copilot で claude 向け?っぽいSkillは使えるのか

巷で(主にXで)流れてくる、こんなSkillが公開されているぞ。というニュースを見ていると、大概サムネにClaudeのロゴもセットで目にする。

Copilot中心で使っている身としては、自分でSkillを作ることはしても、他の人が作ったSkillを導入したことは無かった。

今回はこの辺りを、試してみたい。

参考

gh コマンド で導入

とりあえず Github Copilot に近しい分野からのスキル導入からやってみる。

gh skill コマンドが増えていた。
これは、v 2.90.0以上が必要とのことで、更新した。(もともと2.5.2だった。)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
$ gh skill
Install and manage agent skills from GitHub repositories.

Working with agent skills in the GitHub CLI is in preview and
subject to change without notice.


USAGE
gh skill <command> [flags]

ALIASES
gh skills

AVAILABLE COMMANDS
install: Install agent skills from a GitHub repository (preview)
preview: Preview a skill from a GitHub repository (preview)
publish: Validate and publish skills to a GitHub repository (preview)
search: Search for skills across GitHub (preview)
update: Update installed skills to their latest versions (preview)

INHERITED FLAGS
--help Show help for command

EXAMPLES
# Search for skills
$ gh skill search terraform

# Install a skill
$ gh skill install github/awesome-copilot documentation-writer

# Preview a skill before installing
$ gh skill preview github/awesome-copilot documentation-writer

# Update all installed skills
$ gh skill update --all

# Validate skills for publishing
$ gh skill publish --dry-run

LEARN MORE
Use `gh <command> <subcommand> --help` for more information about a command.
Read the manual at https://cli.github.com/manual
Learn about exit codes using `gh help exit-codes`
Learn about accessibility experiences using `gh help accessibility`

試しにTDDについてのスキルを探して入れてみる。

1
2
3
4
5
6
7
8
9
10
$ gh skill search tdd
✓ Showing 3 skills matching "tdd"

? Select skills to install: [Use arrows to move, space to select, <right> to all, <left> to none, type to filter]
> [ ] tdd mattpocock/skills ★ 20.9k
Test-driven development with red-green-refactor loop. Use when user wants to build features or fix bugs us...
[ ] tdd glebis/claude-skills ★ 128
This skill should be used when the user wants to implement features or fix bugs using test-driven developm...
[ ] tdd ethanolivertroy/public-skills ★ 10
Test-driven development with red-green-refactor loop. Use when user wants to build features or fix bugs us...

3つヒットしたが、今回は一番スターの多いものを入れてみる。
mattpocock/skills

1
2
3
4
5
6
7
8
9
Select skills to install: tdd  mattpocock/skills  ★ 20.9k
Test-driven development with red-green-refactor loop. Use when user wants to build features or fix bugs us...
? Select target agent: [Use arrows to move, type to filter]
> GitHub Copilot
Claude Code
Cursor
Codex
Gemini CLI
...

GitHub Copilot を選択。

1
2
3
? Installation scope:  [Use arrows to move, type to filter]
> Project: install in current repository (recommended)
Global: install in home directory (available everywhere)

プロジェクトに対して導入した。
入れた後のディレクトリ構成は次のようになる。

1
2
3
4
5
6
7
8
9
.agents
└── skills
└── tdd
├── deep-modules.md
├── interface-design.md
├── mocking.md
├── refactoring.md
├── SKILL.md
└── tests.md

どうやら正しく入ったらしい。
試しに、VSCode 上で今現在認識されているskillを教えて と聞いてみると、次のように返ってきた。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
今このセッションで認識している skill は 3 つです。

tdd
テスト駆動開発向けです。Red-Green-Refactor で進める機能追加やバグ修正、テスト先行の実装で使います。

get-search-view-results
VS Code の Search ビューに出ている検索結果を取得するための skill です。

agent-customization
.instructions.md、.prompt.md、.agent.md、SKILL.md、copilot-instructions.md、AGENTS.md など、Copilot/agent のカスタマイズ用ファイルを作成・修正・デバッグするときの skill です。

必要なら次に、

それぞれの skill の使い分け
今の依頼にどの skill を適用すべきか
新しい skill を追加する方法
のどれかをそのまま続けて説明します。

skillを使う

どうやら正しく認識されているらしい。
試しに、以下のプロンプトでCLIを作らせてみる。

1
2
3
4
- フィナボッチ数列の指定個数目を出力するCLIを作成。
- 使用ランタイムDeno
- 使用言語TypeScript
- 十分検討し実装すること

出力は以下の通り。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
要件は「Deno + TypeScriptで、指定個数目のフィボナッチ数を出力するCLI作成」と理解しました。まずワークスペースを確認し、必要なファイルを作成してから実行検証まで行います。

実装前に、最小構成で動くように現在のワークスペース内容を確認し、作成するファイル構成を確定します。

Read test508-cpilot-test

これからCLI本体と設定ファイルを追加します。大きな指数でも扱えるようにBigInt+高速倍加法で実装し、入力検証とヘルプ表示も入れます。

Edited
main.ts+118-1
続けて、実行しやすいようにDenoタスク定義を追加します。

Edited
deno.json+6-1
実装の妥当性確認として、ヘルプ表示と通常計算・ゼロ基準計算の3パターンを実行検証します。

計算ロジック差分も確認するため、境界値として0入力の挙動を追加で検証します。

最後にTypeScript上の問題が出ていないかを確認して、完成状態を確定します。

Checked workspace, no problems found

仕上げとして、読みやすさ向上のために不要な中間変数だけ小さく整理します。

Edited
main.ts+1-2
最終確認として、修正後の実行結果が変わらないことを1回だけ再検証します。

実装完了しました。Deno + TypeScriptで、指定した番号のフィボナッチ数を出力するCLIを作成し、実行検証まで済ませています。

作成ファイル:

main.ts
deno.json
実装内容:

位置指定でフィボナッチ数を出力するCLI
大きな値にも対応できるよう、BigInt + 高速倍加法で計算
引数検証を実装
ヘルプ表示を実装
1-basedをデフォルトにし、--zero-basedオプションも対応
使い方:

deno run main.ts 10
deno run main.ts --zero-based 10
deno task fib -- 10
deno run main.ts --help
確認結果:

deno run main.ts 10 → 55
deno run main.ts --zero-based 0 → 0
deno run main.ts 0 → バリデーションエラー終了(想定どおり)
エディタ上のエラー検査でも問題なし
必要なら次に、簡単な自動テスト(例: 既知の値 1,2,10,50 など)も追加できます。

GPT-5.3-Codex • 0.9x

最終的に main.ts と deno.json が作成されている。TDD的に先にテストを作るようなことはされなかった。
tddスキルは適用されていない様子。

今度は、明確にスキルを呼び出すように別セッション同様の指示をする。

1
2
3
4
5
/tdd
- フィナボッチ数列の指定個数目を出力するCLIを作成。
- 使用ランタイムDeno
- 使用言語TypeScript
- 十分検討し実装すること

このようにすると、実装にあたり質問をしてきた。

このあともう1つ質問してきて以下のような回答と共に実装が行われた。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
依頼内容を理解しました。まずTDDスキルの指示をファイルから読み込み、要件に沿った進め方を確定してから、Deno + TypeScriptのCLIを実装します。

Read skill tdd

TDD手順に従うため、先にCLIの公開インターフェースと優先テスト対象だけ短く確認します。合意後すぐにRED→GREENで実装に入ります。

Analyzing your answers...

Q: CLIの入出力は次の案で進めてよいですか?
A: この案でOK
Q: 最優先で担保したい挙動を選んでください。
A: 正常系: 小さいnで正しい値を返す, 異常系: n未指定・非整数・0以下をエラーにする
仕様合意を確認しました。次にワークスペースを確認して、テスト1件ずつRED→GREENでCLIを組み立てます。

これから最初のREDとして、CLIの正常系1件だけを検証する統合テストとDeno設定を追加します。まず失敗を確認してから最小実装に進みます。

最初のREDは確認できましたが、失敗理由は仕様ではなく権限不足でした。テストが本来の失敗を示すように、テスト実行権限だけ先に整えて再実行します。

次はGREENです。最小実装として、引数1つをn番目とみなしフィボナッチ値を1行出力するCLI本体を追加します。

1つ目のサイクルはGREENになりました。次のREDとして、引数未指定時にエラー終了する挙動をテスト追加して失敗を確認します。

REDを確認しました。ここで最小修正として入力検証を追加し、未指定時はエラーメッセージをstderrへ出して終了コード1にします。

次の挙動として、非整数と0以下の入力をエラーにするテストを追加します。既存の検証ロジックで通るかを確認し、必要なら追加修正します。

主要な異常系は通過しました。正常系の境界として n=1 を追加で固定し、仕様の抜けを埋めて最終リファクタに入ります。

最後にGREEN状態で軽くリファクタします。テスト重複をヘルパー化して可読性を上げ、再テストで回帰がないことを確認します。

実装と検証は完了し、テストも全通過しています。最後に変更ファイルを確認して、要点を短くまとめて共有します。

Deno + TypeScriptで、指定したn番目のフィボナッチ数を出力するCLIをTDDで実装しました。
合意した仕様どおり、正常系と異常系を統合テストで先に確認し、RED→GREEN→Refactorで進めています。

変更点

Denoタスク設定を追加: deno.json
CLI本体を実装: main.ts
CLIの振る舞いテストを追加: fib_cli_test.ts
実装した振る舞い

正常系
n=7 で 13 を出力
n=1 で 1 を出力
異常系
n未指定でエラー終了
非整数入力でエラー終了
0以下入力でエラー終了
エラー時は stderr に n must be a positive integer. を出力し、終了コード1
検証結果

deno task test を実行し、5件すべて成功
変更ファイルの型・構文エラーなし
実行方法

単発実行: deno task fib 7
テスト実行: deno task test
次に必要なら対応できます

ヘルプ表示(-h, --help)の追加
BigInt対応(大きいn向け)
CI用ワークフロー追加(GitHub Actionsなど)

実装としても、src/main.ts と src/fib_cli_test.ts が作成されている。
tddスキルが適用されている様子。
少なくともテストコードも作られている。(順番だけ見るとアプリ実装が先だったようではある。)

skillを勝手に使わせる

先の確認では明示的に/tdd と指示して初めて活用された。
もちろん、現在読み込まれているskillを確認すれば答えるが、明確な呼び出しが要求されている。
これは全体のCopilotのAgentに対して、使うように指示をしていないというのがある。

今度は、以下のプロンプトにより、copilot-instructions.mdを作らせる。

1
2
3
copilot-instructions.md を作成
記述内容
スキル tdd を活用した開発をさせる
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
依頼内容は、ワークスペースに copilot-instructions.md を新規作成し、tdd スキルを活用するよう指示を記述することだと理解しました。まず既存ファイル有無を確認してから、最小構成で作成します。

作成方針を正確に合わせるため、まずカスタム指示ファイル作成のワークフロー定義(関連スキル)を読み込みます。

Read skill agent-customization

次に、配置先を誤らないようワークスペース内に既存の copilot-instructions.md がないか確認し、なければ推奨位置に新規作成します。

既存ファイルがないことを確認しました。.github/copilot-instructions.md を作成し、開発時に tdd スキルを優先利用する指示を明記します。

Created copilot-instructions.md

最後に内容を確認して、要求どおり tdd 活用方針が入っているかチェックします。

Read copilot-instructions.md

作成完了です。
tdd スキルを活用する指示を .github/copilot-instructions.md に追加しました。

記述内容は以下を含みます。

機能開発・バグ修正で tdd スキルを使う
red-green-refactor の厳守
小さな増分で進める
テスト実行不可時の理由と実行コマンド提示
GPT-5.3-Codex • 0.9x

で、以下が作成された copilot-instructions.md の内容です。

.github/copilot-instructions.md
1
2
3
4
5
6
7
8
9
# Copilot Instructions

- Use the `tdd` skill for feature development and bug fixes.
- Follow a strict red-green-refactor cycle:
1. Write a failing test first.
2. Implement the minimum code to pass the test.
3. Refactor while keeping tests green.
- Prefer small, incremental commits aligned to each TDD cycle.
- If tests cannot be executed locally, explain why and provide the exact test command to run.

若干skillに書いてあることと同じことを繰り返し書いているように見えて、skillが正しく読まれ使われているのか判断が難しいので修正させる。

1
所謂TDD開発自体についての記述はskill に移譲させてください。
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
意図は把握しました。copilot-instructions.md からTDD手順の詳細を外し、tdd スキル参照だけに寄せる形へ最小修正します。

Read skill agent-customization

これから対象ファイルの文言を、TDD手順の説明を含まない「スキル委譲」表現に差し替えます。

Edited
copilot-instructions.md+2-6
反映後の内容を確認し、要望どおり「TDD詳細はskillへ移譲」の状態になっているかチェックします。

Read copilot-instructions.md

要望どおり修正しました。
TDDそのものの手順説明は instructions から外し、skill へ委譲する形に変更済みです。

変更ファイル:
copilot-instructions.md

現在の内容:

Use the tdd skill for feature development and bug fixes.
Delegate all TDD workflow details to the tdd skill.
Do not restate TDD process details in this instruction file.
GPT-5.3-Codex • 0.9x

.github/copilot-instructions.md の内容は、tddスキルへの参照になっている。

.github/copilot-instructions.md(改)
1
2
3
4
5
# Copilot Instructions

- Use the `tdd` skill for feature development and bug fixes.
- Delegate all TDD workflow details to the `tdd` skill.
- Do not restate TDD process details in this instruction file.

ここで改めて以下のプロンプトで、実装させる。

1
2
3
4
- フィナボッチ数列の指定個数目を出力するCLIを作成。
- 使用ランタイムDeno
- 使用言語TypeScript
- 十分検討し実装すること
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
計画:

fibonacci.ts — コアロジック (fibonacci(n: bigint): bigint)
main.ts — CLIエントリポイント
fibonacci_test.ts — テスト
BigIntを使い大きな値にも対応。反復実装(再帰はスタックオーバーフローを回避)。

Starting: Tracer bullet: fib(0) test → impl (2/8)

RED — トレーサーバレット(fib(0) テスト):

GREEN — fibonacci(0) を実装:

Created fibonacci.ts

1/1 tests passed (100%)

Starting: fib(1) base case test → impl (3/8)

RED — fib(1) テスト:

Edited
fibonacci_test.ts+4-0
0/1 tests passed (0.00%)

GREEN — fib(1) を実装:

Edited
fibonacci.ts+1-0
2/2 tests passed (100%)

Starting: fib(n) general case test → impl (4/8)

RED — fib(10) 一般ケーステスト:

Edited
fibonacci_test.ts+4-0
0/1 tests passed (0.00%)

GREEN — 反復実装:

Edited
fibonacci.ts+6-1
3/3 tests passed (100%)

Starting: Large number (BigInt) test → impl (5/8)

RED — 大きな数のテスト (fib(100)):

Edited
fibonacci_test.ts+4-0
4/4 tests passed (100%)

既にGREEN。BigInt反復実装は大きな数に対応済み。

RED — エラーハンドリング(負の数):

0/1 tests passed (0.00%)

GREEN — バリデーション追加:

Edited
fibonacci.ts+1-0
5/5 tests passed (100%)

Starting: CLI entry point (main.ts) (7/8)

CLI エントリポイント作成:

全テスト実行 & 動作確認:

5/5 tests passed (100%)

全テストGREEN、CLI動作確認完了。

完成したファイル構成
fibonacci.ts — コアロジック
main.ts — CLIエントリポイント
fibonacci_test.ts — テスト (5件)
設計のポイント
項目 内容
BigInt使用 任意精度整数。fib(100) = 354224848179261915075 も正確に計算
反復実装 再帰ではなくループでO(n)計算、スタックオーバーフロー回避
入力バリデーション 正規表現で非負整数かチェック → BigInt() に渡す
数列定義 0-indexed: F(0)=0, F(1)=1, F(2)=1, ... F(10)=55
使い方

deno run main.ts 10 # → 55
deno run main.ts 100 # → 354224848179261915075
deno run main.ts 0 # → 0

Claude Sonnet 4.6 • 0.9x

正しくtddスキルが呼び出され、TDD的な文脈で実装が想定通りされた。

ただ実装物の様子は、先の明示的に呼び出したときと異なっている。
実装としては、ディレクトリを掘らずに、main.ts fibonacci.ts fibonacci_test.ts が作成された。

claude 向けっぽい文脈のSkillを入れてみる

claude 向けっぽい文脈のSkillを入れてみる と書いたが、調べてみると必ずしもそういった定義はされていなかった。
README.md を見ると、GitHub Copilot CLI 向けの記述もある。

obra/superpowers

以下指示通り導入した。

1
2
3
4
5
$ copilot plugin marketplace add obra/superpowers-marketplace
Marketplace "superpowers-marketplace" added successfully.

$ copilot plugin install superpowers@superpowers-marketplace
Plugin "superpowers" installed successfully. Installed 14 skills.

プロンプトで確認をしたものの、今現在認識されている skill を聞いてみると認識されているように見えない。

しかし、コマンドの通りcopilotCLIに入って開いてみると、/superpowers:~~~ のコマンドが入っているので認識はされている様子。

VS Code上では、/superpowers:~~~ が認識されていない様子。
Copilot CLIで呼び出したコマンドがSESSIONSでもあがるので誤解をしていた。

ごにょごにょして使わせると、以下のようなsuperpowers的なものが作成されたが、superpowersそのものではない。

.agents\skills\superpowers\SKILL.md
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
---
name: superpowers
description: "Software development workflow skill. Use when brainstorming an approach, writing a plan, executing a plan, reviewing code, debugging failures, or verifying that work is actually complete."
---
# Superpowers

## Purpose

Use this skill to apply a disciplined software development workflow in this workspace.

This local version captures the parts of Superpowers that are useful in VS Code Copilot Chat here:

- clarify the goal before coding
- write a small plan before larger changes
- execute work in small verified steps
- review for bugs and regressions before declaring success
- verify completion with real checks

Keep feature development and bug fixes aligned with the `tdd` skill.

## When To Use It

Use this skill when the task involves any of the following:

- brainstorming requirements or design options
- turning an approved idea into an implementation plan
- carrying out a multi-step change
- reviewing code for correctness and regression risk
- debugging a failing behavior or test
- checking whether work is actually complete

## Workflow

### 1. Clarify

Before editing, identify the concrete anchor:

- failing test
- target file
- target symbol
- broken command
- visible user-facing behavior

State one local hypothesis and one cheap check that could disconfirm it.

### 2. Plan

For non-trivial work, write a short step-by-step plan before broad changes.

The plan should:

- stay close to the controlling code path
- prefer the smallest testable change
- name the validation to run after the first substantive edit

### 3. Execute

Make small, local changes.

Rules:

- prefer root-cause fixes over cosmetic patches
- avoid widening scope without evidence
- after the first substantive edit, run the narrowest useful validation immediately
- if validation fails, repair the same slice before expanding scope

### 4. Review

Before finishing, check for:

- behavioral regressions
- missing validation
- mismatches between the request and what was changed
- hidden assumptions that should be stated clearly

When asked for a review, findings come first and should be ordered by severity.

### 5. Verify Completion

Do not declare success without at least one concrete verification step when the environment permits it.

Prefer this order:

1. focused behavior check
2. narrow test
3. narrow typecheck or lint
4. diff review only if nothing executable exists

## Interaction Notes

- Keep communication concise and concrete.
- Explain tradeoffs only when they affect the result.
- Prefer direct implementation over long proposals unless the user is explicitly asking for design discussion.
.github\copilot-instructions.md
1
2
3
4
5
6
7
# Copilot Instructions

- Use the `superpowers` skill for brainstorming, planning, multi-step execution, code review, debugging, and completion verification.
- Use the `tdd` skill for feature development and bug fixes.
- Keep `tdd` focused on the red-green-refactor loop and use `superpowers` for the broader workflow around it.
- Delegate all TDD workflow details to the `tdd` skill.
- Do not restate TDD process details in this instruction file.

現時点で、superpowers をCopilotChatで呼び出すことは、失敗している。

npx skills で導入

npx skillscoreyhaines31/marketingskills を入れてみる。

これは、【保存版】Claude Skills & GitHub リポジトリ 厳選50選 — AI開発に本当に使えるツールだけまとめた で紹介されていたもの。

1
npx skills add https://github.com/vercel-labs/agent-skills

途中で、どのエージェントに入れるか聞いてくるので、GitHub Copilot を選びたいところだが、選択肢にない。
仕方ないので、Clause Code を選択して入れる。
すると、.agents/skills 以下に作成される。プロンプトから認識されているか聞いてみると、認識されている様子。

まとめ

状況をまとめるにおそらく、現状はこうなる。

gh skill で紹介されるもの /plugin で紹介されるもの npx skills で紹介されるもの
Copilot Chat O X O
Copilot CLI O O O

以上、Copilot で「主立ってClaude Code向けに紹介されるSkill」は使えるのか確認した。
ちょっと微妙なところを感じましたが、一カ月も立てば状況もかわっているのも知れません。
Agentic AI Foundation(AAIF)なんかで、この辺りも標準化してくれるのが望ましいと感じられます。
pluginの使用はCopilot CLIでは動く予感を感じましたが、結局実体がどこにあるのかわからないという状況もあるので、これも必要そうです。
また、npx skills で導入の際に、選択肢が非常に多く標準化がこれから進む分野になっていることを感じます。

では。