dsh-codebase-memory

DeepSeek Harness plugin: persistent per-workspace codebase knowledge memory (memory_store / memory_recall / memory_forget / memory_list tools, /memory command), stored in a project-file JSON store.

AI 分析

核心用途是为代码库提供持久化的工作区记忆存储与检索工具。适合进行代码开发、需要 Agent 记住项目上下文的程序员。必要条件是 DSH 配置文件需提供 tools 和 commands 服务。

パッケージ

dsh-codebase-memory

バージョン

0.1.0

ライセンス

MIT

最終更新

2026/08/13

Configuration

- id: codebase-memory
  name: 'dsh-codebase-memory'
  config:
    dirName: .dsh-memory        # memory directory inside each workspace
    fileName: memory.json       # memory file name
    maxEntries: 2000            # hard cap; oldest entries dropped beyond it
    defaultLimit: 5             # default recall/list result limit
    workspaceHint: true         # one-line system-prompt hint about the memory
    injectOnStart: false        # inject recent memories into new sessions
    injectCount: 5              # entries included in the session-start injection
    injectMaxChars: 2000        # char cap for the injected digest

Usage

The model calls the tools itself (e.g. "remember how retryPolicy works" → memory_store, "have we seen this before?" → memory_recall). For humans:

/memory                → 8 most recent entries + total
/memory retry policy   → keyword search

dsh-codebase-memory

AI 分析

インストール

ドキュメント

Configuration

Usage