andrew-yangy/gru-ai

# Technical Summary Implements a 15-step multi-agent pipeline (triage, audit, debate, clarify, plan, build, review, ship) where role-scoped agents accumulate institutional memory across directives through a persistent context tree; each agent operates in isolated clean windows to prevent confirmation bias, with mandatory cross-agent reviews backed by bash script verification that prevent self-review and enforce mechanical separation between builder and evaluator. Built in TypeScript, integrates with git for isolated branching and version control, and targets solo founders automating full development cycles from ideation through deployment without human hands on code.