DeepSeek Harness Memory Plugin: dsh-observational-memory

This plugin is designed to address the loss of contextual detail caused by compression in long DeepSeek Harness sessions. It continuously maintains an “observation–reflection–cleanup” pipeline in the background to consolidate session details before compression. Core features include compression rendering with zero model calls, evidence recovery via `recall`, and a session rollback button. Installation must meet specific version requirements, while configuration supports threshold control, model overrides, and debug options, making it suitable for long-conversation scenarios and helping ensure information continuity.