Repository files navigation

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

Agentic is a context engineering tool that assistsOpenCode 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 enhanceOpenCode'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

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

npm install -g agentic-cli# orbun add -g agentic-cli

git clone https://github.com/Cluster444/agentic.gitcd agenticbun run buildbun installbun link# Makes 'agentic' command available globally

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

agentic pull -g

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

cd~/projects/my-appagentic pull

1. Use theticket command to work with the agent to build out ticket details
2. Use theresearch command to analyze the codebase from the ticket details
3. Use theplan command to generate an implementation plan for the ticket using the research
4. Use theexecute command to implement the changes
5. Use thecommit command to commit your work
6. Use thereview 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.

• Usage Guide - Complete guide to using Agentic
• Development Workflow - Detailed workflow phases

• Agentic CLI - Command-line tool reference
• Commands - Available OpenCode commands
• Agents - Specialized AI subagents

• Thoughts Directory - Knowledge management system
• Architecture Docs - System design documentation

• Bun runtime
• OpenCode CLI
• Git

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

MIT License - seeLICENSE file for details