YouMind
Anmelden

20 Ways to Reduce AI Rework Using Claude's Public Skills

@asahi_ai_x
JAPANISCH24. Okt. 2026
101K
12
3
0
25

TL;DR

This article presents 20 specific prompt strategies derived from Claude's public GitHub skills to reduce AI rework. It covers techniques for clarifying instructions, setting constraints, verifying facts, and automating routine business tasks like reporting and contract review.

I'll teach you how to ask AI effectively to reduce "rework," using Claude's public skills as a reference.

How do you usually ask AI to do work for you?

Actually, Claude's public skills contain excellent examples of how to make requests, which are very helpful.

In this article, I've selected 20 that fit our work and introduced them as Japanese prompts.

Let's begin.

Skills are "Instruction Manuals for AI"

A skill is like an "instruction manual" that tells the AI how to proceed with work and what rules to follow.

For example, just asking "Summarize the meeting content" might not yield the desired result.

But if you ask, "Summarize decisions, pending items, and next steps on one A4 page," the final form becomes clear.

Skills are packed with these kinds of clever ways to ask AI.

The references used this time are from GitHub (a site for sharing files), specifically these five:

【Two Official Anthropic Repos】

・knowledge-work-plugins: Instruction manuals for sales, legal, accounting, HR, etc.

・claude-quickstarts: Examples of AI usage, such as morning information organization.

【Three Individual Developer Repos】

・agency-agents: Settings for AI with specialized roles like planning, sales, and inspection.

・agent-skills: A collection of skills by Addy Osmani summarizing AI work processes.

・agents: A collection of various AIs and skills by Seth Hobson.

あさひ|教頭先生のAI活用術 - inline image

Source: GitHub (anthropics/knowledge-work-plugins)

あさひ|教頭先生のAI活用術 - inline image

Source: GitHub (anthropics/claude-quickstarts)

Here is the list 👇️

【Top 20 Picks for Specific Situations!】

・Want to reduce revision requests to AI → ①〜⑦

・Want to simplify morning checks → ⑧

・Want to create weekly/monthly reports → ⑨

・Want to search internal information → ⑩

・Want to check contracts or quotes → ⑪〜⑮

・Want to get proposals approved → ⑯⑰

・Want to handle negotiations, interviews, or cross-department adjustments → ⑱〜⑳

Let's look at each one.

7 Picks to Reduce Rework | Organizing How You Ask AI

First, here are 7 "basics of asking" that can be used for any job.

One reason AI responses don't match your image is starting work without defining the final form or rules.

Using these 7 methods can reduce the hassle of repeated revisions.

① Have AI Ask You One Question at a Time

Reference: interview-me

Useful when you haven't fully organized what you want to ask AI.

For example, just saying "I want to create materials for an internal study session" doesn't clarify the target audience or purpose.

So, have the AI ask you questions one by one, like "Who is this for?" or "What do you want to convey?"

This allows you to organize necessary conditions before starting work.

【Example Request】

text
1Before creating the materials, please ask me one question at a time.
2Once the completion image is clear, please start creating.

② Prevent Misalignment with a "Spec Memo" Before Creation

Reference: spec-driven-development

It's terrible if you ask AI to create 10 slides and then realize, "This isn't what I wanted!"

So, have it create a blueprint summarizing "Purpose, Materials, Final Form, and What Not to Do" before starting work.

Since humans confirm before work begins, major reworks are less likely.

【Example Request】

text
1Before working, summarize Purpose, Materials, Final Form, and What Not to Do in a one-page A4 spec memo.
2Please wait for my approval before starting creation.

③ Decide Rules to Follow First

Reference: constraint-driven-development

To reduce revisions like "Too many characters" or "Missing required fields," decide rules upfront.

Conditions like "Under 800 characters," "3 headings," or "Do not use unverifiable numbers."

Additionally, have the AI verify compliance after completion.

【Example Request】

text
1Declare 3 rules to follow before working.
2After completion, verify if each rule was followed.

④ Write Based Only on Facts in the Source Material

Reference: source-driven-development

AI sometimes answers plausible but non-existent information from source materials.

This is called "Hallucination."

To prevent errors especially in numbers and dates, it's effective to have the AI cite relevant parts of the source material first.

If information is missing, do not fill it in; write "Unconfirmed."

【Example Request】

text
1Quote the original text from the attached material first, then summarize.
2Write 'Unconfirmed' for any information not found in the material.

⑤ Have It Doubt the Output Before Submission

Reference: doubt-driven-development

If you ask AI to "Check the text," it often replies, "It looks well-summarized."

So, have it check from a harsh perspective.

The process is: "Extract claims → Verify evidence → Consider counterarguments → Revise."

This makes it easier to find issues you might have missed.

【Example Request】

text
1Verify the claims and evidence in this document, and list counterarguments or issues.
2Then provide revised suggestions.

⑥ Create a "Inspector" That Doesn't Pass Easily

Reference: Reality Checker

AI tends to evaluate your ideas positively.

But before starting a new project, you want problems more than praise.

So, have the AI act as a "Strict Inspector."

Set the initial evaluation to "Needs Improvement" and establish a rule where it doesn't pass until sufficient evidence is gathered.

【Example Request】

text
1You are a strict inspector.
2List at least 3 problems with this plan.
3Do not pass anything lacking sufficient basis.

⑦ Make Long Manuals Readable

Reference: code-simplification

Internal manuals tend to get long as they are updated.

But simply asking to "Shorten it" risks deleting important rules.

So, have it remove duplicate explanations and extra text while keeping necessary conditions.

【Example Request】

text
1Cut this manual in half.
2Keep mandatory conditions and only remove duplicates and redundant explanations.

【Save This】Prompt Combining All 7 Concepts

Especially recommended is the combination of "Question → Spec Memo → Evidence Verification."

Using the following prompt allows you to align the completion image with AI before starting work.

text
1Proceed with the following [Work Request].
2Do not create the final draft immediately.
3STEP 1:
4Ask up to 3 questions, one at a time, about missing information.
5STEP 2:
6Based on answers, summarize 'Purpose, Final Form, What Not to Do, Acceptance Criteria' and wait for my approval.
7STEP 3:
8After approval, create using only facts from the materials.
9Do not guess unverified information; write 'Unconfirmed'.
10
11[Work Request]
12(Write the materials or work content you want created here)

Don't let it create immediately; align the direction first.

This single trick is the first step to reducing rework with AI.

あさひ|教頭先生のAI活用術 - inline image

8 Picks to Simplify Daily Work | From Morning Checks to Contract Reviews

Next, here are 8 ways to ask AI for daily tasks.

Let AI help with time-consuming tasks like email checks, weekly reports, and contract reviews.

⑧ Summarize Morning Checks in 5 Lines

Reference: Daily Brief

Recommended for those who spend 30 minutes checking emails and chats upon arrival.

Have AI summarize updates since yesterday.

The key is to focus not on all communications, but on "What I need to do today."

【Example Request】

text
1Review updates since yesterday and summarize what I need to do today in 5 lines.
2Write [Unconfirmed] for any information you could not verify.

Even without integrating with internal tools, you can paste update notes to use this.

⑨ Make Weekly/Monthly Reports Easy to Understand

Reference: status-report

Just listing "What I did this week" in a report makes it hard for bosses to grasp the situation.

So, have AI categorize progress into three levels.

・Green: On track

・Yellow: Needs attention

・Red: Action required

【Example Request】

text
1Turn these notes into a weekly report for my boss.
2Show progress in Green/Yellow/Red and summarize decisions needed at the end.

You can turn work notes into reports that are easy for bosses to judge.

⑩ Search Scattered Internal Information

Reference: search

"Where was that travel expense rule written?"

Searching through emails, chats, and shared folders is tough.

Have AI find the necessary information.

Crucially, ask for the source of the information, not just the answer.

【Example Request】

text
1Find the travel expense reimbursement rules and tell me the file name and heading of the source.

If integrated with internal tools, it can search multiple places. If not, attach the regulation files.

⑪ Highlight Concerning Clauses in Contracts

Reference: triage-nda, review-contract

Do you ever wonder which parts of a contract to check?

Have AI check against your company's standards.

For NDAs, classify into three categories:

・Green: Matches company standards

・Yellow: Needs confirmation/negotiation

・Red: Consult legal team

【Example Request】

text
1Classify this NDA clause-by-clause into Green/Yellow/Red based on the attached company contract standards.
2For Yellow and Red, provide reasons and suggested amendments.

Organize points to check before consulting the legal team.

⑫ Analyze Budget vs. Actuals Variance

Reference: variance-analysis

"Expenses increased by 500k yen over budget this month."

This alone doesn't explain why.

Have AI analyze by splitting into "Quantity, Unit Price, Temporary Factors."

【Example Request】

text
1Explain the variance between budget and actuals by splitting into Quantity, Unit Price, and Temporary Factors.
2Tell me if any data is missing.

Helpful for explaining "Why did it increase?" in monthly meetings.

⑬ Reconcile Invoices with Order Data

Reference: reconciliation

Comparing invoices and order data line-by-line is tedious.

AI can help.

Key point: Don't let it assume similar product names are the same item.

Categorize results into "Match, Discrepancy, No Match."

The original skill reconciles ledgers with bank statements; here we apply the logic to invoices and orders.

【Example Request】

text
1Reconcile all invoice and order data lines, classifying into Match, Discrepancy, or No Match.
2Do not omit; output all lines.

Checking counts and totals helps find oversights.

⑭ Turn Your Job into a Procedure Manual

Reference: process-doc

"Only I know how to do this job."

Are there such tasks?

Prepare for transfers or leave by having AI create a procedure manual from your explanation.

Summarize "Required Documents, Steps, Common Exceptions, Contacts."

【Example Request】

text
1Based on my explanation, create a procedure manual summarizing Required Documents, Steps, Common Exceptions, and Who to Ask When Unsure.

Smooths handovers and preserves veteran employees' experience.

⑮ Compare Multiple Quotes Under Same Conditions

Reference: vendor-review

Should we hire Company A or B?

Deciding by price alone risks hidden costs later.

Have AI compare by "Cost, Risk, Switching Burden."

【Example Request】

text
1Compare Company A and B's quotes by Cost, Risk, and Switching Burden.
2Mark unknown items as 'To Confirm'.

Organizes points to check before contracting, useful for negotiations and approvals.

【Save This】Create Report Memos Bosses Can Judge in 1 Minute

Common to these 8 is organizing info into a "form that leads to the next decision."

The following prompt works for daily business memos too.

text
1Turn the following [Business Memo] into a report memo my boss can judge in 1 minute.
21. Omit background or opinions
32. Organize into 'Decisions', 'Pending Items (Options & Deadlines)', 'Unconfirmed/To Verify'
43. Do not guess numbers or causes not in the memo
5
6[Business Memo]
7(Paste memo here, replacing proper nouns with 'Company A', 'Mr. B')

Just pasting meeting notes turns them into reports where bosses instantly grasp the situation.

あさひ|教頭先生のAI活用術 - inline image

5 Picks to Move People | For Approvals, Negotiations, Interviews, Adjustments

Finally, 5 picks for explaining to others or getting decisions made.

Introducing ways to ask AI that make proposals, negotiations, and interviews clearer and lead to action.

⑯ Make Materials Executives Can Judge in 30 Seconds

Reference: Executive Summary Generator

For executive reports, clarity of conclusion matters more than volume.

Summarize materials briefly in order: "Situation, Key Findings, Impact, Proposal, Next Step."

【Example Request】

text
1Summarize this material into one A4 page for executives.
2Start with 3 lines: Conclusion, Impact on Numbers, Decision Needed.

Organize long materials into forms judged quickly.

⑰ Create Proposals Explaining "Why Now"

Reference: Proposal Strategist

Proposals should convey "Why is this needed now," not just features.

Have AI compare "Loss if ignored" vs "Change after implementation."

【Example Request】

text
1Create a proposal for this project.
2First, compare the loss if left alone vs the change after execution.

Makes benefits easier to understand, increasing persuasiveness.

⑱ Design Questions for Negotiations or 1-on-1s

Reference: Discovery Coach

Feel unsure what to ask in negotiations or 1-on-1s?

Have AI design the question sequence.

Recommended flow: "Current State → Problem → Impact of Ignoring → Ideal."

【Example Request】

text
1Create 4 questions for tomorrow's 1-on-1, ordered by Current State, Problem, Impact of Ignoring, Ideal.

Helps organize concerns and hopes while listening.

⑲ Write Polite Follow-ups Without Friction

Reference: Project Shepherd

No reply from another department despite requests?

Chasing repeatedly feels awkward.

Have AI write short texts including "Previous Agreement, Requested Action, Deadline."

【Example Request】

text
1Write a follow-up email to another department.
2Include previous agreement, requested decision, and deadline in a polite, short text.

Providing options makes it easier for them to reply.

⑳ Create Interview Questions and Evaluation Sheets

Reference: interview-prep

Scattered questions/criteria make fair comparison hard.

Define abilities to evaluate first, then have AI create questions and sheets.

E.g., for Sales: "Listening, Problem Finding, Proposal Skills."

【Example Request】

text
1Decide 3 abilities to evaluate for a Sales role interview, and create questions and a 5-level evaluation sheet for each.

Useful for hiring, promotions, and evaluations.

【Save This】Identify Weaknesses Before Submitting Proposals

Before submitting proposals/approval requests, have AI act as a "Strict Executive."

Pre-check issues likely raised in meetings.

text
1You are a decisive executive strict on numbers and field burden.
2Read the following [Proposal Memo] and identify issues before submission.
31. 3-line summary readable in 30 seconds
4(Conclusion, Impact on Numbers, Approval Sought)
52. 5 questions likely raised in meeting
6(Profitability, Field Burden, Alternatives, Stop Conditions, Unconfirmed Data)
73. List of info to verify before submission
8Do not invent numbers not in memo; write 'Unconfirmed'.
9
10[Proposal Memo]
11(Paste proposal memo here)

Find weaknesses before submission, not after being criticized.

This reduces rework on proposals and approval documents.

あさひ|教頭先生のAI活用術 - inline image

3 Shared Mindsets Across All 20 Picks

We introduced 20 ways.

Actually, all share common traits to reduce AI rework.

Especially important are these 3:

① Define the Finished Product, Not Just "Do Anything"

"Help me as an excellent assistant" is vague.

Better: "Summarize this material into one A4 page for executives."

Clearly state what and in what form you want created.

② Set "Acceptance Criteria" Before Work

"Under 800 chars," "3 headings," "Only verified numbers."

Define acceptance criteria upfront.

agent-skills includes tables preventing AI from skipping steps.

Verification conditions are set from the start.

Meaning: Don't end with "probably okay"; define what confirms completion.

③ Don't Let It Fill Gaps Guessingly

AI guesses unverified info.

State rules upfront:

・Don't guess numbers not in materials

・Write "Unconfirmed" for unverifiable things

This reduces risk of baseless info mixing in.

claude-quickstarts Daily Brief design also has a rule:

Report unread info as "Could not read," not "No problem."

【Save This】Convert Daily Tasks to "Low-Rework Prompts"

Apply these 3 mindsets to daily work.

Just write one time-consuming task in the following prompt.

text
1Convert the following [Daily Task] into a prompt resistant to rework.
2Include these 4:
31. Finished Product (What, in what form)
42. What Not to Do (Unnecessary explanations/guesses)
53. Rules to Follow (Handling unverified info)
64. Post-completion Checklist (3 items)
7If info is missing, ask questions instead of guessing.
8
9[Daily Task]
10(Write one time-consuming task here)

E.g., input "Create weekly report every Friday."

Convert usual tasks into "Request Templates" easy for AI to handle.

あさひ|教頭先生のAI活用術 - inline image

3 Things to Check Before Using at Work

These 20 picks can be tried in chat screens immediately.

However, note these 3 points when using for work:

① Do Not Input Confidential or Personal Info

Do not give AI confidential/personal info without company permission.

When practicing, replace client names with "Company A", staff with "Mr. B".

② Do Not Install External Tools Arbitrarily

Check company rules before installing skills/plugins on work PCs.

These prompts require no installation.

③ Humans Must Perform Final Verification

AI outputs are drafts.

Always have humans verify critical info like numbers and contract terms.

【Save This】Improve Your Usual Prompts

text
1Audit the following [My Prompt].
21. Is the finished product clear?
32. Are exclusions stated?
43. Is there a rule to mark unknowns as 'Unconfirmed'?
5List one missing condition and suggest improvement.
6
7[My Prompt]
8(Paste your usual prompt here)

Just add one missing condition to your current prompt; no need to rebuild from scratch.

Try Just One of the 20 Picks First

This time, I introduced 20 ways to reduce AI rework using Claude's public skills as reference.

These can be tried immediately in ChatGPT, Claude, or Gemini.

Try just one prompt that fits your work first.

Gradually reduce rework with AI and spend time on more important tasks!

Please save this article to refer back when needed.

Thank you for reading to the end!

I publish AI utilization tips for ChatGPT, Claude, Gemini daily.

Follow for tips usable today.

@asahi_ai_x

【Open Chat for Learning AI Utilization!】

🎁Joining Benefits🎁

✅50 AI Prompts to Speed Up Work

✅Templates for Asking AI

✅ChatGPT/Claude/Gemini Usage Cheat Sheet

Lurking is OK!

Take a peek.

Join here 👇

https://line.me/ti/g2/2OjHnH9mrx_sE3tQhyWSpAEOR-zOhMPhJfqXwg

あさひ|教頭先生のAI活用術 - inline image

▼Asahi Official LINE

"Asahi Official LINE" delivers latest AI news for ChatGPT, Claude, Gemini, immediate utilization tips, and prompts for work/side jobs.

Add friend for deals and benefits ☺️

https://lin.ee/R8kmC3c

💼Consultations via DM!

※Article based on GitHub public info as of Oct 24, 2026. Stars/content may change.

・Anthropic Official knowledge-work-plugins

https://github.com/anthropics/knowledge-work-plugins

・Anthropic Official claude-quickstarts

https://github.com/anthropics/claude-quickstarts

・agency-agents (msitarzewski)

https://github.com/msitarzewski/agency-agents

・agent-skills (addyosmani)

https://github.com/addyosmani/agent-skills

・agents (wshobson)

https://github.com/wshobson/agents

In YouMind remixen

Turn one viral article into a full content workflow

Collect the source, decode the pattern, create assets, draft the story, and distribute from one AI workspace.

Explore YouMind
Für Creator

Verwandle dein Markdown in einen sauberen 𝕏-Artikel

Wenn du eigene Langtexte veröffentlichst, wird die 𝕏-Formatierung von Bildern, Tabellen und Codeblöcken mühsam. YouMind macht aus einem ganzen Markdown-Entwurf einen sauberen, sofort postbaren 𝕏-Artikel.

Markdown zu 𝕏 testen

Mehr Muster zum Entschlüsseln

Aktuelle virale Artikel

Mehr virale Artikel entdecken