v0.11.0

New Features

Clean Up Merged Command

Added a palette-only batch command, Clean up merged. It scans all done rows regardless of the current filter or search: worktrees or branch-only rows merged into main, plus worktrees whose PR is merged or closed. It opens a confirmation dialog with counts for worktrees to remove and branches to delete, representative affected rows, and the exact commands that will run.

The batch stays conservative: worktree removal uses only git worktree remove (never --force), branch deletion uses only git branch -d (never -D). Only clean, unlocked, non-root, non-active worktrees are removed; dirty, locked, active, root, detached, prunable, and not-yet-loaded rows are ignored. Worktree branches are deleted only when merged into main, while PR merged/closed worktrees not merged into main are removed but keep their local branch. Approved before_delete hooks run once per removed worktree; a hook failure counts as that row's failure and skips its removal while the rest of the batch continues. Full success shows a summary; partial failure keeps a result dialog open with counts and reasons. When the batch deletes branch refs with known SHAs, the success message offers u to restore all deleted refs.

Bug Fixes

• Fixed the PR column staying silently empty on large repos. The single gh pr list call requesting statusCheckRollup forced GitHub's GraphQL API to resolve CI for all 200 PRs, returning HTTP 504. The fetch is now split: a fast branch→PR mapping without statusCheckRollup, and CI status fetched lazily per open PR.
• Fixed the PR fetch being cancelled by a hard 2s timeout on large repos (listing 200 PRs takes ~4-5s). The timeout is now a named 15s constant.
• Fixed delete dialog errors overflowing the frame. Git delete failures now wrap across lines instead of hard-truncating, advisory git hint lines are dropped, the error block is capped, and the dialog body is clipped to the terminal height.

Improvements

• Cut initial PR latency on normal repos. When the local branch count is small (≤40), each branch is queried with gh pr list --head in a parallel worker pool, so number, state, and CI all arrive together in ~1s instead of ~5s+. Above the threshold, the single list call with lazy CI is used to avoid a large request fan-out.