# 바른 글 (Bareun Geul) - Korean spelling and spacing checker

> 한국어 맞춤법, 띄어쓰기 검사기와 글자 수 세기. Paste a Korean text - a cover-letter answer (자기소개서), a work email,
> a report or a notice. The page counts it four ways (공백 포함, 공백 제외, bytes with Hangul as 2, UTF-8), shows how
> much of a limit is used, and marks the errors that need no context - 몇일, 금새, 됬, 되요, 안되요, 할수 있다, 하는것,
> 읍니다, doubled !! and ?? - fixing them in one press, free. A metered run reads every sentence for what needs context
> (particles, endings, auxiliary verbs, dependent nouns, punctuation) and returns each correction with the wrong span,
> the fix, its kind, a level (확실, 권장, 제안) and a one-line reason; the page checks every correction against the text.
> Built on the SkillSafe skill @daleseo/grammar-checker (MIT).

https://bareun-geul.skillsafe.ai/

## What is free
- Counts: characters with and without spaces (a line break counts as one), bytes with Hangul and Hanja as 2 and everything else as 1 (the way many job-application sites count), UTF-8 bytes, sentences and paragraphs; a limit in any of the three units with what is left or over.
- Rules that need no context, each marked in place with its reason: spelling (몇일 to 며칠, 금새 to 금세, 역활 to 역할, 어떻해 to 어떡해, 됬 to 됐, 되요 to 돼요, 읍니다 to 습니다, 왠 to 웬 outside 왠지, -ㄹ께 to -ㄹ게), negation (지 안다 to 지 않다), spacing (할수 있다, 하는것), doubled marks. Context-dependent cases are left to the run.
- One press applies the certain fixes; the text with them can be copied.
- One-press limit presets (500, 700, 1,000, 1,500), a warning for invisible characters pasted from Word or HWP (zero-width spaces, soft hyphens) that inflate the count, and a warning when most sentences end in a register other than the one selected.
- Three samples (a cover-letter answer, a work email, a library notice) replay saved replies at no cost and cover the three verdicts.

## The metered task
One task, `proof`, on gpt-terra, metered per run. Input string fields: `task`, `genre`, `style` (격식체 or 비격식체), `sentences` (numbered s1, s2, ...) and `facts` (counts, the limit and the page's rule hits). The reply is one JSON object: `verdict` (clean, polish or fix), `corrections` (each `sentence`, `wrong`, `right`, `kind`, `level`, `why`) and `note`. The page keeps a correction only when its wrong span is found in that sentence, does not overlap another and changes something; it checks that every certain rule hit is included, that the verdict matches the levels, and that no number or English word was changed. Kept corrections can be applied to the text at once, copied, or downloaded as .txt, .md or .json.

## Links
- API tutorial: https://bareun-geul.skillsafe.ai/api.html
- Source skill: https://skillsafe.ai/skill/@daleseo/grammar-checker