メモリ・ナレッジ · DeepSeek Harness プラグイン — 25 ページ

Edit DeepSeek Harness global AGENTS.md instructions and memory directly in Web settings.

Knowledge-territory map for DeepSeek Harness: session conversations grow into a living map of concept 'cities' clustered into discipline 'continents', with typed routes and user-built 'bridges' (small containers holding your own notes and cross-discipline cards). AI proposes adjacencies; you light the connection. Data flows over an in-memory Connection RPC channel and a local markdown vault, never the session log.

DSH 插件审核器：在新插件加入 profile 前扫描组合兼容性——重复工具注册、entry id 冲突、peer 版本不匹配、记忆插件唯一性、渠道插件凭据，输出风险报告，预防启动崩溃。

DeepSeek Harness context manager: auto-records every turn, injects priority-sorted memory and pinned cross-session notes into each model request, and folds real conversation history to save tokens. NOTE: manual multi-step install — copy the three package folders into profiles/node_modules and add the provided cordis.patch.yml rows, then restart DSH; not a standard one-command install.

Cross-disciplinary association popup for DeepSeek Harness: pick the disciplines you care about, then draw a card — the LLM offers one concrete structural link between another field and what you're currently working on. A tool for thought (a hook, never a finished analysis), surfaced faint and ignorable over an in-memory Connection RPC channel, never the session log.

Focus board for DeepSeek Harness agents: durable, model-maintained notes in the session workspace that pin the objective, constraints, and decisions across compaction and sessions — with structured entry tags (kind/status), a grouped summary view, automatic context injection, an archive on clear, and an optional read-only web panel

Verification-gap probe for DeepSeek Harness: a backend plugin that watches each user turn, judges steer/ride/neutral with a cheap LLM, and accumulates how far the AI's autonomous progress has outrun what the user has actually verified — surfacing a faint, ignorable prompt (over an in-memory Connection RPC channel, never the session log) when they've drifted out of grasp.

把「记忆」带进 DeepSeek Harness：极简文本记忆系统，双轨记忆（流水日志 + 人物/产品实体画像），大模型为核心驱动。无需安装其他软件，无需编译，无第三方依赖。

记一笔 — meaning-marking popup for DeepSeek Harness: like e-book marginalia, jot in your own words what a moment meant to you. Each mark snapshots the moment (immune to context compaction) into a searchable personal ledger that outlives any session. The AI keeps the log; you keep what mattered. A tool for thought, never written for you, over an in-memory Connection RPC channel + a plugin-owned durable file — never the session log.

Session distillation memory plugin for DeepSeek Harness: distills completed turns into durable knowledge entries surfaced as memory-<slug> skills

Structured long-term memory system for DeepSeek Harness -- unit-based brief/full pairs with auto-injection

节奏台 — a floating overlay hub for the pace-popups series. A draggable, position-persisted bar mounted on the app-global shell.overlay slot; v1a manages which pace popups (grasp / crosslens / jiyibi) are shown via a shared localStorage flag the popups honor live. Later versions host richer panels (记一笔 ledger browsing + read-only re-read). Zero-build hand-written client; no backend state in v1a.

Shared visual skin for the dsh-pace-popups suite: a no-UI plugin that injects one set of design tokens (--pp-*) and utility classes (.pp-*) into the page. Editorial 'lieflat' texture — hairline strokes, generous whitespace, ledger rails, flat surfaces, brightness-carries-hierarchy — while keeping the suite's color (discipline hues + a single gold accent). Load it once; every other popup consumes the same tokens.

Local knowledge base RAG for DeepSeek Harness -- index project files and search for relevant context

Semantic long-term memory for DeepSeek Harness: embedding-based retrieval over a persistent cross-session memory store, with model-facing tools, proactive per-question recall, automatic conversation summarization, and auto-selected embedding provider (apiKey present → API, otherwise local).

DSH-native Sirchmunk settings plugin for graph, cluster, monitoring, history, and privacy workflows.

System resource monitor for the DeepSeek Harness bottom status bar: CPU / memory / disk / uptime, updated live, click to refresh.

LAN project memory synchronization for DSH Desktop

Translation toolkit for DeepSeek Harness agents: segmentation, term extraction, durable glossary and tone memory, cross-segment consistency checks, source-target QA, heuristic quality scoring, tone guides, and a durable translation memory

DSH 插件：春笋（ChunSun）驱动的 AI 原生项目交付 —— 需求/Run/Step/场景/用例/工作记忆 + 验收闭环（pm = project management）

DSH extension bundle: presentation-deck expert — persona + LibreOffice Impress MCP (9 tools) + matplotlib chart kernel + live browser preview. Zero npm dependencies.

L0~L3 分层蒸馏记忆插件 for DeepSeek Harness：自动捕获对话（L0）、抽取原子记忆（L1）、整合场景块（L2）、蒸馏核心画像/团队方法论（L3），并在模型步骤前自动召回注入。移植自 MemoryCore (TencentDB Agent Memory) 的管线设计。

DSH 持久化项目大脑：理解代码架构、记住项目历史、跨 Session 保持上下文

DeepSeek Harness plugins for project-level persistent context: memory consolidation (MEMORY.md + CONTEXT.md), low-frequency autolearn for project skills, session archives with a mechanical index, and automatic handoff.