What Is the Main Purpose of the k‑skill Repository? A Complete Guide to Korean AI Agent Skills

The k‑skill repository provides a modular collection of AI agent "skills" that enable Claude Code, Codex, OpenCode, and similar tools to perform Korean‑specific tasks like booking trains, querying public databases, and automating government document workflows.

The k‑skill open‑source project solves a critical gap in AI agent tooling: the lack of reliable, maintainable implementations for region‑specific workflows. Developed by NomaDamas, this repository packages Korean data retrieval and transaction automation into discrete, installable units that any compatible agent can invoke without custom development.

Core Architecture: Four Layer Stack

The repository's design prioritizes composability over monoliths. Each capability lives in isolation, making it possible to install only what you need.

Skill Packages

Every capability resides in its own folder at the repository root—srt-booking/, zipcode-search/, k-dart/, and others. A valid skill contains:

• skill.json – machine‑readable manifest declaring inputs, output schema, and runtime requirements
• Optional implementation scripts (Python, Node.js, or browser‑runtime helpers)
• SKILL.md – human documentation explaining usage and constraints

This structure appears in skills like k-skill-dart/skill.json, which declares Korean securities market query capabilities.

Unified CLI Adapter

The top‑level k‑skill‑cli reads all skill manifests and exposes them as standard commands. Installation follows npm patterns as documented in README.md lines 12‑20:

# Install every skill globally

npx --yes skills add NomaDamas/k-skill --all -g

# Install only the SRT train booking skill

npx --yes skills add NomaDamas/k-skill --skill srt-booking -g

# Execute a skill directly

k-skill zipcode-search "강남구 역삼동"

The --skill flag provides granular control, while --all bundles the complete ecosystem.

Proxy Layer (packages/k-skill-proxy)

Free Korean public APIs often require authentication keys that agents should not handle directly. The repository ships packages/k-skill-proxy/src/server.js—a lightweight forwarding server that signs requests server‑side. As noted in README.md lines 10‑11, this keeps credentials out of client code and agent context windows.

# Query Korean stock data through the proxy

import requests

proxy_url = "http://localhost:3000/korean-stock-search"
resp = requests.get(proxy_url, params={"query": "삼성전자"})
print(resp.json())

// Fetch building register data via proxy
const fetch = require('node-fetch');

(async () => {
  const res = await fetch('http://localhost:3000/building-register-search', {
    method: 'POST',
    body: JSON.stringify({ address: '서울 강남구' })
  });
  console.log(await res.json());
})();

Runtime Support

Two specialized runtimes handle authentication‑heavy scenarios:

• k‑skill‑browser‑runtime – Provides authenticated browser sessions for skills requiring logged‑in web UIs (e.g., yebigun‑training, court‑payment‑order‑assistant)
• Dolshoi runtime – Alternative automation layer for complex multi‑step workflows

What k‑skill Enables: Task Categories

The repository's README.md (lines 31‑78) catalogs skills across functional domains:

Category	Example Skills
Transportation	srt-booking – Korea's high‑speed rail reservations
Address & Geography	zipcode-search – Korean postal code lookups
Financial Markets	k-dart – Corporate disclosure database queries
Public Records	building-register-search – Real estate registry access
Government Services	court-payment-order-assistant – Legal fee automation

Each entry links to its SKILL.md for detailed parameters and edge‑case handling.

Key Files and Their Roles

File Path	Purpose
README.md	Installation guide, compatibility matrix, and feature table
{skill-name}/skill.json	Runtime contract: inputs, outputs, dependencies
{skill-name}/SKILL.md	Human documentation with example prompts
packages/k-skill-proxy/src/server.js	API key management and request forwarding
packages/k-skill-cli/	Generated command stubs (via npm run generate:skill-stubs)
packages/k-skill-browser-runtime/	Headless browser automation for session‑based sites

Summary

• k‑skill packages Korean‑specific AI capabilities as discrete, installable units
• Each skill self‑describes via skill.json and documents itself via SKILL.md
• The k‑skill‑proxy layer abstracts API authentication away from agents
• Browser and Dolshoi runtimes handle complex login flows without agent exposure
• Installation uses standard npm patterns through the skills CLI

Frequently Asked Questions

How do I install only specific skills from the k‑skill repository?

Use the --skill flag with the skills add command: npx --yes skills add NomaDamas/k-skill --skill zipcode-search -g. This avoids pulling unnecessary dependencies.

Why does k‑skill use a proxy server instead of direct API calls?

The packages/k-skill-proxy server handles API key injection server‑side. This prevents credentials from appearing in agent context windows or client‑side code, improving security for free public APIs that require authentication.

What runtimes are available for skills requiring browser automation?

The repository provides k‑skill‑browser-runtime for standard logged‑in workflows and Dolshoi runtime for more complex automation scenarios. Both abstract session management away from the calling agent.

Where can I find documentation for a specific skill?

Each skill folder contains SKILL.md with detailed usage instructions. The root README.md links to all skill documentation in its feature table (lines 31‑78).

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}