Qwen Code v0.21.13 hardens its /review agent workflow against review loops with a round-5 severity cutoff and worktree lease locks

GitHub (QwenLM/qwen-code)·medium signalXBlueskyLinkedInCopy link

Released Aug 17, 2026, v0.21.13 rebuilds the /review skill around dedicated platform subcommands (meta, fetch-diff) instead of raw gh commands issued through prompt prose (#9096), and caps posted suggestions to Critical findings after round 5 to stop review loops (#9118). Operationally it adds worktree leases that act as locks so concurrent sessions cannot destroy an active review state (#9211), auto-wipe-and-retry on corrupted checkouts (#9220), and session IDs plus diff hashes so agent work is credited across interrupted runs (#9091). This is a rare public look at the failure modes of running review agents at scale — loops, concurrent state destruction, and workspace corruption.