grokkit/docs/user-guide/commit-msg.md

63 lines
2.3 KiB
Markdown
Raw Normal View History

# 📝 Commit Message Guide
The `commit-msg` command generates a conventional commit message based on your currently staged changes. Unlike the `commit` command, it only outputs the suggested message without actually performing the git commit.
### Why use Commit Message?
- **Consistency**: Automatically generate messages following the [Conventional Commits](https://www.conventionalcommits.org/) specification (e.g., `feat(api): add JWT authentication`).
- **Convenience**: Quickly get a descriptive summary of your changes without manually reviewing the diff.
- **Workflow flexibility**: Useful when you want to review, edit, or use the generated message in a custom git command or GUI.
### Basic Usage
```bash
git add .
grokkit commit-msg # Generate message only
grokkit commit-msg -m grok-4 # Use specific model
```
**Output format:** `type(scope): subject\n\nbody`
### Options
| Flag | Description |
|------|-------------|
| `--model`, `-m` | Override the default model (e.g., `grok-4`) |
### How it Works
1. **Diff Retrieval**: Grokkit runs `git diff --cached --no-color` to get all staged changes.
2. **AI Analysis**: It sends the diff to Grok with instructions to generate a conventional commit message.
3. **Streaming Output**: The generated message is streamed directly to your terminal.
### Workflow Example
1. Stage your changes:
```bash
git add main.go internal/grok/client.go
```
2. Generate the message:
```bash
grokkit commit-msg
```
3. Use the output to commit:
```bash
git commit -m "$(grokkit commit-msg)"
```
### Comparison with `commit` command
| Feature | `commit-msg` | `commit` |
|---------|--------------|----------|
| Generates AI message | Yes | Yes |
| Shows diff-based summary | Yes | Yes |
| Asks for confirmation | No | Yes |
| Performs `git commit` | No | Yes |
| Use case | Scripting / Manual control | Interactive git workflow |
### Best Practices
- **Stage only related changes**: To get the most accurate commit message, stage only the changes that belong to a single logical unit of work.
- **Review the output**: AI-generated messages are usually good but may need minor tweaks for complex changes.
- **Model Choice**: Use a reasoning model (`grok-4`) for complex architectural changes to get more detailed body descriptions in the commit message.