Memory Manager EN：把 `.workbuddy/memory/` 做成可计量的 Token 节流阀
Memory Manager EN 是 SkillHub 上的 ai-agent 类 Skill，面向 .workbuddy/memory/，用 tiktoken 计量、硬预算压缩与 throttle 控制进入上下文的记忆体积。支持 --query/--tags 做任务相关性加载、--json 与 --force 便于自动化，含三色预算告警、缓存与并行摘要等性能优化，并内置安全自检。适合 token check、smart load、备份与记忆分析；不替代通用文件或代码工具。安装：skillhub install @indiv-aakook/memory-manager-en。
Read Moredsh-hindsight-bank-router：将 Hindsight 的 bank 身份锚定到远程仓库名
`dsh-hindsight-bank-router` 是一款由 `j62268781-alt` 开源的 DeepSeek Harness (DSH) 插件，旨在解决 Hindsight 记忆库因本地目录名不同而分裂的问题。该插件通过将 Bank 身份锚定到远程仓库名，确保同一仓库在不同目录下的克隆共享同一记忆库。 其核心功能包括：支持声明式优先级链（覆盖用户配置、仓库内配置、远程名、本地目录名）；零子进程开销直接读取 Git 配置；支持中文仓库名；以及安全开放的失败处理机制。默认关闭，需同时开启用户级和工作区开关。该插件通过动态生成 `mapPathToBank` 配置，实现了记忆库的统一管
Read Morehermes-agent 版 OpenClaw 平替自进化技能：让 Agent 从任务里沉淀可复用 Skill
hermes-agent版openclaw 平替自进化技能（@user_a37770d2/hermes-skill，v1.0.0）是 SkillHub 社区 ai-agent 技能包，主打从复杂任务自动创建可复用技能、HOT/WARM/COLD 分层记忆、智能 Nudge 与定时自检，并声明追踪 NousResearch/Hermes-Agent。本文基于 SkillHub 公开简介与安装文档，说明定位、核心能力与 skillhub CLI 安装到 Cursor/Claude Code 等 skills 目录的步骤，并提示以包内 SKILL.md 为准启用；标签 requires_api_key=false。
Read Moredsh-zcode-scribe: Read-Only Memory Room and Path Security Mechanisms of DeepSeek Harness
`dsh-zcode-scribe` is a security memory plugin for DeepSeek Harness (DSH), built on the ZCode memory model. It provides agents with read-only Markdown memory rooms and the `scribe_recall` retrieval tool, while including 52 strict path security rules to effectively prevent directory traversal, Unicode spoofing, and NTFS alternate data stream attacks. The plugin is currently in an early stage, implementing only read-only access and retrieval; write and extraction capabilities are not yet supported.
Read Moredsh-clear-mind: Model-Autonomous Context Compression and Checkpoint Management
`dsh-clear-mind` is a memory management plugin for DeepSeek Harness (DSH) designed to address context-window saturation in long conversations. By providing the `mind_map` and `clear_mind` tools, it enables the root agent to autonomously compress conversation history, replacing redundant information with auditable checkpoints. The plugin also supports proactive threshold alerts, generates GUI tombstone records, and runs entirely locally without requiring network access. It can effectively reduce inference costs while preserving session traceability.
Read Moredsh-obsidian-mem: Transforming Project Memory into an Obsidian Vault
dsh-obsidian-mem is a memory host plugin for DeepSeek Harness, designed to persistently store project documentation and long-term memory in pure Markdown format within an Obsidian vault. It binds repositories using a stable projectId and supports building search indexes, automatic conversation distillation, and using six mem_* tools. The plugin requires DSH >=0.1.5-rc.2 and Node.js >=22.22.2, and the process must be restarted after installation. During use, developers can manage the project list via commands, with data stored in a dedicated local directory. The plugin resolves DSH memory persistence issues and is well suited for scenarios requiring seamless collaboration.
Read Moredsh-tiered-memory: Tiered Memory and Budget Management Plugin
The DeepSeek Harness (DSH) ecosystem introduces the `dsh-tiered-memory` plugin, addressing issues in the original plugin where user, environment, and project contexts could not be distinguished and memory consumption could not be controlled. Ported from Hermes Agent, it implements tiered storage and budget management. Core features include `mem_set`, `mem_get`, `mem_query` for keyword-routed semantic queries, and `tier_summary` for viewing tier usage and eviction candidates. The architecture divides memory into three tiers with independently managed budgets; when a tier is full, writes are rejected and deletion is prompted. The plugin supports command-based installation, has a high code coverage rate (98.5%) and a strict isolated testing process, and is licensed under MIT.
Read MoreDeepSeek Harness Memory Management Plugin: dsh-memento-tab
`dsh-memento-tab` is a session-level memory management plugin for DeepSeek Harness, designed to address the cumbersome nature of command-line management and its side effects. It provides a UI interface that covers the 11 operations of `/memory`, and adds features such as budget monitoring, search filtering, inline rewriting, merging, pending proposals, export/import, and auditing. The plugin does not conflict with upstream and reduces `recall_count` by optimizing the read path. After installation, it appears as the fourth tab at the top of the session, offering a more convenient memory management experience.
Read MoreIntroducing the DeepSeek Harness Session Manager Custom Plugin
The DeepSeek Harness (DSH) memory management plugin `floatinglifetl/dsh-plugin_session-manager-custom` is maintained by FloatingLifeTL. It is currently **paused from development and archived**; the code is no longer updated and may be incompatible with the latest versions, so it is not recommended for production deployment. Because specific features and installation instructions are not provided, users should check the source code permissions themselves before use.
Read MoreDeepSeek Harness Role-Playing Plugin: A DSH Plugin Centered on Role-Playing, with the Desktop Pet as an Additional Feature
ajuwm/dsh-roleplay-plugin is an immersive role-playing plugin for DeepSeek Harness. It uses systematic stat management, multi-character memory, and life interaction systems to effectively resolve agent persona collapse and memory confusion. The plugin supports step-by-step guidance, continuous story saves, multi-character interaction within the same scene, and a desktop floating pet feature, aiming to lower the barrier to building high-quality agents. Installation requires DSH 0.1.1+ and a pnpm environment, while core features do not depend on the desktop pet.
Read Moredsh-mnemos: Controlled Cross-Session Memory Plugin for DeepSeek Harness
The DSH plugin dsh-mnemos is purpose-built to address agents' "context loss" and disorganized memory management. It is a controlled, extensible cross-session memory plugin that ensures data security through a human approval queue and provides memory retrieval, batch distillation, skill transformation, and SQLite/Git-based storage synchronization. Supporting an open bus mechanism for third-party integration, it also offers a `/memory` command and API interface, aiming to build a long-term memory infrastructure for DSH agents.
Read MoreDeepSeek Harness (DSH) Plugin: dsh-hermes-memory
`dsh-hermes-memory` is a native persistent memory plugin for DeepSeek Harness (DSH), designed to address challenges such as agent state management, context retention, and sensitive information security. Built on DSH's storage domain and session query services, it provides fine-grained scopes, including global, user, project, and failure. It supports memory CRUD operations, automatic context injection into sessions, security scanning, and expiration cleanup. The plugin can automatically capture interaction rules and provides management tools for status inspection. The current version is 0.3.0 and does not yet support vector retrieval or a standalone UI, making it suitable for developers who need to maintain context across sessions.
Read MoreIntroducing the DeepSeek Harness Plugin: frank-nf/dsh-memory-nexus
`frank-nf/dsh-memory-nexus` is a plugin designed for DeepSeek Harness (DSH), aimed at addressing pain points in context management and long-term memory integration for complex agent construction and long conversations. It provides a 4-layer memory structure, context compression, prompt orchestration, and enterprise-grade security support. Users need to obtain and install it from the GitHub repository, and it is recommended to review the source code and license before installation to ensure safety.
Read MoreDeepSeek Harness Plugin: dsh-gentle-engram
`dsh-gentle-engram` is a DeepSeek Harness plugin that bridges Engram's persistent memory capabilities into DSH workflows through Cordis lifecycle hooks. It implements features such as tool calling, asynchronous persistence, context injection, and session cleanup, simplifying the integration process. The plugin requires Engram to run and must be installed and configured via npm.
Read Moredsh-memorix-panel: View and Manage the Memorix Memory Bank in the DSH Web GUI
`dsh-memorix-panel` is a visualization plugin for DeepSeek Harness (DSH) designed to address the risks and lack of intuitiveness associated with directly managing the Memorix database. It integrates Memorix's SQLite memory store into the DSH Web GUI and provides a visual management interface. The plugin includes rich features such as database status and statistics, search and filtering, project and memory list browsing, and detail views. Data access uses a "read-only reads, CLI writes" strategy to ensure index synchronization with zero runtime dependencies. It also supports state-machine management of long-term memory (e.g., candidate → approved) and operations such as copying a project name with one click. Installation requires Node.js ≥ 22.18.0 and the Memorix CLI environment; configuration is flexible and read-only mode is supported.
Read Moredsh-memory-manager: Unified Memory Management Plugin for DeepSeek Harness
`dsh-memory-manager` is an AI agent memory management plugin for the DeepSeek Harness plugin ecosystem. It addresses the challenges of limited context windows and the lack of state persistence by providing hierarchical storage capabilities, ranging from short-term sliding windows to long-term vector memory. The plugin supports semantic and hybrid retrieval, AES-256 encrypted storage, and a visualized web-based management interface. By default, it is based on local SQLite with zero cloud dependencies, making it suitable for users who need cross-session retrieval and long-term memory maintenance.
Read Moredsh-self-improvement
The DeepSeek Harness Agent plugin provides agents with a maintainable long-term memory pathway through tiered and weighted mechanisms. It categorizes memory into five types, supports retrieval based on importance and recency, and offers governance features such as deduplication, poisoning protection, and score-based eviction to manage capacity constraints. The plugin also supports automatic retrospection, skill accumulation, and cross-session experience reuse, ensuring that agents retain context after restarts or session switches. Users can manage memory using commands such as `/memory` and `/retro`, and can configure memory scope at either the workspace or global level. It is suitable for complex application scenarios that require long-running operation, automatic experience accumulation, and secure governance capabilities.
Read Moredsh-maestro-sync: A Maestro Harness Plugin for Syncing Memory and Sessions Across Machines
`dsh-maestro-sync` is a DSH Maestro plugin maintained by `ddtcorex` that safely syncs memory and session data across machines. Its core mechanism is “preview before apply”: `--dry-run` generates a `SyncPreview`, and after verification, `--preview-id` and `--confirm` execute the write. Before applying, a Stale-guard mechanism checks whether files have changed, and every write operation creates a `.bak` timestamped backup file to ensure fault tolerance and rollback capability. Additionally, the plugin supports backing up data to S3-compatible storage (such as R2) and strictly excludes sensitive or unnecessary content, including configuration files, caches, and backups.
Read Moredsh-memory-hermes: Hermes-Style Bounded Planning Memory Plugin
`dsh-memory-hermes` is the memory plugin for DeepSeek Harness, designed to address the issue of agent memory bloat. It provides a “bounded curated memory” solution, maintaining refined `MEMORY.md` and `USER.md` files and improving prompt-cache friendliness through frozen snapshots and background introspection mechanisms. The plugin offers core features such as dual-file memory, skill-library consolidation, security scanning, and concurrency safety. With simple configuration and hot-reload support, it effectively manages context and prevents overflow, making it suitable for development scenarios requiring long-term stable state and knowledge accumulation.
Read Moredsh-memory-forget — A Forgetting Engine for AI Agents
dsh-memory-forget is a forgetting engine plugin for AI agents, designed to transform “forgetting” from a failure into a design. It manages the memory lifecycle through TTL, decay, eviction, and auditing mechanisms to solve the context pollution problem. Core features include a pluggable memory bus, Ebbinghaus-curve-based decay, importance management (sliding update mechanism), hard injection budget limits, privacy-first physical deletion policies, and multi-agent scope support. The plugin is maintained by XIAOke8698, released under the MIT License, and has zero dependencies. It supports restoring forgotten memories and provides token accounting. Installation requires Node.js >= 20, and a StoreAdapter must be implemented to ensure data persistence.
Read Moredsh-active-listening: Effective Listening
The `dsh-active-listening` plugin in the DeepSeek Harness (DSH) ecosystem follows the “everything is a plugin” philosophy and focuses on handling conversational context and memory coherence. It provides the core tool `active_listening`, which supports returning structured methods based on scenarios and runs in a pure Node environment without relying on network requests. It can be installed with `dsh plugin add github:uckkk/dsh-active-listening`. The plugin depends on `@deepseek-ai/cordis` (^4.0.1) and `@deepseek-ai/dsh-tools` (>=0.1.0-rc.6) and is licensed under the MIT License.
Read Moredsh-memory-manager: A Persistent Memory Plugin for DeepSeek Harness
`dsh-memory-manager` is a community plugin for DeepSeek Harness, designed to address context loss in agent conversations. It ports ZCode's memory design and supports structured, per-project memory storage. Core features include a memory management panel, automatic injection of memory indexes into conversations to save tokens, and `memory_save/get/list` tools for agents to proactively read and write memories. The plugin supports global memory, importing from ZCode, and automatic indexing, helping agents effectively maintain long-term memory.
Read MoreMind Garden: DeepSeek Harness Long-Term Companion Plugin
Mind Garden is an installable plugin for DeepSeek Harness, designed to address the problems of model memory confusion in long-term interactions and the burden of user memory management. It provides users with a controllable, private “long-term companion” and reflection space, featuring a user-driven memory system, cinematic visual feedback, and nine interconnected reflection spaces. The plugin reuses the core capabilities of Harness, supports photo observation through the DeepSeek Vision model, and allows users to correct the AI’s conversational direction by adjusting the dialogue stance and intent. It is suitable for users who require deep reflection and long-term documentation.
Read MoreIntroduction to the DeepSeek Harness Archive Enhancement Plugin
`dsh-archive-viewer` is an archive-enhancement plugin for DeepSeek Harness (DSH), designed to address the issues of session accumulation and difficult retrieval. The plugin supports automatic periodic archiving of idle sessions, tree-based folder organization, batch operations, and LLM-based summary generation. It provides features such as viewing, restoring, and exporting, and is compatible with any skin. Note that some features, such as session restoration and one-click DSH shutdown, rely on core patch support. Plugin data is stored locally, and archiving policies can be customized through configuration options.
Read MoreDeepSeek Harness Memory Plugin: dsh-song-memory
`$songoao25/dsh-song-memory` is a memory management plugin for DeepSeek Harness, designed to address the need for cross-session state persistence in multi-tool collaboration. The plugin does not rely on external services and can retain state memory outside of conversations. Its specific features have not yet been disclosed. It must be installed via the source code repository and enabled in the configuration file. Because the plugin runs with the permissions of the current process, be sure to review the source code and license before use.
Read Moredsh-memory-manager: Cross-session Memory Management for DSH
dsh-memory-manager is DSH's memory management plugin designed to address the challenge of cross-session context accumulation. It offers a cross-session memory library persisted on disk, supporting features such as Obsidian-compatible folders, immutable snapshots and annotation layers, memory tags, and bidirectional links. Core features include automatic injection of reduction and session summaries, context plan management, six Agent memory tools, force-directed graph dual views, and a comprehensive conversation sidebar panel. The plugin allows users to non-destructively exclude or restore the injection of specific turns, and enables users to toggle features freely via the settings page. Installation requires a local DSH environment. It involves adding the local repository path using the `file:` protocol and restarting, with Node.js 18+ required. During operation, users can add memories to a plan or enable temporary mode via the sidebar panel, ensuring that relevant memories are automatically injected into new sessions.
Read Moredsh-memory-palace: Add Markdown file-based memory to DeepSeek Harness
The DSH plugin `dsh-memory-palace` aims to provide cross-session persistent memory for DeepSeek Harness. It stores memory in human-editable Markdown files, including long-term memory (`MEMORY.md`) and daily logs, and supports a dual-layer memory structure at both user and workspace levels. Core functions include automatic memory reading and writing, automatic migration of expired logs to long-term memory, compatibility with WorkBuddy/CodeBuddy memory directories, and providing tools such as `memory_note` and `memory_read` for memory CRUD operations. The plugin supports a hybrid mode to preserve log evidence and features proactive memory and intelligent summarization capabilities. All configurations can be adjusted graphically via the DSH settings panel. This plugin...
Read Moredsh-hermes-memory: Dual-Repository Memory Plugin for DeepSeek Harness
dsh-hermes-memory is a dual-library memory plugin for DeepSeek Harness (DSH), designed to address the issue of cross-session information discontinuity. It divides memory into two independent libraries: MEMORY.md records environment facts, project conventions, and tool quirks (with a limit of 2200 characters), while USER.md records user preferences, communication styles, and work habits (with a limit of 1375 characters). Both libraries feature independent budgets and eviction mechanisms, with character limits hardcoded. The plugin provides a `memory` tool supporting add, replace, delete, and batch operations, adopting an "exact match first" strategy; when the limit is exceeded, it returns a list of entries for the model to retry merging. It automatically injects a frozen snapshot at the start of a session and prompts for persistence after 10 consecutive turns without writes. Auxiliary tools include search, list, statistics, and debugging functions. Data is stored in local JSON files, suppor...
Read Moredsh-continual-harness: Providing Continuous Refinement and Rollback for DSH Agents
`dsh-continual-harness` is a DeepSeek Harness (DSH) plugin published by `jasen215`, designed to help AI agents continuously update, retain, and refine reusable experiences across multiple sessions. The plugin is licensed under MIT, requires Node 22+ and TypeScript 6.0+, and its core features include state projection, automatic review and refinement, manual refinement with rollback, memory lifecycle management (such as archiving and pinning), and ranking injection. It provides tools such as `harness_refine`, `harness_wrapup`, and `harness_benchmark`, supporting automatic rollback on failure, session wrap-up suggestions, and A/B benchmarking to ensure that bad states do not linger in the context.
Read More