Commit

When to use

Use when you want to commit your changes as clean, atomic micro-commits with conventional commit messages. Instead of one big commit, the skill scopes to the files the current session actually touched, groups related changes, and commits each logical unit separately.

How it works

1. Picks a mode first — before running any git command
2. Builds the set of files edited during this session and diffs it against git status
3. Asks what to do with pre-existing dirty files the session never touched (session only, include all, or pick per file)
4. Analyzes the in-scope diff and recent git log to match the repo’s commit style
5. Groups related files — tests with their source, lockfiles with their manifest, formatting on its own
6. Proposes a conventional commit message per group, then stages by explicit path and commits
7. Offers to push once the commits are in

Modes

• Interactive — proposes each commit for your approval before executing
• Non-interactive — dispatches a background agent that auto-approves sensible commits, skips ambiguous ones, and streams progress back as it goes. It won’t push to main or master.

Scoping

The skill never uses git add ., -A, or -u. Every commit stages explicit paths from the session-scoped file list, so unrelated work in your tree stays untouched. Files that look like secrets are never committed in non-interactive mode, and git hooks are never skipped.

Invoke

/commit:commit
/commit:commit i    # interactive mode
/commit:commit ni   # non-interactive (background)