dmux-workflows

The dmux-workflows skill provides a structured approach to orchestrating multiple AI agent sessions in parallel using dmux, a tmux-based pane manager designed specifically for coordinating agent harnesses. It addresses the challenge of working with complex development tasks that benefit from parallel execution, where a single AI session may be too slow or limited in scope. By enabling multiple agents to run concurrently in isolated panes, it allows users to divide large problems into smaller, independently executable units while maintaining a coordinated workflow.

This skill leverages dmux’s pane-based interface to manage multiple AI tools such as Claude Code, Codex, OpenCode, Cline, Gemini, and Qwen. Users can spawn new panes, assign focused prompts, and later merge outputs back into a central session. It supports structured workflow patterns including research-and-implement separation, multi-file feature development, test-and-fix loops, cross-harness collaboration, and parallel code review pipelines. It also integrates with development best practices such as using git worktrees for isolation when working on overlapping codebases, ensuring safe parallel execution without file conflicts.

dmux-workflows is intended for developers, AI engineers, and teams building complex software systems who want to accelerate delivery through parallel agent coordination. It is particularly useful for tasks like feature development, large-scale refactoring, test generation, documentation writing, and security or performance reviews. By enabling divide-and-conquer execution across multiple AI agents, it improves productivity, reduces iteration time, and provides a scalable structure for managing multi-agent AI development workflows.