multi-model-review

方法论与经验教训手册 — AI 协作实战 50 条踩坑实证。The Methodology & Lessons Learned Handbook: 50 empirically-grounded lessons from AI collaboration practice.

从20个项目中系统提取可复用方法论模式的实验记录。包含10轮正式审查实证（4后端）、58项发现、G5可追溯审计。不成熟框架，诚实的实验记录。

Three tools for running AI coding agents on real work — a deterministic chat front door, a durable campaign runner, and multi-model review with adjudication. They share a doctrine, not a runtime.

pi package: turns a PRD into governance (SOUL/AGENTS/SYSTEM/MEMORY) and that governance into an issue-driven auto-develop pipeline with per-role model routing, independent multi-model review, and hard run budgets.