WHAT NEW IN X10.31 SAT

X 上的 AI 合集档案 · 22 个合集 · 63,170 条

jevVIDEO

Jev 前置筛选任务,把 Opus 成本从 400 美元压到 20

用 Jev 在 Opus 之前做去重和任务筛选,只把合格任务连同上下文交给 Opus,同一份被接受的产出成本降到二十分之一。

Solomon Rojie
@Solomonrojie19 天前收录原推发布于 2026.10.12
Solomon Rojie @Solomonrojie原推发布于 2026.10.12
在 X 上打开原推

Opus 5.5 + Jev just made my workflow 20x more efficient $400 → $20 for the same accepted output. 16 steps to stop burning premium tokens on repeated decisions, unchanged inputs, and pointless retries Jev screens incoming tasks before Opus starts. Code catches duplicates. Selected work reaches Opus with the context and requirements needed to solve it the roadmap: 01 prepare the work task audit → wasted calls → source packet → acceptance criteria 02 control execution API access → job ledger → repeat check → Jev selection 03 improve the result Opus handoff → batched checks → failed criteria → one revision 04 measure the workflow saved results → shadow review → active mode → total cost the loop: input → repeat check → Jev → Opus → checks → review → reuse take a GitHub issue queue. Jev checks whether a report fits the project and contains enough detail. Selected issues reach Opus for investigation and a proposed fix the first deliverable: one issue investigated + a proposed change + test results ready for review an unchanged issue returns its saved status. a failed check sends back the specific requirement, evidence, and previous output every new Opus call has a defined job and a reason to happen the article below breaks down how to connect the pieces ↓

这条是引用着另一条写的:x.com/0xnineinch/status/2108604801411879139

x.com/solomonrojie/status/2109379354367299874采集于 2026.10.12 10:06