jev-planner --agents codex,claude:deep,claude:quick --effort quick=low -o PLAN.md "Add per-user rate limiting to the public API"1 · Draft — Every agent — different providers, or several instances of one — plans alone against the repository, all in parallel; none sees another draft.[jev-planner] Drafting independent plans with Codex, Claude (deep) and Claude (quick)…2 · Jev judges — Jev scores completeness, feasibility and risk coverage, chooses a finalizer, and decides whether a cross-review would materially improve the plan.[jev-planner] Asking Jev for typed quality and routing decisions…3 · Cross-review (only when Jev asks) — When Jev asks for one, each agent sees every other agent's plan and returns a revised, standalone plan — which Jev judges again.[jev-planner] Cross-reviewing the 3 drafts…[jev-planner] Re-evaluating the revised plans with Jev…4 · Final plan — The agent Jev chose merges the strongest ideas into one plan, unless one reviewed plan already stands alone.# then "Synthesizing the final plan with …": the agent Jev chose merges the plans, or one reviewed plan is adopted as it stands[jev-planner] Wrote /home/you/my-app/PLAN.mdjev-planner
npm install -g jev-plannerSeveral minds, one plan
Every plan for the price of the slowest
The agents work in parallel, so each step takes only as long as your slowest agent — and every draft is kept in the run folder, so you already have other plans to compare.
Built by everyone, orchestrated by Jev
Each agent reads the others' plans, corrects their facts about the repository and revises its own — or, in a debate, objects and answers. Jev judges every round, calls for another when it would help, and picks the agent that merges the strongest ideas.
Any mix of agents
Codex, Claude, DeepSeek, Kimi and GLM, in any combination of two or more — even one provider several times over, each instance with its own model and effort.
The four stages, in detail
1Draft
Every agent — different providers, or several instances of one — plans alone against the repository, all in parallel; none sees another draft.
2Jev judges
Jev scores completeness, feasibility and risk coverage, chooses a finalizer, and decides whether a cross-review would materially improve the plan.
3Cross-review
When Jev asks for one, each agent sees every other agent's plan and returns a revised, standalone plan — which Jev judges again.
Only when Jev asks for it.
4Final plan
The agent Jev chose merges the strongest ideas into one plan, unless one reviewed plan already stands alone.