ontology：用类型化知识图为 Agent 做结构化记忆
ontology（@user_b634fbfe/002，v1.0.4）是 SkillHub 社区 ai-agent 类 Skill，用带类型的知识图组织 Agent 记忆与可组合 Skill 的共享状态。支持 Person、Project、Task、Event、Document 等实体的创建与查询、对象关联、约束校验，并将多步动作表达为图变换，适用于 remember、查已知信息、link 依赖与跨 Skill 数据访问等场景。无需 API Key。可通过 SkillHub 安装文档用 CLI 或提示词安装到 Cursor、Claude Code、Codex 等对应 skills 目录。
Read More灵魂存档（Soul Archive）：用本地 JSON 把对话沉淀成可复用的人格与记忆
灵魂存档（@user_dfcfbf47/soul-archive，v3.2.2）是 SkillHub 社区 ai-agent 类 Skill，通过日常对话构建数字人格副本与智能体记忆：会话开始注入人格摘要、跨会话召回、失败模式预警与模式蒸馏，对话结束默认 auto_extract 自动沉淀，数据均为本地明文 JSON且无需 API Key。本文介绍定位、能力与触发词，并给出 skillhub install 安装到 Cursor/Claude Code/Codex skills 目录的步骤与典型对话用法。
Read More项目上下文管理：跨对话与跨项目的 Agent 记忆
SkillHub 社区 Skill「项目上下文管理」（@user_19a6d235/project-context-manager，v1.0.0）面向 ai-agent 场景，提供跨对话、跨项目记忆与渐进式披露，缓解 WorkBuddy 等多项目、多对话下的上下文断裂。本文基于官方简介与 SkillHub 安装文档，说明定位、能力、CLI/提示词安装方式及 Cursor、WorkBuddy 等 skills 目录差异，并给出可复现的使用思路；标注 requires_api_key=false，不编造未公开的脚本与配置。
Read More自我进化助理：把错误与纠正沉淀为 Agent 的长期记忆
自我进化助理是 SkillHub 社区 ai-agent 类 Skill（v1.0.0，@user_298b682c/selfimprovingagentbyhutingyu），通过自我反思与自组织记忆，在命令/API 失败、用户纠正、知识过时或发现更优做法时记录教训，并在重要任务前回顾历史学习。无需 API Key。文章说明其定位、核心能力与适用场景，并给出基于 skillhub CLI 在 Cursor、Claude Code、Codex 等目录下的安装命令及官方安装提示词。
Read More记忆管家 V3.9.0：管理 `.workbuddy/memory/` 的 Token 节流与相关性加载
记忆管家 V3.9.0 是 SkillHub 上的 AI 记忆加速器与 Token 节流阀，面向 .workbuddy/memory/ 目录。核心包括 tiktoken 分词计量、--query/--tags 相关性加载（CJK bigram，可选 embedding）、硬预算闸与 throttle 压实、三色告警与近 7 日均值、--force 与全命令 --json、self-test 安全自检，以及并行摘要与 rank 缓存等性能优化。文章说明安装 skillhub CLI、--dir 指向各 Agent skills 目录的方式，以及 Token 检查、智能加载、清理备份等典型触发场景与使用注意。
Read Moredsh-memory-lite: A Hierarchical Memory Library with On-Demand Retrieval
The DeepSeek Harness plugin `dsh-memory-lite` uses hierarchical Markdown and an on-demand retrieval mechanism to solve the token waste caused by duplicate injection of agent memory. Its design follows the "zero-resident" and "zero-dependency" principles, returning only about 738 characters of matched snippets during queries and saving 92% of tokens compared with full reads. It offers both CLI and DSH plugin modes, supports provenance verification, and is suitable for cost-sensitive scenarios that require precise control over retrieved content.
Read Moredsh-memory-lite: A Lightweight Cross-Session Memory Plugin for DeepSeek Harness
`pacoyi/dsh-memory-lite` is a lightweight cross-session memory plugin for DeepSeek Harness, designed to address context loss after session restarts. It provides a memory tool (save/recall/list/forget) and a settings-page UI, supporting local JSON storage, scope isolation, audit logs, and import/export. The plugin includes an approval gating mechanism to prevent unauthorized writes, making it suitable for developers who need to retain project context across sessions and prioritize security.
Read Moredsh-memory-panel: DeepSeek Harness Long-Term Memory Tool and Visualization Panel
`dsh-memory-panel` is a memory management plugin for DeepSeek Harness (DSH), designed to address context loss after Agent session resets. It provides a set of tools (such as `memory_save` and `memory_recall`) and a visualization panel, supporting memory storage, retrieval, deletion, and category-based filtering. The plugin stores memories in plaintext at `~/.dsh/memory.json` and includes safety mechanisms such as atomic writes, a concurrency queue, automatic backup upon corruption, and growth control. Users can install it via command line and use it after restarting the service. Note that its API currently lacks authentication protection, so storing sensitive data in plaintext is not recommended.
Read Moredsh-persist: DeepSeek Harness Long-Term Memory Plugin
The dsh-persist plugin provides hierarchical memory storage for the DeepSeek Harness Agent via the file system, solving the “losing memory when switching conversations” problem. Core features include conversation-specific memory, cross-project sharing, bge-m3-based semantic search, and a three-state automatic recall mode. All memories are stored locally as plain-text Markdown and are human-editable. Users can read and write memories through the `memory` tool or a UI tab. Installation requires Node.js >= 22.6, and enabling semantic search requires a SiliconFlow API key. The plugin’s default data paths and routing remain unchanged, and it supports privacy controls and remote access configuration.
Read Moredsh-focal: A Privacy-First Task Memory Plugin for DeepSeek Harness
dsh-focal is a privacy-first plugin for DeepSeek Harness designed to address memory management and privacy leakage issues in multi-turn Agent tasks. Implemented based on the FOCAL paper, it filters conversations and isolates tasks locally, records only tool states and path parameters, and removes sensitive information. The plugin provides selective context injection, ensuring the model has access only to necessary context snapshots, with all data stored locally. Developers can manage task memory through the command line or configuration files, making it suitable for scenarios with stringent data security requirements.
Read Moredsh-light-memory: Adding a lightweight, plain-text memory module to DSH Web GUI
dsh-light-memory is a DSH plugin developed by chidaic, providing a lightweight memory system for the DSH Web GUI. Instead of relying on databases, vector stores, or MCP, this plugin adopts four Markdown files—USER.md, PROJECT.md, WORKLOG.md, and CONVENTION.md—to store user preferences, project conventions, operation logs, and memory specifications, thereby achieving memory visibility, editability, and Git management. The core functionality revolves around two actions: append and distill. The former appends temporary operation records to the work log, while the latter precipitates conclusions into user or project memory files at task boundaries. The plugin offers four tool interfaces that support the injection of context through various means, such as systemPrompt, to work with prefixes.
Read Moredsh-memory: Memory with citations based on DSH session logs
`dsh-memory` is a plugin based on DeepSeek Harness (DSH) designed to address memory distortion and tracing difficulties in multi-turn agent conversations. Leveraging DSH's lossless conversation logs, it converts persistent facts into human-auditable Markdown memory files through offline distillation after sessions conclude, storing them in the `~/.dsh/memory` directory. The core advantage lies in every memory being accompanied by a citation, allowing for precise tracing back to original conversation log events; it supports reading content via `memory_read` and viewing original excerpts via `memory_expand`. The plugin provides a compact Recall Index to optimize token usage and supports `user`, `project`, `f...`
Read Moredsh-mnemon: Building a Three-Layer Pluggable Memory Control Plane for DeepSeek Harness
dsh-mnemon is a DSH memory plugin maintained by omdsh-dev. It manages context, project files, and long-term memory through a three-layer division of Runtime, Documents, and Memory Spaces. The third layer can integrate with nine providers, including Mnemon. This article introduces its core capabilities, installation commands under DSH 0.1.1-rc.2, a 5-minute verification process, as well as permissions, data boundaries, and applicable scenarios, enabling quick evaluation for integration after discovery on SkillHub.
Read More