Skill commit-message
✓ VerifiedMenerapkan konvensi pesan commit Git yang terstandarisasi, termasuk struktur, format, scoping, dan metadata trailers.
- Bintang GitHub
- 23
- File
- 1
- Commit terakhir
- 11 Nov 2026
- Lisensi
- MIT
- Tool yang diizinkan
- —
Apa fungsi skill commit-message?
Menerapkan konvensi pesan commit Git yang terstandarisasi, termasuk struktur, format, scoping, dan metadata trailers.
Ringkasan dibuat dari SKILL.md skill ini.
Tentang skill commit-message
Menjaga riwayat commit Git yang terstandarisasi adalah fungsi utama dari skill ini bagi developer maupun automated agent. Skill ini menyediakan framework yang ketat untuk menulis, meninjau, memvalidasi, dan mengubah pesan commit guna memastikan pesan tersebut berfungsi sebagai catatan perubahan yang faktual. Dengan menerapkan struktur spesifik, skill ini memastikan setiap pesan mengomunikasikan alasan di balik modifikasi kode secara jelas tanpa mengulang apa yang sudah terlihat pada diff.
Seorang agent sebaiknya memuat skill ini setiap kali pesan commit dibuat atau diubah. Skill ini menghasilkan teks terstruktur yang terdiri dari scoped subject line, body yang deskriptif, dan metadata trailers. Skill ini menerapkan batasan teknis seperti limit subject 50 karakter, line wrapping 72 karakter untuk body, dan penggunaan simbol ASCII secara eksklusif untuk mencegah kerusakan pada environment. Skill ini juga mengelola persyaratan kompleks seperti deklarasi BREAKING: untuk backward compatibility serta format trailer khusus untuk pelacakan issue dan authorship.
Apa yang bisa dilakukan agent dengannya
- 1Generate a standardized commit messageMenulis pesan commit yang mengikuti struktur wajib berupa subject, body, dan trailers berdasarkan diff yang diberikan.
- 2Format breaking change declarationsMemastikan bahwa commit yang memperkenalkan breaking change menyertakan prefix BREAKING: yang wajib serta panduan migrasi.
Ditulis dari SKILL.md milik skill tersebut. Setiap use case diverifikasi dengan kutipan darinya.
Cara install skill commit-message
Setiap perintah menyalin folder skill ke direktori skills client. Restart client atau mulai sesi baru untuk memuatnya.Install skill commit-message di Plugin
/plugin marketplace add xobotyi/cc-foundry /plugin install git-commit@foundry
Install skill commit-message di Claude Code
~/.claude/skills/commit-message/SKILL.md
$ npx degit xobotyi/cc-foundry/plugins/git-commit/skills/commit-message ~/.claude/skills/commit-messageInstall skill commit-message di Claude Code (project)
.claude/skills/commit-message/SKILL.md
$ npx degit xobotyi/cc-foundry/plugins/git-commit/skills/commit-message .claude/skills/commit-messageInstall skill commit-message di Codex
~/.agents/skills/commit-message/SKILL.md
$ npx degit xobotyi/cc-foundry/plugins/git-commit/skills/commit-message ~/.agents/skills/commit-messageInstall skill commit-message di Gemini CLI
~/.gemini/skills/commit-message/SKILL.md
$ npx degit xobotyi/cc-foundry/plugins/git-commit/skills/commit-message ~/.gemini/skills/commit-messageInstall skill commit-message di Claude.ai
npx degit xobotyi/cc-foundry/plugins/git-commit/skills/commit-message commit-message && zip -r commit-message.zip commit-message # claude.ai → Customize → Skills → + → Create skill → Upload a skill → commit-message.zip
Pengecekan otomatis
Berjalan 16 Nov 2026Laporan lengkap →Lolos cek otomatis kami. Ini bukan audit keamanan: baca skripnya sebelum membiarkan agent menjalankannya. Cara kerja checks · Yang perlu diperbaiki untuk menambah poin
- Source repository terhubung dan dapat dijangkauxobotyi/cc-foundry/plugins/git-commit/skills/commit-message is public and reachablegithub.com/xobotyi/cc-foundry/tree/HEAD/plugins/git-commit/s ↗Lolos
- Lisensi open source (OSI)MITspdx.org/licenses/MIT.html ↗Lolos
- Ada commit atau release dalam satu tahun terakhirLast commit 2026-11-11github.com/xobotyi/cc-foundry/commits ↗Lolos
- Skrip tanpa unduh-lalu-jalankan atau pencurian kredensialNo scripts in the skill folderLolos
- Tidak ada instruksi tersembunyi di SKILL.mdNo override phrases, safeguard switches, credential requests or hidden charactersgithub.com/xobotyi/cc-foundry/blob/HEAD/plugins/git-commit/s ↗Lolos
- Tool yang diizinkan otomatis dibatasiNo pre-approved tools; the agent asks before using onegithub.com/xobotyi/cc-foundry/blob/HEAD/plugins/git-commit/s ↗Lolos
- SKILL.md mengikuti format Agent Skillsthe description does not say when to use the skillgithub.com/xobotyi/cc-foundry/blob/HEAD/plugins/git-commit/s ↗Temuan
SKILL.md
Lihat di GitHubCommit Message Format
A commit message is a factual record of one change. It tells the reader why the change happened. The reader has the diff and the docs. The message adds only what the diff and the docs cannot show.
What Makes a Good Message
Structure
[scope] subject
body
trailers
- Subject — what changed.
- Body — the reason for the change, and the procedure that verifies it.
- Trailers — structured metadata such as references and authorship.
ASCII Symbols
- No em dash and no en dash (—, –). Use
--or-. - No arrows (→, ←, ↔). Use
->,<-, and<->. - No bullet symbols (•, ▸). Use
-or*. - No fancy quotes (“”, ‘’). Use straight quotes.
- No emoji (🐛, ✅, 🚀), unless the project conventions make them necessary.
- No other decorative symbols (©, ™, §, ¶, ∞).
Some environments break these symbols, and the symbols add no information.
Subject Line
- With a scope, for a monorepo:
[scope] <verb> <description> - Without a scope, for a single-purpose repo:
<verb> <description>
Scope (Optional)
[parser]— a top-level project[core/auth]— a nested path[web/api]— a component in a project
Derive the scope from the file paths, not from the file contents. Use the common parent directory, or the most
significant component that you changed. Use one spelling for one component. [myapp/backend] and [myapp/b] in the
same log break the filtering that the scope exists for.
A path can hold a version directory such as v2. Omit the segment while one version is alive: write [core/engine],
because the reader has one candidate. Write [core/engine/v1] and [core/engine/v2] when both versions are alive,
because the segment is then the distinction that matters.
Body
- Why the change was necessary.
- What the subject cannot hold, such as a behavior change that surprises the reader.
- How a reader verifies it later, when that is not obvious. Give the steps that the reader repeats
(
run X, confirm Y). Never give the results that you produced in this session.
Wrap each body line at 72 characters. This is a hard limit. Git tools such as log, format-patch, and email expect
72-character body lines. Trailers are the only exception.
Shape
Each paragraph after the first earns its lines, or it goes. A paragraph earns its lines when it records a fact that the diff and the docs cannot show:
- A second reason that the first paragraph does not carry.
- A
BREAKING:declaration, or a step of the migration path. - A behavior change that the subject cannot predict, such as the new meaning of an absent value.
- A one-line note about the work that follows in the chain. Name that work. Do not write "more changes coming".
The reason itself depends on the kind of change:
- A bug fix — why the defect existed, not only what you corrected.
- A feature — the use case.
- A refactor — the motivation.
- A breaking change — the migration path.
The length follows the number of reasons, not the size of the diff. A refactor of 2000 lines with one reason gets one paragraph. A migration with 6 steps gets the 6 steps. A body of 20 lines is correct only when 20 lines of reasons, declarations, and steps exist. Such a body usually holds a walkthrough, or a second unit that belongs in its own commit.
More than one behavior change. One unit can change several behaviors. Give one line to each change, in one list. Do not give a paragraph to each change:
- <behavior that the caller sees, present tense>.
- <behavior that the caller sees, present tense>
<continuation of the same item, indented two spaces>.
Mark each item with -. A wrapped line then keeps its item visible in git log. Each line gives the behavior that the
caller sees, without the old behavior as a contrast. A line that names a file, a function, or a test inventories the
diff. Delete such a line.
The list does not license a longer body. Several behavior changes under one reason stay in one commit, and the reason keeps its paragraph. Two independent reasons are two commits, whatever the number of behavior changes.
Record, Not Documentation
These signs show that a body became documentation:
- A paragraph for each decision inside the artifact. That is the design doc of the artifact.
- A walkthrough of the new behavior. A paragraph that names a function, a call order, an empty case, a fallback, or a flag that gates the new code reports the procedure, and the diff shows the procedure.
- A list of the surfaces, the files, or the tests that the change touches. The diff already lists them.
- A repeat of what the updated docs say.
Two tests. Apply both to each paragraph.
- Paste the paragraph into the repo's docs. Does it stay true and useful there? Then it belongs in the repo's docs, not in the message.
- Name the item from the earns-its-lines list above that the paragraph carries. A paragraph that matches no item is a walkthrough. Delete it.
Terse Register
The diff carries the what. The message carries the why. A body can use fragments, and it can drop articles where the meaning stays clear. Identifiers, file paths, and error strings stay exact.
The tense carries the timeline. State the current behavior in the plain present tense, because the commit itself
marks the "now". The past tense belongs to the cause of the change, and to nothing else:
Empty input dereferenced nil. Returns empty token list instead.
Never include:
- "This commit does X", "This change...". The diff says what.
- "I", "we". The commit speaks for itself.
- "now", "currently", "previously". The tense already carries the timeline.
- The old behavior as a contrast: "X, where Y used to", "no longer does Y", "changed from Y to X". The diff holds the
previous code. Name the old behavior only as the cause of a fix, or as the thing that a
BREAKING:change removes. - "As requested by...". Use a
Co-Authored-By:trailer, or write nothing. - The scope repeated:
[parser] update parser codebecomes[parser] handle empty input. - Promotional adjectives without data: "great", "amazing", "improved", "better". Say what is better.
- Filler: "just", "really", "basically", "actually", "simply". Connective fluff: "however", "furthermore", "additionally".
- Session artifacts: test counts, lint status, typecheck status, build status, quality-gate status, CI results, and verification stories. Examples: "55 of 55 tests passing", "all checks pass", "verified by building X". These record your session at commit time, not the change. A future reader cannot verify them, and the next commit makes them stale. Steps that a reader repeats are correct in the body. A log of the checks that you ran is not.
Breaking Changes
BREAKING: <what is broken>
<explanation and migration path>
Write the BREAKING: prefix in capital letters. Give a short description of what breaks after it. The paragraphs that
follow explain the reason and the migration path. Write a migration path of more than one step as a numbered list under
a Migration: line.
Trailers (Footer Metadata)
Trailers are structured key-value pairs at the end of the commit message. They use the git-trailer format.
Task:— a link to the issue tracker —Task: https://tracker.example.com/PROJ-123Fixes:— the issue that this commit corrects —Fixes: #456Refs:— related issues or commits —Refs: #123, #124Closes:— the issue that closes at merge time —Closes: #789See:— related documentation —See: docs/auth.mdReviewed-By:— the code reviewer —Reviewed-By: Alice <alice@example.com>Co-Authored-By:— an additional author —Co-Authored-By: Bob <bob@example.com>
Write each trailer key in Title-Case. Keep each value on one line. For more than one value, use a comma-separated list, or repeat the trailer.
Amending Commits
An amended commit replaces the original commit and rewrites the history. The final message must describe what the commit introduces. It must not describe the sequence of revisions.
Common violation: the agent treats the amend as iterative work. It then describes the changes against the original commit: "also update the tests", "fix the edge case from the previous version", "address review feedback". The reader has no access to the original commit, and no context for "also" or "previous".
Correct approach: read the full staged diff after the amend. Then write a message that describes the complete change as one introduction. Ignore the original message and start again.
No AI Attribution
- No "Generated with Claude Code" or a similar sentence
- No "Co-Authored-By: Claude" and no other AI co-author trailer
- No emoji that marks AI generation (🤖, and similar)
- No links to claude.com, anthropic.com, or AI tools
Commits must look like regular developer commits.
- Record, not documentation — the body carries the reason for the change. Rationale, invariants, and behavior descriptions live in the artifact and travel in the diff
- Factual subjects — what changed, not how good it is
- Explain the cause — for a bug fix, say why the code was defective
- Body is essential — a message of one line is rarely acceptable. The default body is one paragraph
- No walkthrough — the body never narrates the new procedure: the call order, the empty case, the fallback, the flag that gates it. The artifact keeps such a fact
- Length follows the reasons — not the size of the diff. Each paragraph after the first earns its lines, or it goes. Two independent reasons usually mean two commits
- Terse register — the diff carries the what, the message carries the why. No "this commit", "I", or "we". The tense carries the timeline. No filler. No promotional adjectives
- No session artifacts — test counts, lint status, CI status, gate status, and verification stories record your session, not the change. "How to verify" means steps that the reader repeats
- BREAKING first — a breaking change starts the body with
BREAKING: - Amends rewrite history — an amended message describes the full change, not the difference
- ASCII symbols — no decorative Unicode. Use the ASCII equivalents
- No AI attribution — commits look like regular developer work
Berkas dalam skill ini
1Pertanyaan seputar skill commit-message
Bagaimana cara install skill commit-message di Claude Code?
Jalankan npx degit xobotyi/cc-foundry/plugins/git-commit/skills/commit-message ~/.claude/skills/commit-message. Claude Code memuat skill dari ~/.claude/skills (semua proyek) atau .claude/skills (satu proyek) dan menggunakan skill saat deskripsinya cocok dengan tugas.
Client mana saja yang bisa menggunakan skill commit-message?
Client apa pun yang membaca format Agent Skills: Claude Code, Codex, Gemini CLI, dan claude.ai (unggah sebagai ZIP). Script dalam skill hanya berjalan di client yang mengizinkan eksekusi kode.
Bagaimana cara menangani breaking changes dalam pesan commit?
Bagian body harus dimulai dengan deklarasi 'BREAKING:' menggunakan huruf kapital sebagai paragraf pertama, diikuti dengan penjelasan dan panduan migrasi.
Apa aturan untuk subject line?
Subject line harus maksimal 50 karakter (maksimal 72), menggunakan imperative mood, dimulai dengan huruf kecil setelah scope, dan tidak diakhiri dengan tanda titik.
Kapan saya harus menggunakan scope dalam pesan commit?
Gunakan scope jika repositori berisi beberapa proyek, package, atau subsystem, atau ketika pengguna perlu memfilter log berdasarkan komponen tertentu.