Guide · Agent skill
How to use the dmux Parallel Agent Workflows skill (dmux-workflows)
Updated · written with AI from the project's own documentation and checked against our catalogue data
The dmux-workflows skill teaches an AI agent how to plan a task as several parallel agent sessions managed by dmux, a tmux-based pane manager. This guide explains what the skill does, how to install it, how to ask for it in plain language, and what you need before you try it. By the end you will be able to split a piece of work into independent panes and know when that split is worth doing.
Quick start
- 1Download and unzip. Download the ZIP from this page and unzip it. Keep the dmux-workflows folder.
- 2Place the skill. Move the folder to ~/.claude/skills/ for personal use or to .claude/skills/ in a project. Alternatively, upload it under Settings → Capabilities → Skills in the Claude apps.
- 3Install dmux and check prerequisites. Install tmux, Node.js with npm, and git, then run npm install -g dmux. Make sure your agent tool has its own account or API key.
- 4Split a task into independent parts. Ask the agent to split the work and name the parts. Confirm that each part touches different files or concerns.
- 5Start panes in dmux. Run dmux, press n for each pane, and enter a bounded prompt.
- 6Review and merge. Press m to read each pane's output, review it, and merge results. Use git worktrees if files overlap.
What the dmux-workflows skill makes the agent do
Agent skills are instruction packs that an AI agent loads when a task matches them. This one covers multi-agent work: when a task can be divided into independent pieces, the agent proposes splitting it across panes, for example research, implementation, tests, and docs.
The skill is mostly guidance and example prompts. It explains how to start each pane with its own prompt, how to merge pane output back, and how to use git worktrees so that agents editing overlapping files do not collide. It does not run automated actions on its own.
- Suggests a split of a task into parallel panes
- Provides example prompts for each pane
- Recommends merging results and using git worktrees for file-conflict-prone work
Install the dmux-workflows skill
Download the ZIP from this page and unzip it. For a personal setup, place the dmux-workflows folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder.
If you use the Claude apps instead, upload the skill under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/dmux-workflows
- Project only: .claude/skills/dmux-workflows inside your project folder
- Claude apps: Settings → Capabilities → Skills, then upload
Install the tools the skill needs
The skill assumes you have tmux, dmux, Node.js and npm, and git. dmux is installed with npm. The source also lists installing tmux with Homebrew on macOS or apt on Linux if it is missing.
You also need an AI agent tool such as Claude Code or Codex, with its own account or API key.
- tmux (brew install tmux on macOS, apt install tmux on Linux)
- dmux, installed globally with npm
- Node.js and npm
- git, needed for worktree workflows
- Claude Code or Codex with its own account or API key
npm install -g dmuxStart a dmux session and create panes
Start dmux from your terminal. Inside dmux, press n to create a new pane and type its prompt. Each pane runs its own agent session. Press m to merge a pane's output back to the main session.
- Press n to create a pane, then type a prompt
- Give each pane a distinct, bounded task
- Press m to read output and merge results
dmux
# Inside dmux, press 'n' and enter a prompt for each pane, e.g.:
# Pane 1: "Implement the auth middleware in src/auth/"
# Pane 2: "Write tests for the user service"
# Pane 3: "Update API documentation"Trigger the skill with natural requests
You do not need special syntax. The skill activates when you ask for parallel work in ordinary words. The source lists phrases such as "run in parallel", "split this work", "use dmux", and "multi-agent".
Describe the parts of the work and say which ones are independent. The clearer the boundaries, the better the suggested split.
- "Split this work across parallel agents using dmux."
- "Run the research and the implementation in parallel panes."
- "Use dmux to review src/api/ for security, performance, and test gaps at once."
Common workflow patterns
The skill includes five patterns. Research plus implementation runs a research pane and an implementation pane together, then feeds the findings into the implementation. Multi-file feature splits schema, API, and UI work into separate panes. Test plus fix runs a watcher pane and a fixer pane. Cross-harness uses different tools, such as Claude Code for security review and Codex for refactoring. Code review pipeline runs security, performance, and coverage reviews of the same folder and merges them into one report.
- Research + Implement
- Multi-File Feature
- Test + Fix Loop
- Cross-Harness
- Code Review Pipeline
Keep panes isolated with git worktrees
When panes may touch the same files, give each one its own worktree and branch. Create the worktrees, start an agent in each folder, and merge the branches when the work is done. The skill also advises reviewing each pane's output before merging.
- Use separate worktrees for file-conflict-prone work
- Review pane output before merging
- Merge branches one at a time when they are finished
git worktree add ../feature-auth feat/auth
git worktree add ../feature-billing feat/billing
# Pane 1: cd ../feature-auth && claude
# Pane 2: cd ../feature-billing && claude
git merge feat/auth
git merge feat/billingWhat people use it for
- Splitting a new feature into schema, API, and UI panes that different agents build at the same time
- Running a security review and a performance review of the same API folder in parallel, then combining them into one report
- Keeping a test watcher running in one pane while a second pane fixes failures it reports
- Using Claude Code for a security review and Codex for a refactor in the same session
Frequently asked questions
How do I install the dmux-workflows skill?
Download the ZIP from this page and unzip it. Place the dmux-workflows folder in ~/.claude/skills/ for personal use, or in a project's .claude/skills/ folder. In the Claude apps, you can upload it under Settings → Capabilities → Skills.
What do I need to use it?
You need tmux, dmux (installed with npm install -g dmux), Node.js and npm, and git for worktree workflows. You also need an AI agent tool such as Claude Code or Codex with its own account or API key.
Is dmux-workflows free?
The skill is listed under the MIT licence. The source does not state prices for dmux or the agent tools you run with it, so check those separately. Each pane is a full agent session that uses tokens.
Does the skill run scripts or use the network?
According to the listing, the skill does not run scripts and does not use the network. It is mainly instructions and example prompts. The dmux and agent tools you install do their own work.
What is the difference between dmux and the Claude Code Task tool?
dmux manages separate agent sessions in tmux panes that you control. The Claude Code Task tool spawns subagents inside a single session. The skill's source lists both and suggests the Task tool for programmatic parallelism within one session.
Why are my dmux panes not responding or producing conflicts?
A pane may be waiting for input, so press m to read its output. Conflicts usually mean panes edited the same files, so give each pane its own git worktree. If token usage is high, run fewer panes.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you already work with a terminal, tmux, and git, and you want to run several Claude Code or Codex sessions on independent parts of a task. Pick something else if you are new to the terminal, your steps depend tightly on each other, or you are watching API costs closely, because every pane is a full agent session.
Download the skill