dsh-statecore

When working with DeepSeek Harness on complex projects, session context often fragments and is lost as conversations deepen. dsh-statecore addresses the pain points of "memory fragmentation" and "context loss," serving not just as a simple storage but as a native memory engine equipped with evidence chain tracking. Its core capability lies in its "native" nature; by automatically ingesting human and assistant messages without requiring model cooperation, it converts conversations into auditable facts. It also supports auto-injecting memory into the model's context and ensures no critical information is lost during session compaction via its digest mechanism. Paired with native tools like remember and why, it provides structured JSON output in Code Mode, significantly boosting the credibility of debugging and reasoning. It is particularly suitable for developers and teams who need to maintain context consistency over long projects, perform complex reasoning, or require audited sources of facts. By sharing a SQLite store, it seamlessly collaborates with MCP clients, enabling the AI to make decisions based on past facts in every interaction, thereby building more intelligent and reliable agent workflows.