TypeSafe's official skill teaches Claude Code to build agents on Jev
Opus 5.5 plans the work while Jev handles every yes/no decision in the loop, falling back to Opus only on low confidence, with a prompt you can hand straight to your agent.
This is absolutely f*cking insane TypeSafe just dropped an official skill that teaches Claude Code to build agents on JEV i genuinely don't understand why nobody is using this yet the trick is stupidly simple: Opus 5.5 plans the work. JEV makes every yes/no call inside the loop the setup: > add the skill so Claude Code reads TypeSafe's live docs before writing any code > list every fork your agent hits: approve or reject, which tool, spam or not > send those forks to JEV with the options spelled out > low confidence -> hand it to Opus 5.5 the easy calls never touch the expensive model Opus gets a tiny pile that actually deserves deeper analysis faster. cheaper. way less compute burned basically, JEV answers the phone so Opus only takes the calls that matter LLMs think. JEV decides. code does send this to your agent: "read what TypeSafe released for JEV: the docs, the repo and the official Claude Code skill. then go through my agent, find every step where it picks yes/no or one option from a list, move those to JEV and keep Opus 5.5 for the hard cases. show me the plan before you change anything"
This post quotes another post: x.com/archiveexplorer/status/2103590048230695408