I stopped asking my AI agents to read the project memory. Now the server does it for them.

Follow-up to I run 9 parallel Claude Code sessions — the bottleneck wasn't the model, it was...