dmux-workflows
by affaan-mdmux-workflows is a guide for orchestrating parallel AI agent sessions with dmux in tmux panes. It helps split research, implementation, testing, and docs across Claude Code, Codex, OpenCode, and similar harnesses so you can manage multi-agent development with less context bottlenecking.
This skill scores 78/100, which means it is a solid listing candidate for directory users who need multi-agent orchestration. It gives a clear activation trigger, concrete pane-based workflows, and enough operational detail to reduce guesswork versus a generic prompt, though users should still verify the external dmux install steps and expect some adoption effort because the repo itself does not include scripts or bundled support files.
- Explicit triggers for parallel or multi-agent work, including user phrases like "run in parallel" and "use dmux"
- Operationally concrete workflow instructions: create panes with 'n', merge with 'm', and route tasks across Claude Code, Codex, OpenCode, Cline, Gemini, and Qwen
- Substantial body content with multiple workflow patterns and constraints, which helps agents choose a usable execution path quickly
- No install command or bundled support files, so users must rely on the external dmux repository and their own setup
- The skill appears workflow-oriented rather than self-contained tooling, so value depends on the agent/harness already supporting dmux-style pane orchestration
Overview of dmux-workflows skill
dmux-workflows is a workflow skill for coordinating multiple AI agent sessions in parallel through dmux, a tmux-based pane manager. It is most useful when one prompt is not enough: you need research, implementation, testing, and documentation handled at the same time, or you want separate agents working on different parts of the same repo without stepping on each other.
What this skill is for
Use the dmux-workflows skill when the job is divide-and-conquer orchestration, not single-shot prompting. The main value is reducing context bottlenecks by splitting a task into independent panes, then merging outputs back into one session.
Who should install dmux-workflows
This skill fits people running Claude Code, Codex, OpenCode, Cline, Gemini, Qwen, or similar harnesses who already work in a terminal and want a repeatable multi-agent setup. It is especially relevant for larger engineering tasks, repo audits, parallel bug fixing, and feature work with clear subproblems.
Key differentiators
The dmux-workflows skill is practical rather than abstract: it shows when to activate dmux, how to branch work across panes, and how to merge results. That makes it more decision-oriented than a generic “parallelize tasks” prompt, but it still depends on you supplying a good task split and a repo-safe workflow.
How to Use dmux-workflows skill
Install dmux-workflows and review the source
Install the skill with npx skills add affaan-m/everything-claude-code --skill dmux-workflows. Then open SKILL.md first, since it contains the actual operating pattern, followed by any linked repo context such as README.md, AGENTS.md, metadata.json, or supporting folders if they exist. For this skill, the repo surface is small, so SKILL.md is the primary source of truth.
Turn a rough goal into a usable dmux prompt
The best dmux-workflows usage starts with a task split, not a vague request. Instead of “fix this feature,” give the skill something like: “Use dmux to split this into research, implementation, tests, and docs. Keep the work isolated per pane and merge only after each branch has a concrete result.” Clear subtask boundaries improve output quality because each agent can act independently.
Suggested workflow for parallel execution
A typical dmux-workflows guide looks like this: identify 2–4 independent tracks, assign each pane one track, keep prompts specific to files or outcomes, then merge only after the results are checked for conflicts and overlap. Good fits include “research the bug cause,” “patch the code,” and “write regression tests” in separate panes. Poor fits are tasks that require constant back-and-forth or a single shared context.
Practical input tips
Name the target repo, the exact files or subsystem, the success criteria, and the constraints before invoking the skill. If you want the skill to behave well in a dmux-workflows for Multi-Agent Systems setup, include what each agent should not touch, what can run in parallel, and what the final merge should preserve. The stronger the boundary, the less the panes will duplicate work.
dmux-workflows skill FAQ
Is dmux-workflows only for advanced users?
No. The skill is beginner-friendly if you are comfortable using the terminal and can describe a task in parts. You do not need to design a full agent framework; you mainly need to provide a task that can be split into meaningful chunks.
When should I not use dmux-workflows?
Do not use it for tiny edits, single-file changes, or tasks where one model pass is enough. If the work depends on one chain of reasoning, parallel panes can slow you down instead of helping.
How is this different from a normal prompt?
A normal prompt asks one agent to do everything sequentially. The dmux-workflows skill is about orchestration: it helps you assign separate goals to separate panes, which is useful when the work can be isolated and merged later without losing correctness.
Does it fit every agent harness?
It is designed around dmux’s tmux-pane model, but the workflow ideas transfer to other harnesses that can run multiple sessions. If your environment cannot manage concurrent panes cleanly, the skill is less useful.
How to Improve dmux-workflows skill
Give cleaner task splits
The biggest quality gain comes from better decomposition. State which branch is research, which is implementation, which is testing, and which files are owned by each pane. Avoid asking two panes to edit the same file unless you have a clear merge plan.
Provide merge criteria up front
Say what “done” means before the panes start: passing tests, no API changes, docs updated, or a specific bug reproduced and fixed. In dmux-workflows install workflows, this reduces ambiguity and makes the final m merge step more reliable.
Watch for duplicate effort and context drift
The most common failure mode is two panes solving the same problem from slightly different angles. Prevent that by giving each pane a narrow scope, a shared source of truth, and a short summary format for results. If the first pass is noisy, rerun with stricter boundaries and smaller prompts.
Iterate with sharper prompts
After the first run, refine the task by adding the exact subsystem, file paths, or constraints that mattered most. For dmux-workflows, better inputs usually beat bigger prompts: one clear objective per pane, one expected artifact per pane, and one merge rule for the whole session.