Hi everyone,
I've been experimenting with multi-agent workflows for coding, and I kept running into the common issue of agents losing context or merging bad code without proper reviews. To solve this, I built OpenCode Nexus, an open-source, shareable multi-agent workflow plugin for OpenCode.
It sets up a structured pipeline with four specialized agents and strong context-preservation defaults so you don't have to micromanage the AI.
Key Features:
4 Specialized Agents:
Orchestrator: Primary controller that delegates tasks.
Implementer: Subagent that writes the code.
Spec Reviewer: Ensures the implementation matches the original requirements.
Code Reviewer: Checks for code quality and best practices.
Context Preservation: Prevents context loss by persisting state in durable files under .opencode/ (plans, context, tasks, and handoffs) instead of relying solely on transient chat history.
Two-Stage Review Process: Enforces spec compliance first, then code quality.
Git-Aware: Uses feature branches for precise review boundaries (git diff main...feature/task-N). Final branch integration is kept as an explicit user decision.
Fully Customizable Models: Models aren't hardcoded. You can use your own API keys, but it defaults to highly capable, low-cost open-source models (like DeepSeek V4 and MiniMax M3).
You can check out the code, installation instructions, and workflow skills here:
๐ GitHub: https://github.com/mohammad154/opencode-nexus I'd love to hear your feedback, feature requests, or contributions! Let me know what you think.