deepseekplugins

DeepSeek Harness plugin

dsh-skill-indexer

DSH host bundle plugin: builds a two-level (category + skill) recall index over all local SKILL.md files and routes a natural-language intent to the best-matching skills, with three-state (hit /

Memory
Jump to install

Source facts

Repository
NattoCB/dsh-skill-indexer
Latest update
Aug 20, 2026
Category
Memory
GitHub stars
0
Format
plugin
Catalog evidence
Upstream dsh.bundle evidence
Evidence path
package.json#dsh.bundle
Checked against
0.1.0-rc.8
Upstream check date
2026-08-20

This evidence comes from the upstream catalog. This site has not installed, run, or security-reviewed the plugin.

Install

Start with a prompt that asks an agent to review the GitHub repository and source. Switch to the command if you want to install it yourself.

Copy this prompt into DSH, Codex, or another agent and ask it to review the GitHub repository and source first.

Do not install or run any commands yet. Read this plugin's GitHub repository, README, and relevant source code. Then answer the questions below clearly and directly so I can decide whether it fits my needs:

1. What is this plugin, and what problem does it solve?
2. Who is it for, and what are its typical use cases?
3. How is it used after installation? Include one minimal example.
4. What known limitations or privacy, security, compatibility, or maintenance risks does it have?
5. Give a clear recommendation: recommend, conditionally recommend, or do not recommend, with reasons.

Distinguish statements documented by the repository, inferences from source code, and unknowns. If evidence is insufficient, say so explicitly. Do not guess or simply repeat the README.

GitHub: https://github.com/NattoCB/dsh-skill-indexer
Plugin: dsh-skill-indexer
Author: NattoCB

Check the source files

Read the README and other files from this plugin directory before installing.

File explorer3 files
README.mdSource · read only

dsh-skill-indexer

<!-- Hero --> <div align="center"> <b style="font-size: 1.15em;">让 agent 用一句话意图，从本机所有 SKILL.md 里召回对的技能。</b><br /><br /> <a href="LICENSE"><img alt="License" src="https://img.shields.io/badge/License-MIT-yellow.svg" /></a> <a href="https://img.shields.io/badge/node-%3E%3D18-339933"><img alt="Node" src="https://img.shields.io/badge/node-%3E%3D18-339933" /></a><br /><br /> <img alt="两级召回索引" src="https://img.shields.io/badge/-两级召回索引-4d6bfe" /> <img alt="意图路由" src="https://img.shields.io/badge/-意图路由-4d6bfe" /> <img alt="三态回退" src="https://img.shields.io/badge/-三态回退-4d6bfe" /> <img alt="零第三方依赖" src="https://img.shields.io/badge/-零第三方依赖-4d6bfe" /> <img alt="只读源目录" src="https://img.shields.io/badge/-只读源目录-4d6bfe" /> <img alt="哈希增量" src="https://img.shields.io/badge/-哈希增量-4d6bfe" /><br /><br /> <b>接入面</b> —— <code>skill_index</code> 工具 / <code>/skill-index</code> 命令 / <code>cordis.patch.yml</code> </div>

> 一个 DSH host bundle 插件：把本机所有 SKILL.md（默认 ~/.workbuddy/skills + $DSH_HOME/skills）建成两级（大类 → 技能）召回索引，用一句话意图路由到最匹配的技能，命中/低置信/未命中三态回退。接入面：skill_index 模型工具、/skill-index 命令、cordis.patch.yml 配置——纯本地、只读源目录、零第三方运行时依赖。 > > A DSH host bundle plugin that builds a two-level (category → skill) recall index over every local SKILL.md and routes a natural-language intent to the best-matching skills — three-state (hit / low / miss) fallback, read-only against the source trees, zero third-party runtime deps.

✨ 功能一览

  • 🗂️ 两级召回索引：扫描 ~/.workbuddy/skills 与 $DSH_HOME/skills，解析 SKILL.md frontmatter（手写 YAML 子集，block scalar 全量解析），抽取触发词/关键词并归纳大类，产出 skills-index.json（UTF-8、无 ASCII 转义、2 空格缩进）。
  • 🎯 一句话意图路由：L1 大类打分（关键词覆盖 × IDF，pos_weight=1/(1+start/10)）命中大类，L2 类内技能排序（词表 = keywords ∪ triggers ∪ name，trigger 命中 ×1.5）；短查询靠单个词即可命中（「把这周的ETF拉出来画个图」→ etf）。
  • 🔁 三态回退：hit 命中 / low 低置信（回退全库 flat top-k）/ miss 未命中（明确告知）；默认 θ1=0.35、θ2=0.12。
  • ⚡ 零第三方运行时依赖：scan/parse/extract/classify/index/route 只用 node:fs/path/crypto，唯一外部包是宿主自带的 @deepseek-ai/dsh-tools（defineTool）。
  • 🔒 只读 + 哈希增量：不改动技能源目录，索引与 usage.log 只写 dataDir；scan 按内容 hash 对比，未变技能保留 usage_count/last_used，无变更时提示 "(no changes)"。
  • 🧩 人工大类覆盖：categories.yaml 可对指定技能直接指定大类，优先级高于内置词表归纳。

Quick Start

前置

  • DSH（DeepSeek Harness）web profile
  • Node.js ≥ 18

安装

dsh plugin --profile web add github:NattoCB/dsh-skill-indexer

或手工把 cordis.patch.yml 的 insert 段合并进目标 profile 的 patch，然后 pnpm install。

运行

安装后 host 启动即自动建索引。手动控制：

/skill-index scan                        # 重建索引（无变更时提示 "no changes"）
/skill-index query 把这周的ETF拉出来画个图  # 意图召回
/skill-index categories                  # 打印大类表
/skill-index status                      # 索引状态与各 root 计数

agent 侧直接调用模型工具：

skill_index(query="把这个PDF转成markdown")
# → { state, intent, categories, hits, text }；text 为可直接注入的上下文

开发 / 测试：

npm test          # node --test，43/43 通过
npm run demo      # 扫描默认 roots 并路由示例意图

Configuration

通过 cordis.patch.yml 的 config 段配置：

KeyDefaultMeaning
roots[]（缺省 ~/.workbuddy/skills + $DSH_HOME/skills）扫描根目录；路径数组或 [{path, type}]
theta10.35命中阈值（max_c ≥ θ1 → hit）
theta20.12未命中阈值（max_c < θ2 → miss）
dataDir$DSH_HOME/skill-indexerskills-index.json 与 usage.log 输出目录
categoriesYaml''可选 categories.yaml 人工大类覆盖，优先级最高

categories.yaml 示例：

overrides:
  git-finish-work: ["git-workflow", "delivery"]
  pdf-to-markdown: docs

设计要点

  • 匹配原语：子串匹配（英文小写化去标点、中文直接子串），不是 token 集合交集；idf = ln((N+1)/(df+1)) + 1。
  • 确定性：全流程无随机数，同输入字节级一致（除 generated_at）。
  • id 规则：唯一名 → id = name；跨 root 重名 → name@type（如 pdf-to-markdown@dsh）；categories[].skills 与 skills[].id 引用闭合。
  • 源规格：skill-indexer PRD v1.1（2026-08-19，本地文档未公开）——两级索引、三态回退、哈希增量与验收标准 A1–A5 / B1–B5 / C1–C4。
  • 非目标：无常驻进程、无 HTTP 服务、无 UI、无 embedding/GraphRAG。

已知限制

  • 中文关键词按「连续中文串」抽取，长串会粘住短词（如「当面对高杠杆」粘住「高杠杆」），短查询召回偏弱。
  • 内置大类词表（10 类）是 MVP 覆盖；跨机部署前建议在目标语料上做 θ1/θ2 网格标定并增补词表。

---

<div align="center">

GitHub · Issues · MIT License

</div>

Back to searchOpen GitHub repository

More related plugins

View category