What it may solve
Three-layer local memory for DeepSeek Harness: L1 frozen-snapshot memory (MEMORY.md/USER.md injected into the system prompt, prefix-cache friendly) + L2 searchable knowledge base (facts + SQLite FTS5, Chinese search via sliding-window tokenization) + L3 import from Hermes/Claude Code/Codex/WorkBuddy with dedup. Auto-injects relevant facts on the first step or memory trigger words (agent/pre-step) and after skill loads (tools/post-execute); data stays on the local drive.
Imported third-party catalog description; not a Registry verification conclusion.