Commit

/commit

Context

• User hint: $ARGUMENTS
• Pre-flight: !bash ${CLAUDE_SKILL_DIR}/scripts/pre-flight.sh
• Current branch: !git branch --show-current 2>/dev/null || echo "N/A"
• Change analysis: !bash ${CLAUDE_SKILL_DIR}/scripts/analyse-changes.sh 2>/dev/null || echo "N/A"
• Full diff: !git diff HEAD 2>/dev/null || echo "N/A"
• Recent commits: !git log --oneline -10 2>/dev/null || echo "N/A"
• Effort level: !bash ${CLAUDE_SKILL_DIR}/scripts/effort-level.sh

Instructions

Not a git repository

If the git check above shows "NOT_A_GIT_REPO":

1. Tell the user this directory is not a git repository.
2. Run git init to initialise one.
3. Stage all files by name (review the directory contents first — never use git add -A or git add .). Skip files that look like secrets or build artefacts (.env, node_modules/, dist/, etc.).
4. Create the initial commit with the message chore: initial commit plus the Co-Authored-By trailer (see references/message-format.md).
5. Run git status to confirm success, then stop.

No changes detected

If the change analysis shows "No changes detected", tell the user and stop.

If the user hint is non-empty, use it to guide which files to stage and what the commit message should convey. The hint may describe the scope ("the /parallel changes"), the intent ("fix typo in README"), or both. It is a hint, not a literal commit message — still apply Conventional Commits format and the atomicity rules below.

Step 1 — Stage and verify atomicity

Use the change analysis above (directory groups and atomicity signal) as the starting point for your atomicity assessment.

1. Review the directory grouping and the full diff to determine whether changes form one or more logical units of work. If a user hint is present, use it to identify which subset of changes to stage.
2. The "and" test: draft a one-line summary. If it contains "and" connecting two independent actions (e.g., "improve X and enable Y"), it's two commits — split them. A shared topic or session does not make unrelated changes atomic.
3. If everything is a single logical change: stage all modified files by name (e.g., git add file1 file2) and proceed.
4. If changes should be split into multiple commits: explain the split, listing which files/hunks belong to each commit. Stage the first logical unit (use git add <file> or git add -p <file> for partial staging) and proceed with that commit. Then continue to the next unit — repeat steps 1–3 for each remaining unit until all changes are committed.
5. Never use git add -A or git add . — always stage files by name to avoid accidentally including untracked files.

Step 2 — Compose commit message

Read references/message-format.md (in this skill's directory) for the full Conventional Commits format, type table, and Co-Authored-By trailer spec.

Draft the commit message following that format. Use the recent commits above for style consistency.

• Scope: derive from the directory grouping in the change analysis. If changed files share a common directory or component, use it as scope. Omit scope only when changes are genuinely cross-cutting or all files are in the repository root with no directory structure.
• Effort level: use the effort level from context above. Include it in the Co-Authored-By trailer unless it shows N/A.

Step 3 — Validate and commit

1. Validate the draft message before committing:

   cat <<'EOF' | bash ${CLAUDE_SKILL_DIR}/scripts/validate-message.sh
   <full draft message here>
   EOF
   

2. If validation fails, fix the issues and re-validate.
3. Stage files and create the commit using a HEREDOC for correct formatting.
4. Run git status after committing to confirm success.

Safety rules

• Never amend a previous commit unless the user explicitly asks for --amend.
• Never skip hooks (--no-verify) unless the user explicitly asks.
• Never force-push unless the user explicitly asks — and warn before doing so.
• If a pre-commit hook fails: fix the issue, re-stage, and create a NEW commit. Do not amend — the failed commit never happened, so amending would modify the wrong commit.

Notes

• This skill is project-agnostic — it works in any repository.
• Projects can override this by creating .claude/skills/commit/SKILL.md at project level.