Run Claude Code in parallel.

A worktree and terminal per task, status from Claude Code's own hooks, and the ouijit command in its system prompt. Nothing to configure.

What Ouijit adds

Knows OuijitThe ouijit CLI reference is appended to the system prompt at launch, so Claude can create tasks, move cards, open its plan beside itself and stage review comments without being told how.

Live statusClaude Code's own hooks report when a turn starts and ends. Ouijit passes them in a settings overlay at launch, merged with yours, so your own config is untouched. The card shows thinking or ready, and a sound or notification fires when it finishes.

SandboxedRun it with its permission checks off inside Ouijit's sandbox and the kernel holds it to the task's worktree. The dev server beside it stays on the host.

Hooks

What the project runs when a task changes status. Set them in project settings, or ask the agent to.

# Start · a task moves to In Progress
claude "$OUIJIT_TASK_DESCRIPTION"

# Continue · a task already in progress is reopened
claude -c

The start hook with its permission checks off, for a project that runs inside the sandbox.

claude --dangerously-skip-permissions "$OUIJIT_TASK_DESCRIPTION"

Questions

Can it sign in inside the sandbox?

Yes. The sandbox lets it reach its config directory, the OS keychain and claude.ai. Everything else outside the worktree stays out of reach.

Does this replace Claude Code's own sandbox?

It covers more. Claude Code's sandbox wraps shell commands; Ouijit's wraps the whole terminal, including file tools, MCP servers and hooks. The two can run together.

Does it change Claude Code anywhere else?

No. The claude on PATH in an Ouijit terminal is a wrapper around the real one, and only there.

Supported agents