After your AI agent finishes a coding task, the work just sits there. Uncommitted changes, no documentation of what was learned, no record of architectural decisions. You end up doing the housekeeping yourself or it never gets done.

Smart Commit runs automatically after every coding task and handles two things: keeping your CLAUDE.md up to date and making clean git commits with good messages.

It uses a complexity scoring system to decide what level of housekeeping each task deserves. A one line typo fix gets a simple commit and nothing else. A multi file refactor that touches three directories gets a descriptive commit message with a body, plus a CLAUDE.md update capturing what was learned about the codebase.

For CLAUDE.md, it focuses on things that would save a future session from rediscovering the hard way: architectural decisions, non obvious debugging insights, environment quirks, new conventions. It skips updates when the change is self explanatory from the code itself.

For commits, it follows conventional commit format (feat, fix, refactor, etc.) with scoped messages that describe intent, not just which files changed. It stages files selectively instead of blindly running git add, so it won't accidentally commit your .env or build artifacts. It pushes after every commit so your work is always synced.

The skill also handles edge cases cleanly: dirty working trees with someone else's changes, monorepos with per package documentation, branch conflicts that need human judgment, and projects that aren't git repos at all.

Works with Claude Code, Cursor, Codex CLI, VS Code Copilot, and any agent that supports the SKILL.md standard.