evo-subagent

When orchestrating complex tasks with large language models, teams often struggle with disorganized subagent roles, messy cross-project knowledge, and the need to reconfigure models with every invocation. The evo-subagent plugin is built to solve these core pain points. It provides stable, predictable routing by binding agent roles to specific provider/model pairs and maintaining isolated, role-specific knowledge bases, ensuring a consistent execution environment.

The plugin's core strength lies in its precise routing and evolutionary mechanism. It defines roles via Markdown files within a project and strictly validates the backend model before a child agent is ever created, preventing misconfiguration upfront. Each agent's knowledge evolves over time, recording verified commands and lessons learned, with priority markers to manage context efficiently. Crucially, all bindings and data are strictly isolated by project workspace, preventing knowledge contamination between different initiatives.

If you are a developer, research team, or AI application builder relying on DeepSeek Harness to construct automated workflows—and you need to manage multiple specialized agents like code reviewers or researchers—this plugin offers a robust foundational architecture. It is ideal for users who prefer a declarative approach to managing complex agent systems and want their agents' capabilities to continuously improve within a project context.