Skip to content

Launcher reference

The three launchers discover the same repository set, synchronize shared agent configuration, prepare optional user/task context, and then start their editor.

Launcher Agent Repository wiring
./claude.sh Claude Code Adds each discovered repository directory
./codex.sh Codex Adds each discovered repository directory
./cursor.sh Cursor Agent Generates a multi-root workspace

Shared flags

Flag Effect
--sibling Discover repositories beside Spine rather than below repos/
--user-context PATH Use a specific local user/task context directory
--no-user-context Skip user and task context for this launch
--task-selection=launcher Use the default shell selector
--task-selection=auto Reuse the selected task without prompting
--task-selection=chat Defer selection to the agent when interactive

Arguments not consumed by Spine pass through to the agent CLI. Cursor also accepts the legacy --sync flag as a no-op because synchronization is already automatic.

Launchers require at least one discovered ecosystem repository. Run ./setup.sh first, and keep layout flags consistent. For the generated-source boundary, see source ownership.