stark

What it does

Conductor is a live release-review room that solves the slow, siloed nature of pre-production checks. Instead of routing a release through one reviewer at a time, it launches four AI specialists simultaneously: security, correctness, product impact, and reliability. Each specialist investigates the same release goal, publishes evidence into one shared room, and the system produces an auditable ship-or-hold decision a human can inspect, copy, or export.

How the agents run concurrently

Conductor creates four Mozaik createAgent participants inside one shared @mozaik-ai/core runtime. They share a RuntimeState containing the review goal, agent statuses, findings, event trace, and verdict; all agents start independently without waiting for another to finish. They publish semantic events through Mozaik’s event bus, and the browser receives the evolving shared state through SSE—making overlapping work, evidence convergence, and the final decision visible in real time. The repository includes the rewritten README, submission copy, no root license file, and no API key. I left unrelated pre-existing Python/CAID files out of the GitHub repo.