Parallel development in tmux* with Git worktrees

Parallel development in tmux with git worktrees*

📖 Documentation ·

Install·

Quick start·

Commands·

Changelog

Giga opinionated zero-friction workflow tool for managing git worktrees and tmux windows as isolated development environments. Perfect for running multiple AI agents in parallel without conflict.

Philosophy: Build on tools you already use. tmux/zellij/kitty/etc. for windowing, git for worktrees, your agent for coding - workmux ties them together.

• Also supports kitty, WezTerm, and Zellij as alternative backends.

📖 New to workmux? Read the introduction blog post for a quick overview.

Tip

I'm building aven, a local-first task manager for power users and agents who live in the terminal.

Parallel workflows. Work on multiple features the same time, each with its own AI agent. No stashing, no branch switching, no conflicts.

One window per task. A natural mental model. Each has its own terminal state, editor session, dev server, and AI agent. Context switching is switching tabs.

Automated setup. New worktrees start broken (no .env

, no node_modules

, no dev server). workmux can copy config files, symlink dependencies, and run install commands on creation.

One-command cleanup. workmux merge

handles the full lifecycle: merge the branch, delete the worktree, close the tmux window, remove the local branch.

Terminal workflow. Build on your terminal setup instead of yet another agentic GUI that won't exist next year. If you don't have one yet, tmux might be worth picking up.

New to worktrees? See Why git worktrees?

• Create git worktrees with matching tmux windows in a single command ( add

) - Merge branches and clean up everything (worktree, tmux window, branches) in one command ( merge

) Dashboardfor monitoring agents, reviewing changes, and sending commandsSidebarfor a persistent, at-a-glance view of all agents across tmux windowsDelegate tasks to worktree agentswith the/worktree

skillDisplay agent status in tmux window names- Automatically set up your preferred tmux pane layout (editor, shell, watchers, etc.)

• Run post-creation hooks (install dependencies, setup database, etc.)
• Copy or symlink configuration files ( .env

,node_modules

) into new worktrees Sandbox agentsin containers or VMs for enhanced securityAutomatic branch name generationfrom prompts using LLM- Shell completions

"I've been using (and loving) workmux which brings together tmux, git worktrees, and CLI agents into an opinionated workflow."

— @Coolin96[🔗]

"Thank you so much for your work with workmux! It's a tool I've been wanting to exist for a long time."

— @rstacruz[🔗]

"It's become my daily driver - the perfect level of abstraction over tmux + git, without getting in the way or obscuring the underlying tooling."

— @cisaacstern[🔗]

"I have to mention workmux at every opportunity because it's the perfect glue between worktrees, agents and tmux windows."

— @dedbrizz[🔗]

curl -fsSL https://raw.githubusercontent.com/raine/workmux/main/scripts/install.sh | bash
brew install raine/workmux/workmux

Other methods (Cargo, mise, Nix) #

Cargo (requires rustup):

cargo install workmux

mise:

mise use -g cargo:raine/workmux

Nix (flake and home-manager setup):

nix profile install github:raine/workmux

For manual installation, see pre-built binaries.

Note

workmux requires a terminal multiplexer. Make sure you have tmux (or WezTerm / Kitty / Zellij) installed and running before you start. See My tmux setup if you need a starting point.

Initialize configuration (optional):

workmux init

This creates a

.workmux.yaml

file to customize your workflow (pane layouts, setup commands, file operations, etc.). workmux works out of the box with sensible defaults, so this step is optional. - Create a new worktree and tmux window:

workmux add new-feature

This will:

• Create a git worktree at <project_root>/../<project_name>__worktrees/new-feature

• Copy config files and symlink dependencies (if configured) - Run any setup commandspost_create

• Create a tmux window named wm-new-feature

(the prefix is configurable) - Set up your configured or the default tmux pane layout

• Automatically switch your tmux client to the new window

• Create a git worktree at

Do your thing - Finish and clean up****Local merge: Runworkmux merge

to merge into the base branch and clean up in one step.PR workflow: Push and open a PR. After it's merged, runworkmux remove

to clean up.

workmux uses a two-level configuration system:

Global(~/.config/workmux/config.yaml

): Personal defaults for all projectsProject(.workmux.yaml

): Project-specific overrides

Project settings override global settings. When you run workmux from a subdirectory, it walks upward to find the nearest .workmux.yaml

, allowing nested configs for monorepos. See the Monorepos guide for details. For post_create

and file operation lists (files.copy

, files.symlink

), you can use "<global>"

to include global values alongside project-specific ones. Other settings like panes

are replaced entirely when defined in the project config.

~/.config/workmux/config.yaml

:

nerdfont: true # Enable nerdfont icons (prompted on first run)
merge_strategy: rebase # Make workmux merge do rebase by default
merge_keep: true # Keep worktree, window, and branch after merge by default
agent: claude

panes:
  - command: <agent> # Start the configured agent (e.g., claude)
    focus: true
  - split: horizontal # Second pane with default shell

.workmux.yaml

:

post_create:
  - '<global>'
  - mise use

files:
  symlink:
    - '<global>' # Include global symlinks (node_modules)
    - .pnpm-store # Add project-specific symlink

panes:
  - command: pnpm install
    focus: true
  - command: <agent>
    split: horizontal
  - command: pnpm run dev
    split: vertical

For a real-world example, see workmux's own .workmux.yaml.

Most options have sensible defaults. You only need to configure what you want to customize.

Option	Description	Default
main_branch
Branch to merge into	Auto-detected
base_branch
Default base branch for new worktrees	Current branch
worktree_dir
Directory for worktrees (absolute or relative). Supports ~ and {project} .
<project>__worktrees/
window_prefix
Prefix for tmux window/session names	wm-
mode
Tmux mode (window or session )
window
window_placement
New tmux window placement (after_current or rightmost )
after_current
agent
Default agent for <agent> placeholder
claude
agents
Named agent commands (

{}

merge_strategy

merge

, rebase

, squash

)merge

merge_keep

workmux merge

by defaultfalse

theme

custom colors)default

(auto dark/light)| Option | Description | Default | |---|---|---| worktree_naming | How to derive names from branches | full | worktree_prefix | Prefix for worktree directories and windows | none |

worktree_naming

strategies:

full

: Use the full branch name (slashes become dashes)basename

: Use only the part after the last/

(e.g.,prj-123/feature

→feature

)

Define your tmux pane layout with the panes

array. For multiple windows in session mode, use windows instead (they are mutually exclusive).

panes:
  - command: <agent>
    focus: true
  - command: npm run dev
    split: horizontal
    size: 15

Each pane supports:

Option	Description	Default
command
Command to run (see

focus

false

zoom

focus: true

)false

split

horizontal

or vertical

)size

percentage

<agent>

: resolves to the configured agent (fromagent

config or--agent

flag)

Built-in agents (claude

, gemini

, codex

, opencode

, kiro-cli

, vibe

, pi

, omp

) are auto-detected when used as literal commands and receive prompt injection automatically, without needing the <agent>

placeholder or a matching agent

config:

panes:
  - command: 'claude --dangerously-skip-permissions'
    focus: true
  - command: 'codex --yolo'
    split: vertical

Each agent receives the prompt (via -p

/-P

/-e

) using the correct format for that agent. Auto-detection matches the executable name regardless of flags or path.

Define reusable pane arrangements in the layouts

map and select one at add-time with -l/--layout

:

layouts:
  design:
    panes:
      - command: <agent>
        focus: true
      - command: <agent:codex>
        split: vertical
  review:
    panes:
      - command: <agent>
workmux add my-feature -l design

When -l

is used, the layout's panes

replace the top-level panes

for that worktree. All other config (hooks, files, agent, etc.) comes from the top-level as usual. The -l

flag cannot be combined with --agent

.

New worktrees are clean checkouts with no gitignored files (.env

, node_modules

, etc.). Use files

to automatically copy or symlink what each worktree needs:

files:
  copy:
    - .env
  symlink:
    - .next/cache # Share build cache across worktrees

Both copy

and symlink

accept glob patterns.

To re-apply file operations to an existing worktree (e.g., after updating the config), run workmux sync-files

from inside the worktree. Use --all

to sync all worktrees at once.

Run commands at specific points in the worktree lifecycle, such as installing dependencies or running database migrations. All hooks run with the worktree directory as the working directory (or the nested config directory for nested configs) and receive environment variables: WM_HANDLE

, WM_WORKTREE_PATH

, WM_PROJECT_ROOT

, WM_CONFIG_DIR

.

WM_CONFIG_DIR

points to the directory containing the .workmux.yaml

that was used, which may differ from WM_WORKTREE_PATH

when using nested configs.

Hook	When it runs	Additional env vars
post_create
After worktree creation, before tmux window opens	—
pre_merge
Before merging (aborts on failure)	WM_BRANCH_NAME , WM_TARGET_BRANCH
pre_remove
Before worktree removal (aborts on failure)	—

Example:

post_create:
  - direnv allow

pre_merge:
  - just check

Customize the icons shown in tmux window names:

status_icons:
  working: '🤖' # Agent is processing
  waiting: '💬' # Agent needs input (auto-clears on focus)
  done: '✅' # Agent finished (auto-clears on focus)

Agents in "working" status that produce no pane output for 10 seconds are automatically detected as interrupted.

Set status_format: false

to disable automatic tmux format modification

• Worktrees are created in <project>__worktrees

as a sibling directory to your project by default - If no panes

configuration is defined, workmux provides opinionated defaults:- For projects with a CLAUDE.md

file: Opens the configured agent (seeagent

option) in the first pane, defaulting toclaude

if none is set. - For all other projects: Opens your default shell.

• Both configurations include a second pane split horizontally

• For projects with a post_create

commands are optional and only run if you configure them

Use the panes

configuration to automate environment setup. Unlike post_create

hooks which must finish before the tmux window opens, pane commands execute immediately within the new window.

This can be used for:

Installing dependencies: Runnpm install

orcargo build

in a focused pane to monitor progress.Starting services: Launch dev servers, database containers, or file watchers automatically.** Running agents**: Initialize AI agents with specific context.

Since these run in standard tmux panes, you can interact with them (check logs, restart servers) just like a normal terminal session.

Running dependency installation (like pnpm install

) in a pane command rather than post_create

has a key advantage: you get immediate access to the tmux window while installation runs in the background. With post_create

, you'd have to wait for the install to complete before the window even opens. This also means AI agents can start working immediately in their pane while dependencies install in parallel.

panes:
  - command: pnpm install && pnpm run dev

  - command: <agent>
    split: horizontal
    focus: true

Here's how workmux organizes your worktrees by default:

~/projects/
├── my-project/               <-- Main project directory
│   ├── src/
│   ├── package.json
│   └── .workmux.yaml
│
└── my-project__worktrees/    <-- Worktrees created by workmux
    ├── feature-A/            <-- Isolated workspace for 'feature-A' branch
    │   ├── src/
    │   └── package.json
    │
    └── bugfix-B/             <-- Isolated workspace for 'bugfix-B' branch
        ├── src/
        └── package.json

Each worktree is a separate working directory for a different branch, all sharing the same git repository. This allows you to work on multiple branches simultaneously without conflicts.

You can customize the worktree directory location using the worktree_dir

configuration option (see Configuration options). The value supports ~

for the home directory and a {project}

placeholder that resolves to the main worktree's directory name. This lets a single global config namespace every repo's worktrees under one root, e.g. worktree_dir: ~/.workmux/{project}

.

For faster typing, alias workmux

to wm

:

alias wm='workmux'

• Create a new worktree and tmux windowadd

• Merge a branch and clean up everythingmerge

• Rebase a worktree branch onto its base branchrebase

• Remove worktrees without mergingremove

• List all worktrees with statuslist

• Open a tmux window for an existing worktreeopen

• Close a worktree's tmux window (keeps worktree)close

• Restore worktree windows after a crashresurrect

• Get the filesystem path of a worktreepath

• Show TUI dashboard of all active agentsdashboard

• Toggle a compact agent status sidebar in tmuxsidebar

• Exit tracked agent processes older than a thresholdreap-agents

• Edit the global configuration fileconfig edit

• Generate configuration fileinit

• Manage sandbox backends (container/Lima)sandbox

• Clean up stale Claude Code entriesclaude prune

• Generate shell completionscompletions

• Show detailed documentationdocs

Creates a new git worktree with a matching tmux window and switches you to it immediately. If the branch doesn't exist, it will be created automatically.

<branch-name>

: Name of the branch to create or switch to, a remote branch reference (e.g.,origin/feature-branch

), or a GitHub fork reference (e.g.,user:branch

). Remote and fork references are automatically fetched and create a local branch with the derived name. Fork references derive the local branch asuser-branch

(e.g.,someuser:feature

creates local branchsomeuser-feature

). Optional when using--pr

.

--base <branch|commit|tag>

: Specify a base branch, commit, or tag to branch from when creating a new branch. Overridesbase_branch

config. Defaults tobase_branch

from config, then the currently checked out branch.--pr <number>

: Checkout a GitHub pull request by its number into a new worktree.- Requires the gh

command-line tool to be installed and authenticated. - The local branch name defaults to the PR's head branch name, but can be overridden (e.g., workmux add custom-name --pr 123

). - If that local branch already exists and has no worktree, it is reused.

• Requires the -A, --auto-name

: Generate branch name from prompt using LLM. SeeAutomatic branch name generation.--name <name>

: Override the worktree directory and tmux window name. By default, these are derived from the branch name (slugified). Cannot be used with multi-worktree generation (--count

,--foreach

, or multiple--agent

).-b, --background

: Create the tmux window in the background without switching to it. Useful with--prompt-editor

.-w, --with-changes

: Move uncommitted changes from the current worktree to the new worktree, then reset the original worktree to a clean state. Useful when you've started working on main and want to move your branches to a new worktree.--patch

: Interactively select which changes to move (requires--with-changes

). Opens an interactive prompt for selecting hunks to stash.-u, --include-untracked

: Also move untracked files (requires--with-changes

). By default, only staged and modified tracked files are moved.-p, --prompt <text>

: Provide an inline prompt that will be automatically passed to AI agent panes.-P, --prompt-file <path>

: Provide a path to a file whose contents will be used as the prompt.-e, --prompt-editor

: Open your$EDITOR

(or$VISUAL

) to write the prompt interactively.--prompt-file-only

: Write the prompt file to the worktree without injecting it into agent commands. No agent pane is required. Useful when your editor has an embedded agent that reads.workmux/PROMPT-*.md

directly.-l, --layout <name>

: Use a named pane layout from config instead of the default panes. Cannot be combined with--agent

.-a, --agent <name>

: The agent(s) to use for the worktree(s). Can be specified multiple times to generate a worktree for each agent. Overrides theagent

from your config file.-W, --wait

: Block until the created tmux window is closed. Useful for scripting when you want to wait for an agent to complete its work. The agent can signal completion by runningworkmux remove --keep-branch

.-o, --open-if-exists

: If a worktree for the branch already exists, open it instead of failing. Similar totmux new-session -A

. Useful when you don't know or care whether the worktree already exists.-s, --session

: Create a tmux session instead of a window. SeeSession modefor details.--config <path>

: Use an alternate config file for this invocation. Still merges with global config.--fork

: Fork the last conversation from the current worktree into the new one. The agent resumes with the forked conversation context. Use--fork=<session-id>

to fork a specific session (prefix matching supported). Currently supports Claude Code.

These options allow you to skip expensive setup steps when they're not needed (e.g., for documentation-only changes):

-H, --no-hooks

: Skip runningpost_create

commands-F, --no-file-ops

: Skip file copy/symlink operations (e.g., skip linkingnode_modules

)-C, --no-pane-cmds

: Skip executing pane commands (panes open with plain shells instead)

• Determines the handle for the worktree by slugifying the branch name (e.g.,feature/auth

becomesfeature-auth

). This can be overridden with the--name

flag. - Creates a git worktree at <worktree_dir>/<handle>

(theworktree_dir

is configurable and defaults to a sibling directory of your project) - Runs any configured file operations (copy/symlink)

• Executes post_create

commands if defined (runs before the tmux window opens, so keep them fast) - Creates a new tmux window named <window_prefix><handle>

(e.g.,wm-feature-auth

withwindow_prefix: wm-

) - Sets up your configured tmux pane layout

• Automatically switches your tmux client to the new window

workmux add user-auth

workmux add existing-work

workmux add hotfix --base production

workmux add origin/user-auth-pr

workmux add origin/feature/foo

workmux add feature/parallel-task --background

workmux add feature/long-descriptive-branch-name --name short

workmux add my-feature -o
workmux add --pr 123

workmux add fix/api-bug --pr 456

workmux add someuser:feature-branch
workmux add feature/new-thing --with-changes -u

workmux add fix/bug --with-changes

workmux add feature/partial --with-changes --patch
workmux add feature/ai --prompt "Implement user authentication with OAuth"

workmux add feature/testing -a gemini

workmux add feature/refactor --prompt-file task-description.md

workmux add feature/new-api --prompt-editor

workmux add feature/task -P task.md --prompt-file-only
workmux add docs-update --no-hooks --no-file-ops --no-pane-cmds

workmux add quick-fix --no-file-ops
workmux add feature/api --wait -p "Implement the REST API, then run: workmux remove --keep-branch"

for task in task1.md task2.md task3.md; do
  workmux add "task-$(basename $task .md)" --wait -P "$task"
done

When you provide a prompt via --prompt

, --prompt-file

, or --prompt-editor

, workmux automatically injects the prompt into panes running the configured agent command (e.g., claude

, codex

, opencode

, gemini

, kiro-cli

, vibe

, pi

, omp

, or whatever you've set via the agent

config or --agent

flag) without requiring any .workmux.yaml

changes:

• Panes with a command matching the configured agent are automatically started with the given prompt.
• You can keep your .workmux.yaml

pane configuration simple (e.g.,panes: [{ command: "<agent>" }]

) and let workmux handle prompt injection at runtime.

This means you can launch AI agents with task-specific prompts without modifying your project configuration for each task.

If your editor has an embedded agent (e.g., neovim with an agent plugin), use --prompt-file-only

to write the prompt to .workmux/PROMPT-<branch>.md

without requiring an agent pane. Your editor can then detect and consume the file on startup. This can also be set permanently in config with prompt_file_only: true

.

The --auto-name

(-A

) flag generates a branch name from your prompt using an LLM. The tool used depends on your configuration:

auto_name.command

is set: uses that command as-isconfig.agent

is a known agent (claude

,gemini

,codex

,opencode

,kiro-cli

,vibe

,pi

,omp

): uses the agent's CLI with a fast/cheap model- Neither: falls back to the CLI toolllm

workmux add -A

workmux add -A -p "Add OAuth authentication"

workmux add -A -P task-spec.md

When agent

is configured (e.g., agent: claude

), workmux automatically uses that agent's CLI for branch naming. No additional setup is required beyond having the agent installed.

If no agent is configured and no auto_name.command

is set, workmux uses the llm

CLI tool:

pipx install llm

Configure a model (e.g., OpenAI):

llm keys set openai
llm install llm-ollama

If you set auto_name.command

, llm

is not required.

When an agent is configured, these commands are used automatically:

Agent	Auto-name command
claude
claude --model haiku -p
gemini
gemini -m gemini-2.5-flash-lite -p
codex
codex exec --config model_reasoning_effort="low" -m gpt-5.1-codex-mini
opencode
opencode run
kiro-cli
kiro-cli chat --no-interactive
pi
pi -p
omp
omp -p

To override back to llm

when an agent is configured, set auto_name.command: "llm"

.

Optionally configure auto-name behavior in .workmux.yaml

:

auto_name:
  model: 'gemini-2.5-flash-lite'
  background: true # Always run in background when using --auto-name
  system_prompt: |
    Generate a concise git branch name based on the task description.

    Rules:
    - Use kebab-case (lowercase with hyphens)
    - Keep it short: 1-3 words, max 4 if necessary
    - Focus on the core task/feature, not implementation details
    - No prefixes like feat/, fix/, chore/

    Examples of good branch names:
    - "Add dark mode toggle" → dark-mode
    - "Fix the search results not showing" → fix-search
    - "Refactor the authentication module" → auth-refactor
    - "Add CSV export to reports" → export-csv
    - "Shell completion is broken" → shell-completion

    Output ONLY the branch name, nothing else.

To use a specific tool, set auto_name.command

. The command string is split into program and arguments, and the composed prompt is piped via stdin.

auto_name:
  command: 'claude -p'

auto_name:
  command: 'llm'

Option	Description	Default
command
Command for branch name generation (overrides agent profile)	Agent profile or llm CLI
model
LLM model to use with the llm CLI (ignored when command set)
llm 's default
background
Always run in background when using --auto-name
false
system_prompt
Custom system prompt for branch name generation	Built-in prompt

Recommended models for fast, cheap branch name generation (with llm

):

gemini-2.5-flash-lite

(recommended)gpt-5-nano

workmux can generate multiple worktrees from a single add

command, which is ideal for running parallel experiments or delegating tasks to multiple AI agents. This is controlled by four mutually exclusive modes:

• ( -a

,--agent

): Create a worktree for each specified agent. - ( -n

,--count

): Create a specific number of worktrees. - ( --foreach

): Create worktrees based on a matrix of variables. stdin: Pipe input lines to create worktrees with templated prompts.

When using any of these modes, branch names are generated from a template, and prompts are templated with variables. Single-worktree prompts are passed through literally, so common syntax like GitHub Actions ${{ ... }}

does not need to be escaped.

-a, --agent <name>

: When used multiple times, creates one worktree for each agent.-n, --count <number>

: Creates<number>

worktree instances. Can be combined with a single--agent

flag to apply that agent to all instances.--foreach <matrix>

: Creates worktrees from a variable matrix string. The format is"var1:valA,valB;var2:valX,valY"

. All value lists must have the same length. Values are paired by index position (zip, not Cartesian product): the first value of each variable goes together, the second with the second, etc.--branch-template <template>

: AMiniJinja(Jinja2-compatible) template for generating branch names.- Available variables: {{ base_name }}

,{{ agent }}

,{{ num }}

,{{ index }}

,{{ input }}

(stdin), and any variables from--foreach

. - Default: {{ base_name }}{% if agent %}-{{ agent | slugify }}{% endif %}{% for key, value in foreach_vars %}-{{ value | slugify }}{% endfor %}{% if num %}-{{ num }}{% endif %}

• Available variables: --max-concurrent <number>

: Limits how many worktrees run simultaneously. When set, workmux creates up to<number>

worktrees, then waits for any window to close before starting the next. Requires agents to close windows when done (e.g., via prompt instruction to runworkmux remove --keep-branch

).

When generating multiple worktrees, any prompt provided via -p

, -P

, or -e

is treated as a MiniJinja template. You can use variables from your generation mode to create unique prompts for each agent or instance. For ordinary single-worktree add

commands, prompt text is not templated.

Instead of passing --foreach

on the command line, you can specify the variable matrix directly in your prompt file using YAML frontmatter. This is more convenient for complex matrices and keeps the variables close to the prompt that uses them.

Format:

Create a prompt file with YAML frontmatter at the top, separated by ---

:

Example 1: mobile-task.md

---
foreach:
  platform: [iOS, Android]
  lang: [swift, kotlin]
---

Build a {{ platform }} app using {{ lang }}. Implement user authentication and
data persistence.
workmux add mobile-app --prompt-file mobile-task.md

Example 2: agent-task.md

(using agent

as a foreach variable)

---
foreach:
  agent: [claude, gemini]
---

Implement the dashboard refactor using your preferred approach.
workmux add refactor --prompt-file agent-task.md

Behavior:

• Variables from the frontmatter are available in both the prompt template and the branch name template
• All value lists must have the same length, and values are paired by index position (same zip behavior as --foreach

) - CLI --foreach

overrides frontmatter with a warning if both are present - Works with both --prompt-file

and--prompt-editor

You can pipe input lines to workmux add

to create multiple worktrees. Each line becomes available as the {{ input }}

template variable in your prompt. This is useful for batch-processing tasks from external sources.

Plain text: Each line becomes {{ input }}

echo -e "api\nauth\ndatabase" | workmux add refactor -P task.md

JSON lines: Each key becomes a template variable

gh repo list --json url,name --jq -c '.[]' | workmux add analyze \
  --branch-template '{{ base_name }}-{{ name }}' \
  -P prompt.md

This lets you structure data upstream with jq

and use meaningful branch names while keeping the full URL available in your prompt.

Behavior:

• Empty lines and whitespace-only lines are filtered out
• Stdin input cannot be combined with --foreach

(mutually exclusive) - JSON objects (lines starting with {

) are parsed and each key becomes a variable {{ input }}

always contains the raw line- If JSON contains an input

key, it overwrites the raw line value

workmux add my-feature -a claude -a gemini -p "Implement the new search API integration"

workmux add my-feature -n 2 -p "Implement task #{{ num }} in TASKS.md"

workmux add my-feature --foreach "platform:iOS,Android" -p "Build for {{ platform }}"

workmux add my-feature --foreach "agent:claude,gemini" -p "Implement the dashboard refactor"

workmux add testing --prompt-file task.md

echo -e "auth\npayments\napi" | workmux add review -A -P review.md

Combine stdin input, prompt templating, and concurrency limits to create a worker pool that processes items from an external command.

Example: Generate test scaffolding for untested files

find src/utils -name "*.ts" ! -name "*.test.ts" | \
  workmux add add-tests \
    --branch-template '{{ base_name }}-{{ index }}' \
    --prompt-file generate-tests.md \
    --max-concurrent 3 \
    --background

find ...

lists files without tests (one per line) piped to stdin--branch-template

uses{{ index }}

for unique branch names--prompt-file

uses{{ input }}

to pass each file path to the agent--max-concurrent 3

limits parallel agents to avoid rate limits--background

runs without switching focus

Merges a branch into a target branch (main by default) and automatically cleans up all associated resources (worktree, tmux window, and local branch).

Tip

** merge vs remove**: Use

merge

when you want to merge directly without a pull request. If your workflow uses pull requests, use to clean up after your PR is merged on the remote.

remove

[branch-name]

: Optional name of the branch to merge. If omitted, automatically detects the current branch from the worktree you're in.

--into <branch>

: Merge into the specified branch instead of the main branch. Useful for stacked PRs, git-flow workflows, or merging subtasks into a parent feature branch. If the target branch has its own worktree, the merge happens there; otherwise, the main worktree is used.--ignore-uncommitted

: Commit any staged changes before merging without opening an editor--keep

,-k

: Keep the worktree, window, and branch after merging (skip cleanup). Useful when you want to verify the merge before cleaning up.--cleanup

: Clean up after merging, overridingmerge_keep: true

.--notification

: Show a system notification on successful merge. Useful when delegating merge to an AI agent and you want to be notified when it completes.

By default, workmux merge

performs a standard merge commit (configurable via merge_strategy

). You can override the configured behavior with these mutually exclusive flags:

--rebase

: Rebase the feature branch onto the target before merging (creates a linear history via fast-forward merge). If conflicts occur, you'll need to resolve them manually in the worktree and rungit rebase --continue

.--squash

: Squash all commits from the feature branch into a single commit on the target. You'll be prompted to provide a commit message in your editor.

If you don't want to have merge commits in your main branch, use the rebase

merge strategy, which does --rebase

by default.

merge_strategy: rebase

To keep the worktree, window, and branch after every merge unless overridden, set:

merge_keep: true

Use workmux merge --cleanup

to clean up for a single merge when this default is enabled.

• Determines which branch to merge (specified branch or current branch if omitted)
• Determines the target branch ( --into

or main branch from config) - Checks for uncommitted changes (errors if found, unless --ignore-uncommitted

is used) - Commits staged changes if present (unless --ignore-uncommitted

is used) - Merges your branch into the target using the selected strategy (default: merge commit)

• Deletes the tmux window unless keep behavior is enabled via --keep

ormerge_keep: true

• Removes the worktree unless keep behavior is enabled via --keep

ormerge_keep: true

• Deletes the local branch unless keep behavior is enabled via --keep

ormerge_keep: true

When you're done working in a worktree, simply run workmux merge

from within that worktree's tmux window. The command will automatically detect which branch you're on, merge it into main, and close the current window as part of cleanup.

workmux merge user-auth

workmux merge

workmux merge user-auth --rebase

workmux merge user-auth --squash

workmux merge user-auth --keep
workmux remove user-auth  # clean up later when ready

workmux merge feature/subtask --into feature/parent

Rebases a worktree branch onto its saved base branch. If the branch does not have a saved local base branch, workmux rebases onto the configured main branch.

[name]

: Optional worktree name or branch. If omitted, workmux detects the current worktree from the current directory.

• Determines which worktree branch to rebase
• Reads the saved base branch recorded when the worktree was created
• Falls back to the configured main branch when the saved base is not a local branch
• Runs git rebase <base>

inside the worktree - Leaves the worktree, window, and branch in place

workmux rebase user-auth
workmux rebase

Removes worktrees, tmux windows, and branches without merging (unless you keep the branches). Useful for abandoning work or cleaning up experimental branches. Supports removing multiple worktrees in a single command.

[name]...

: One or more worktree names (the directory names). Defaults to current directory name if omitted.

--all

: Remove all worktrees at once (except the main worktree). Prompts for confirmation unless--force

is used. Safely skips worktrees with uncommitted changes or unmerged commits.--gone

: Remove worktrees whose upstream remote branch has been deleted (e.g., after a PR is merged on GitHub). Automatically runsgit fetch --prune

first.--force

,-f

: Skip confirmation prompt and ignore uncommitted changes--keep-branch

,-k

: Remove only the worktree and tmux window while keeping the local branch

workmux remove

workmux remove experiment

workmux rm feature-a feature-b feature-c

workmux rm -f old-work stale-branch

workmux rm old-work

workmux remove --keep-branch experiment

workmux rm -f experiment

workmux rm --gone

workmux rm --gone -f

workmux rm --all

Renames a worktree's directory, its tmux window or session, and the per-worktree workmux metadata. Optionally also renames the underlying git branch.

[old-name]

: Optional current worktree name. Defaults to the current worktree when run from inside one.<new-name>

: The new handle (directory name and tmux window/session base name).

--branch

,-b

: Also rename the underlying git branch to match<new-name>

. Fails if the worktree is on a detached HEAD.

workmux rename feature-new

workmux rename feature-old feature-new

workmux rename feature-old feature-new --branch

Rename is non-destructive: uncommitted changes and untracked files are preserved. The main worktree cannot be renamed. Collisions (existing target path, existing tmux target, or existing branch) are rejected before any changes are made.

Lists all git worktrees with their agent status, multiplexer window status, and merge status. Supports filtering by worktree handle or branch name.

[worktree-or-branch...]

: Filter by worktree handle (directory name) or branch name. Accepts multiple values. When omitted, shows all worktrees.

--pr

: Show GitHub PR status for each worktree. Requires thegh

CLI to be installed and authenticated. Note that it shows pull requests' statuses withNerd Fonticons, which requires Nerd Font compatible font installed.--json

: Output as JSON. Produces a JSON array of objects with fields:handle

,branch

,path

,is_main

,mode

,has_uncommitted_changes

,is_open

,created_at

.

workmux list

workmux list --pr

workmux list --json

workmux list my-feature
workmux list feature-auth feature-api
BRANCH      AGE  AGENT  MUX  UNMERGED  PATH
main        -    -      -    -         ~/project
user-auth   2h   🤖     ✓    -         ~/project__worktrees/user-auth
bug-fix     3d   ✅     ✓    ●         ~/project__worktrees/bug-fix
api-work    1w   -      ✓    -         ~/project__worktrees/api-work

• AGE shows how old the worktree is (e.g., 2h

,3d

,1w

,2mo

) - AGENT shows the current agent status (see status tracking):🤖

= working,💬

= waiting for input,✅

= finished- Multiple agents per worktree show a count (e.g., 2🤖 1✅

)

✓

in MUX column = multiplexer window exists for this worktree●

in UNMERGED column = branch has commits not merged into main-

= not applicable

Opens the global configuration file (~/.config/workmux/config.yaml

) in your preferred editor. Uses $VISUAL

, $EDITOR

, or falls back to vi

. Creates the file with commented-out defaults if it doesn't exist yet.

Prints the path to the global configuration file. Useful for scripting.

Prints the default configuration file with all options documented. Useful for discovering available options or piping to an AI agent for context.

Generates .workmux.yaml

with example configuration and "<global>"

placeholder usage.

Opens or switches to a tmux window for a pre-existing git worktree. If the window already exists, switches to it. If not, creates a new window with the configured pane layout and environment. Accepts multiple names to open several worktrees at once.

[name...]

: One or more worktree names (the directory name, which is also the tmux window name without the prefix). Optional with--new

when run from inside a worktree.

-n, --new

: Force opening in a new window even if one already exists. Creates a duplicate window with a suffix (e.g.,-2

,-3

). Useful for having multiple terminal views into the same worktree. In tmux, the duplicate appears immediately to the right of the window where the command runs.-s, --session

: Open in session mode, overriding the stored mode. Persists the mode change for subsequent opens. Cannot be combined with--new

. Only supported with tmux.--config <path>

: Use an alternate config file for this invocation. Still merges with global config.--run-hooks

: Re-runs thepost_create

commands (these block window creation).--force-files

: Re-applies file copy/symlink operations. Useful for restoring a deleted.env

file.-p, --prompt <text>

: Provide an inline prompt for AI agent panes.-P, --prompt-file <path>

: Provide a path to a file containing the prompt.-c, --continue

: Resume the agent's most recent conversation in this worktree. Injects the appropriate flag for the configured agent (e.g.,--continue

for Claude,--resume

for Gemini).-e, --prompt-editor

: Open your editor to write the prompt interactively.--prompt-file-only

: Write the prompt file without injecting it into agent commands.

• Verifies that a worktree with <name>

exists. - If a tmux window exists and --new

is not set, switches to it. - Otherwise, creates a new tmux window (with suffix if duplicating). In tmux, duplicate windows appear immediately to the right of the window where the command runs.

• (If specified) Runs file operations and post_create

hooks. - Sets up your configured tmux pane layout.

• Automatically switches your tmux client to the new window.

workmux open user-auth

workmux open user-auth --new

workmux open --new

workmux open user-auth --session

workmux open user-auth --continue

workmux open user-auth --continue -p "Continue implementing the login flow"

workmux open user-auth --run-hooks

workmux open user-auth --force-files

workmux open user-auth api-refactor bugfix-login

Closes the tmux window for a worktree without removing the worktree or branch. This is useful when you want to temporarily close a window to reduce clutter or free resources, but plan to return to the work later.

[name]

: Optional worktree name (the directory name). Defaults to current directory if omitted.

workmux close user-auth

workmux close

To reopen the window later, use workmux open.

Tip: You can also use tmux's native kill-window command (default: prefix + &

) to close a worktree's window with the same effect.

Restores worktree windows after a tmux or computer crash. Uses persisted agent state files to detect which worktrees had active agents before the crash, then reopens them with --continue

to resume agent conversations.

--dry-run

: Show what would be restored without doing it.

workmux resurrect --dry-run

workmux resurrect

• Reads agent state files from ~/.local/state/workmux/agents/

• Matches each state file's working directory to a git worktree in the current repo

• Skips worktrees that are already open or no longer exist

• Opens each matched worktree with --continue

to resume the agent

Re-applies file operations (copy and symlink from files

config) to existing worktrees. Useful when you add new entries to the files

config or a symlink was accidentally deleted.

--all

: Sync all worktrees instead of just the current one.

workmux sync-files

workmux sync-files --all

Prints the filesystem path of an existing worktree. Useful for scripting or quickly navigating to a worktree directory.

<name>

: Worktree name (the directory name).

workmux path user-auth

cd "$(workmux path user-auth)"

cp config.json "$(workmux path feature-branch)/"

Opens a TUI dashboard showing all active AI agents across all tmux sessions. Useful for monitoring multiple parallel agents and quickly jumping between them.

-d, --diff

: Open the diff view directly for the current worktree. Useful when you want to quickly review uncommitted changes without navigating through the agent list.-P, --preview-size <10-90>

: Set preview pane size as percentage (larger = more preview, less table). Default: 60.-s, --session

: Filter to only show agents in the current session. Useful for session-per-project workflows where each session maps to a different repository.-t, --tab <agents|worktrees>

: Open directly on the specified tab.

Important

This feature requires agent status tracking to be configured. Without it, no agents will appear in the dashboard.

Key	Action
1 -9
Quick jump to agent (closes dashboard)
Tab
Toggle between current and last agent
d
View diff (opens WIP view)
o
Open PR in browser
p
Peek at agent (dashboard stays open)
s
Cycle sort mode
/
Filter agents by name
F
Toggle session filter
f
Toggle stale filter (show/hide stale)
i
Enter input mode (type to agent)
Ctrl+u
Scroll preview up
Ctrl+d
Scroll preview down
+ /-
Resize preview pane
Enter
Go to selected agent (closes dashboard)
j /k
Navigate up/down
:
Open command palette
q /Esc
Quit

The bottom half shows a live preview of the selected agent's terminal output. The preview auto-scrolls to show the latest output, but you can scroll through history with Ctrl+u

/Ctrl+d

. Press i

to enter input mode and type directly to the agent without leaving the dashboard.

#: Quick jump key (1-9)** Project**: Project name (from__worktrees

path or directory name)Agent: Worktree/window name** Git**: Diff stats showing branch changes (dim) and uncommitted changes (bright). Shows a rebase icon when a rebase is in progress.Status: Agent status icon (🤖 working, 💬 waiting, ✅ done, or "stale")** Time**: Time since last status change** Title**: Claude Code session title (auto-generated summary)

Press s

to cycle through sort modes:

Priority(default): Waiting > Done > Working > Stale** Project**: Group by project name, then by priority within each project** Recency**: Most recently updated first** Natural**: Original tmux order (by pane creation)

Your sort preference persists in the tmux session.

Press F

to toggle the session filter. When active, only agents in the current session are shown. This is useful for session-per-project workflows where each session maps to a repository. You can also start the dashboard with --session

to default to session filtering. The preference persists across sessions.

Press /

to activate the name filter. Type to filter the agent list by project or worktree name (case-insensitive). Press Enter

to accept the filter and return to normal navigation, or Esc

to clear the filter. When a filter is active, it is shown in the footer bar.

Press f

to toggle between showing all agents or hiding stale ones. The filter state persists across dashboard sessions within the same tmux server.

Press d

to view the diff for the selected agent. The diff view has two modes:

WIP- Shows uncommitted changes (git diff HEAD

)review- Shows all changes on the branch vs main (git diff main...HEAD

)

Press Tab

to toggle between modes. The footer displays which mode is active along with diff statistics showing lines added (+) and removed (-).

Key	Action
Tab
Toggle WIP / review
a
Enter patch mode (WIP only)
j /k
Scroll down/up
Ctrl+d
Page down
Ctrl+u
Page up
c
Send commit command to agent
m
Trigger merge and exit dashboard
:
Open command palette
q /Esc
Close diff view

Patch mode (a

from WIP diff) allows staging individual hunks like git add -p

. This is useful for selectively staging parts of an agent's work.

When delta is installed, hunks are rendered with syntax highlighting for better readability.

Key	Action
y
Stage current hunk
n
Skip current hunk
u
Undo last staged hunk
s
Split hunk (if splittable)
o
Comment on hunk (sends to agent)
j /k
Navigate to next/previous hunk
:
Open command palette
q /Esc
Exit patch mode

Press y

to stage the current hunk and advance to the next. Press n

to skip without staging. The counter in the header shows your progress (e.g., [3/10]

).

Press s

to split the current hunk into smaller pieces when there are context lines between separate changes. Press u

to undo the last staged hunk.

Press o

to comment on the current hunk. This sends a message to the agent including the file path, line number, the diff hunk as context, and your comment. Useful for giving feedback like "This function should handle the error case".

Add to your ~/.tmux.conf

for quick access:

bind C-s display-popup -h 30 -w 100 -E "workmux dashboard"

bind C-w display-popup -h 30 -w 100 -E "workmux dashboard --tab worktrees"

Then press prefix + Ctrl-s

to open the dashboard as a tmux popup.

Toggles a live agent status sidebar on the left or top edge of all tmux windows. By default, each sidebar pane shows active agents across all tmux sessions with live status updates, providing an always-visible overview without taking over the full screen like the dashboard. Use workmux sidebar filter session

to show only agents in the current tmux session.

workmux sidebar                 # Toggle sidebar on/off (all sessions)
workmux sidebar --session       # Toggle current session only, or opt out of global mode
workmux sidebar --position top  # Override configured placement for this toggle

The sidebar displays:

• Status icon (working/waiting/done with spinner animation)
• Project and worktree name (e.g. myproject/fix-bug

) - Elapsed time since last status change

Key	Action
j /k
Navigate up/down
Enter
Jump to agent
g /G
Jump to first/last
v
Toggle layout mode
f
Toggle session filter
q
Quit sidebar

With tmux mouse mode enabled (set -g mouse on

), click an agent row or top-bar chip to jump to its pane, or scroll to navigate the list.

When the global sidebar is active, workmux sidebar --session

hides it in the current tmux session only. Run the same command again to show it in that session again while keeping the global sidebar active elsewhere.

Configure placement, width, and layout in .workmux.yaml

:

sidebar:
  position: left # "left" (default) or "top"
  width: 40 # left width in columns, or "15%" for percentage
  layout: tiles # left only: "compact" or "tiles" (default)

The left sidebar defaults to 10% of terminal width, clamped between 25 and 50 columns. Widths above 80 columns use the default width so tmux pane expansion recovers automatically.

Use workmux sidebar --position top

or --position left

to override the configured placement when enabling the sidebar.

bind C-t run-shell "workmux sidebar"

Then press prefix + Ctrl-t

to toggle the sidebar.

Note:The sidebar is currently tmux-only. When enabled, a sidebar pane is created in every existing window, and new windows automatically get one via a tmux hook.

Shows tracked agent processes whose last state update is older than a threshold. By default this is a dry run, though it still reconciles workmux state while checking live panes.

workmux reap-agents
workmux reap-agents --hours 48
workmux reap-agents --hours 24 --force

Use --force

to interrupt matching agents and remove their workmux agent state after they exit.

Commands for managing sandbox functionality. See the sandbox guide for full documentation.

Command	Description
sandbox pull
Pull the latest container image from the registry
sandbox build
Build the container image locally
sandbox shell
Start an interactive shell inside a sandbox
sandbox agent
Run the configured agent in a sandbox with RPC support
sandbox stop
Stop running Lima VMs
sandbox prune
Delete unused Lima VMs to reclaim disk space
sandbox install-dev
Cross-compile and install workmux into sandboxes (dev)

Removes stale entries from Claude config (~/.claude.json

) that point to deleted worktree directories. When you run Claude Code in worktrees, it stores per-worktree settings in that file. Over time, as worktrees are merged or deleted, it can accumulate entries for paths that no longer exist.

• Scans ~/.claude.json

for entries pointing to non-existent directories - Creates a backup at ~/.claude.json.bak

before making changes - Removes all stale entries

• Reports the number of entries cleaned up

• Only removes entries for absolute paths that don't exist

• Creates a backup before modifying the file

• Preserves all valid entries and relative paths

workmux claude prune
- Removing: /Users/user/project__worktrees/old-feature

✓ Created backup at ~/.claude.json.bak
✓ Removed 3 stale entries from ~/.claude.json

Generates shell completion script for the specified shell. Completions provide tab-completion for commands and dynamic branch name suggestions.

<shell>

: Shell type:bash

,zsh

, orfish

.

workmux completions zsh

See the Shell Completions section for installation instructions.

Displays this README with terminal formatting. Useful for quick reference without leaving the terminal.

When run interactively, renders markdown with colors and uses a pager (less

). When piped (e.g., to an LLM), outputs raw markdown for clean context.

You can ask an agent to read the docs and configure workmux for you:

> run `workmux docs` and configure workmux so that on the left pane
  there is claude as agent, and on the right side neovim and empty
  shell on top of each other

⏺ Bash(workmux docs)
  ⎿  <p align="center">
       <picture>
     … +923 lines

⏺ Write(.workmux.yaml)
  ⎿  Wrote 9 lines to .workmux.yaml

⏺ Created .workmux.yaml with the layout:
  - Left: claude agent (focused)
  - Right top: neovim
  - Right bottom: empty shell

Workmux can display the status of the agent in your tmux window list, giving you at-a-glance visibility into what the agent in each window doing.

• 🤖 = agent is working
• 💬 = agent is waiting for user input
• ✅ = agent finished (auto-clears on window focus)

Agent	Status
Claude Code	✅ Supported
OpenCode	✅ Supported
Codex	✅ Supported*
Copilot CLI	✅ Supported*
Pi	✅ Supported*
Oh My Pi	✅ Supported
Gemini CLI	✅ Supported
Kiro

Tracking issueNotes:

Codex: No 💬 waiting state** Copilot CLI**: No 💬 waiting state** Pi**: No 💬 waiting state** Kiro**: Hooks support is messy: requires a custom agent since the default can't be edited

Run workmux setup

to automatically detect Claude Code, Copilot CLI, OpenCode, Pi, Oh My Pi, and other supported agent CLIs, install status tracking hooks, and install skills:

workmux setup

You can also run specific parts: workmux setup --hooks

or workmux setup --skills

. For Claude Code, CLAUDE_CONFIG_DIR

is respected for both hook and skill installation.

Workmux will also prompt you on first run if it detects an agent without status tracking or skills configured.

Workmux automatically modifies your tmux window-status-format

to display the status icons. This happens once per session and only affects the current tmux session (not your global config).

If you prefer manual setup:

Claude Code: install the workmux status plugin:

claude plugin marketplace add raine/workmux
claude plugin install workmux-status

Or manually add the hooks to ~/.claude/settings.json

. See .claude-plugin/plugin.json for the hook configuration.

Copilot CLI: copy the hooks to your repository:

mkdir -p .github/hooks/workmux-status
curl -o .github/hooks/workmux-status/hooks.json \
  https://raw.githubusercontent.com/raine/workmux/main/.github/hooks/workmux-status/hooks.json

Note: Copilot hooks are per-repository. The waiting state is not supported due to limitations in the Copilot CLI hooks implementation.

OpenCode: download the workmux status plugin:

mkdir -p ~/.config/opencode/plugins
curl -o ~/.config/opencode/package.json \
  https://raw.githubusercontent.com/raine/workmux/main/resources/opencode/package.json
curl -o ~/.config/opencode/plugins/workmux-status.ts \
  https://raw.githubusercontent.com/raine/workmux/main/resources/opencode/plugins/workmux-status.ts

Restart OpenCode for the plugin to take effect.

Oh My Pi: copy the workmux status extension to your global OMP extensions directory:

mkdir -p ~/.omp/agent/extensions
curl -o ~/.omp/agent/extensions/workmux-status.ts \
  https://raw.githubusercontent.com/raine/workmux/main/.omp/extensions/workmux-status.ts

Restart omp for the extension to take effect.

You can customize the icons in your config:

status_icons:
  working: '🔄'
  waiting: '⏸️'
  done: '✔️'

If you prefer to manage the tmux format yourself, disable auto-modification and add the status variable to your ~/.tmux.conf

:

status_format: false
set -g window-status-format '#I:#W#{?@workmux_status, #{@workmux_status},}#{?window_flags,#{window_flags}, }'
set -g window-status-current-format '#I:#W#{?@workmux_status, #{@workmux_status},}#{?window_flags,#{window_flags}, }'

Use workmux last-done

to quickly switch to the agent that most recently finished its task or is waiting for user input. Repeated invocations cycle through all completed and waiting agents in reverse chronological order.

Add a tmux keybinding for quick access:

bind-key L run-shell "workmux last-done"

Then press prefix + L

to jump to the last completed or waiting agent, press again to cycle to the next oldest, and so on.

Use workmux last-agent

to toggle between your current agent and the last one you visited. This works like vim's Ctrl+^

or tmux's last-window

• it remembers which agent you came from and switches back to it. Pressing it again returns you to where you were.

This is available both as a CLI command and as the Tab

key in the dashboard.

Add a tmux keybinding for quick access:

bind Tab run-shell "workmux last-agent"

Then press prefix + Tab

to toggle between your two most recent agents.

workmux can run agents inside containers (Docker/Podman/Apple Container) or Lima VMs, isolating them from your host. Agents are restricted to the project worktree; sensitive files like SSH keys, AWS credentials, and other secrets are not accessible. This lets you run agents with --dangerously-skip-permissions

without worrying about what they might touch on your host.

Sandboxing is transparent: status indicators, the dashboard, spawning new agents, and merging all continue to work normally across the sandbox boundary.

Container (Docker/Podman/Apple Container)	Lima VM
Isolation
Process/VM-level	Machine-level (virtual machine)
Persistence
Ephemeral (new container per session)	Persistent (stateful VMs)
Toolchain
Custom Dockerfile or host command proxying	Built-in Nix & Devbox support
Network
Optional restrictions (domain allowlist)	Unrestricted

Container is a good default: simple to set up and ephemeral, so no state accumulates between sessions. Choose Lima if you want persistent VMs with built-in Nix/Devbox toolchain support.

sandbox:
  enabled: true

The pre-built container image is pulled automatically on first run. For Lima, the VM is created and provisioned on first use.

Both backends support:

Host command proxying: Run specific commands (build tools, linters) on the host from inside the sandbox viahost_commands

configExtra mounts: Mount additional host directories into the sandbox (read-only by default)** Git identity**: Youruser.name

anduser.email

are automatically injected so git commits work without exposing your full~/.gitconfig

Credential sharing: Agent credentials are shared between host and sandbox** Network restrictions**(container only): Block outbound connections except to approved domains

See the sandbox guide for full setup, configuration, and security details.

By default, workmux creates tmux windows within your current session. With session mode, each worktree gets its own tmux session instead. This allows each worktree to have multiple windows.

Add to your config:

mode: session

Or use the --session

flag:

workmux add feature-branch --session

Persistence: The mode is stored per-worktree. If you create a worktree with--session

, subsequentopen

/close

/remove

commands automatically use session mode for that worktree.Navigation: Aftermerge

orremove

, workmux switches you back to the previous session.

Use the windows

config to launch multiple windows in each session. Each window can have its own pane layout. This is mutually exclusive with the top-level panes

config.

mode: session
windows:
  - name: editor
    panes:
      - command: <agent>
        focus: true
      - split: horizontal
        size: 20
  - name: tests
    panes:
      - command: just test --watch
  - panes:
      - command: tail -f app.log

Each window supports:

Option	Description	Default
name
Window name (if omitted, tmux auto-names from command)	Auto
panes
Pane layout (same syntax as top-level panes )
Single shell

focus: true

works across windows: the last pane with focus set determines which window is selected when the session opens.

tmux only: Session mode is currently only supported for the tmux backend.** No duplicates**: Unlike window mode which supports opening multiple windows for the same worktree (-2

,-3

suffixes), session mode creates one session per worktree.

Here's a complete workflow:

workmux add user-auth


workmux merge user-auth

workmux add api-endpoint

workmux list

workmux turns a multi-step manual workflow into simple commands, making parallel development workflows practical.

git worktree add ../worktrees/user-auth -b user-auth
cd ../worktrees/user-auth
cp ../../project/.env.example .env
ln -s ../../project/node_modules .
npm install

tmux new-window -n user-auth
tmux split-window -h 'npm run dev'
tmux send-keys -t 0 'claude' C-m

cd ../../project
git switch main && git pull
git merge --no-ff user-auth
tmux kill-window -t user-auth
git worktree remove ../worktrees/user-auth
git branch -d user-auth
workmux add user-auth


workmux merge

Run multiple AI agents simultaneously, each in its own worktree.

workmux add refactor-user-model -p "Refactor the User model to use composition"
workmux add add-search-endpoint -p "Add a /search endpoint with pagination"

workmux dashboard

workmux merge refactor-user-model
workmux merge add-search-endpoint

Tip

Use -A

(--auto-name

) to generate branch names automatically from your prompt, so you don't have to think of one. See Automatic branch name generation.

Git worktrees let you have multiple branches checked out at once in the same repository, each in a separate directory. This provides two main advantages over a standard single-directory setup:

Painless context switching: Switch between tasks just by changing directories (cd ../other-branch

). There's no need togit stash

or make temporary commits. Your work-in-progress, editor state, and command history remain isolated and intact for each branch. - True parallel development: Work on multiple branches simultaneously without interference. You can run builds, install dependencies (npm install

), or run tests in one worktree while actively coding in another. This isolation is perfect for running multiple AI agents in parallel on different tasks.

In a standard Git setup, switching branches disrupts your flow by requiring a clean working tree. Worktrees remove this friction. workmux

automates the entire process and pairs each worktree with a dedicated tmux window, creating fully isolated development environments. See Before and after for how workmux streamlines this workflow.

While powerful, git worktrees have nuances that are important to understand. workmux is designed to automate solutions to these, but awareness of the underlying mechanics helps.

Gitignored files require configurationConflictsPackage manager considerations (pnpm, yarn)Rust projectsPort conflicts in monoreposSymlinks and.gitignore

trailing slashes

When git worktree add

creates a new working directory, it's a clean checkout. Files listed in your .gitignore

(e.g., .env

files, node_modules

, IDE configuration) will not exist in the new worktree by default. Your application will be broken in the new worktree until you manually create or link these necessary files.

This is a primary feature of workmux. Use the files

section in your .workmux.yaml

to automatically copy or symlink these files on creation:

files:
  copy:
    - .env # Copy environment variables
  symlink:
    - .next/cache # Share Next.js build cache

Note: Symlinking node_modules

can be efficient but only works if all worktrees share identical dependencies. If different branches have different dependency versions, each worktree needs its own installation. For dependency installation, consider using a pane command instead of post_create

hooks - this runs the install in the background without blocking the worktree and window creation:

panes:
  - command: npm install
    focus: true
  - split: horizontal

Worktrees isolate your filesystem, but they do not prevent merge conflicts. If you modify the area of code on two different branches (in two different worktrees), you will still have a conflict when you merge one into the other.

The best practice is to work on logically separate features in parallel worktrees. When conflicts are unavoidable, use standard git tools to resolve them. You can also leverage an AI agent within the worktree to assist with the conflict resolution.

Modern package managers like pnpm

use a global store with symlinks to node_modules

. Each worktree typically needs its own pnpm install

to set up the correct dependency versions for that branch.

If your worktrees always have identical dependencies (e.g., working on multiple features from the same base), you could potentially symlink node_modules

between worktrees. However, this breaks as soon as branches diverge in their dependencies, so it's generally safer to run a fresh install in each worktree.

Note: In large monorepos, cleaning up node_modules

during worktree removal can take significant time. workmux has a special cleanup mechanism that moves node_modules

to a temporary location and deletes it in the background, making the remove

command return almost instantly.

Unlike node_modules

, Rust's target/

directory should not be symlinked between worktrees. Cargo locks the target

directory during builds, so sharing it would block parallel builds and defeat the purpose of worktrees.

Instead, use sccache to share compiled dependencies across worktrees:

brew install sccache

Add to ~/.cargo/config.toml

:

[build]
rustc-wrapper = "sccache"

This caches compiled dependencies globally, so new worktrees benefit from cached artifacts without any lock contention.

When running multiple services (API, web app, database) in a monorepo, each worktree needs unique ports to avoid conflicts. For example, if your .env

has hardcoded ports like API_PORT=3001

and VITE_PORT=3000

, running two worktrees simultaneously would fail because both would try to bind to the same ports. Simply copying .env

files won't work since all worktrees would use the same ports.

Solution: Use a post_create

hook to generate a .env.local

file with unique ports. Many frameworks (Vite, Next.js, CRA) automatically load .env.local

and merge it with .env

, with .env.local

taking precedence. For plain Node.js, use multiple --env-file

flags where later files override earlier ones.

Create a script at scripts/worktree-env

:

#!/usr/bin/env bash
set -euo pipefail

port_in_use() {
  lsof -nP -iTCP:"$1" -sTCP:LISTEN &>/dev/null
}

find_port() {
  local port=$1
  while port_in_use "$port"; do
    ((port++))
  done
  echo "$port"
}

hash=$(echo -n "$WM_HANDLE" | md5 | cut -c1-4)
offset=$((16#$hash % 100))

api_port=$(find_port $((3001 + offset * 10)))
vite_port=$(find_port $((3000 + offset * 10)))

cat >.env.local <<EOF
API_PORT=$api_port
VITE_PORT=$vite_port
VITE_PUBLIC_API_URL=http://localhost:$api_port
EOF

echo "Created .env.local with ports: API=$api_port, VITE=$vite_port"

Configure workmux to copy .env

and generate .env.local

:

files:
  copy:
    - .env # Copy secrets (DATABASE_URL, API keys, etc.)

post_create:
  - ./scripts/worktree-env # Generate .env.local with unique ports

For plain Node.js (without framework support), load both files with later overriding earlier:

{
  "scripts": {
    "api": "node --env-file=.env --env-file=.env.local api/server.js",
    "web": "node --env-file=.env --env-file=.env.local web/server.js"
  }
}

Each worktree now gets unique ports derived from its name, allowing multiple instances to run simultaneously without conflicts. The .env

file stays untouched, and .env.local

is gitignored.

See the Monorepos guide for alternative approaches using direnv.

If your .gitignore

uses a trailing slash to ignore directories (e.g., tests/venv/

), symlinks to that path in the created worktree will not be ignored and will show up in git status

. This is because venv/

only matches directories, not files (symlinks).

To ignore both directories and symlinks, remove the trailing slash:

- tests/venv/
+ tests/venv

On first run, workmux prompts you to check if a git branch icon displays correctly. If you have a Nerd Font installed, answer yes to enable nerdfont icons throughout the interface, including the tmux window prefix.

To change the setting later, edit ~/.config/workmux/config.yaml

:

nerdfont: true # or false for unicode fallbacks

If your project uses direnv for environment management, you can configure workmux to automatically set it up in new worktrees:

post_create:
  - direnv allow

files:
  symlink:
    - .envrc

By default, Claude Code prompts for permission before running commands. There are several ways to handle this in worktrees:

Share permissions across worktrees

To keep permission prompts but share granted permissions across worktrees:

files:
  symlink:
    - .claude/settings.local.json

Add this to your global config (~/.config/workmux/config.yaml

) or project's .workmux.yaml

. Since this file contains user-specific permissions, also add it to .gitignore

:

.claude/settings.local.json

Named agent profiles

Define named agents in your global config when an agent needs a wrapper command, extra arguments, or environment variables. Simple string aliases still work:

agents:
  cc-work: "claude"
  cc-personal: "env CLAUDE_CONFIG_DIR=~/.claude-personal claude"
  cc-bedrock: "env -u CLAUDE_CODE_USE_BEDROCK -u AWS_REGION AWS_PROFILE=prod claude"
  cc-yolo: "claude --dangerously-skip-permissions"
  cod: "codex --yolo"

Structured profiles make the same kind of config easier to maintain because arguments and environment are not written as one long shell string:

agents:
  cc-personal:
    type: claude
    command: claude
    env:
      CLAUDE_CONFIG_DIR: ~/.claude-personal
      ANTHROPIC_AUTH_TOKEN:
        from_env: ANTHROPIC_AUTH_TOKEN
  cod-mini:
    type: codex
    command: codex
    args:
      - exec
      - -m
      - gpt-5.1-codex-mini

To skip prompts entirely, define a named agent with the Claude skip-permissions flag:

agents:
  cc-yolo:
    type: claude
    command: claude
    args:
      - --dangerously-skip-permissions

Reference a named profile per project with agent: cc-yolo

, pass it with -a cc-yolo

, or use it directly in a pane with <agent:cc-yolo>

. Structured profiles support command

, args

, env

, and type

fields.

The /worktree

skill lets you delegate tasks to parallel worktree agents directly from your conversation. A main agent on the main branch can act as a coordinator: planning work and spinning up worktree agents for each task.

📝 See this blog post for a detailed walkthrough of the workflow.

> /worktree Implement user authentication
> /worktree Fix the race condition in handler.go
> /worktree Add dark mode, Implement caching  # multiple tasks

See the Skills guide for more skills including /merge

, /rebase

, /coordinator

, and /open-pr

.

To enable tab completions for commands and branch names, add the following to your shell's configuration file.

For bash, add to your .bashrc

:

eval "$(workmux completions bash)"

For zsh, add to your .zshrc

:

eval "$(workmux completions zsh)"

For fish, add to your config.fish

:

workmux completions fish | source

• Rust (for building)
• Git 2.5+ (for worktree support)
• tmux (or an alternative backend)

While tmux is the primary and recommended backend, workmux also supports alternative terminal multiplexers:

(experimental) - For users who prefer WezTerm's features. Thanks toWezTerm@JeremyBYUfor contributing this backend.(experimental) - For users who prefer kitty terminal. Requireskittyallow_remote_control

andlisten_on

configuration.(experimental) - For users who prefer Zellij. Detected automatically viaZellij$ZELLIJ

.

workmux auto-detects the backend from environment variables ($TMUX

, $WEZTERM_PANE

, $KITTY_WINDOW_ID

, or $ZELLIJ

). Session-specific variables are checked first, so running tmux inside kitty correctly selects the tmux backend. Set $WORKMUX_BACKEND

to override detection.

workmux is inspired by wtp, an excellent git worktree management tool. While wtp streamlines worktree creation and setup, workmux takes this further by tightly coupling worktrees with tmux window management.

For managing multiple AI agents in parallel, tools like claude-squad and vibe-kanban offer dedicated interfaces, like a TUI or kanban board. In contrast, workmux adheres to its philosophy that tmux is the interface, providing a native tmux experience for managing parallel workflows without requiring a separate interface to learn.

Bug reports and feature suggestions are always welcome via issues or discussions. Large and/or complex PRs, especially without prior discussion, may not get merged. Thanks for contributing!

See CONTRIBUTING.md for development setup.

tmux-tools— Collection of tmux utilities including file picker, smart sessions, and moretmux-file-picker— Pop up fzf in tmux to quickly insert file paths, perfect for AI coding assistantstmux-bro— Smart tmux session manager that sets up project-specific sessions automaticallygit-surgeon— Non-interactive hunk-level git staging for AI agentsclaude-history— Search and view Claude Code conversation history with fzfconsult-llm— Consult other AI models from your agent workflowtmux-agent-usage— Display AI agent rate limit usage in your tmux status bar

source & further reading

github.com — original article