magna

Multi-agent team orchestration playbook for non-trivial coding work, with AGENTS.db as the storage and recall middleware. Use when acting as Team Lead in a Claude Code session that warrants subagent orchestration — starting or resuming a multi-round build or refactor, running an investigate-then-fix defect loop, executing a hypothesis-sweep experiment loop, dispatching specialized subagents (Topic Director, Synthesizer, Scout, Architect, Builder, Verifier, Investigator, Historian), picking an operating mode (Mode 1 experiment loop, Mode 2 build/refactor, Mode 3 defect fix), briefing an agent, deciding what to do after a Verifier round, handling cross-repo branch coordination, judging whether to surface to the user, writing session retros, or managing the AGENTS.db layered context store (base/user/delta/local) that backs the team's durable knowledge. Trigger on phrases like "team lead", "dispatch", "spawn", "subagent", "Builder/Verifier/Director/Scout/Architect/Investigator/Synthesizer/Historian", "next round"