dsh-project-memory：跨会话的工程记忆
dsh-project-memory 是 DeepSeek Harness 上的 host-plane 记忆插件，为每个项目持久化根因、失败尝试与约束等结构化工程结论，在 prompt 组装阶段自动 recall，并提供 memory_search、memory_record、memory_correct 等工具。一次 dsh plugin add 即可对所有 preset 生效，无需单独选模式。本文基于官方 README 介绍安装命令、记录 schema、召回逻辑与适用注意。
Read Moredsh-project-memory: Add a verifiable, approval-required project memory for DeepSeek Harness
A common challenge in collaborative agent development is that the technical selections, constraints, and pitfalls encountered in one session are lost when the conversation moves to the next. You either have to re-explain everything from scratch each time, or let the model write to the memory file on its own. The risk of the latter is the lack of validation, meaning the content written by the model may no longer align with the current code. `dsh-project-memory` addresses this issue by transforming persistent decisions, constraints, agreements, lessons learned, and facts into versioned project records. However, the model can only propose these entries; whether they are committed to the records is decided by the human user. The philosophy behind DeepSeek Harness (DSH) is "Everything is a plugin." This plugin represents an extension of memory capabilities along this very line of thinking. # What is it? The positioning of `dsh-project-memory` can be summarized in a single sentence.
Read Moredsh-project-memory: Project memory plugin for DeepSeek Harness
dsh-project-memory is a persistent project memory plugin designed specifically for the DeepSeek Harness (DSH) agent. It resolves context loss issues in cross-session development by storing documents, code symbols, experience records, and task status on disk. Core capabilities include support for document memory (such as PDF/Markdown) and code symbol extraction with source code references, alongside optional integration of TypeScript for semantic enhancement. It employs a content hash-based automatic refresh mechanism and a read-time memory mechanism to achieve cross-linking between documents and code. Furthermore, it provides precise recall based on the BM25 algorithm with blind spot warnings and supports the selective filtering of experience records. Additionally, the TaskBridge functionality supports the binding, renaming, and status viewing of cross-session tasks. This plugin runs in a Node.js environment with no vector database dependency and is suitable for...
Read Moredsh-project-memory: Project-level persistent memory plugin for deepseek-harness
`dsh-project-memory` is a zero-dependency memory plugin for deepseek-harness, designed to address the issue of information not being preserved across sessions. It writes project-stable knowledge into Markdown files in the local `memory/` directory (with `MEMORY.md` as the index), facilitating manual review, Git auditing, and version control. The plugin provides four core tools: `memory_load_context` for loading memory, `memory_upsert` for writing or updating, `memory_finalize` to solidify task conclusions, and `memory_forget` to delete expired records (while protecting the index file). Features include sensitive information interception (e.g., passwords, tokens), workspace-aware root directory resolution, and r...
Read More