How to run crewbit without git worktrees

Use git.disable: true when your task does not involve a locally cloned git repository or when worktree creation is unnecessary overhead.

When to use this

• Non-git projects — the repository uses SVN, Mercurial, or a custom monorepo toolchain that does not support git worktrees.
• Tasks with no codebase — writing, research, planning, or data analysis where the agent only needs to call APIs or produce files, not commit code.
• Remote infrastructure work — the agent interacts with remote systems (cloud APIs, databases) and there is no local codebase to check out.
• OpenCode in ephemeral CI containers — a fresh container already contains a clean checkout; creating a second worktree is redundant.

Steps

1. Add git.disable to your workflow file

provider: jira

providers:
  jira:
    baseUrl: https://your-org.atlassian.net
    projectKey: PROJ
    transitionIds:
      start: "21"
      done: "31"
    issueTypes:
      task: "10001"

transitions:
  research:
    from: Ready
    command: /research

git:
  disable: true

2. Start the daemon as usual

pnpm start -- ./your-workflow.yaml

crewbit will pick up issues and run the agent directly in the directory where the command was started. No worktree branch is created and no cleanup is needed after the session.

What changes when git.disable: true

Behaviour	Default	With git.disable: true
Working directory for agent	isolated git worktree	cwd where crewbit started
Temporary branch created	yes (crewbit-<issueKey>)	no
Worktree cleaned up after session	yes	n/a
git.worktreePrefix used	yes	ignored
git.branchPattern used	yes	ignored

Combining with the OpenCode runner

git.disable works with both the claude and opencode runners. For OpenCode, the ?directory= query parameter sent to the server will point to the cwd instead of a worktree path.

runner: opencode

opencode:
  baseUrl: http://localhost:4096

git:
  disable: true