用 Jev 先筛队列再喂给 Opus
1200 份报告里只有 60 份值得查,作者用 Jev 先过滤,再让 Opus 只处理被选中的任务以省成本。
STOP SPENDING 20x MORE ON OPUS 5.5 THAN YOU NEED TO 1,200 incoming reports. 60 worth investigating. Jev filters the queue before every submission becomes an expensive Opus task start in shadow mode. your workflow keeps running while Jev checks the same reports and logs what it would investigate, skip, or send for review after a week, compare its choices with the actual work. see which calls you could have avoided and which important reports Jev would have missed then let it route the tasks. Opus investigates the selected reports, while code tracks completed work and catches unchanged submissions you’re cutting the queue before it reaches the expensive model my article below covers the setup and how to measure whether that filter earns its place ↓
这条是引用着另一条写的:x.com/0xnineinch/status/2108604801411879139