How to use the Git Commit Helper agent skill (ce-commit)

What the Git Commit Helper skill makes the agent do

When you ask an agent to commit your work, this skill sets out a careful sequence. The agent first checks the repository state, including the current branch, recent commit history and uncommitted changes. If it is on a detached HEAD or on the default branch, it creates a feature branch before committing.

The agent matches the commit message style your project already uses. If there is no clear convention, it looks at recent log messages, and if that gives no pattern it uses conventional commits in the form type(scope): description. The subject line should be imperative and describe what is now possible or fixed, not list the files changed.

The agent stages only the named files it is committing, never everything at once with git add -A or git add. If the changed files split into distinct concerns, it can make two or three separate commits. The skill makes local commits only. It does not push or open pull requests.

• Checks git status, the diff against HEAD, the current branch and the last 10 commits
• Creates a feature branch when committing from main, master, the default branch or a detached HEAD
• Uses the message style of your project, then the recent log pattern, then conventional commits
• Stages named files only and commits with an explicit file list
• Reports the commit hash and subject when finished

Install the ce-commit skill

Download the ZIP for this skill from this page and unzip it. The skill is a folder that contains a SKILL.md file. Put that folder in one of these locations, depending on how you want to use it.

For all of your projects, place it in your personal skills folder at ~/.claude/skills/. To use it only in one repository, place it in that project's .claude/skills/ folder. In the Claude apps you can also upload the skill under Settings → Capabilities → Skills instead of copying files by hand.

• Personal (all projects): ~/.claude/skills/
• Single project: the .claude/skills/ folder inside the repository
• Claude apps: Settings → Capabilities → Skills, then upload

Check what you need before you start

The skill needs git installed and available on your command line. It also needs a git repository that has changes to commit. If the folder is not a git repository, the skill stops.

The GitHub CLI (gh) is optional. The skill uses it only as a fallback to find the default branch name when the remote default cannot be read from origin/HEAD.

• git installed and available on the command line
• A git repository with changes to commit
• GitHub CLI (gh): optional, used only as a fallback to find the default branch name
• The skill does not run scripts and does not use the network

How to trigger the skill with natural requests

You do not need a special command. Ask the agent in plain language to commit or save your changes. The skill is written to activate when you ask to commit or save staged or unstaged changes with a repository-appropriate message.

Phrases that match the skill's purpose include asking the agent to commit your changes, save your staged work, or commit what you have changed. You can add context such as a message style you want, but the skill's own rules still apply unless you override them.

• "Commit my changes."
• "Save what I've staged with a good commit message."
• "Commit the login fix in this repo using our usual message style."
• "Make a commit for the unstaged changes in the API folder."

What happens during a commit: an example flow

Suppose you ask the agent to commit a bug fix while you are on main. The skill tells the agent to create a feature branch named from the change content, then re-check the branch. It then stages only the specific files for the fix and writes the message to a file outside the repository before committing with that file and an explicit file list.

If your staged set already contains files you did not intend to commit, the explicit file list keeps them out of this commit. The agent reports the hash and subject so you can confirm the result.

• Subject example from the skill: Fix double-submit on checkout
• Subject example from the skill: Add per-subscription mute (U3), where U3 is a plan unit ID when one is in hand
• If the caller specifies exclude paths, those files stay uncommitted and the report says so

Safety notes: scripts, network and what gets committed

The skill runs no scripts and uses no network access, according to the source. Its work is local git commands. Because it creates branches automatically and commits without asking, review the result after it finishes.

The skill never pushes and never opens a pull request. For that full ship flow, the source points to the companion ce-commit-push-pr skill. If someone else's work is ported into your commit, the skill adds Co-authored-by trailers only when the identity is recorded in the source commits or patch, and never guesses an email address.

• No scripts run and no network is used
• Local commits only: no push, no pull request
• Creates a feature branch without asking when committing from main or detached HEAD
• Never uses git add -A or git add .
• Adds Co-authored-by trailers only for identities recorded in the source

Tips for better results

Keep your staged set intentional. The skill stages named files and commits only those, but anything already staged may still matter to your review, so check git status before you ask. If you want a particular message style, say so in your request. If you want to review each message before it is written, this skill is not the right fit, because it tells the agent not to ask before branching.

• Stage or leave changes as you intend and check git status first
• State your preferred message convention in the request if it differs from the repo log
• Use ce-commit-push-pr when you also want to push and open a pull request
• Review the new branch name and commit subject after the agent reports back