Guide · Agent skill
How to use the Git Commit with Conventional Commits agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The git-commit skill is a set of instructions that teaches an AI agent how to make Git commits in the Conventional Commits format, such as 'feat(auth): add token refresh'. It is for developers who want consistent commit messages without writing each one by hand, and for teams that need a large set of changes split into logical commits. After reading this guide you will know how to install the skill, how to ask for a commit in plain language, and which safety limits the skill sets for the agent.
Git Commit with Conventional Commitsgithub/awesome-copilot · MIT Download the skill Quick start
- 1Download and unzip the skill. Download the skill ZIP from this page and unzip it so the folder contains SKILL.md.
- 2Place the folder. Move the git-commit folder into ~/.claude/skills/ for personal use, or into .claude/skills/ inside your project for team use. In the Claude apps, upload the ZIP under Settings, then Capabilities, then Skills.
- 3Make sure Git and shell access are ready. Confirm that Git is installed and that your agent is allowed to run shell commands for git.
- 4Ask for a commit in plain words. Tell the agent to commit your changes, split them into commits, or type /commit.
- 5Review the result. Check the new commit with git log -1 --stat and confirm that no secret files were included.
What the git-commit skill makes the agent do
An agent skill is a folder with a SKILL.md file that tells an AI agent how to handle one kind of task. This skill covers making Git commits. When you ask for a commit, the agent reads your repository's commit rules and recent messages, examines the diff, groups related files, writes a Conventional Commits message, and runs git commit itself.
The skill uses the Conventional Commits format, which starts with a type, an optional scope and a short summary. The types it lists include feat, fix, docs, style, refactor, perf, test, build, ci, chore and revert. Breaking changes are marked with an exclamation mark before the colon or with a BREAKING CHANGE footer.
- Reads commitlint configuration, CONTRIBUTING.md and recent messages before writing
- Stages files in logical groups so one commit holds one change
- Writes the message in the imperative mood, for example 'add' not 'added'
- Runs git commit and then checks the result with git log and git status
Install the git-commit skill
Download the skill ZIP from the skill's page in this directory and unzip it. The skill folder must contain SKILL.md. Then place the folder in one of two locations. Use ~/.claude/skills/ to make the skill available in all your projects. Use the .claude/skills/ folder inside a single project to share it with that project only.
If you use the Claude apps instead of a local folder, upload the ZIP under Settings, then Capabilities, then Skills.
- Personal install: ~/.claude/skills/git-commit/
- Project install: .claude/skills/git-commit/ inside your repository
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Trigger the skill with natural requests
You do not need a special command. The skill is meant to be used when you ask for a commit, a split into several commits, or a conventional commit. You can also type /commit, which the skill description names as a trigger.
The skill is not meant for requests that only want a message. If you want the message drafted without a commit, say so clearly, because the skill says not to use it in that case.
- 'Commit my changes with a conventional commit message.'
- 'Split these changes into three logical commits.'
- '/commit'
- 'Write a commit message for the auth changes but do not commit yet.' (this request is outside the skill)
Example workflow
Suppose you have changed the login code and also fixed a parser bug. You ask the agent to commit. The skill checks git status and the diff, unstages files that belong to other changes, and stages each group separately. It then runs one git commit per group.
For a one-line message the agent runs a command like the one below. For a message with a body and a footer, each paragraph is passed with its own -m option so Git adds the blank lines.
- Check the result with git log -1 --stat after each commit
- Long messages can be written to a temporary file and committed with git commit -F
git commit -m "feat(auth): add token refresh"
git commit -m "fix(parser): accept empty input" -m "An empty file caused a crash." -m "Closes: #123"What the skill needs and how it runs
The skill requires Git and an agent that is allowed to run shell commands. Its allowed tool is limited to git commands. Only the git add --resolved option needs a newer Git release, so the skill's compatibility note should be checked if you use it. The skill does not use the network.
The skill file lists SKILL.md as its only file. It runs git commands through the agent's shell rather than bundled scripts, so you can read every command it will run in the SKILL.md text before you install it.
- Requires: Git and shell access for the agent
- Network: not used according to the skill metadata
- Licence: MIT
Safety rules built into the skill
The skill sets clear limits. It does not push, open pull requests, rebase, or change commit history unless you explicitly ask. It does not use --no-verify unless you tell it to, so Git hooks still run. If a hook fails, the agent fixes the problem and runs the commit again instead of using --amend.
The skill also tells the agent not to commit secret files such as .env files, credentials.json or private keys, and not to change the Git configuration. Repository rules override the skill's own rules.
- No pushing, and no force push to the default branch
- No git reset --hard, git rebase or the experimental git history command
- No git add -p or git add -i, because these wait for human input
- Do not commit empty changes; the agent reports that there is nothing to commit
Tips for better results
Keep the repository's commit rules in place, such as a commitlint configuration or CONTRIBUTING.md, because the skill reads them first. If your team uses a different list of types, the skill says to use the repository's list. Review the first few commits the agent makes, since the skill notes that repository conventions should be checked when the agent may misread them.
If you want changes split into several commits, name the groups you expect. This helps the agent stage the right files together.
- Review git log -1 --stat after each commit
- Use a project install when a team shares one set of conventions
- Say 'do not commit' if you only want a message
What people use it for
- Turning a day of mixed edits into several clean commits with type and scope
- Keeping a team's history consistent with feat, fix and docs prefixes
- Marking a breaking API change with an exclamation mark and a BREAKING CHANGE footer
- Committing a conflict resolution after the files are fixed, using git add --resolved
Frequently asked questions
How do I install the git-commit agent skill?
Download the skill ZIP from this page and unzip it. Place the git-commit folder in ~/.claude/skills/ for all your projects, or in .claude/skills/ inside one project. In the Claude apps, upload the ZIP under Settings, then Capabilities, then Skills.
What does the git-commit skill require?
It requires Git and an agent that is permitted to run shell commands for git. Only the git add --resolved option needs a newer Git release. The skill metadata says it does not use the network.
Is the git-commit skill free?
The skill is licensed under the MIT licence, as stated in its metadata. This page does not state any price, so check the directory listing for any other terms.
Does the skill push my code or open pull requests?
No. The skill says not to push, open a pull request, rebase or change commit history unless you ask for it. It also does not force push to the default branch.
Can I use this skill if I only want a commit message?
No. The skill says not to use it when you want only a commit message and no commit. Ask the agent for the message in a normal chat instead.
What if my repository uses different commit rules?
The skill reads commitlint configuration, CONTRIBUTING.md and recent messages first, and says the repository rule overrides the skill. Check the first few commits the agent makes to confirm it followed your rules.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
This skill suits developers and teams who want Conventional Commits messages written and committed by an agent, with clear limits on pushing and history changes. It is less suitable if you only want drafted messages, or if you need an agent that pushes, opens pull requests or rebases, since the skill excludes those actions.
Download the skill