Introduction

In long-term projects, architectural decisions, historical attempts, failed approaches, and next-step plans are often easily forgotten or scattered across conversation history. dsh-missher-memory is designed to address this issue. It provides reviewed, project-scoped memory functionality for DeepSeek Harness, ensuring that agents can reliably reconstruct context.

Plugin Overview

dsh-missher-memory is a standalone DeepSeek Harness plugin package maintained by developer Missher12. Its core value lies in using a “review” mechanism to recover architecture, decisions, progress, failed attempts, and follow-up steps, helping developers maintain coherence during long conversations.

Core Features

This plugin includes the following core capabilities:

  • Project-level memory review: Structured management and review of project-scoped memory.
  • Context recovery: Supports recovering architectural designs, decision records, project progress, past failed approaches, and follow-up steps.
  • Indexed retrieval: Provides efficient memory index retrieval.
  • Data merge strategy: Supports reversible deduplication, avoiding memory redundancy.
  • Automatic and manual retrieval: Supports automatic memory retrieval, while also providing a candidate review and approval workflow to ensure memory accuracy.

Installation and Configuration

Prerequisites

Before installing, ensure the environment meets the following dependency versions:

  • DeepSeek Harness: 0.1.5-rc.2
  • Cordis: 4.0.2
  • Node.js: ^22.19.0 or >=24.0.0

Installation Steps

Use the official installation command to add the plugin to the current web profile:

dsh plugin --profile web add https://github.com/Missher12/dsh-missher-memory/releases/download/v0.3.1/dsh-missher-memory-0.3.1.tgz

After installation, run the following command to confirm that the plugin is active:

dsh --profile web --dump-config

Usage Example

After installation and project binding, you can call the memory_search tool in a session to perform retrieval. The following is a typical usage example:

memory_search({ query: "packaged smoke", scope: "project", limit: 5 })

Parameter description:
* query: The text content to search.
* scope: Memory scope; options include project (project memory) or personal (personal memory).
* limit: The maximum number of results to return.

Notes and Applicable Scenarios

  • Data storage location: Plugin data is stored under the $DSH_HOME/missher-memory/ directory and primarily managed by the state.db file.
  • Database compatibility: The plugin does not modify DeepSeek Harness core code, and it does not copy or alter existing legacy databases.
  • Platform support: Currently validated only on macOS (Intel/Apple Silicon), Windows x64, and Linux x64. Windows ARM and Linux ARM platforms are not yet supported.
  • Runtime behavior: The plugin runs with the permissions of the current DSH process and does not involve Python or native builds, unless required by specific components.

This plugin is suitable for advanced development scenarios that require long-term maintenance of complex projects and that rely on agents to accurately reference historical decisions and architectural approaches.

Conclusion

By combining human review with automatic retrieval, dsh-missher-memory provides DeepSeek Harness with a reliable project memory layer. Its open-source nature (MIT license) allows developers to freely inspect the source code and configuration details. For more details, see the GitHub repository or the DSH community catalog.