Agent 협업 · DeepSeek Harness 플러그인 — 76 페이지

CLS-style replay consolidation pass with LLM summarizer and deterministic fallback

Mneme context backend: compile state and recall evidence into agent context at session start and pre-step

Mneme skill backend: lower active policy nodes into the DSH skill registry

Mneme tool backend: memory_remember/recall/list/forget/trace model tools

RecMem-gated LLM claim extraction pass with replay-logged raw output

Feedback-directed optimization: recall/confirm signals update salience and retrieval weights

Graph candidate index over claim/relation facts with PPR-style spreading activation

Mneme replay search backend: sessionQuery FTS hits mapped back to ledger nodes with literal fallback

Vector candidate index as a disposable projection cache with deterministic fallback

Mneme memory intermediate representation, ctx.memory engine, and pure compiler folds

MemGPT-style layered allocation policy

A-MEM-style note linking: shared subjects become relation nodes

Mneme front-end pass: lower DSH session events into typed memory IR nodes

Mneme optimize pass: TTL forgetting, contiguous-run consolidation, and catalog budget

Session projection pass: memory/catalog and memory/stats over DSH sessionProjections

Mneme solve pass: version resolution, evidence conditions, and activation analysis

Create, search, install, and apply versioned DeepSeek Harness plugins and profiles

DSH tools for planning and applying reproducible Plugin Hub Profiles through the local dsh-hub CLI

Version resolution and profile planning for DSH Hub

Portable DSH bundle, profile, and Hub API schemas

Incremental diff reviewer for DeepSeek Harness — Web UI review panel + /review command. 增量代码审查插件：checkpoint 增量队列 + 审查意见注入 agent.

DSH Plugins 4U 总插件：在设置中展示本仓库的自定义插件目录

DSH Skin Studio — 皮肤中心（浏览 / 试穿 / 应用 / 上传皮肤 + 内置 7 款皮肤主题 + /skins 静态资源路由）

TypeScript type definitions for DSH Skin Studio