Comment by swrly

Parallel workers in git worktrees is clever. We have a similar pattern with fan-out/fan-in — you can split an array across parallel agent executions and collect results. The SQLite-structured JSON output approach is interesting for coordination. We use template variables + scratchpad (Redis-backed shared state) for inter-agent data flow. Different trade-offs — yours gives you more control at the cost of more infrastructure.