AI Commit

Skill: ai-commit

What I do

I provide expertise in creating properly attributed commits for AI-generated code using the project's standard workflow. I ensure every commit is atomic, follows conventional commit formats, and includes mandatory co-authoring attribution.

When to use me

• When creating new commits for code generated or modified by AI
• When you need to split changes into atomic, logical units
• When attributing work to both the human developer and the AI agent

Core principles

1. Atomic commits: Each commit must represent a single, logical change. Do not bundle unrelated fixes or features together.
2. Standard workflow: Always write your commit message to a temporary file first, then use the project's make target for execution.
3. Proper attribution: Include the Co-authored-by trailer for the AI model used to maintain a clear audit trail.
4. Conventional format: Use clear types like feat, fix, docs, or refactor to categorise changes.

Patterns & examples

Workflow for a new commit:

1. Stage your changes with git add.
2. Write the message to a file, for example /tmp/commit.txt.
3. Run make ai-commit FILE=/tmp/commit.txt.

Example commit message in /tmp/commit.txt:

feat: add user authentication middleware

Implement JWT validation for all protected routes to ensure secure access.

Co-authored-by: Claude <claude@anthropic.com>

Using fixup commits: For small corrections to a previous, unpushed commit, use git commit --fixup=<hash> to keep history clean before a final squash.

Anti-patterns to avoid

• ❌ Direct git commit: Skipping the make ai-commit target loses consistent formatting and attribution.
• ❌ Bloated commits: Bundling multiple logical changes makes code reviews difficult and rollbacks risky.
• ❌ Missing trailers: Failing to include co-authoring information breaks the project's attribution rules.

KB Reference

~/vaults/baphled/3. Resources/Knowledge Base/AI Development System/Skills/Git/AI Commit.md

Related skills

• git-master: For advanced history search and planning
• git-advanced: For rebase and history management
• clean-code: To ensure the committed code meets quality standards