What is opencode-agentic-workflows?

This project provides a ready made structure for running many AI coding agents together in an organized way, built on top of a tool called OpenCode. Instead of having multiple AI agents work independently and unpredictably, this setup uses one main orchestrator agent that a person talks to directly. That orchestrator then delegates specific jobs, like planning, writing code, reviewing, or debugging, to a set of thirteen specialized hidden helper agents behind the scenes. The system is deliberately structured as a strict hierarchy rather than agents freely talking to each other. The orchestrator routes tasks to the right specialist, a planner agent handles breaking work into tracks and stages with readiness checks, and finished work goes through an independent review step using three separate reviewer agents running different underlying models, whose opinions get combined by a fourth agent into one final judgment. There is also a verifier role that checks the work actually meets the goal after review is done. To use it, you clone the repository into your own project, set which AI model you want the system to use in a configuration file, and run the opencode command. You can also assign different models to different specialist roles if you want cheaper models handling simpler tasks like planning and more capable models handling actual coding work. A newer addition is a cautious self improvement system, where the agents can propose small changes to their own instructions and skills, but only within a narrow, low risk area of the project files. Any proposed change first gets reviewed by an analysis agent and then scored by a separate evaluation agent against safety and reliability criteria before it is allowed to be applied, and it is explicitly blocked from ever touching core routing, permissions, or governance files on its own.