# Commit

> Use after completing work to create atomic conventional commits grouped by layer. Supports --analyze, --push, --pr, --merge.

- **Type:** Skill
- **Install:** `agentstack add skill-ramonclaudio-skills-commit`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [ramonclaudio](https://agentstack.voostack.com/s/ramonclaudio)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [ramonclaudio](https://github.com/ramonclaudio)
- **Source:** https://github.com/ramonclaudio/skills/tree/main/plugins/commit/skills/commit

## Install

```sh
agentstack add skill-ramonclaudio-skills-commit
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Git Workflow - Atomic Commits

ultrathink

Commit code changes using atomic commits with conventional commit format. Each commit is one independent logical change that can be reverted without breaking other functionality.

## Voice

Follow `~/.claude/rules/voice.md` for all authored text. Commit-specific rules:
- Lowercase after colon, no period at end
- Specific verbs: add, fix, extract, drop, rename, move, split, wire, swap
- PR bodies: start with what changed and why, plain bullet list, no filler
- PR titles: same format as commit messages
- No `**Bold:** description` lists in PR bodies
- Backtick all filenames, paths, commands, config keys, and code references
- Never mention Claude/Anthropic/AI

## Git State (auto-populated)

- **Current branch:** !`git branch --show-current`
- **Status:** !`git status --short`
- **Staged changes:** !`git diff --cached --stat`
- **Unstaged changes:** !`git diff --stat`
- **Full diff:** !`git diff HEAD`
- **Recent commits:** !`git log --oneline -10`

## Arguments

- `$ARGUMENTS` containing `--analyze`: Only run Phase 1 analysis, output groupings without committing
- `$ARGUMENTS` containing `--push`: Run Phases 1-3 on current branch, then push directly (no branch, no PR)
- `$ARGUMENTS` containing `--pr`: Continue through Phase 4 (push and create PR)
- `$ARGUMENTS` containing `--merge`: Run Phase 5 only (merge specified PR and cleanup) - expects PR# as next arg
- No arguments: Run Phases 1-3 (analyze, commit, verify)

## Workflow

### Phase 1 - Analysis

Analyze the git state above. Group changes by layer + type. Verify independence.

Track each commit group as a task. Subject: affected files. Description: the commit message.

Output analysis in `` tags.

### Phase 2 - Execution

**GPG detection:** Before the first commit, check git config:
```bash
git config --get commit.gpgsign
```
- `true`: Use `git commit -S -m` for all commits
- Empty/unset: Use `git commit -m` (no `-S` flag). Log once: "GPG not configured, commits will not be signed."

**Branch creation (with `--pr` flag):**
```bash
BASE_BRANCH=$(git branch --show-current)
git checkout -b type/description-in-kebab-case
```
Store `BASE_BRANCH` for use in Phase 4. This ensures:
- Commits go to feature branch, not the working branch
- PR targets the correct base branch (not always main)

For each commit group (TaskUpdate status: `in_progress` -> `completed`):
```bash
git add [files]
```

Validate the message before committing: run `${CLAUDE_SKILL_DIR}/scripts/validate-commit-msg.sh "message"`. The script checks conventional format, lowercase description, no trailing period, and max 72 chars. If it fails, fix the message before proceeding.

```bash
git commit [-S] -m "type(scope): description"
```

Output in `` tags.

### Phase 3 - Verification

For each commit, verify:
- Atomic: Single concern, independently revertable
- Clear: Specific about what changed
- Signed: Used `-S` flag (if GPG available)

If verification fails:
```bash
git reset --soft HEAD~1
# Fix and recommit
```

### Phase 3.5 - Push (with `--push`)

Push all commits directly to the current branch. No branch creation, no PR.

```bash
git push
```

If the branch has no upstream, set it:
```bash
git push -u origin HEAD
```

After push, output summary and stop. Do NOT continue to Phase 4.

### Phase 4 - Pull Request (with `--pr`)

Push branch and create PR targeting `BASE_BRANCH` from Phase 2:
```bash
git push -u origin HEAD
gh pr create --base "$BASE_BRANCH" --title "type(scope): description" --body "$(cat ` (creates new commit) |
| Need to find old state | `git reflog` then `git reset --hard ` |

## Gotchas

- Don't amend published commits. Use `git revert` instead.
- GPG agent can timeout during batch commits. If signing fails mid-batch, check `gpg-agent` status before retrying.
- `--push` fails silently if no upstream is set. Use `git push -u origin HEAD` for new branches.
- Empty commits happen when all changes in a group were already staged. Verify with `git diff --cached --stat` before committing.
- PR body format: plain bullets, no `**Bold:** description` pattern. Start with what changed and why.

## Constraints

- Execute actual bash commands (not suggestions)
- Use Bash tool for all git operations
- Stop and report errors immediately
- Group by layer: data -> backend -> UI -> config -> docs
- Commit in dependency order
- Verify independently revertable
- Sign commits with `-S` flag when GPG is available
- Be terse in confirmations
- Never mention Claude, Claude Code, Anthropic, or AI in commits, PR bodies, or any output
- Never use co-authored-by or "generated by" attributions
- All text must pass as human-written. Re-read before submitting. If it sounds like a chatbot wrote it, rewrite it.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [ramonclaudio](https://github.com/ramonclaudio)
- **Source:** [ramonclaudio/skills](https://github.com/ramonclaudio/skills)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-ramonclaudio-skills-commit
- Seller: https://agentstack.voostack.com/s/ramonclaudio
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.