Coordinate, don’t just tile
Attributed messaging, teams, tasks, approvals, and shared live context—not a window manager pretending to be orchestration.
◇Terminal-native multi-agent coordination
Launch, observe, and coordinate coding agents across different harnesses—without giving up control of your codebase.
go install github.com/clishakehq/clishake/cmd/clishake@latest
One session across your favorite agent CLIs
◇Built for real collaboration
clishake adds the coordination substrate that multiple coding agents actually need: identity, shared state, safe workspaces, and a human in control.
Attributed messaging, teams, tasks, approvals, and shared live context—not a window manager pretending to be orchestration.
Every editing agent gets its own branch and Git worktree. Overlaps are surfaced; writes are never silently lost.
Approvals remain with the lead. Every action has an actor, timestamp, subject, and durable audit trail.
Agents live in a dedicated tmux server. Close the terminal, come back later, and reconcile exactly what happened.
Six harnesses today, a clean extension seam tomorrow. Capabilities are reported only when a harness truly supports them.
◇Simple mental model
Assign, observe, approve
Route · persist · supervise
Work in isolated branches
◇From zero to coordinated
Initialize clishake inside any project, add the harnesses you already use, and start routing work. Supported on macOS and Linux, and on Windows through WSL2.
Read the quickstart →macOSLinuxWindows via WSL2

clishake init to two working agents
/ask flow · exact plan shown before execution◇Natural language, explicit control
clishake can translate intent into a whitelisted command plan using the AI CLI already on your machine. It shows every command before anything runs.
Open source. Terminal-native. Human-controlled.