agent-team

Agent Team addresses the problem of pushing every task through one overloaded parent agent or inherited subagent setup. It lets you build a team of independent AI agents inside DeepSeek Harness, where each member has its own model, session, context window, permissions, reasoning mode, and tool scope instead of carrying the same broad configuration everywhere.

The plugin is built around explicit division of labor. You can create reusable assistant profiles for roles such as planning Leader, coding member, tester, reviewer, documentation writer, or commit assistant, then choose different providers and models for each one. A Leader coordinates the work by creating tasks, assigning members, tracking progress, and collecting results, while members execute in separate conversations and collaborate through a shared Workspace.

This is useful for developers and teams who want multi-agent workflows in DeepSeek Harness without turning every small task into an expensive general-model job. It fits users who need parallel role-based sessions, controlled permissions, mixed model providers, and shared file collaboration, while keeping planning, execution, and verification cleanly separated.