What problem does it solve? Multi-agent orchestration sessions lose track of what each agent delivered, learned, and should improve next, making it impossible to run feedback loops or audit turn-by-turn progress. This Skill standardizes per-agent turn summaries and supervisor consolidation so every turn produces truthful, persisted, machine-readable records. ## Core Features & Use Cases - Agent turn summaries: Emit record-summary JSON capturing what was done, learned, files changed, blockers, and 1-2 self-suggested improvements per turn. - Supervisor consolidation: Merge child agent summaries into a canonical supervisor_turn_report with turn status, cross-agent insights, and a recommended orchestration strategy. - Bounded persistence: Store full entries in .git/.orchestration/events.jsonl and compact last-5-turn projections in context.json, never publishing to repo-memory/. - Use Case: After a Supervisor coordinates Researcher and Backend Engineer agents on a feature, use this Skill to record each agent's summary JSON, consolidate them into one executive briefing with a Ready/Blocked/Done status, and feed the last 5 turns into the next routing decision. ## Quick Start Ask the agent to end its turn with a record-summary JSON and an executive briefing note covering what was delivered, learned, and the next step.