10 个 Jev 开源项目覆盖 agent 循环每个环节
按提示前、路由、搜索、执行前、测试、提交六个环节,列出 10 个用 Jev 做是/否判断的开源项目及仓库地址。
Nobody is talking about this, but 10 open-source Jev builds already cover every hook of the agent loop Got a Jev key and no idea what to build? Don't build. Steal: Before the prompt → jev-skill-router: picks one skill out of everything you've installed, or none. Starts in shadow mode, only logs http://github.com/shimo4228/jev-skill-router Routing the call → jev-opus: asks Jev how much effort the next Opus 5.5 step needs: low, medium or high http://github.com/WXK-AI/jev-opus → codex-jev-router: picks the model and reasoning effort for every Codex subagent, falls back to Sol when unsure http://github.com/suenot/codex-jev-router Searching → jev-scout: an MCP server that scores every query, result and page for relevance and credibility http://github.com/kierandotai/jev-scout → jegrep: semantic grep in Rust, no embedding index, no daemon http://github.com/can1357/jegrep Before anything runs → is-malicious: flags suspicious lines in source, config and CI before execution http://github.com/luantak/is-malicious Testing → pytest-jev: plain-English claims, the test passes only above 0.8 http://github.com/allebee/pytest-jev → hunch: if Hunch.likely?("fraudulent") in Ruby, a branch on a typed answer http://github.com/carldaws/hunch Committing and finishing → jev-commit: checks that the commit message actually matches the diff http://github.com/valentynkit/jev-commit → jev-belay: won't let the agent say "done" without proof http://github.com/valentynkit/jev-belay Ten builds. One loop. A yes or no at every hook Bookmark it, the full Jev setup is below ↓
这条是引用着另一条写的:x.com/0xkiyoro/status/2107135100911563017