What problem does it solve? Re-dispatching a parallel coder task that already has a worktree lane can fail because of stale worktrees or leftover branches, and blindly deleting them risks destroying work owned by another active session. ## Core Features & Use Cases - Provisioning-first cleanup: Re-dispatches through the standard coder/worktree path so built-in provisioning pre-cleans safe, stale same-lane worktrees and branches automatically. - Ownership verification: Reads .swarm/session/state.json and checks delegationChains to confirm no other active session owns the lane before any manual removal. - Surgical removal: Targets only .swarm-worktrees/<session>/<task> via git worktree remove and git worktree prune, deleting the swarm/lane/<session>/<task> branch only after confirming it is stale, with force deletion requiring explicit human approval. - Use Case: A coder task failed mid-run and must be retried; this protocol verifies lane ownership, removes only the confirmed stale worktree and branch, and verifies the branch list is empty before retrying. ## Quick Start Before re-dispatching the failed coder task, follow the worktree retry cleanup protocol to verify lane ownership and remove only the stale worktree and branch.
