Reflecting National Institute of Korean Language-affiliated rules, Bareun Hangeul (formerly the Pusan National University spell/grammar checker) surface is used to perform final correction of Korean sentences. It is used as the final review step for AI-generated writing.
This skill ports NomaDamas k-skill
office-korean-spell-check(MIT) to cowork (the skill name is organized without the prefixkorean-spell-check). It is based on the public web checker, and final contextual judgment is confirmed by a human.
Policy first — check before use
https://nara-speller.co.kr/old_speller/는 Non-commercial use guidance and free for individuals and students wording are specified.https://nara-speller.co.kr/robots.txt는/is allowed, but/test_speller/is prohibited.- Therefore, this skill is used only for user-led final review, low-frequency requests, document, email, and README correction purposes.
- It is not used for large-scale batch processing, SaaS backend integration, or unauthorized resale or re-exposure within commercial services not used. In such cases, first consider contacting the provider or entering a paid API contract.
When to use / When not to use
Use: final correction of text outputs such as blogs, newsletters, copy, and contracts, README review, and rule-based rechecking after AI correction.
Prohibited use: mass transmission of code blocks, logs, and English-heavy text, external transmission of original texts containing much sensitive information, and commercial mass processing.
Prerequisites
- Internet connection, Python 3.10+
- Browser-type User-Agent + Python stdlib
urllibPOST는old_speller/resultsreturns the check result HTML (a normal fetch to bypass Cloudflare may return 403).
Workflow
1. Receive text or file path
- Direct text input → immediate check
- File check → only UTF-8 text/Markdown. If there are many code blocks, first ask the user whether they want to narrow the scope.
2. Keep requests conservative
- Default chunk size about 1500자
- Minimum between chunks 1초 pause
- Do not run too many files at once.
3. Check request (run inline stdlib code directly)
It works with only the Python standard library, without a separate script. Browser-type User-Agent + urllib POST
bypasses Cloudflare, and the errInfo JSON array embedded in the response HTML is parsed and organized into 원문 / 교정안 / 이유.
Run the code below as is, but split long texts into 1500-character chunks and place a 1-second pause between chunks (no external dependencies required).
import json, re, time, urllib.request, urllib.parse
URL = "https://nara-speller.co.kr/old_speller/results"
UA = "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/120 Safari/537.36"
def check_chunk(text: str) -> list[dict]:
body = urllib.parse.urlencode({"text1": text}).encode("utf-8")
req = urllib.request.Request(URL, data=body, headers={"User-Agent": UA,
"Content-Type": "application/x-www-form-urlencoded"})
html = urllib.request.urlopen(req, timeout=30).read().decode("utf-8", "replace")
# 응답 HTML의 data 배열(errInfo)에서 원문/교정안/도움말 추출
m = re.search(r"data\s*=\s*(\[.*?\]);", html, re.S)
items = json.loads(m.group(1)) if m else []
return [{"original": it.get("orgStr", ""),
"suggestions": [c for c in (it.get("candWord") or "").split("|") if c],
"reason": it.get("help", "")} for it in items]
def spell_check(text: str, size: int = 1500) -> list[dict]:
out = []
for i in range(0, len(text), size):
out += check_chunk(text[i:i+size])
if i + size < len(text):
time.sleep(1) # 청크 간 1초 휴지 (저빈도 정책 준수)
return out
# 파일 검사: text = open("README.md", encoding="utf-8").read()
print(json.dumps(spell_check("아버지가방에들어가신다."), ensure_ascii=False, indent=2))
This skill replaced the reference to the missing script (korean_spell_check.py) with Principle A (prompt/inline stdlib code).
errInfo/dataField names and response structures may change depending on changes to the public checker, so if parsing fails, check the response HTML and adjust the field names.requests·BeautifulSoupNo installation of external packages is required.
4. Change-focused response
Recommended response order:
- Corrected full sentence/paragraph
- List of major changes
- For each change,
원문/교정안/이유 - If necessary, a phrase such as "The result is based on the public web checker, and final contextual judgment is to be confirmed by a human"
Example JSON:
{
"original": "아버지가방에들어가신다",
"suggestions": ["아버지가 방에 들어가신다"],
"reason": "띄어쓰기, 붙여쓰기, 음절 대치와 같은 교정 방법에 따라 수정한 결과입니다."
}
Related skill chaining — recommended just before ai-slop-reviewer
Standard text deliverable chain:
{콘텐츠 생성 스킬} → ai-slop-reviewer → korean-spell-check → 사용자 최종 검토
- before:
moai-marketer:content-blog,moai-marketer:content-newsletter,moai-marketer:content-copywriting,moai-marketer:content-sns-content - before:
moai-coworker:ai-slop-reviewer— Review AI patterns (excessive adjectives, repetition, translationese) first - after: The user publishes after final contextual review
Done when
- First confirmed the public surface policy (non-commercial, low-frequency).
- For long text, chunking (about 1,500 characters) was applied.
- The results were organized around
원문 / 교정안 / 이유with a focus on - It was made clear that this is not high-frequency or commercial use.
Notes
- guide:
https://nara-speller.co.kr/guide/ - main UI:
https://nara-speller.co.kr/speller/ - old UI / form post:
https://nara-speller.co.kr/old_speller/,https://nara-speller.co.kr/old_speller/results - robots:
https://nara-speller.co.kr/robots.txt