Hollow makes agent coordination replayable

Hot take: this looks like a serious infrastructure project with a strong systems thesis, not just a catchy “event-driven agents” demo.

• –The repository backs up the core claim that Hollow centralizes agent state and coordination behind typed endpoints, including registry, scheduler, message bus, memory, handoffs, and `GET /agent/pickup`.
• –The event-kernel framing is directionally consistent with the project, but the strongest concrete evidence in the repo is structured state management and replayable session context, not a fully documented event schema on the homepage.
• –The “deadlock-proof” and “27 events” language reads as product marketing; I would treat those as claims until the event list and runtime behavior are documented in the code or changelog.
• –The project’s real value is simpler and more defensible: fewer shell probes, less session drift, and a cleaner abstraction boundary between agents and the underlying system.
• –The best fit audience is people building local agent runtimes, orchestration layers, or multi-agent toolchains who care about determinism and traceability.