Skip to content
Agent Kanban

Identity and repository access

What stays on the runner and what the workspace stores.

GitHub sign-in identifies you by your GitHub account ID. The hosted service exchanges a short-lived authorization code, reads your profile, and sets a secure HTTP-only session cookie. Signing out revokes that session.

Each account has its own workspace. Pairing creates an independent machine credential using an expiring one-time code. The runner opens outbound HTTPS connections and starts no HTTP listener. Only repositories you add locally appear in its inventory. Revoking a machine stops its access to the job queue.

The cloud stores repository names, branch names, tasks, provider status, recent run activity, verification output and patch previews. Those records can contain source code. Each preview contains up to 200,000 characters of patch and the latest 120 activity entries. Full worktrees and transcripts stay on the machine.

Each coding CLI uses its local authentication. Acceptance commands run as your OS user in the task worktree. A Git worktree isolates changes to a branch. OS permissions and each CLI’s execution settings govern access to the machine.

Opening a pull request uses the runner’s local GitHub credentials and pushes the reviewed branch. Signing in to the hosted board requests no repository scope. A local merge refuses a dirty checkout.