New advertisers: 50% off your first month with code FIRSTMONTHSee slots
Synthesys-Lab avatar

agentize

Synthesys-Lab/agentize

AI-powered planning, implementation, and review commands for Claude Code that chain issues to code.

version
1.1.9
GitHub stars
55

What it does

  • Adds commands like /ultra-planner, /issue-to-impl, /code-review, and /pull-request, plus /sibyl and /team-review
  • Covers the workflow from plan to GitHub issue, implementation, code review, and pull request
  • Uses multi-agent debate through ultra-planner and mega-planner, with proposal critique and reduction agents
  • Designed for developers who want human approval only at the planning stage, with gh CLI integration

AI-powered development workflow with planning, implementation, and review commands

Add marketplace

/plugin marketplace add Synthesys-Lab/agentize

Run it inside Claude Code once, then install the plugin below.

Install

/plugin install agentize@agentize

Run it inside Claude Code after adding the marketplace.

Files

Discovered from source

Commit Message

This skill instructs AI agent on how to commit staged changes to a git repository with meaningful commit messages.

Inputs

The commit skill takes the following inputs:

  • The purpose of the commit, either a delivery or a milestone
    • Milestone is the only commit that can bypass pre-commit hooks
    • Milestone can only happen on a development branch
  • The staged files to be committed
    • The commit message should clearly describe the changes made. If the changes are less than 20 lines, a short commit message is sufficient. Otherwise, a full commit message is required.
  • If available, the related milestone or issue number
    • As per our naming convention, the development branch should be named issue-<number>-<brief-title>, so you can find the issue number from the branch name.

Full Commit Message

The commit message should follow the structure below:

[tag]: A brief summary of the changes of this commit.

path/to/file/affected1: A brief description of changes made to this file.
path/to/file/affected2: A brief description of changes made to this file.
...

If needed, provide addtional context and explanations about the changes made in this commit.
It is preferred to mention the related Github issue if applicable.

A milestone commit is always on a development branch associated with a issue. If it is a milestone, additionally add the following information:

  1. Add [milestone] before the tag.
  2. Mention the issue number after the brief summary, e.g., A milestone to issue #42.
  3. Briefly summarize the test case status, e.g. 35/42 test cases passed.
    • Milestone is to react to a big issue breaking down to smaller steps.
    • Thus, it is important to tract the progress, and is the only case allowing bypassing pre-commit hooks.

Short Commit Message

The commit message should follow the structure below:

[tag]: A brief summary of the changes of this commit.

A short message is always for a delivery commit.

Tags

A git-msg-tags.md file should appear in {ROOT_PROJ}/docs/git-msg-tag.md which defines the tags related to the corresponding modules or modifications. The AI agent MUST refer to this file to select the appropriate tag for the commit message. If not, reject the commit, and ask user to provide a list of tags in docs/git-msg-tag.md, by showing the example format below:

Please provide a docs/git-msg-tags.md, which can be as simple as the following example:

# Git Commit Message Tags
- `[core]`: Changeing the core functionality of the project.
- `[docs]`: Changing the documentation.
- `[tests]`: Changes test cases.
  - Use it only when solely changing the test cases! Do not mix with other changes with tests!
- `[build]`: Changes related to build scripts or configurations.

Ownership

DO NOT claim the co-authorship of the commit with the user in the message. It is the user who is FULLY responsible for the commit.

Pre-commit Check

When committing the changes, this skill should faithfully follow the input on if it is a milestone to use --no-verify or not. If it is a milestone, the commit MAY bypass pre-commit hooks. If it is a delivery commit, the commit MUST NOT bypass pre-commit hooks! DO NOT use pre-existing issue as an excuse to bypass pre-commit in any case!

From the Synthesys-Lab/agentize README

AI-powered SDK for Software Development

Tests

Prerequisites

Required Tools

  • Git - Version control (checked during installation)
  • Make - Build automation (checked during installation)
  • Bash - Shell interpreter, version 3.2+ (checked during installation)
  • GitHub CLI (gh) - Required for GitHub integration features
    • Install: https://cli.github.com/
    • Authenticate after installation: gh auth login
    • Used by: /setup-viewboard, /open-issue, /open-pr, GitHub workflow automation
  • Python 3.10+ - Required for permission automation module, otherwise you can have infinite yes to prompt!
    • Use Python venv or anaconda to manage a good Python release!
    • Requires PyYAML (pip install pyyaml) for YAML configuration parsing

Recommended Libraries

  • Anthropic Python Library - For custom AI integrations (optional)
    • Install: pip install anthropic
    • Note: Not required for core SDK functionality, but recommended if you plan to extend or customize AI-powered features

Verification

After installing prerequisites, the installer will automatically verify git, make, and bash availability. GitHub CLI authentication can be verified with:

gh auth status

Quick Start

Agentize is an AI-powered SDK that helps you build your software projects using Claude Code powerfully. It is splitted into two main components:

  1. Claude Code Plugin: Automatically registered during installation when claude CLI is available. See Tutorial 00a: Claude UI Setup for details.
  2. CLI Tool: A source-first CLI tool to help you manage your projects using Agentize. See Tutorial 00: CLI Quickstart for the CLI workflow.
curl -fsSL https://raw.githubusercontent.com/SyntheSys-Lab/agentize/main/scripts/install | bash

Then add to your shell RC file (~/.bashrc, ~/.zshrc, etc.):

source $HOME/.agentize/setup.sh

See docs/feat/cli/install.md for installation options and troubleshooting.

Upgrade: Run lol upgrade to pull the latest changes.

Your First 15 Minutes

After installation, the installer creates ~/.agentize.local.yaml in your home folder. This file controls which AI backends are used for planning and implementation.

The 5-Step Agentize CLI Workflow

  1. Configure (already done) - confirm ~/.agentize.local.yaml exists and adjust backends if needed

  2. Clone with worktrees:

    wt clone https://github.com/org/repo.git myproject.git
    

    wt clone sets up a bare repository and leaves you in trees/main.

  3. Plan your first feature:

    lol plan --editor
    

    Review the GitHub issue it creates.

  4. Implement the plan:

    lol impl <issue-number>
    
  5. Navigate between worktrees:

    wt goto <issue-number>
    wt goto main
    

See Tutorial 00: CLI Quickstart for a full walkthrough.

Troubleshoot

If you encounter any issue during the usage. For example:

  1. It asks you for permission on a really simple operation.
  2. It fails to automatically continue on a session.

Enable debug mode in your .agentize.local.yaml:

handsoff:
  debug: true

Then re-run the command. This will give you a detailed log in either

  • /path/to/your/project/.tmp/handsoff-debug.log or
  • $HOME/.agentize/.tmp/handsoff-debug.log Paste your logs on issue for me (@were) to debug!

For further help, please visit our troubleshooting guide.

Core Philosophy

Minimizing human intervention by artifact centric.

  • Session-centric: People tell AI what to do, and wait until it ends. Then give feedback until they are satisfied. Human looping in too much limits the scalability.
  • Artifact-centric: People tell AI what to do, and AI produces a plan first. Plan is the ONLY phase that human can intervene. After the plan is approved, AI will execute the plan and produce the code merge for human to review.

A clear separation between human, AI, and formal language.

  • Humans are for the intention of development, including providing feature requirements, approving plans, and code merges.
  • AI is the worker of software development for both making the plan, and maintaining the codebase, including tests, documentation, and code quality.
  • Formal language is for the coordination and orchestration between AI, and other systems, e.g. Github Issues, Pull Requests, and CI/CD pipelines.
    • I (@were) found that skills are promising for AI to synthesize fixed code to interact with such systems, but these flows are more fixed and formal than I expected --- putting them in formal language (e.g. Python scripts, or YAML configuration) is more transparent and faster to execute the whole workflow.

Workflow:

See our detailed workflow diagrams:

Legend: Red boxes represent user interventions (providing requirements, approving/rejecting results, starting sessions). Blue boxes represent automated AI steps.

Tutorials

Learn Agentize in 15 minutes with our step-by-step tutorials (3-5 min each):

  1. CLI Quickstart - Learn the core CLI workflow in 15 minutes
  2. Claude UI Setup - Set up the Claude Code plugin and slash commands
  3. Ultra Planner - Primary planning tutorial (recommended)
  4. Issue to Implementation - Complete development cycle with /issue-to-impl and /code-review
  5. Advanced Usage - Scale up with parallel development workflows

Project Organization

agentize/
├── .claude-plugin/         # Plugin root (use with --plugin-dir)
│   ├── marketplace.json    # Plugin manifest
│   ├── commands/           # Claude Code commands
│   ├── skills/             # Claude Code skills
│   ├── agents/             # Claude Code agents
│   └── hooks/              # Claude Code hooks
├── python/                 # Python modules (agentize.*)
├── docs/                   # Documentation
│   └── git-msg-tags.md     # Commit message conventions
├── src/cli/                # Source-first CLI libraries
│   ├── wt.sh               # Worktree CLI library
│   └── lol.sh              # SDK CLI library
├── scripts/                # Shell scripts and wrapper entrypoints
├── templates/              # Templates for SDK generation
├── tests/                  # Test cases
├── Makefile                # Build targets for testing and setup
└── README.md               # This readme file

Details

Marketplace
Synthesys-Lab/agentize
Source
.claude-plugin/
Author
SyntheSys-Lab
Category
Development
License
MIT
Keywords
workflow, tdd, sdd, handsoff

Similar to agentize

More Development Tools