Atomic Commits Skill

Transform uncommitted changes into clean, logical commits using iterative analysis.

Workflow

1. Discover

git status
git diff --stat
git diff --name-only

2. Bucket by Functionality

Group files into logical buckets:

3. Order by Dependency

1. Infrastructure first (others depend on it)
2. Feature/Refactor (core logic)
3. Integration last (wiring it together)

4. Commit Each Bucket

For cleanly separated files:

git add path/to/bucket/files
git commit -m "type(scope): description"

For mixed changes in one file: Use granular-git skill.

Commit Messages

Follow Existing Patterns First

Check git log --oneline -20 for established conventions:

• Scope names (e.g., backend, frontend, auth)
• Type usage (e.g., feat, fix, refactor)
• Message style

Fallback: Conventional Commits

<type>(<scope>): <description>

Description Rules

1. Imperative mood: "add" not "added"
2. 50 chars max
3. Lowercase after scope
4. No trailing period

Decision Checklist

Before each commit:

• Changes are logically related
• No unintended files
• Dependencies satisfied
• Message follows existing pattern

Merge Conflicts

When resolving, prefer the implementation that:

1. Has better error handling
2. Includes new features (Task, tracing)
3. Follows updated patterns

See Also

• references/bucket-types.md - Detailed bucket categories
• references/decision-tree.md - When to split/merge commits
• Granular-git skill - For precise file-level commits