dmux-workflows

Orchestrates parallel AI agent sessions using dmux tmux pane management.

Updated Mar 26, 2026
One-click install
npx skills add https://github.com/inuishan/PET --skill dmux-workflows-inuishan
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: dmux-workflows
Source: https://github.com/inuishan/PET/tree/main/.codex/skills/dmux-workflows
Command: npx skills add https://github.com/inuishan/PET --skill dmux-workflows-inuishan

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires dmux, tmux.

What problem does it solve? Coordinating multiple AI agent sessions manually is error-prone and slow when a task can be split into independent tracks. This Skill provides proven patterns for running parallel agent workflows with dmux, a tmux pane manager for AI agent harnesses. ## Core Features & Use Cases - Parallel Agent Orchestration: Create and manage multiple agent panes across Claude Code, Codex, OpenCode, Cline, Gemini, and Qwen from a single dmux session. - Workflow Patterns: Apply five ready-made patterns including research-plus-implement, multi-file feature splits, test-and-fix loops, cross-harness delegation, and parallel code review pipelines. - Git Worktree Isolation: Prevent merge conflicts by running agents in separate git worktrees when tasks touch overlapping files. - Use Case: Split a billing feature into three panes — one for database migrations, one for API endpoints, and one for UI components — then merge the results and integrate in the main pane. ## Quick Start Ask the agent to split the current task into independent subtasks and run them in parallel dmux panes, then merge the results back into the main session.

Frequently Asked Questions about dmux-workflows

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run multiple AI agents in parallel with dmux?▼

Install dmux with npm install -g dmux, start a session by running dmux, then press n to create a new pane and type a prompt for each agent. Press m to merge a pane's output back into the main session when it finishes.

What tasks should I parallelize across agent panes?▼

Parallelize only independent tasks with clear boundaries, such as separate files, research versus implementation, or distinct review perspectives. Avoid splitting tasks that depend on each other's output, and keep total panes under five or six to control token usage.

Which AI agent harnesses does dmux support?▼

dmux supports Claude Code, Codex, OpenCode, Cline, Gemini, and Qwen. Each pane runs its own full agent session, so you can mix harnesses in a cross-harness workflow for different task types.

How do I avoid merge conflicts with parallel agents?▼

Use git worktrees to isolate file changes per pane by running git worktree add for each branch, then launch agents inside their own worktree directories. Merge the branches together after all panes complete their work.

Why is a dmux pane not responding?▼

A pane may be waiting for input from its agent session rather than being stuck. Use the m key to read the pane's output and check its state before assuming it has failed.

What are the limitations of parallel agent workflows?▼

Each pane consumes API tokens as a full agent session, so costs scale with pane count. Parallelism also fails for dependent tasks, and overlapping file edits cause merge conflicts unless worktrees are used.