yansera/dsh-living-memoir 插件介绍
`yansera/dsh-living-memoir` 是 DeepSeek Harness (DSH) 的跨会话记忆插件，旨在解决传统追加式记忆导致的膨胀与冗余问题。它将记忆存储为本地 Markdown 文件，利用宿主 LLM 能力进行整体重写与整理。 该插件采用三层结构管理记忆，支持自动记录、手动编辑及工具调用。通过防呆机制和元数据（如置信度、钉住状态）优化记忆质量，防止内容丢失或重复。用户可在设置中选择改写模型与语言，且所有数据完全本地化，支持版本控制。
Read MoreDeepSeek Harness 记忆插件：dsh-memory-dreaming
# dsh-memory-dreaming 插件：将对话转化为长期记忆 **前言**：在 DeepSeek Harness (DSH) 的开发与使用中，会话历史膨胀导致检索困难是常见痛点。`dsh-memory-dreaming` 插件专为解决这一问题设计，它不依赖 Channel Core，直接在本地维护结构化记忆，将验证过的对话转化为日常日记、梦境整合、每周复盘和可检索的长期事实。 **核心功能**： * **增量式处理**：每次处理最多 7 条新日记，回溯 30 天，仅处理归档副本，不重扫原始对话。 * **日常与复盘**：自动整理 `memory/YYYY-MM-DD.md` 并定期合
Read MoreDSH Hermes Memory: A Lightweight Hierarchical Memory Plugin for DeepSeek Harness
DSH Hermes Memory is a lightweight long-term memory plugin for DeepSeek Harness (DSH), designed to address context history bloat and forgetting issues. It manages agent context without requiring a separate database by using hierarchical storage (such as user preferences and project facts) and automatic refinement and compression mechanisms. It supports privacy encryption, tool-based query management, and compatibility with the DSH environment, helping developers quickly build agents with long-term memory capabilities.
Read Moredsh-workspace-memories: Persistent Workspace Memory for DeepSeek Harness
`dsh-workspace-memories` is a plugin for DeepSeek Harness (DSH) designed to address the limitations of stateless agents and static updates to `AGENTS.md`. It provides workspace-level persistent memory through external storage, with features such as sidebar management, agent-initiated saving, and automatic injection into system prompts. The plugin supports configuring either home or workspace storage modes and includes limits on file naming, size, and quantity. A new session must be started after installation to update the tool list.
Read Moredsh-deepbrain: Bringing the DeepBrain Memory Plugin to DeepSeek Harness
The DeepSeek Harness plugin, dsh-deepbrain, introduces DeepBrain’s structured memory capabilities to agents. Instead of storing raw documents, it stores refined “judgments” and evidence chains, and provides six streamlined MCP tools that enable agents to retrieve source-attributed citations, query person information, and obtain topic summaries. The plugin emphasizes real-time invocation, no caching, no write access unless specifically requested, and strict API permission management, aiming to address the lack of reliable verification in agent memory. It is suited to developers working with meeting records and interview content.
Read Moredsh-context-tree: Reusable Trajectory Tree Context for DeepSeek Harness
`dsh-context-tree` is a DeepSeek Harness plugin maintained by wr-web, designed to address the lack of intuitive management when reusing context across sessions. It converts completed turns into visualized, branchable checkpoints, enables resumption from any precise point, and controls the recall scope to reduce redundant exploration. Through a bounded recall mechanism (limiting node lifetime and size) and token accounting, the plugin makes context reuse measurable and is suited for development scenarios that require long-term maintenance of agent sessions.
Read Moredsh-echo-memory: A Cross-Session Persistent Memory Plugin for DeepSeek Harness
dsh-echo-memory is a DSH-specific cross-session persistent memory plugin maintained by JxaMe, designed to solve agent memory interruption issues. It adopts a minimalist architecture with local-only JSON storage and does not depend on vector stores or databases. It provides five tools (such as save, retrieve, and delete) for agents to invoke, supporting memory saved from natural conversations and AI-suggested memories. It uses the BM25F algorithm for on-demand recall and features workspace isolation and corruption self-healing. The plugin also offers a browser-based floating UI that supports drag-to-select text saving and memory management, making it well suited for DSH developers who require long-term memory and data privacy protection.
Read Moredsh-error-improvement: Transforming User-Confirmed Experience into Persistent Agent Memory
The `dsh-error-improvement` plugin is designed to address the problem of agents repeatedly making mistakes in DeepSeek Harness by enhancing robustness through persistent memory mechanisms. It operates independently of external services, using pure file-based storage to convert agent execution results into persistent memories. Its core workflow includes monitoring errors and generating drafts that require mandatory user confirmation in Settings before taking effect; offering Assist, Strict, and Off injection modes; detecting repeated errors and enforcing interception; and promoting validated recipes into skills. All data is stored locally, must be enabled through profile configuration, and is suitable for agent development scenarios requiring long-term memory and high reliability.
Read Moredsh-plastic-memory: A Memory Plugin for DeepSeek Harness
`dsh-plastic-memory` is a memory management plugin for DeepSeek Harness, designed to address redundancy and contamination in long-term agent memory. Core features include five memory categories, workspace isolation with global read-only protection, a multi-layered governance system, causal tracing, and persistent Markdown storage. The toolset covers memory saving, search, snapshot management, and full-store health scanning. The plugin supports customizing memory types and governance policies through configuration, making it suitable for developers building agents with long-term coherence. It is currently in Beta and requires a Node.js environment.
Read Moredsh-memory (Cairn): Persistent Memory for DeepSeek Harness
`dsh-memory` is a memory plugin for DeepSeek Harness (DSH), designed to address context loss after agent sessions end. It provides persistent cross-session storage with three-tier memory scopes: global, project, and user. The plugin leverages the BM25 algorithm for highly relevant retrieval and supports summary-first injection as well as step-level automatic recall. It also features automatic identity updates, security scanning, a deduplication pipeline, and two-tier lifecycle management, along with a management interface and manual review mechanism, making it suitable for complex task development that requires long-term context maintenance.
Read Moredsh-ontology: A Typed Ontology Memory Plugin for DeepSeek Harness
dsh-ontology is a structured knowledge framework plugin for DeepSeek Harness, designed to address fragmentation and inconsistencies in agent textual memory. It includes a TBox (schema definitions) and an ABox (fact assertions), supporting schema-constraint-based validation, automatic inference (e.g., transitivity and symmetry), and dependency-aware retraction. The plugin enables agents to define classes, relations, and assert facts, upgrading bag-of-texts memory into a controlled, reliable knowledge graph, suitable for scenarios that require handling complex dependencies.
Read MoreDSH Plugin: satan9394/dsh-persistent-memory — Cross-Session Persistent Memory Compression
satan9394/dsh-persistent-memory is a memory plugin for DeepSeek Harness, designed to address the challenges of difficult cross-session memory retention and high token consumption for agents. Inspired by the design of claude-mem, it provides cross-session compressed storage, lifecycle hook capture, three-tier retrieval (search/timeline/details), and progressive disclosure. The plugin supports private tags and hybrid search to significantly reduce token usage. There is currently no official installation command, so users must manually access the plugin directory or GitHub repository to deploy it.
Read Moredsh-memories: Dual-Ledger Cross-Session Memory Plugin for DeepSeek Harness
The `dsh-memories` plugin in DeepSeek Harness uses a dual-ledger system to address agents' challenges in cross-session memory and progress tracking. It automatically extracts long-term facts and project progress, and provides noise filtering, memory decay, and backup capabilities, while supporting LLM integration and manual command control. Developers can manage the ledgers through slash commands or tool functions, making it suitable for complex project scenarios that require long-term context maintenance.
Read Moredsh-daoing-memory: Self-Evolving Memory Plugin for DeepSeek Harness
`dsh-daoing-memory` is a persistent memory system designed for DeepSeek Harness (DSH) agents, intended to address signal flooding and hallucination contamination in conventional approaches. The plugin builds trust through an "earning" mechanism, classifies memories into semantic (facts and interests) and experiential (lifecycle experiences) types, and ensures data accuracy and traceability through an audit ledger and append-only logging. It provides end-to-end tooling for recording, extraction, recall, and auditing, with support for human intervention via a workbench. Developers must manually install the plugin and skills and restart the service; agents can then collaborate to build memories, converting only verified experiences into high-trust assets.
Read Moredsh-diary-detective (Cyber Diary Detective): A warm mirror reflecting your diary.
# Preface When building agents using frameworks like DSH (DeepSeek Harness) that adopt the "everything is a plugin" philosophy, you must implement long-term memory yourself: deciding where to store it, how to compress it, and how to retrieve it next time—every step involves manual intervention. Diaries represent a typical scenario: they require archiving the original text while ensuring the model can grasp key points when reviewing them the next day, without starting with a cold, statistical report. Here, we introduce dsh-diary-detective (Cyber Diary Detective), which packages diary I/O, memory compression, and gentle feedback into a single plugin. The project is maintained by SageChenAI and is licensed under the MIT License. # What is it One-sentence positioning: A warm mirror that reads diaries, reveals blind spots, and provides gentle feedback. It performs two distinct layers of functions: 1. Mechanical Layer: Storing spoken diaries as files and reading back the original text
Read Moredsh-agent-memory: Agent-driven long-term memory for DSH
dsh-agent-memory is an MIT-licensed long-term memory plugin (v0.3) built for DeepSeek Harness (DSH), aiming to provide cross-session structured memory capabilities. It adopts a hierarchical storage architecture, dividing memory into L1 facts (preferences and environment), L2 knowledge (reusable content), L3 contexts (timeline of experiences), and a summary layer, while supporting routing for both global and workspace scopes. The plugin exposes tools such as remember, recall, memory_browse, update/forget, and memory_stats to agents, supporting multi-dimensional retrieval and time pyramid browsing. Its key features include automatic time bucket compression and integration with dsh-agent-compact, automatically consolidating che...
Read MoreDeepSeek Harness Plugin `dsh-negative-ledger`: Recording Falsified Paths and Automatically Invalidating Them Upon Evidence Changes
`@akslcw/dsh-negative-ledger` is a community plugin for DeepSeek Harness (DSH), designed to persistently record an agent's "falsified paths" and prevent repeated invocation of failed commands, attempts to read missing files, or calls to unavailable APIs. This plugin is not a memory system, cache, or bug tracker; instead, it transforms failed attempts into traceable, invalidatable negative facts by binding evidence and retry conditions. Its core mechanisms include: identifying repeated attempts via fingerprints based on command or file path; offering three modes—`warn` (default, with added contextual hints), `block` (intercepts before execution), and `off`; and automatically invalidating conclusions when dependent evidence changes to allow retries. The plugin supports SQLite (multi-process/WAL) and JSONL (single-process) backends.
Read MoreImplementing Traceable Knowledge Graph Memory for DeepSeek Harness with graph-memory
Graph-memory is a DSH memory-type plugin maintained by adoresever, which extracts dialogues into TASK/SKILL/EVENT knowledge graphs, supporting cross-session semantic recall and local SQLite storage. This article introduces its core capabilities, directory page installation commands, source tarball installation steps, gm_* tools and vector retrieval configuration, as well as the plugin permissions, beta version, and the application boundaries of the 75% context compression benchmark.
Read More