dsh-learn-wiki

DeepSeek Harness Memory Low risk

Knowledge base with a corrective acquisition loop for DeepSeek Harness: when the agent keeps hitting the same failure, a background worker fetches the web under a rate limit, distills the result into a staging area, and only a two-stage commit makes it recallable — pages without sources are refused.

DSH 的纠错式补料知识库：agent 反复撞同一堵墙时，后台 worker 在限流下联网抓取、蒸馏结果先落暂存区，只有两段式 commit 才能进入召回——无来源的页面一律拒绝。

How to install

DeepSeek Harness dsh plugin add github:Dayi-Z/dsh-learn-wiki

About

dsh-learn-wiki DSH 的「边做边学」知识库插件：**工作前自动检索 → 卡住时后台限流联网补料 → 蒸馏落暂存 → 两段式 commit 升入知识库 → 下次自动命中**。它把 CRAG（Corrective RAG）的纠错回路从"单次问答"搬到 agent 循环上：知识库检索不到就自己去学，学到的东西沉淀下来，下次就不用再学。它解决什么问题绝大多数"记忆插件"是**静态管道**——检索不到就检索不到，不会去学。结果就是同一个知识缺口在项目里反复踩，每次都要重新上网查。本插件的增量只有一个，但是关键的那个：**"反复撞同一堵墙"是一个可观测的信号，会触发补料并沉淀。** ★ 触发器是**挣扎**，不是"检索未命中"。原设计确实用未命中，但实测那个信号太廉价 ——任何新话题都会未命中，于是系统为聊到的每件新鲜事都去联网。未命中仍可用（gapTrigger: 'miss' | 'both'），只是**默认关**。三层知识架构 | 层 | 载体 | 角色 | 谁写 | |---|---|---|---| | **L1** | 独立 Markdown 仓库 | 精选知识页，强命中直接注入 | 本插件蒸馏 + 人工编辑 | | **L2** | Hindsight | 原始情景记忆，语义召回 | Hindsight（由既有 hindsight 插件负责） | | …

GitHub ↗

Recommendation signals

47 Tool quality · Based on stars, downloads, maintenance, security and docs

– User interest · Adjusted by in-site views, install copies and download clicks

47 Overall

0views

0unique visitors

0install copies

0download clicks

0outbound clicks

Links

GitHub ↗ Report issue ↗

Basic safety check

Findings: curated 收录但无 npm 包/安装命令
Sources: curated:awesome-dsh-plugin.com, curated:awesome-dsh-plugin/awesome-dsh-plugin
Topics: deepseek, dsh, dsh-plugin, knowledge-base, memory, rag

Related plugins

Claude Code

Score82

get-shit-done

gsd-build/get-shit-done

A light-weight and powerful meta-prompting, context engineering and spec-driven development system for Claude Code by TÂCHES.

☆ 64.5K ↓ 48.4K Memory ↗

MCP ServerClaude Code

Score80

markdownify-mcp

zcaceres/markdownify-mcp

A Model Context Protocol server for converting almost anything to Markdown

☆ 3.0K ↓ 1.3K Memory ↗

DeepSeek Harness

Score80

hindsight

vectorize-io/hindsight/tree/main/hindsight-integrations/coding-agents

Hindsight, agent memory that learns: long-term project memory with auto recall and retain, knowledge pages, deep reflection, and per-repo memory banks.

☆ 23.6K ↓ 119.1K Memory ↗

MCP Server

Score79

airtable-mcp-server

domdomegg/airtable-mcp-server

🗂️🤖 Airtable Model Context Protocol Server, for allowing AI systems to interact with your Airtable bases

☆ 456 ↓ 13.1K Memory ↗