beastmode

Multi-agent orchestration framework for high-intensity feature implementation. Routes work across model tiers: frontier models (Claude Fable, Kimi 3, Opus/Codex) own design, architecture, and review sign-off, while economy models (MiniMax M3, Qwen/Gwen) handle implementation and mechanical validation in isolated worktrees, with a self-improving learning loop that promotes lessons back into skills. Harness-agnostic: works with Ultraswarm, GSD, delegate_task, or manual orchestration.

Pi-specific harness adapter for the universal Beastmode MofA orchestration framework (see the `beastmode` skill, v2.1). Composes five installed pi packages into the universal loop — pi-goal as the in-session goal engine, pi-dynamic-workflows as the fan-out / model-router, pi-loop-police as the anti-spin circuit breaker, pi-permission-system as the worker-contract enforcer, rpiv-todo + pi-telegram as live visibility. Use when Luis asks for beastmode, a Mixture-of-Agents run, autonomous goal execution, planner/worker/watcher, or a phase takeover inside a pi session.