Плагины

Глубокая интеграция с Mnemon: локальная трёхуровневая память (Runtime Memory, извлекаемые Documents, курируемые Memory Spaces).

Model-driven context compression (Active Context Pruning) for DeepSeek Harness: the model decides when and what to compress.

Adds persistent execution contracts, recursive work graphs, critical clarification, and evidence gates for long or underspecified Harness tasks.

DeepSeek Harness 动态上下文管理插件（Dynamic Context Pruning for dsh），对标 opencode-dcp