Skills Vibe-Switch

Name: Vibe-Switch
Author: openclaw

tmux for AI Agents — Orchestrate multiple AI coding agents in parallel with one command. Run Claude Code, Codex CLI, and Gemini CLI simultaneously on isolated Git branches with seamless context handoff.

install

source · Clone the upstream repo

git clone https://github.com/openclaw/skills

Claude Code · Install into ~/.claude/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/openclaw/skills "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/brianjhang/vibe-switch" ~/.claude/skills/openclaw-skills-vibe-switch && rm -rf "$T"

OpenClaw · Install into ~/.openclaw/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/openclaw/skills "$T" && mkdir -p ~/.openclaw/skills && cp -r "$T/skills/brianjhang/vibe-switch" ~/.openclaw/skills/openclaw-skills-vibe-switch && rm -rf "$T"

manifest: skills/brianjhang/vibe-switch/SKILL.md

Vibe-Switch — tmux for AI Agents

Orchestrate multiple AI coding agents in parallel with one command. Each agent runs in its own isolated Git worktree and branch. When one agent finishes, hand off the context seamlessly to another.

Install

npm install -g vibe-switch

Quick Start

# Start agents in parallel
vibe run "Implement JWT auth middleware" --agent gemini
vibe run "Build responsive login page" --agent codex

# Monitor all agents
vibe status
vibe watch

# Hand off context between agents
vibe handoff vibe/gemini-c3d5 --to codex -m "API is ready at /api/auth"

# Cleanup
vibe stop --all
vibe clean

Command Reference

Command	Description
`vibe run "<task>" --agent <agent>`	Start an agent on an isolated Git branch + worktree
`vibe status`	Display all tasks, status, and branches
`vibe watch`	Real-time multi-agent output streaming
`vibe log <branch> [-f]`	View or follow logs for a specific task
`vibe stop [branch\|--all]`	Stop one or all agents
`vibe handoff <branch> --to <agent>`	Transfer context to another agent
`vibe summary <branch>`	View task summary, logs, and diff stat
`vibe clean`	Clean up completed tasks, logs, and worktrees
`vibe agents`	List installed agent adapters
`vibe doctor`	Diagnose environment and agent availability
`vibe init`	Create project config `.vibeswitch.json`
`vibe config [key] [value]`	View or set project config

Supported Agents

Agent	Command	Sandbox	Best For
Claude Code	`claude`	No	Architecture, code review, broad reasoning
Codex CLI	`codex`	Yes	Implementation, refactors, test-driven changes
Gemini CLI	`gemini`	No	Exploration, cross-checking, documentation
Antigravity	`antigravity`	No	Full-stack development, complex tasks
OpenClaw	`openclaw`	No	General purpose AI coding

Architecture

Git Worktree Isolation: Each agent runs in a dedicated branch and directory — no file conflicts.
Adapter Pattern: Modular design for easy integration of new AI agent CLIs.
JSON File Storage: Task metadata persisted locally at
```
~/.vibe-switch/
```
— no database needed.

Common Workflow Patterns

Parallel Development

vibe run "Implement auth API and migration" --agent claude --branch vibe/auth-api
vibe run "Build auth UI and form validation" --agent codex --branch vibe/auth-ui
vibe run "Write integration tests for login" --agent gemini --branch vibe/auth-tests
vibe status
vibe watch

Context Handoff

# Gemini finished the API → hand off to Codex for the frontend
vibe handoff vibe/auth-api --to codex -m "API complete. Build the client-side integration."

Agent Capability Matrix

Agent	Local Files	Network/SSH	Sandbox
claude	✅	✅	No
codex	✅	❌	Yes
gemini	✅	✅	No

Important: Codex runs in a strict sandbox — no SSH, no external APIs. Do not assign deployment or network-dependent tasks to Codex.

Key Constraints

Must be run from inside a Git repository.
Worktrees are based on the current branch HEAD — uncommitted changes are NOT visible to spawned agents. Always
```
git add && git commit
```
before dispatching.
Right-size your tasks: if it takes less than 5 minutes and touches one file, do it yourself.