// Creates atomic Conventional Commits. Use when committing code changes, splitting hunks into revertable units, or writing detailed commit messages.
Guides t-wada Red-Green-Refactor TDD. Use when implementing features, fixing bugs, or refactoring logic with strict test-first development.
Resolves missing CLI tools. Use when a command is unavailable, a shell reports command not found, or a tool must be run without installing it globally.
Prevents and handles GitHub API rate limits during Nix commands. Use when running nix flake, nix run, nix build, nix shell, or comma against GitHub-backed inputs.
Consults Codex CLI for a second opinion on implementation plans, code reviews, or problem-solving. Use when an independent perspective from a different agent is needed before a significant decision.
Spawns parallel task agents to explore a codebase area. Use when researching unfamiliar code, auditing a subsystem, or gathering diverse perspectives before a design decision.
Runs the full PR workflow — creates a feature branch, commits, pushes, and opens the pull request. Use when the user asks to create or open a PR ("create a PR", "push this up and open a PR").
| name | commit |
| description | Creates atomic Conventional Commits. Use when committing code changes, splitting hunks into revertable units, or writing detailed commit messages. |
Arguments:
You are an expert git commit architect creating fine-grained, independently revertable commits following Conventional Commits specification.
Before committing, inspect the current state:
git status --short
git diff HEAD
git log --oneline -10
Revertability First: Each commit must be revertable independently without breaking other functionality. Prefer smaller, granular commits over large groupings. Split by hunks within files, not just entire files.
PR branches are normally squash-merged, so do not compress review work with git commit --amend by default. Keep review fixes as small follow-up commits that can be reverted independently. Amend only for unpublished local mistakes or when the user explicitly asks.
Tiny commits are expected. A single review comment, one wording correction, one reference-file extraction, one symlink sync, or one generated formatting pass can each be its own commit when independently revertable.
Tiny does not mean incomplete. For moves, renames, or extractions, one commit must include both sides of the operation: remove or update the old location, add the new location, update references, and sync generated links if required. Never commit only the destination of a move while leaving the source/reference cleanup for a later commit.
For concrete good and bad examples, read references/revertable-commits.md.
git diff <file>git apply --cached -v <patch>git show HEADNEVER use git add -p or git add --interactive - Claude Code cannot handle interactive commands.
Use git apply --cached -v to stage precise non-interactive patches. Read references/git-apply.md when a patch fails, needs whitespace handling, or must be staged without touching unrelated hunks.
Use standard git history commands to understand intent before committing. Prefer targeted commands such as git log --follow -- <file>, git show <commit> -- <file>, and git blame <file>. Match the repository's existing commit granularity, scopes, and explanation style.
<type>(<scope>): <subject>
<body>
<footer>
Types: feat, fix, docs, style, refactor, perf, test, build, ci, chore, revert
Body should explain:
chore(xxx): format or just chore: formatAfter all commits are complete, push to remote. Let repository git hooks run; if pre-commit or pre-push runs format, sync, lint, typecheck, or tests, treat those hooks as part of the normal validation path and fix any failures in a new small commit.
Read references/push.md for the exact upstream check and push commands.