Composer-1 Unreliably Creates "Context Summary" After Hitting Token Limit

It’s worth noting that I don’t experience this issue with other models like Claude Opus, Sonnet, etc.—they handle context continuation more reliably. The problem appears to be specific to Composer-1.