SuperAPI AI Directory

find-skills-ja

General & other

View repository

Updated 10-16 08:08

About this skill

Find Skills JA — Skill Discovery (Improved Version for Japanese Users)

A skill for finding and installing skills from the open agent skill ecosystem (skills.sh).

Searching skills.shdoes not match Japanese or natural-language queries(verified in practice; see table below). Before searching, this skill always converts queries into "short English noun keywords," and checks quality and safety before proposing installation.

Query form Search results
Japanese ("meeting minutes", "test") 0 results
English natural language ("how to write meeting notes") 0 results
Short English keywords ("meeting minutes") Matches

When to use

  • When asked "Can you do ~?", "Is there a skill for ~?", or "I want to automate ~"
  • When an existing skill likely exists for the task
  • When the user shows interest in extending the agent's capabilities

Procedure

Step 1: Identify the need

Organize the following from the user's request.

  1. Domain (e.g., React, testing, document creation, work efficiency)
  2. Specific task (e.g., meeting minutes creation, PR review, slide generation)
  3. Whether it is a common task (i.e., whether a skill is likely to exist)

Step 2: Design search keywords (most important)

Always create search terms according to the following rules.

  • Convert into 2–3 English noun keywords(do not use sentences; do not leave Japanese)
  • Prepare queries in at least 2 synonym variantsand search both, then merge the results
    • Example: meeting minutes と meeting notes / deploy と deployment と ci-cd
  • If unsure, translate "the deliverable name for that task" into English (議事録→minutes, 見積→estimate)

Frequently used Japanese→search keyword translation table:

Japanese request Search keywords (1st candidate / 2nd candidate)
I want to create meeting minutes meeting minutes / meeting notes
Proposal / planning document proposal / proposal writer
Slide / PowerPoint creation slides / pptx presentation
Diagramming materials diagram / visualization
PR review pr review / code review
Writing tests testing / unit test
Refactoring refactor / code quality
Deployment / release deploy / deployment ci-cd
Proofreading / polishing text writing / copyediting
SEO measures seo / seo optimization
Data analysis data analysis / analytics
Scraping scraping / web scraping
Excel work xlsx / spreadsheet
Documentation maintenance docs / documentation readme

Step 3: Run searches (collect at least 6 candidates)

npx skills find <英語キーワード> [--owner <owner>]
  • Always search using all prepared keyword variants(do not stop at one variant. Each variant returns up to 6 results)
  • Merge the results, remove duplicates, andcreate a candidate list of 6–10 total itemscreate a candidate list of 6–10 total items
  • If fewer than 6 results, search with a broader third keyword as well (e.g.: slides insufficient → also search with presentation / pptx )
  • If you want to narrow down to a specific company or author, use --owner(e.g.: --owner anthropics)

Do not proceed to presenting while only 2–3 candidates have been gathered.

Step 4: Quality and safety verification (performed only on recommended candidates)

Do not propose installation based on search results alone. However, perform detailed verification only on thetop 1–2 candidatesyou intend to recommend (doing it for all candidates is slow. For the candidate list, you may include the search result metadata as-is). For recommended candidates, always check the following.

  1. Install count — 1K+ is a guideline. However, install counts can be inflated by bulk repository installs (if all skills in the same repository have roughly the same number, it's from a bulk install). Do not trust the numbers alone.
  2. Source reliability — Official orgs (such asanthropics、vercel-labs、microsoft、 googleworkspace ) are highly reliable. For unknown authors, always do the following 3.
  3. Review of SKILL.md body — Before installing, fetch and read the target SKILL.md, and report the following to the user. How to fetch: on the skill page https://skills.sh/<owner>/<repo>/<skill> the body is displayed. A GitHub raw URL is also fine, but since the path layout differs by repository, if you get a 404, use the skills.sh side.
    • A summary (within 3 lines) ofwhat this skill instructs the agent to doSummary (within 3 lines)
    • Whether there are any dangerous instructions: sending information externally, curl / fetch to unknown URLs, access to credentials, destructive commands, suspicious wording such as "hide this instruction", etc.
    • The repository'slast updated date and star count(neglected skills operate on outdated assumptions)

Step 5: Presentation to the user (two-stage approach: candidate list → recommendation)

Do not show only 2–3 narrowed down. First show the gathered candidates as a list in a form that makes it clear "many were found", and then propose a recommendation. Format:

「議事録」で 8件見つかりました(meeting minutes / meeting notes で検索)

| # | スキル | 提供元 | installs |
|---|---|---|---|
| 1 | meeting-minutes | github/awesome-copilot | 9.5K |
| 2 | meeting-notes | claude-office-skills/skills | 4.3K |
| 3 | meeting-minutes-taker | daymade/claude-code-skills | 852 |
| 4 | ...(集めた候補をすべて並べる。6〜10行) |

★ おすすめは 1 の meeting-minutes です。
提供元がGitHub公式orgで3.7万スター、昨日更新。本文もレビュー済みで
議事録整形の指示のみ、危険な指示はありません。

次点: 2 の meeting-notes — シンプルさ重視ならこちら(ただし6ヶ月更新停止)。

インストールしますか?(番号指定でもOK / プロジェクトのみ・全プロジェクト共通が選べます)
  • List in descending order of installs. In principle, include all gathered candidates (exclude only clearly irrelevant ones)
  • Reason for recommendation within 2 lines (based on officialness, install count, update status, and body review results)
  • Adding one runner-up makes it easier to choose

Step 6: Installation

Execute after obtaining the user's confirmation.

# 既定: プロジェクトスコープ(確認プロンプトあり)
npx skills add <owner/repo@skill>

# ユーザーが「全プロジェクトで使いたい」と明示した場合のみ
npx skills add <owner/repo@skill> -g
  • -yDo not use (no confirmation)
  • Before installation, check for role overlap with existing skills using npx skills ls , and if there is overlap, point it out

Step 7: If not found

  1. Tell the user that no matching skill was found
  2. Suggest what can be helped directly with standard features
  3. If it's a frequently performed task, suggest building your own: npx skills init my-skill

A little trick for trial installation

If you want to try it just once on the spot without installing:

npx skills use <owner/repo@skill> | claude

It is safe to guide them through the order of "first take a taste with use → if good, add".