qwen-code Now Files a Blocking Finding When a Review Loop Stops Converging

GitHub·medium signalXBlueskyLinkedInCopy link

PR #9596, merged 2026-08-22, makes each review round hand over two integers: how many findings first appeared this round, and how many of those trace back to the previous round's fixes. On the second consecutive round where most new work was work the previous round created, the command composes its own blocking finding asking the author to split the change or reconsider the approach. The motivating analysis across six multi-round takeover PRs found roughly a third of post-initial findings were introduced by the immediately preceding fix round, dominated by guards added with no test of their own.