local-model

Auto-continue for DeepSeek Harness: when a local model hits its output-token cap, automatically send "continue" so the reply keeps flowing