Git Commit
Create atomic commits by staging the right files, analyzing the staged diff, composing a conventional commit message, and optionally pushing.
Workflow
1) Pre-flight + context (single call)
Run all checks and context collection in one bash call:
git rev-parse --is-inside-work-tree \
&& ! test -d "$(git rev-parse --git-dir)/rebase-merge" \
&& ! test -f "$(git rev-parse --git-dir)/MERGE_HEAD" \
&& ! test -f "$(git rev-parse --git-dir)/CHERRY_PICK_HEAD" \
&& git symbolic-ref HEAD \
&& git status --short --branchIf any check fails, stop with a clear error and suggested fix.
Arguments: $ARGUMENTS
2) Parse arguments
- Flags:
- --all commit all changes - --deep deep analysis, breaking changes, concise body - --push push after commit - --close <issue_numbers> append Closes #N trailers for listed issues (comma/space-separated)
- Value arguments:
- Type keyword (any conventional type) overrides inferred type - Quoted text overrides inferred description
3) Stage + read diff
- If
--all:
- If no changes at all: error "No changes to commit" - If unstaged changes exist: git add -A - If already staged: proceed
- Otherwise (atomic commits):
- Session-modified files = files edited in this session - Currently staged files: git diff --cached --name-only - For staged files NOT in session-modified set: git restore --staged <file> - For session-modified files with changes: git add <file> - If none: error "No files modified in this session"
- Unrelated changes: session-modified files may contain pre-existing uncommitted changes (hunks not from this session). Include the entire file—partial staging is impractical. Never revert, discard, or
git checkoutunrelated changes. - Read the staged diff once:
git diff --cached - Log staged files with status (A/M/D)
4) Analyze + compose message
Read the staged diff and produce the commit message in a single pass.
Type inference — determine from behavior:
| Behavior | Type |
|---|---|
| New functionality | feat |
| Bug fix / error handling | fix |
| Code reorganization, no behavior change | refactor |
| Documentation | docs |
| Tests | test |
| Build system (webpack, vite, esbuild) | build |
| CI/CD pipelines | ci |
| Dependencies | chore(deps) |
| Formatting / whitespace only | style |
| Performance | perf |
| Reverting previous commit | revert |
| AI config (CLAUDE.md,.claude/,.gemini/,.codex/) | ai |
| Other maintenance | chore |
Explicit type keyword in arguments takes precedence over inference.
Scope — infer only when path makes it obvious (lowercase).
Unrelated hunks — ignore pre-existing changes when determining type/scope/description. If unrelated changes are in the same file as session changes, they are included in the commit scope but should not influence the message.
Message format:
- Subject line (<= 50 chars):
type(scope): descriptionortype: description - Imperative mood ("add" not "added"), lowercase, no period
- Describe what the change does, not which files changed
- Body: hyphenated lines for distinct changes; skip for trivial changes
If --deep:
- Deep semantic analysis; detect breaking changes
- Infer scope from code structure even when path isn't clear
- Body: 2-3 hyphenated lines max, focus on WHY
- Breaking change:
BREAKING CHANGE:+ one-line migration note - Check for GitHub issues in the chat transcript; add
Closes #123
If --close:
- Append a
Closes #Nline for each issue number provided - Multiple issues: one
Closes #Nper line in the body/trailer
5) Commit
- Use
git commit -m "subject"(add-m "body"only if body is non-empty) - Output: commit hash + subject + file count summary
- If failed: show error + suggest fix
6) Push (if --push)
- If upstream exists:
git push - If no upstream:
git push -u origin HEAD - If failed: show error + suggest fix (pull/rebase first, set upstream, check auth)