Skip to content

Cluster444/agentic

Repository files navigation

Agentic

npm version License: MIT

Modular AI agents and commands for structured software development with OpenCode.

What It Does

Agentic is a context engineering tool that assists OpenCode in producing reliable software improvements. Agentic is a workflow management system for AI-assisted software development using OpenCode. It provides:

  • Context Management: Organized "thoughts" directory structure for storing architecture docs, research, plans, and reviews
  • Modular AI Agents & Commands: Pre-configured prompts and specialized subagents that enhance OpenCode's capabilities through task decomposition and context compression
  • Structured Development Workflow: A phased approach (Research → Plan → Execute → Commit → Review) for handling tickets and features
  • Distribution System: A CLI tool to distribute agent/command configurations to projects via .opencode directories

Purpose

The system aims to:

  • Make AI-assisted development more systematic and reproducible
  • Reduce context window usage through specialized subagents
  • Maintain project knowledge over time (architecture decisions, research, implementation history)
  • Provide guardrails for AI agents through structured workflows

Quick Start

Installation

From bun/npm (Recommended)

npm install -g agentic-cli
# or
bun add -g agentic-cli

From Source

git clone https://github.com/Cluster444/agentic.git
cd agentic
bun run build
bun install
bun link  # Makes 'agentic' command available globally

Deploy globally

This will pull all agents/commands into your global ~/.config/opencode/ directory.

agentic pull -g

Deploy to Your Project

This will pull all agents/commands into a local .opencode directory.

cd ~/projects/my-app
agentic pull

Development Workflow

  1. Use the ticket command to work with the agent to build out ticket details
  2. Use the research command to analyze the codebase from the ticket details
  3. Use the plan command to generate an implementation plan for the ticket using the research
  4. Use the execute command to implement the changes
  5. Use the commit command to commit your work
  6. Use the review command to verify the implementation

Between each phase it is important to inspect the output from each phase and ensure that it is actually in alignment with what you want the project do be and the direction it is going. Errors in these files will cascade to the next phase and produce code that is not what you wanted.

In OpenCode, these commands are invoked with a slash: /ticket, /research, /plan, /execute, etc. Most of these commands want the ticket in question that you want to review, exceptions are ticket itself, and commit/review. Ticket you give an actual prompt that describes what you're trying to do, and commit/review are meant to work in the context window that you ran execute in so that it has all of the details of how the process itself went.

Documentation

Getting Started

Core Components

Project Structure

Requirements

Contributing

This project is in active development. Contributions, ideas, and feedback are welcome!

License

MIT License - see LICENSE file for details

About

An agentic workflow tool that provides context engineering support for opencode

Topics

Resources

License

Stars

Watchers

Forks

Packages

No packages published