十个用 Jev 做决策层的开源仓库合集
Opus 写代码、Jev 决定什么能进模型：浏览器 agent、编码 agent、代理路由、看门狗、代码审查等十个可 fork 的项目。
Opus writes the code. Jev decides what is even allowed to reach the model. Ten repos where Jev makes the call and the agent just does the work. Same split every time: Opus does the work, Jev makes the call, code enforces it. Fork any of them. -jev-ultrafast https://github.com/browser-use/jev-ultrafast A browser agent. Jev picks the action and the exact DOM element. A separate model types text only when something actually needs typing. -mu https://github.com/qybaihe/mu A Pi coding agent. Jev covers 38 decision points, including which chunks of a tool result are allowed into context. -Jevonian https://github.com/xinyao27/jevonian A local OpenAI / Anthropic proxy. One Jev call per request picks the model and the thinking level. -pi-warden https://github.com/DevMortimer/pi-warden A watchdog. It catches rule breaks, risky actions, stuck loops, and fake "done" claims. -Oko https://github.com/bartlomein/oko Jev checks whether each code chunk does what the agent asked for. Only accepted chunks stay, and they are served over MCP. -jev-pref https://github.com/doeixd/jev-pref Write project preferences once in jev-pref.json. Jev checks every change against them and feeds the findings back to the agent. -taste-lint https://github.com/mblode/taste-lint Catches AI slop in UI, copy, and agent instructions before it ships. -jev-review https://github.com/devagrawal09/jev-review A staged code-review flow with a local dashboard. Jev decides every step. -opencode-jev-orchestrator https://github.com/aaronshaf/opencode-jev-orchestrator The parent model stays fixed. Jev spots the hard turns and hands them to temporary subagents. -yoshi https://github.com/compozy/yoshi A Claude Code / Codex proxy. Jev prunes the request context and leaves the tool-call structure untouched. Fork any of these Jev builds, then read the full Jev setup below.
这条是引用着另一条写的：x.com/hanakoxbt/status/2101709924828934222