AgentSkillsCN

pkm

TechPKM GitHub 仓库搜索——覆盖 1000 多个精选的 AI/ML 项目摘要。 激活触发条件: - 显式触发:/pkm <查询> - 关于 GitHub 仓库、库或开源工具的问题 - “查找/搜索……相关的仓库”、“……相关项目”、“推荐库” - 技术关键词 + 搜索意图(例如,“LangChain 替代方案”、“RAG 实现工具”) 功能特点: - 支持模糊搜索,具备容错能力 - 支持韩语与英语双语搜索 - 可按语言(Python、TypeScript 等)与标签进行筛选 - 返回结果附带相关性得分,并配有摘要信息

SKILL.md
--- frontmatter
name: pkm
description: |
  TechPKM GitHub Repository Search - Searches through 1000+ curated AI/ML project summaries.

  Activation triggers:
  - Explicit: /pkm <query>
  - Questions about GitHub repos, libraries, or open-source tools
  - "Find/search repositories for...", "... 관련 프로젝트", "라이브러리 추천"
  - Technical keyword + search intent (e.g., "langchain alternatives", "RAG 구현 도구")

  Capabilities:
  - Fuzzy search with typo tolerance
  - Korean/English bilingual search
  - Filter by language (Python, TypeScript, etc.) and tags
  - Returns relevance-scored results with summaries
invocation: user

PKM Search Skill

You are now enhanced with the PKM (Personal Knowledge Management) search capability for TechPKM.

What This Skill Does

This skill searches through 1000+ curated GitHub repository summaries in TechPKM. Each repo has been analyzed and summarized in Korean with metadata including:

  • Programming language
  • Tags (in Korean)
  • Aliases
  • Concise summary

When to Use

Automatically activate when users:

  1. Ask about GitHub repositories or open-source tools
  2. Request library/framework recommendations
  3. Search for specific technical implementations
  4. Use phrases like "깃헙", "프로젝트", "라이브러리", "레포"
  5. Ask about AI/ML tools, frameworks, or implementations

How to Search

Basic Search

bash
cd /Users/imgeunseog/Documents/Github/TechPKM/.claude/skills/pkm
python scripts/search.py "<query>"

With Filters

bash
# Filter by language
python scripts/search.py "langchain" --language Python

# Filter by tags (comma-separated)
python scripts/search.py "RAG" --tags "오픈소스,LLM"

# Limit results
python scripts/search.py "embedding" --limit 5

# JSON output for programmatic use
python scripts/search.py "agent" --json

Combined Example

bash
python scripts/search.py "자연어처리" --language Python --tags "LLM" --limit 10

Search Features

1. Fuzzy Matching

Tolerates typos: "langchan" → finds "langchain"

2. Korean/English Translation

Automatically expands queries:

  • "NLP" → also searches "자연어처리"
  • "딥러닝" → also searches "deep learning"
  • "RAG" → also searches "검색 증강 생성"

3. Scoring System

Results ranked by relevance (0-100):

  • Exact repo name match: +100
  • Alias match: +80
  • Exact tag match: +60
  • Partial tag match: +40
  • Summary contains query: +30
  • Fuzzy similarity: up to +25

Index Management

Rebuild Index (if needed)

bash
python scripts/build_index.py

The index auto-detects changes:

  • New files added to /github/
  • Modified files
  • Deleted files

Force rebuild:

bash
python scripts/build_index.py --force

Response Format

When presenting results, use this format:

markdown
## PKM Search Results: "<query>"

X repositories found:

### 1. owner/repo-name
**Language**: Python | **Score**: 95/100
**Tags**: Tag1, Tag2, Tag3
**URL**: https://github.com/owner/repo

> Summary of the repository's purpose and capabilities...

### 2. ...

Example Interactions

User: "langchain 대안 뭐 있어?" Action: Search for "langchain" and related frameworks

User: "RAG 구현할 때 쓸만한 Python 라이브러리" Action: python scripts/search.py "RAG" --language Python --limit 10

User: "음성 인식 오픈소스" Action: python scripts/search.py "음성 인식" (will also find "speech recognition")

User: "Find TypeScript AI agent frameworks" Action: python scripts/search.py "AI agent" --language TypeScript

Important Notes

  1. Always run search first before claiming no results exist
  2. Try multiple queries if initial search yields few results
  3. Use Korean terms for better matches (most tags are in Korean)
  4. Check the URL before sharing - all repos link to github.com
  5. Summarize relevantly - don't just list all results, highlight the most relevant ones

Troubleshooting

If search returns no results:

  1. Try simpler/shorter query
  2. Try the Korean equivalent term
  3. Remove filters and search broadly first
  4. Check if index exists: ls data/index.json
  5. Rebuild index if needed: python scripts/build_index.py --force