I built a governance layer for CrewAI (pip install crewai-governance)

37% of multi-agent failures are coordination breakdowns. Not capability issues -- coordination issues. Agents finish and their work vanishes. Two agents do the same task without knowing it. New runs repeat old mistakes. I built crewai-governance to fix the three most common ones: - Exit reports: structured JSON after every crew run (what each task did, what worked, what failed, token usage) - Overlap detection: before kickoff, scans active crews and warns if mandates overlap - Knowledge inheritance: automatically injects prior run summaries into new crew context This came out of building a full governance framework realizing nobody will adopt 39 sections of governance rules, but they might adopt 3 features that solve real pain. What coordination problems are you hitting with multi-agent systems? What would you actually want from a governance layer?