Function handle_despawn_worker 

pub(crate) fn handle_despawn_worker(
    workspace: &Arc<Workspace>,
    project_key: &ProjectKey,
    label: &str,
    force: bool,
    respond: Sender<DespawnResult>,
)

Handle a Command::DespawnWorker (the agents__despawn MCP tool): the lead’s clean-close gesture. Unlike handle_close_worker it also cleans up the worker’s git worktree.

Order matters: the worktree dirty-check runs BEFORE any teardown, so a dirty worker (uncommitted/untracked or unpushed) is blocked without being killed (unless force). Teardown then releases the session, which drops the worker’s command sender; the disconnect itself runs asynchronously on the worker’s own task and can take up to the 5s close-wait budget to reap the child. The worktree removal runs on the blocking pool (see DespawnCleanup) only after that task has finished dying - its exit path reaps the session’s own child (waits it out, then SIGKILLs), so that child is dead before the removal takes the tree (#1767; descendants that outlive a SIGKILLed child are #2337) - without holding the caller’s runtime, and bounded by TEARDOWN_WAIT so a teardown that never answers cannot wedge the reply. A post-teardown worktree-removal failure is surfaced as a warning in the DespawnResult but never rolls back the kill - teardown and worktree cleanup are independent.