each
16
17■ SWOT Analysis (Our perspective)
18- Strengths: Strengths our company has over competitors
19- Weaknesses: Points where our company is inferior
20- Opportunities: Opportunities for our company arising from competitor moves
21- Threats: Threats to our company
22
23■ Direction of Differentiation
24- Identify one thing "competitors are not doing"
25- One line on the action our company should take in that area
26
27## Constraints
28- Base only on public information. Clearly state guesses as "guesses"
29- Limit each item to 3 bullet points
30- Use a format that can be used as-is in management meetings

■ SNS posting no longer starts from zero every time

After the analysis is done, it's the phase to communicate those results.

Skills used: Content Expansion + Tone Unification

For marketing staff, "creating 5 patterns of SNS posts from one blog post" is a standard task.

But making it short for X, formal for LinkedIn, processing it for newsletters...

Just rewriting to match the tone of each medium takes 2 hours. Moreover, if shared across a team, the writing style becomes inconsistent, and brand unity collapses.

If you pass one blog post to the Content Repurposing Engine in the original article, it will convert it into four patterns: X post, LinkedIn, newsletter intro, and video script. You no longer need to rewrite from scratch.

The Tone & Style Enforcer is a skill that, if you have it read your brand tone guide, will automatically convert text written by anyone into a unified tone following the guide.

Work that took 2 hours to match each medium ends in 15 minutes. You can focus on content strategy judgments like "which message should I emphasize this week?"

I've created a skill that combines these two.

1---
2name: SNS Content Expansion
3description: Used when generating posts for each SNS in bulk from blog posts or internal reports
4---
5
6# SNS Content Expansion Skill
7
8## Input
9Text from source blog posts, press releases, internal reports, etc.
10
11## Output
12Simultaneously convert for the following 4 media
13
14■ For X (formerly Twitter)
15- Condense key points within 140 characters
16- Include a hook in the first line that makes it "personal"
17- Do not use bullet points; compose with a series of short sentences
18
19■ For LinkedIn
20- Polite tone (Desu/Masu)
21- Add one paragraph at the beginning to supplement industry context
22- End with one question to encourage comments
23
24■ For Newsletter Intro
25- Structure to lead to the main body with "Read more here"
26- Summary where readers can judge "it's worth reading" in 3 lines
27
28■ For Internal Slack Sharing
29- Start with "In 3 lines"
30- One line at the end for action proposals to relevant departments
31
32## Constraints
33- Do not change the original article's claims. Only change the angle
34- Complete one medium at a time so tones don't mix
35- Do not add hashtags

We've looked at 8 skills across 4 business challenges, and some of you may have noticed.

What they all have in common is the pattern of "tasks disappearing so you can focus on judgment." The task of thinking about structure, the task of organizing information, the task of collecting data, the task of rewriting for each medium. By shifting this "hand-moving time" to Skills, the ratio of "brain-using time" increases dramatically.

And while skills work well individually, they change things even more when combined. Research with Deep Research, organize with Knowledge Structuring, and summarize into a proposal with SCQA. With this chain, the entire flow of "Research → Organize → Propose" is automated.

Implementation takes only 5 minutes with Copy-Paste

For those wondering, "So, how do I put those skills in?" here are the specific steps.

Create a folder named .claude/skills/ inside your project folder. It takes 1 minute.

Copy and paste the .md file of the skill you want to use into that folder and save it. 2 minutes is plenty.

If you start Claude Code, it will automatically read and use the skills. Including confirmation, it's 2 minutes.

The point is that you can "accumulate systematically as folders." One entry is one area of expertise. Ten entries are ten areas of expertise. Since Skills build up on the file system, the AI gets smarter the more you use it.

Try putting in just one first. The moment you put one in, you should be able to experience the difference between the "world of writing prompts every time" and the "world where instructions accumulate with Skills."

The Greatest Weapon is the Thinking Method of "Skill-ifying Your Own Work"

The 8 skills introduced so far are just "ready-made products."

What's truly powerful is the thinking method itself of turning your own work into skills.

The original article includes two meta-skills, Workflow Automation Agent and Skill Creator, which make "making it yourself" possible.

Let's look at the process of skill-ifying "creating meeting minutes" specifically.

First, decompose the task into steps with the Workflow Automation Agent. For meeting minutes, it's "Get recording text → Identify speakers → Extract key points → Identify action items → Format." 5 steps.

Next, define the input and output for each step. Input is recording text. Output is structured minutes, and the structure is decided as 3 parts: "Key Points," "Decisions," and "TODO List."

Then, convert it to an .md file with the Skill Creator. Here is the actually completed skill.

1---
2name: Meeting Minutes Structuring
3description: Used when creating structured meeting minutes from recording text or rough notes
4---
5
6# Meeting Minutes Structuring Skill
7
8## Input
9Receive meeting recording text or rough notes
10
11## Processing Steps
121. Identify and organize speakers
132. Extract key points of the discussion chronologically
143. Separate decisions and pending items
154. Identify next actions with owners
16
17## Output Format
18
19### Meeting Overview
20- Date / Participants / Purpose in one line each
21
22### Key Points (Chronological)
23- 3-5 bullet points of discussed content
24
25### Decisions
26- Describe with the set of "What," "By when," "Who"
27
28### TODO List
29- Table format of Owner / Content / Deadline
30
31## Constraints
32- Do not change nuances when summarizing speakers' opinions
33- Categorize items ending in "to be considered" as pending, not decisions
34- Keep it to a volume that can be read within 5 minutes

From next time, with just one word, "Make meeting minutes," the same quality of minutes will come back every time.

I'll also introduce the skill-ification of "proposal review."

Decomposing the task: "Structure check → Check numerical evidence → Identify expected questions → Improvement suggestions." 4 steps. Input is proposal text, output is review comments from 4 perspectives.

1---
2name: Proposal Self-Review
3description: Used for self-reviewing proposals or plans before submission
4---
5
6# Proposal Self-Review Skill
7
8## Input
9Receive the full text of the proposal
10
11## Review Perspectives (4)
12
13### 1. Structure Check
14- Is the conclusion first?
15- Is it in SCQA structure? (Point out if not)
16- Is the role of each section clear?
17
18### 2. Numerical Evidence Check
19- If numbers appear, ask "What is the source of this number?"
20- Are premises explicitly stated?
21- Are comparison targets appropriate?
22
23### 3. Expected Question Identification
24- List 3 questions a boss or decision-maker might ask
25- One line of proposed answer for each question
26
27### 4. Improvement Suggestions
28- Identify the one weakest section and suggest improvements
29- Narrow it down to a level of "it will pass if you just fix this one point"
30
31## Constraints
32- Review comments within 1 page total
33- Do not write "good points." Focus only on points to improve

If you keep it as an .md file, just by saying "Review this proposal," the same perspective of review will come back every time. You can pass the AI review yourself before asking a senior for a review.

There is one important boundary here.

What is easy to skill-ify are "routine tasks that proceed with the same steps every time." Creating minutes, standard reports, checklist-type reviews. These tasks are perfectly compatible with Skills.

On the other hand, "tasks that require judgment in a completely different context every time," such as deciding the direction of a new business, cannot be completed with Skills alone. Human judgment is necessary there.

But that's exactly why it's valuable.

By leaving routine work to Skills, you increase the time for judgments that only humans can make. In a previous post, I said, "Same tool, same price. What makes the difference is the design of the environment," and Skills are exactly the specific method for designing that "AI environment."

Every time you add one skill, the AI's area of expertise increases by one. Ten for ten areas. If you put in 50, it's no exaggeration to call it "your own dedicated AI team."

When the accumulation of Skills exceeds a certain threshold, the moment comes when it changes from individual task support to a "business-wide OS." The lineage of concepts shown at the beginning: Prompt → CLAUDE.md → Skills → Personal OS. That final destination lies ahead of the reader building up their own skills.

Summary

• The essence of the original article saved by 20,000 people is "the end of the era of writing prompts every time." There is a mechanism called Skills where you teach the AI once and it remembers forever.
• The content of Skills is just a text file. Zero programming knowledge is fine. Anyone who can structure and explain their own work can use it.
• I've selected 8 skills for non-engineers from the original 22. Introduced them by 4 business challenges: "Proposals," "Information Organization," "Research," and "Content Expansion."
• In each skill, "tasks" turn into "judgments." Proposal structure 2 hours → 10 mins, meeting summary 30 mins → 2 mins, competitive research half a day → 30 mins, content expansion 2 hours → 15 mins.
• With Claude Code, implementation takes 5 minutes with copy-paste. The AI automatically reads them just by putting them in a folder.
• Not just using ready-made products, but the thinking method of "skill-ifying your own work" is the greatest weapon. Both meeting minutes and proposal reviews can be skill-ified.
• One skill for one area of expertise. Accumulating them makes it "your own dedicated AI team," and eventually a Personal OS that runs your entire business.

In this article, I introduced 6 skills in a form you can copy and paste as is. Try one first.

For those who found this article helpful:

UT Claude Code Lab (@ClaudeCode_UT) is an account run by current University of Tokyo students who are Claude Code enthusiasts. Joint agent development with major companies is also underway.

Follow to receive information like this 👇

■ Practical Claude Code techniques including Skills

■ Selection and explanation of the overseas AI front line

■ This is the only place to get skills and tools developed seriously by UT students for free

If you're interested, please follow and check it out ❗️

[Official LINE started!]

I will provide skill delivery and limited information via the official LINE in the future! Let's register!

https://lin.ee/quYxNMc