Dots 四代理协作 + Opus 5.5 修 628 个 GitHub issue
作者用 Dots 里四个代理分工排查、复现、验收，再让 Claude Code 的 Opus 5.5 写补丁，628 个 issue 收敛成 16 个通过测试的修复，比单跑 Opus 快 3.8 倍、便宜 3.6 倍。
THIS IS GETTING F*CKING INSANE, DOTS + OPUS 5.5 TURNED 628 GITHUB ISSUES INTO THE SAME 16 TESTED FIXES FOR $34.80 INSTEAD OF $126.40 I set up four agents in Dots and one Claude Code agent running Opus 5.5 meet the team: → Scout found the relevant files and missing details in each report → Echo grouped duplicate reports around the same failure → Trace reproduced the bugs and handed Claude the failing steps → Atlas checked the diffs and test results, then returned anything unresolved → Claude wrote the patches and regression tests they passed findings between them. Claude received a focused bug report with the evidence already attached one handoff connected 19 export failures. different filenames, different users, the same timeout path Echo grouped the reports. Trace reproduced the failure. Claude patched the shared code. Atlas checked that the regression test failed before the change and passed after it across the backlog: 628 issues → 79 distinct problems → 31 reproduced bugs → 16 passing fixes I compared both workflows against the same backlog and checks: → Opus 5.5 investigates each report: 18h 40m / $126.40 in API costs → four Dots agents coordinate, Opus 5.5 patches: 4h 52m / $34.80 in API costs same 16 passing fixes. about 3.8x faster and 3.6x cheaper in this modeled comparison I came back to: > 16 tested patches > reproduction steps for each fix > before-and-after test results > unresolved issues kept separate I reviewed the diffs before anything merged full playbook below ↓
这条是引用着另一条写的：x.com/mika_systems/status/2107101689878630885