Smart Commit Skill

When I say "commit this" or "generate a commit message," follow this workflow:

1. Intelligence Gathering

• Run git diff --cached to see exactly what is being committed.
• Identify Change Type: Determine if it is a feat, fix, docs, style, refactor, test, or chore.
• Analyze Scope: Identify the specific module or component affected (e.g., api, models, ui).

2. Generate the Message

Follow the Conventional Commits specification:

<type>(<scope>): <short summary in imperative mood>

[optional body: explain the WHY, not just the WHAT]

[optional footer: relate to Issue IDs, e.g., Closes #123]

don't add Co-Authored-By on message

3. Execution

1. Present the generated message to the user.
2. Run git commit -m "[message]" (or -m "[subject]" -m "[body]" for multi-line).

4. The Gotcha

Warning: Remind the user if they are committing too many unrelated changes at once (non-atomic commits). If the diff is too large, suggest splitting it into two separate commits.

Why this works for you

• Context Aware: Because it runs git diff --cached, it knows exactly what you've staged, unlike generic AI tools that look at your whole folder.
• Structured: It forces the "Conventional Commit" format, which makes your git log look professional and machine-readable.

Example Interaction

User: "Claude, commit these Pydantic changes."

Claude:

Proposed Message: feat(models): implement Pydantic v2 validation for User schema

Updated UserRecord to inherit from BaseModel and added EmailStr validation for the email field. This prevents malformed data from reaching the service layer.