In DeepSeek Harness (DSH)’s plugin-based architecture, retaining and retrieving an agent’s long-term memory is a common challenge for developers. Relying solely on conversation context cannot solve information loss across sessions, while introducing external memory systems often comes with high token consumption. The satan9394/dsh-persistent-memory plugin provides a cross-session persistent memory compression solution aimed at reducing the cost of maintaining long-term memory for agents.

This is a memory-oriented DSH skill maintained by user satan9394. The plugin borrows design concepts from thedotmack/claude-mem 86k★ and focuses on resolving fragmented memory and context overflow in multi-turn conversations.

Core Features

The plugin mainly provides the following capabilities:

  • Cross-session persistent memory compression: Compresses and stores scattered conversational information, enabling it to be retrieved in later sessions.
  • Lifecycle hooks for observation capture: Automatically captures and records key observations through lifecycle hooks.
  • 3-level retrieval mechanism: Provides three retrieval paths—search, timeline, and details—with the goal of saving 10x tokens.
  • Progressive disclosure: Discloses memory content gradually based on conversational needs, rather than loading everything at once.
  • Private tags: Supports adding private tags to memories for organized management.
  • Hybrid search: Combines multiple search strategies to improve retrieval efficiency.

Installation and Enablement

This plugin currently does not provide an official installation command. According to DSH ecosystem conventions, users should install it through the following steps:

  1. Visit the plugin directory: https://www.skillhub.cn/plugins/satan9394/dsh-persistent-memory
  2. Refer to the GitHub repository: https://github.com/satan9394/dsh-persistent-memory
  3. Follow the directory guidance or repository instructions.

Typical Usage

Due to a README retrieval failure, specific usage examples are currently unavailable. Based on the plugin name and core features, the usage flow likely includes capturing, compressing, and storing memories, followed by layered retrieval calls.

Application Scenarios and Notes

  • Target users: Developers building DSH agents with long-term memory capabilities.
  • Permission notice: The plugin runs with the permissions of the current DSH process, so make sure the relevant environment configuration is correct.
  • Compliance: Before installation, check the source code and license information (currently missing) and confirm compliance with usage requirements before deployment.

Summary

For developers who want low-cost and efficient memory management in DSH, satan9394/dsh-persistent-memory provides a complete toolchain from capture to retrieval. Although the current documentation and examples are still incomplete, its 3-level retrieval and progressive disclosure design offer a useful reference for optimizing the context window.