📄️ Agents
Agents are specialized AI workers with their own identity, system prompt, tool access, skills, model overrides, and connection mode. The UI writes the same files described here, so anything created in Settings > Agents can also be reviewed and versioned on disk.
📄️ Skills
Skills are portable, reusable agent capabilities defined as instruction files. They extend what your agents know without code changes — and they work across DotAgents, Claude Code, Cursor, and any tool that reads the .agents/ directory.
📄️ Knowledge & Notes
Knowledge gives your agents durable, local context across sessions. In the .agents protocol, the mixed-content container is knowledge, markdown artifacts are notes, and the small runtime-injected subset are working notes.
📄️ Multi-Agent Delegation
DotAgents supports multi-agent orchestration through the Agent Client Protocol (ACP). Your main agent can delegate tasks to specialized sub-agents — each running as a separate process with its own capabilities.