dsh-project-memory

In the DeepSeek Harness development workflow, models can often propose excellent code solutions but struggle to establish persistent, reliable project memory. dsh-project-memory is designed to address this pain point by transforming team decisions, lessons learned, and facts into versioned project records, ensuring critical information is never lost when sessions end. The core strength of this plugin lies in its "human approval" boundary mechanism; candidate memories generated by the model must undergo strict review via the /memory approve command to enter the final memory bank. Additionally, it employs an evidence-based design, recording source code relative paths, hashes, and line numbers rather than copying snippets directly, ensuring freshness and accuracy through local, deterministic search. It is ideal for software development teams that require high collaboration and knowledge retention, as storing memories as Git-friendly JSON files allows team members to easily share project context and manage memory changes using version control, maintaining consistency and alignment in complex development environments.