memory-systems: A Design Guide for Agent Persistent Memory and Retrieval Architectures
memory-systems is the Memory System Design guide in the Agent Skills for Context Engineering collection, covering cross-session semantic memory, entity consistency, temporal validity, graph/vector and hybrid retrieval, and memory consolidation. The article explains its positioning, activation boundaries, memory hierarchy and framework selection table, and principles for consolidation and JIT loading. It also provides official examples using Mem0, temporal graphs, and Cognee, along with loading methods via Claude Code and npx skills, emphasizing starting with a shallow design, validating it with LoCoMo/LongMemEval, and then deepening the architecture.
Read Moredsh-memj: A Memory System Based on Project Logic
`dsh-memj` is a memory system plugin for DeepSeek Harness (DSH), designed to address cross-session context loss. It aggregates knowledge along the project's main line, supports a three-tier structure covering project, topic, and global levels, and enables knowledge handoff and archival through the `/memj-handoff` command. The plugin supports Markdown storage and frontend hot-loading for editing prompts and knowledge structures, allowing users to customize rules or enable the model to self-evolve through prompt iteration. It requires no external dependencies; once installed, it provides standardized documentation consolidation and multi-dimensional retrieval, offering developers a complete closed loop for building long-term maintained project knowledge bases.
Read Moredsh-cognitio: An Error-Correction-Driven Cognitive Architecture Plugin
dsh-cognitio is a plugin-based memory and governance tool for DeepSeek Harness, designed to address the unreliability of AI memory. It implements hierarchical memory (facts, rules, etc.) and mechanical reminder mechanisms via SQLite, automatically injecting rules at critical moments. Key features include self-correcting evolution, user arbitration and approval workflows, local data storage, and mandatory reference library lookup. Users must install it via the CLI and configure environment variables; all accumulated content must be approved before taking effect, ensuring data privacy and auditability of decisions.
Read Moredsh-plugin-xt-memory: DeepSeek Harness Cross-Session Self-Learning Memory Plugin
`dsh-plugin-xt-memory` is a plugin for DeepSeek Harness designed to address cross-session memory needs in long conversations. It provides memory management across two dimensions: injection and tools. The injection component uses the BM25 algorithm to retrieve memory snippets based on relevance scores, while the tool component provides read and write interfaces. In the background, it asynchronously extracts key information from conversations and deduplicates and merges it. Data is persistently stored in local Markdown files and supports an automatic archival mechanism. Developers can adjust parameters such as budget and extraction frequency through configuration options to enable long-term memory capabilities for agentic applications.
Read MoreThe AdaptiveContextPlane (ACP) Plugin for DeepSeek Harness (dsh): A Governed Long-Term Memory System
The `dsh-adaptive-context` (ACP) plugin is designed to address long-term memory management pain points for agents in the DSH ecosystem. It transforms memory management into an “auditable ledger” through features such as a SQLite evidence ledger, write and read gates, user permission controls, and context orchestration, ensuring that memories are authoritative, accurate, and traceable. The plugin supports multi-source routing, style promotion, temporal truth queries, and audit export, enabling a complete record of the evolution of user preferences and corrections. Installation requires adding the plugin via the CLI and configuring an explicit data directory, after which a restart is required for the changes to take effect. It provides a standardized memory governance solution for long-running agents with high consistency requirements.
Read Moredsh-project-context: DeepSeek Harness Project-Level Persistent Context Plugin
`dsh-project-context` is a plugin that turns DeepSeek sessions into a searchable project context. It manages project memory through four stages (archiving, organizing, consolidation, and handover), supports both manual and automatic triggering, and ensures that session content remains continuous in long-term projects. The plugin is MIT-licensed, uses file locking to ensure data safety, and is ideal for developers who need to maintain complex project contexts.
Read MoreL1-L2-L3 Three-Tier Memory Agent Foundation: reatcat/l123-harness
The core philosophy of the DeepSeek Harness (DSH) plugin `reatcat/l123-harness` is "everything is a plugin." It provides capabilities such as an L1-L3 three-tier memory system, gate approvals, and event logging. The plugin bridges lifecycle events via Hooks to enable L1 resident memory, automatic L2 index injection, and L3 detail management, while supporting TDD workflows and dual-host compatibility. The installation command is `dsh plugin --profile web add github:reatcat/l123-harness`. Note that local development should use a tarball package to avoid dependency issues, and the `PreCompact` and `SessionEnd` events are not currently implemented.
Read MoreDeepSeek Harness Memory Plugin: Intelligent Context and Personalized Recommendations for DSH
`ly028716/dsh-memory-plugin` is a memory system plugin designed specifically for DeepSeek Harness (DSH). It persists data locally in JSON, addressing the difficulty Agents have in remembering context and preferences. The plugin offers core features such as intelligent recommendations, tool tracking, preference memory, project context management, and session history, supporting data import/export and local privacy protection. Developers can make explicit calls through the `ctx.memory` API. By default, the plugin does not automatically collect data; users retain full control over local memory, effectively improving development workflow efficiency.
Read Moredsh-lan-memory: Lan – Memory and Personality System (DeepSeek Harness Memory and Personality Plugin)
`dsh-lan-memory` is a memory and persona plugin for DeepSeek Harness, designed to address context loss, memory confusion, and persona drift in agents. It adopts a three-tier memory architecture (resident, retrieval, and experience library), a SOUL static baseline, and a MOOD state pool to enable persistent cross-session memory and a stable persona. Built on pure JSONL file storage with zero external dependencies, it supports GUI-based online configuration editing and one-click organization, making it suitable for agent development scenarios with strong requirements for privacy and long-term consistency.
Read MoreDeepSeek-Harness-Hanako-Memory: Manage dsh assistant presets via a web page, featuring a built-in Hana-style memory system
moononnn/DeepSeek-Harness-Hanako-Memory is a plugin for the DeepSeek Harness (dsh) ecosystem (npm package name: dsh-assistant-manager, version 0.4.1, MIT License). It integrates the memory system of Hana (openhanako) with assistant configuration into dsh, providing a web interface to manage assistant presets and eliminating the hassle of manual YAML editing. Core features include selecting four archetypes (Hanako/Butter/Ming/Kong) when creating a new assistant, which automatically generates template identities and personalities and saves them as dsh presets. It supports full-field editing, avatar cropping and uploading, deletion protection, setting as the main assistant, and sorting. The memory section provides a master switch, add/delete pinned memories, and management for today/week/lifetime.
Read More