메모리 및 지식 · DeepSeek Harness 플러그인 — 49 페이지

Thread session memory for dsh — lossless event capture + status-card injection (dsh plugin, standalone repo).

TiddlyWiki 5 as the DSH persistent knowledge base: tiddlywiki_* agent tools (search/get/put/delete + git sync), a full TiddlyWiki editor embedded in the GUI center column, a floating quick-note widget in the chat area, and git-based sync with auto-commit. Mounts via the official dsh plugin system — no DSH source changes.

分层记忆 + 预算管理(Tiered memory with per-tier budgets)。 移植 Hermes Agent 的记忆纪律:dsh 生态约 65 个记忆插件全是单层存储,本插件是 **分层 + 预算**概念的首个实现。

TiMEM long-term memory for DeepSeek Harness: MCP bridge (credentials from ctx.credentials) + bundled memory/rule/knowledge skills

听雪 · 双模式虚拟生命系统：LanceDB 自建向量记忆库 + 自建实体关系图谱 + 小 LLM 语义推理，聊天模式与 agent 模式双模式，全部交互在 QQ（dsh-notifier）。

DSH Desktop Windows titlebar assistant: (1) insets the conversation center column so full-column plugins (taskboard, ssh, im, better-sidebar, ...) clear the frameless titlebar (36px drag strip + top-right caption controls), and (2) a configurable titlebar strip that renders user-defined feeds in the freed top space — static text, HTTP + regex, user JS functions (sandboxed), or built-in system metrics (CPU/memory/network/disk/battery/uptime), with long content paginated automatically or via hotkeys. No DSH source changes, uninstall to revert.

Automatic TMCRA recall, injection, and role-separated writeback for DeepSeek Harness

System information tool for DeepSeek Harness: OS, CPU, memory, uptime, and network facts the agent can read when diagnosing the local environment

Per-project memory for DeepSeek Harness: auto-creates memory/PROJECT.md, DECISIONS.md and SESSIONS.md per project, loads them into every session, and records durable outcomes by classification.

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

Long-term memory for DeepSeek Harness in Claude Code's format: a private store, an overlay that also reads Claude Code's memories, or one store shared with Claude Code, with provenance on every file.

Let the AI know you. 让模型主动记下你的判断依据（原则/红线/工作方式），跨会话复用 —— 有容量、会遗忘、每条都说得出为什么记。

vibeweaver 技能的 DeepSeek Harness 插件封装：渐进披露契约段 + 机械门禁（assert_artifacts 证据检查）+ 编码任务自动激活 + 回合守卫。

常驻视觉服务：直连视觉模型（默认 opencode-go/minimax-m3，回退 zai-coding-cn/glm-4.6v）。describe_image / subagent_vision 工具 + 粘贴图片自动转译（llm/stream 钩子）+ 输入框视觉状态小胶囊与详情页（活动日志：指令/思考过程/输出）。零子代理、零 agent 上下文开销，按会话记忆窗支持视觉追问与验收。

A DSH visual workbench for Obsidian-style knowledge workspaces and local Comfy MCP workflows.

外贸获客插件 for DeepSeek Harness (dsh): 3-layer Google lead search → contact extraction → ICP-aware rule+AI scoring → email finder/verifier → cold emails (spintax, daily cap, dry-run, threading, open/click tracking) → IMAP reply scan with AI classification → follow-up sequences → CRM pipeline → SOP stage machine with human approval gates → knowledge base → website monitoring → inbox warmup → deliverability checks → cron jobs → WhatsApp broadcast/media/quotes. 零依赖。

dsh-dream-skin 的附属插件：在「设置 → 外观」里按图片记忆调整壁纸位移（X/Y 像素微调）

dsh-dream-skin 的附属插件：在「设置 → 外观」里按图片记忆调整壁纸位移（X/Y 像素微调）

无人值守定时任务调度器：cron/固定间隔/一次性三种调度，到点在 web profile 进程内拉起全新 root agent 执行任务指令（全栈继承 agentrouter/wb-memory/skills/hooks），会话落盘可续聊；每任务可钉模型与 Agent 预设（预设被删/损坏时三级降级跑完并留痕）；提供 cron_* 模型工具、设置页完整管理与侧边栏速览。零 npm 运行时依赖、零遥测。

把 WorkBuddy 的文件化记忆机制带入 DeepSeek Harness：分层注入（全局画像 + 工作区置顶 PINNED + 摘要/相关条目 + 今日日志 + 近 N 日回顾）；记忆源为各 dsh 工作区 .workbuddy/memory/，纯 Markdown 无向量库；含 auto-digest 流水记录、会话收尾小结、auto-distill 日志提炼、memory_search 原生工具（长期记忆+近期日志双源检索）、超阈值 LLM 整理（同轮反馈重试）与 TTL/回收站治理，并在设置面板提供管理界面。

DSH Web 界面的半透明红色关闭按钮：SVG 图标、可拖动、记忆位置，二次确认后优雅关闭当前 DSH 进程并关闭本页

Adds Chinese web-fiction planning, file-backed memory, chapter workflow, consistency review, and style profiling skills to DeepSeek Harness agents.

DeepSeek Harness 插件：写作档位（T1–T4 风格 × L1–L5 严谨）、项目记忆（.mem 三区 + 守卫）、去 AI 味后处理（whale_polish）——双半体（host 工具/注入 + 浏览器快捷条/工具卡）。

基于 **DeepSeek Harness (DSH) + Obsidian** 的个人知识库体系（源自 Karpathy LLM Knowledge Bases）：LLM 负责编译与维护，人类负责选材、审核与提问。