dsh-fresh-start

As conversations in DeepSeek Harness (DSH) grow to hundreds of thousands of messages, the system often suffers from severe performance bottlenecks. With the live event tree constantly expanding, building context becomes sluggish and consumes excessive memory, causing noticeable lag during subsequent operations. To solve this issue, the /fresh command streamlines a complex workflow into a single click. It first forces a compaction of the current session and then uses an LLM to distill the history into a natural language summary. This summary serves as the seed for a new session, inheriting the existing environment while automatically archiving the old one to free up resources, allowing for a seamless transition back to work. This plugin is essential for developers, researchers, and heavy users who manage long-running projects or large-scale data. It ensures a smooth experience even in massive sessions and features robust fallback strategies—such as proceeding with archiving even if summarization fails—to guarantee you never lose progress due to memory constraints.