@agent-smith/feat-git — AI-Powered Git Commit Management

Smart git commit message generation powered by language models. Part of the Agent Smith toolkit.

Features

• 🤖 AI-Generated Commit Messages — Automatically generates meaningful commit messages from your changes
• 🎯 Smart Context Awareness — Analyzes staged/unstaged modifications to understand what you're committing
• ✏️ Custom First Line — Provide a custom first line while letting AI generate the body
• 💬 Additional Instructions — Give the model extra context or constraints for message generation
• 📦 Package-Aware — Include package names in generated commit messages
• 🔄 Interactive Confirmation — Review and confirm generated messages before committing
• ⚙️ Model Flexibility — Use any supported language model with customizable templates
• 🔧 File-Specific Commits — Target specific files or directories for focused commits

Documentation

For AI Agents

• Codebase Summary — Architecture, key files, and patterns for the feat-git plugin
• Git Plugin — Complete plugin documentation and usage guide

For Humans

• Git Plugin — Usage guide and configuration

Installation

npm i -g @agent-smith/feat-git

Configuration

Add the plugin to your config.yml file:

plugins:
  - "@agent-smith/feat-git"

Then run the configuration command to update the CLI:

lm conf

Quick Start

Generate a commit message for all staged changes:

lm commit

The model will analyze your changes and generate a commit message, then prompt you to confirm.

Usage

Basic Commit

In a git repository with uncommitted or staged modifications:

lm commit

File-Specific Commit

Target specific files or directories:

lm commit src

Custom First Line

Provide your own first line while AI generates the body:

lm commit --msg "Custom first line"

Additional Instructions

Give the model extra context or constraints:

lm commit --instructions "Omit the version number change"

Package-Aware Messages

Include a package name in the generated first line:

lm commit --pkg "git plugin"

Model Configuration

Using a Specific Model

The default model is qwencoder 30b a3b. To use another model:

lm commit -m mymodel

Model with Template

Specify both model and template:

lm commit -m "gemma3:4b-it-q8_0" -t "gemma"

Automatic Template Detection

Omit the template name and it will be guessed based on the model:

lm commit -m "deepseek-coder-v2:16b-lite-instruct-q8_0"

In this example, the Deepseek v2 template will be automatically detected.

Format: -m modelname -t templatename

Command Reference

| Command | Description | |---------|-------------| | lm commit | Generate and confirm a commit message for all staged changes | | lm commit <path> | Generate commit message for specific files/directories | | lm commit --msg "<line>" | Use custom first line for the commit message | | lm commit --instructions "<text>" | Provide additional instructions to the model | | lm commit --pkg "<name>" | Include package name in generated first line | | lm commit -m <model> | Use a specific language model | | lm commit -m <model> -t <template> | Use model with specific template |

Important Notes

• 📌 Requires an active git repository with modifications
• 🔒 The command prompts for confirmation before executing the commit
• 🤖 A language model must be available and configured in Agent Smith
• ⚠️ Unstaged changes must be staged first or passed as arguments
• 🔧 Ensure the plugin is properly configured via lm conf before use

License

MIT