You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Flight worker commits to a harness-created worktree branch instead of the flight branch #547
A visual-flight worker committed its work to a second, session-created branch instead of the flight branch the dispatch placed.
What happened (0.49.0, tmux rung): the dispatch created the flight worktree on planwright/flight/<id> (nested under the tower's own worktree). The worker's Claude Code session then ran in a separate worktree, .claude/worktrees/flight-<id>, on a branch named worktree-flight-<id>, both apparently created by the harness's own worktree isolation rather than by planwright. The worker committed there. Because the flight branch was checked out in the other worktree, it couldn't switch to it. It only reported the mismatch when it parked, and later cherry-picked onto the flight branch on operator direction.
Why it matters: the flight branch is what the tower reads for landing evidence, and the record lands there. Work on a side branch looks to the tower like "no landing yet", and it leaves an extra worktree and branch to clean up.
Expected: the worker session starts inside the flight worktree on the flight branch, or detects the mismatch at startup and parks before doing any work.
Possible cause
The worker's launch directory and the dispatch-placed worktree may differ, so the harness's worktree-on-launch behaviour makes a fresh worktree instead of reusing the placed one. Two worktrees existed side by side: the planwright-placed one at the flight branch and the harness one at worktree-flight-<id>.
Activity