在 Claude Code 前加 Jev 路由与命令守卫

CyrilXBT @cyrilXBT原推发布于 2026.10.11

在 X 上打开原推

i can’t believe people still send every task to Opus one Jev call in front of your agent and every task finds its tier: → Jev reads the task and picks small, mid or frontier → Haiku and Sonnet take the easy work → Opus 5.5 only gets the hard stuff → a Jev guard checks every bash command before it runs the router costs $0.042 per million tokens. output is free. paste this into claude code 👇 claude plugin marketplace add typesafe-ai/skills claude plugin install typesafe@typesafe-ai Add a Jev router and a Jev command guard to this project. Router: before each task, ask Jev (jev-latest) one choice question (small, mid, frontier) and one noul question (plan before editing?). Map small to Haiku 5.5, mid to Sonnet 5.5, frontier to Opus 5.5. Low confidence routes up a tier, never down. Guard: a PreToolUse hook on Bash that asks Jev if the command is destructive. Under 0.35 run it, 0.35 to 0.7 ask me, over 0.7 deny. Read TYPESAFE_API_KEY from env, time out after 4 seconds, and fall back to asking me. Show me the diff before changing anything. All the facts in both posts come from published sources: the install commands, Jev’s pricing, the force push and rm example scores, and Opus at $4 / $20.

这条是引用着另一条写的：x.com/1373665408193036293/status/210869352490011…