Memory and knowledge · DeepSeek Harness plugins — Page 17

会话记忆 → OKF 知识沉淀插件:预测驱动的神经自我学习记忆系统。把会话中高价值内容按 OKF v0.1 规范沉淀为长期记忆,跨会话自动唤起。Session-to-OKF memory plugin with neuro-self-learning (predictive recall, uncertainty-driven capture, reinforcement feedback).

配置打包器（Agent config packer for DeepSeek Harness）：把本地 Agent 资产（Skills/会话/Profile/设置/记忆）按模块打包成 zip——迁移、分享、恢复、差异对比、隐私安全扫描，全部可选。v0.2.5：记忆模块语义澄清（包内只有 Markdown 镜像，运行时 SQLite 真库不在包内；中英 README 增「记忆怎么迁移」）+ 客户端 bundle 迁入 lib/；全链路异步 IO（fs/promises + execFile，不再阻塞事件循环，对外 API 一律返回 Promise）+ /packer/api/* 防护（鉴权默认 fail-closed：只用官方 connection.requestRejection，拿不到就 403；请求体上限 8MB 与速率限制 60 次/分钟在完整缓冲请求体之前判定；错误文案脱敏）、测试补齐；v0.2.3：恢复安全加固（目标路径 containment / 缺指纹 fail-closed / 备份+原子替换+失败回滚 / tar 成员白名单 / SQLite 库默认排除）+ 隐私扫描补齐（Unix 与 UNC 路径、无引号密钥与密钥形状、全量计数、无扩展名文本）+ peerDependenciesMeta 标记宿主内建 peer 为 optional；v0.2.2：管理面板按打包工作流设计语言（阶段流程条/器材面板/diff 色带/橙琥珀青品牌色）+ 深色适配（DSH 主题跟随）

DSH web plugin: per-host personalization — global custom instructions, local long-term memory (collect / retain / integrate / manage), and reply tone presets.

Long-term memory plugin for DeepSeek Harness: persistent cross-session markdown memory with a soul/persona file, auto-injected at every session start, plus remember/recall/consolidate/forget workflows and portable migration tooling.

DSH (DeepSeek Harness) QQ 机器人长期记忆插件：对话→知识图谱→向量召回，回合前注入相关记忆，回合后后台抽取入库。Zero-dep graph + vector long-term memory for dsh-onebot-qq.

Conversation history recall for DeepSeek Harness: search the original text of every session (like searching chat records) — literal/fuzzy/semantic retrieval, fully local & offline. AI never forgets what you told it.

Evidence-based experience learning for DeepSeek Harness

SiYuan (思源笔记) for DeepSeek Harness: bridges the local SiYuan MCP endpoint and registers its official note tools as mcp__siyuan__*.

Microsoft SkillOpt-Sleep integration for DeepSeek Harness: give your dsh agent a nightly sleep cycle that harvests past sessions, replays recurring tasks, and consolidates validated skills behind a held-out gate.

以角色为中心的 DSH 记忆、状态与会话插件

DSH 通用学习 Agent「学习伙伴」：文档式笔记插件——Obsidian vault 多根检索、文件夹规划与硬门禁落盘、微目录与覆盖度、学习进度与跨会话记忆

DSH teacher plugin: Socratic tutor that leads you to answers from a markdown question set, tracks knowledge gaps in-session, and retests them on a spaced-repetition schedule.

User preference memory for DeepSeek Harness: persisted user profile with memory_get/memory_update tools and system-prompt injection

In-process graph memory for DeepSeek Harness, backed by TriviumDB.

Video as a surface that actually plays. A right-bar tab type (row `video`, kind `video`) for MP4, MOV, WebM, MKV, AVI, WMV, FLV, OGV, MPEG-TS and 3GP: the film streams through dsh-media's Range-capable route (so a 2 GB file seeks instead of being read into memory), plays in a pane that fits it, and carries a facts panel with the container, every stream, the chapters as jump targets and what a browser can and cannot decode. When a container or codec is one no browser opens, the tab offers the one action that fixes it - a remux or a transcode through dsh-media's ffmpeg, into a cached MP4 it then plays - with ffmpeg's own progress while it runs. Client-only: the bytes, the probe and the conversion all belong to dsh-media, which says so in a sentence when it is not installed. Alpha.

VRChat creator toolkit for DeepSeek Harness (DSH): a 414-page VRChat knowledge base with proactive red-line injection, 9 purpose-built tools, and 6 bundled skills. Covers Udon/UdonSharp, networking and Persistence, Avatar/PhysBone, Shader pipelines and Quest limits; can test whether an API is exposed in Udon, lint generated C#/Shader against hard constraints, inspect the local Unity install and license, scaffold a new World or Avatar project, and drive VPM packages through vrc-get. Injects nothing unless the vrchat preset is selected. / VRChat 创作者工具链：414 页知识库 + 红线主动注入 + 9 个专用工具 + 6 个随包技能。覆盖 Udon/UdonSharp、网络同步与 Persistence、Avatar/PhysBone、Shader 管线与 Quest 限制；能查 Udon API 是否暴露、扫描生成代码的硬约束违例、检查本机 Unity 与许可证、从零创建 World/Avatar 工程、并通过 vrc-get 管理 VPM 包。未选用 vrchat 预设时不注入任何文本。

Playable world-model toolkit for DeepSeek Harness: look at frames, compare pred vs GT, watch the action track, name the 3D / pixel / latent route, and RSI the research loop.

Eyes for text-only DeepSeek on DeepSeek Harness: image understanding via Antigravity IDE quota (default, flash/pro), OpenAI-compatible VLM endpoints, Gemini API, or local Ollama — model-invokable vision tool, wrapper adapters for deepseek/opencode-go, evidence memory, and a polished client panel.

陪伴型机器人的情感记忆层：事件编码、情绪加权遗忘、联想召回、关系与技能，还会承认自己记不清。An emotion-weighted memory layer for long-running companion bots.

huahua-dsh-memory-os：一站式 AgentOS（DeepSeek Harness）——分层记忆/自我进化/技能/待办/COI 调度/AIMS 身份角色记忆管理/人事部专家调度深度融合；让 Agent 数据裂变生长。

知识银河 — DeepSeek Harness 知识收录看板插件：学习 / 学术 / 日常三种模式，对话收录与定时收录，分组看板与时间轴，单星系星图与跨模式知识联系。

Cross-session project memory for DeepSeek Harness: seven-layer SQLite memory, first-turn snapshot injection, per-message keyword hits, memory_remember/search/project tools, automatic reflection with reflection-fold UI, and nightly dream consolidation.

Task-conditioned Chat and Work session memory for DeepSeek Harness

Generic stateless dialogue compression that mimics human memory — forgetting-curve bounded context for LLM conversations, with a ready-to-use adapter module for DeepSeek Harness (DSH).