Five CLAUDE.md rules let Jev pick the execution plan first
Connecting Jev in CLAUDE.md lets it decide the model, effort level and number of agents before each task, then hand execution to Sonnet 5.5.
JEV + SONNET 5.5 is INSANE The trick is 5 rules in your CLAUDE. md Connect Jev, then paste this: 1. Before each task, ask Jev to pick a work plan from the approved list: which model, which effort level, one agent or two. If Jev isn't confident, use the default plan. 2. Clear coding tasks: Sonnet 5.5, medium effort. Hard reasoning or strict JSON: raise the effort. 3. Only start a second agent for separate research or a review I asked for. Give it the task, the files, the limits, and what done looks like. 4. Do exactly what was asked. No extra features, files, tests, refactors or docs I didn't request. Only stop to ask me if you're blocked or about to do something that needs my approval. 5. Before saying you're done, run a test, build or type check that proves the change works. If nothing can run, say what's missing and why. Same task, two timers. Try it on your next session. follow @cyrilXBT
This post quotes another post: x.com/1373665408193036293/status/210474869077881…