10 个真正用上 Jev 的 GitHub 仓库

rody @0x_rody原推发布于 2026.09.25

在 X 上打开原推

Everyone is talking about JEV. Almost nobody knows how to actually use it. These 10 GitHub repos fix that. 1. vercel/eve Vercel's open agent framework. Ships Jev as the default model in its experimental evaluate path. https://github.com/vercel/eve 2. smithersai/smithers Agentic TypeScript workflow framework with a Jev session checker wired into the workflows. https://github.com/smithersai/smithers 3. typesafe-ai/skills The official agent skill for Claude Code and Codex: primitives, patterns, and how to structure your evals. https://github.com/typesafe-ai/skills 4. awlevin/typesafe-computer-use Computer use for about $0.0002 a step: OCR the screen, classify the next action, click. https://github.com/awlevin/typesafe-computer-use 5. realZachi/pg-jev Ask your Postgres tables questions in plain language. Judgments run inside the database. https://github.com/realZachi/pg-jev 6. kitze/skillbox Self-hosted versioned skills library for agents, with Jev picking which skill to load. https://github.com/kitze/skillbox 7. y0usaf/pi-jev Decision layer for the Pi coding agent: a measured tool-call gate plus typed answers on demand. https://github.com/y0usaf/pi-jev 8. typesafe-ai/system-one-adapter-python Official drop-in client backed by OpenAI or Anthropic. Build the shape now, swap in Jev later. https://github.com/typesafe-ai/system-one-adapter-python 9. dbreunig/building-with-jev-skill A skill for writing and improving programs that call Jev. The part most repos skip. https://github.com/dbreunig/building-with-jev-skill 10. itsmostafa/typesafe-mcp Single Go binary MCP server that exposes Jev judgments to Claude Desktop, Claude Code and Codex. https://github.com/itsmostafa/typesafe-mcp After going through these, the pattern is pretty clear: LLMs generate. Agents execute. JEV decides. That decision layer is what most agent stacks are still missing. Save this if you want to integrate Jev into your workflow ↓

这条是引用着另一条写的：x.com/0x_rody/status/2103165281149354256