/commit — Intelligent Git Commit Workflow

You've been working across two features simultaneously. The files are tangled — tests for one analysis mixed with documentation for another — and pushing them as one commit will make the history unreadable. git reset --soft can undo it after the fact, but /commit gets the grouping right on the first try.

When you need this

• Your working tree has changes spanning multiple logical units of work
• You want well-organized commit messages that explain what changed and why
• You're about to push and want a clean, readable history

What it does — and what it won't

/commit analyzes your staged and unstaged changes, groups them into logical commits by topic, writes clear commit messages, and executes them in sequence. It shows you the proposed groupings and asks for confirmation before running — you stay in control.

Unlike git add --patch (which works line-by-line within a single file), /commit groups changes by topic across multiple files — creating one commit per logical unit of work.

Note: /commit does not automatically stage untracked files. You choose which files are in scope. It will show you git diff --cached --name-only before committing so you can verify nothing unexpected is included.

Worked example

Maya finishes a session where she touched her DiD specification script, updated two reference files, and edited her methods chapter. She has 11 files in her working tree across two distinct workstreams.

/commit

/commit groups the changes into three logical commits: (1) feat(scripts): add staggered DiD estimator — 3 R scripts; (2) docs(references): update parallel trends documentation — 2 reference files; (3) docs(chapter2): revise econometric methods section — 1 chapter file. It presents the groupings, waits for her confirmation, then executes all three.

Try it

/commit