commit skill
✓ VerifiedGenerates Conventional Commits messages, performs commits, and optionally pushes changes to a remote repository.
- GitHub stars
- 16
- Files
- 2
- Last commit
- Nov 10, 2026
- License
- MIT
- Allowed tools
- —
What does the commit skill do?
commit generates Conventional Commits messages, performs commits, and optionally pushes changes to a remote repository.
Summary generated from the skill's SKILL.md.
About the commit skill
Automating the creation of standardized git commit messages is what this skill enables for developers working within a git repository. When an agent detects staged changes, it analyzes the git diff and recent logs to generate a message following the Conventional Commits specification. This includes selecting appropriate types like feat, fix, or refactor, determining scopes from file paths, and identifying breaking changes.
Users interact with the tool by requesting a commit, which triggers a preview of the generated message. The agent then provides options to commit the changes, commit and push to a remote repository, edit the message, or cancel the operation. If the --push flag is provided, the agent skips the manual approval step and immediately executes both the commit and the push to the current branch or its upstream. The process relies on standard git commands such as git status, git diff, and git log to gather necessary context.
What an agent can do with it
- 1Generate a standardized commit messageAn AI agent can analyze staged changes to write a message following the Conventional Commits format.
- 2Commit and push changes automaticallyDevelopers can use a specific flag to bypass the confirmation prompt and push directly to a remote repository.
- 3Review and edit commit contentThe tool allows users to inspect the proposed message and make revisions before finalizing the commit.
Written from the skill's SKILL.md. Every use case is checked against a quote from it.
How to install the commit skill
Each command copies the skill folder into the client's skills directory. Restart the client or start a new session to load it.Install the commit skill in Plugin
/plugin marketplace add yagizdo/quiver /plugin install quiver@quiver
Install the commit skill in Claude Code
~/.claude/skills/commit/SKILL.md
$ npx degit yagizdo/quiver/skills/commit ~/.claude/skills/commitInstall the commit skill in Claude Code (project)
.claude/skills/commit/SKILL.md
$ npx degit yagizdo/quiver/skills/commit .claude/skills/commitInstall the commit skill in Codex
~/.agents/skills/commit/SKILL.md
$ npx degit yagizdo/quiver/skills/commit ~/.agents/skills/commitInstall the commit skill in Gemini CLI
~/.gemini/skills/commit/SKILL.md
$ npx degit yagizdo/quiver/skills/commit ~/.gemini/skills/commitInstall the commit skill in Claude.ai
npx degit yagizdo/quiver/skills/commit commit && zip -r commit.zip commit # claude.ai → Customize → Skills → + → Create skill → Upload a skill → commit.zip
Automated checks
Ran Nov 15, 2026Full report →Passed our automated checks. That is not a security audit: read the scripts before you let an agent run them. How checks work · What to fix for more points
- Source repository linked and reachableyagizdo/quiver/skills/commit is public and reachablegithub.com/yagizdo/quiver/tree/HEAD/skills/commit ↗Pass
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit 2026-11-10github.com/yagizdo/quiver/commits ↗Pass
- No download-and-run or credential theft in scriptsNo scripts in the skill folderPass
- No hidden instructions in SKILL.mdNo override phrases, safeguard switches, credential requests or hidden charactersgithub.com/yagizdo/quiver/blob/HEAD/skills/commit/SKILL.md ↗Pass
- Pre-approved tools are scopedNo pre-approved tools; the agent asks before using onegithub.com/yagizdo/quiver/blob/HEAD/skills/commit/SKILL.md ↗Pass
- SKILL.md follows the Agent Skills formatthe description does not say when to use the skillgithub.com/yagizdo/quiver/blob/HEAD/skills/commit/SKILL.md ↗Finding
SKILL.md
View on GitHubGather Git Context
!`git rev-parse --is-inside-work-tree 2>/dev/null || echo "NO_GIT"`
!`git status --short 2>/dev/null || echo "NO_GIT"`
!`git diff --cached 2>/dev/null || echo "NO_GIT"`
!`git log --oneline -10 2>/dev/null || echo "NO_COMMITS"`
Instructions
Step 0 -- Git Availability
If any gather-context block above returned NO_GIT, this directory is not a git repository.
Print: > No git repository detected. /commit requires a git repo.
Stop here.
Silently determine which case applies — do not show the case label or decision logic to the user:
- No changes: If
git statusis empty → tell the user and stop. - Nothing staged: If
git diff --cachedis empty butgit statusshows changes → tell the user to stage files and run/quiver:commitagain, then stop. - Staged changes exist: Proceed silently to Commit Message Generation.
Commit Message Generation
Analyze git diff --cached and the recent log. Generate a Conventional Commits message following these rules:
Type — Select one based on primary intent:
| Type | When to Use |
|---|---|
feat |
New feature or capability visible to users |
fix |
Bug fix |
docs |
Documentation only (README, comments, JSDoc) |
style |
Formatting, whitespace, semicolons — no logic change |
refactor |
Code restructuring with no behavior change |
perf |
Performance improvement |
test |
Adding or fixing tests |
build |
Build system or external dependencies (npm, Makefile) |
ci |
CI/CD configuration (GitHub Actions, CircleCI) |
chore |
Maintenance tasks (version bumps, tooling config) |
revert |
Reverting a previous commit |
Scope — From file paths: single directory → its name, single file → domain name, cross-cutting → omit.
Breaking changes — If the change removes/renames public APIs, changes signatures, alters data formats, or removes CLI flags/commands, add ! after type/scope and a BREAKING CHANGE: footer.
Subject line — Imperative mood, lowercase after colon, no period, ≤72 chars total, describe what not how.
Body — Almost never. Only for breaking changes or when the subject alone would be misleading. Wrap at 72 chars. Never include file lists — git log --stat shows this.
Footers — BREAKING CHANGE: if applicable. Refs: #issue if relevant.
Include a body or footers only for breaking changes or multi-type changes where the subject alone is genuinely ambiguous. Default to subject-only. Don't add Co-authored-by or attribution footers unless explicitly requested.
Output
Important: The user must see the full commit message before answering the prompt. Print it to the chat stream and keep the AskUserQuestion question itself to one short line -- see Default (no flag) below.
Flag: --push
If $ARGUMENTS contains "push" (e.g., /quiver:commit --push), skip the AskUserQuestion step entirely. Instead, show the generated commit message, then immediately execute Commit & Push (commit + push to remote) without prompting.
Default (no flag)
Two steps, both mandatory: print the message, then ask. Do not paste the commit message into the AskUserQuestion question field -- that field is rendered as a single truncated line by some Claude Code surfaces, so a multi-line message is cut off or dropped entirely and the user is asked to approve something they cannot read.
Step 1 -- print the message as normal chat output, verbatim inside a fenced block, so the user sees exactly what git commit will receive:
{type}({scope}): {subject}
{body lines, if the message has a body}
Step 2 -- ask, and wait for the answer:
- Question:
"Commit this?"-- one short line. No newlines, no message text, no ANSI escape codes.AskUserQuestionrenders the question in a TUI box that prints escape sequences literally instead of styling them. - Header: "Action"
- Options:
- Commit — "Commit with this message"
- Commit & Push — "Commit and push to remote"
- Edit — "Revise the message"
- Cancel — "Abort without committing"
Printing the message is not approval. It is only the readable copy of what the prompt is about, and it exists because the prompt cannot display it. Never run git commit without an answer from this AskUserQuestion. The user asking for a commit in their message is not the answer either -- that is what put the skill on this step. --push is the only path that skips the prompt.
Commit Execution
On commit or commit & push:
- Commit using a HEREDOC:
git commit -m "$(cat <<'EOF' {full commit message} EOF )" - Verify:
git log --oneline -1andgit status --short.
If commit & push, also:
- Detect branch:
git branch --show-current - Push: if
git rev-parse --abbrev-ref @{upstream} 2>/dev/nullsucceeds →git push, otherwise →git push -u origin {branch} - Verify push exit code is 0.
Output:
Committed:
{short hash}{subject} Branch:{branch name}Pushed to:origin/{branch}(only if pushed)
On edit: Ask what to change, revise, re-present, and ask again.
On cancel:
Commit cancelled. Staged changes are intact.
Stop here.
Error Handling
If git commit fails, show the error verbatim and suggest the user fix the issue and re-run /quiver:commit. Never retry automatically or use --no-verify.
Files in this skill
2commit skill questions
How do I install the commit skill in Claude Code?
Run npx degit yagizdo/quiver/skills/commit ~/.claude/skills/commit. Claude Code loads skills from ~/.claude/skills (all projects) or .claude/skills (one project) and uses a skill when its description matches the task.
Which clients can use the commit skill?
Any client that reads the Agent Skills format: Claude Code, Codex, Gemini CLI and claude.ai (upload as a ZIP). Scripts in the skill run only where the client allows code execution.
What happens if I have changes that are not staged?
The tool will notify you to stage your files and run the command again.
Can I skip the confirmation prompt?
Yes, by using the --push flag, the tool will generate the message and immediately commit and push without asking.
What if the directory is not a git repository?
The tool will detect the lack of a git repository and stop execution.