commit-ko

commit-ko — commit message naturalization

From a Korean commit message draft, remove only the AI-specific bureaucratic and translationese vocabulary, without changing a single character of the meaning, scope, or type.

Input

• Commit message draft (1-line subject + optional body). If there are multiple, process each one.
• If it is in Conventional Commits format (type(scope): subject, 예: fix(auth): 로그인 실패 시 재시도 로직을 수행함), leave the type/scope prefix as is, and take only the Korean part after the colon as the target for polishing.
• (Optional) If the user provides git diffor a change summary along with it — refer to it only for the purpose of cross-checking that the meaning has not been distorted during polishing, not for adding new content.

Iron rules (inheriting the same spirit as humanize-korean)

1. Meaning, scope, and type are invariant — Never change the commit type such as feat/fix/docs/refactor/chore, the scope, or the factual relationship of what was done.
2. No inserting new content — Do not invent reasons, effects, or figures that were not in the original. Change only vocabulary and endings.
3. Do not change it toward being overly polite or overly formal either — If a commit-conventional ending like "~함" is already natural, do not upgrade it to "~하였습니다".

Processing procedure

1. references/commit-lexicon.md.
2. In the commit message, look for bureaucratic verbs (수행/진행/실시, etc.), Sino-Korean nominalizations (-성/-적/-화), overly polite endings, verbose noun phrase patterns, and English literal translation constructions(by-passive "~에 의해", double passive "~되어지다", have literal translations like "가지고 있다"). However, loanwords established as development terms such as 업데이트, 리팩터링, 버그 are not targets — the issue is not the origin of the vocabulary but the syntactic structure.
3. Replace only the patterns found according to the lexicon's prescriptions. If none of those patterns are present, guide with "This is already a natural commit message" and end (do not force changes).
4. Preserve unique tokens such as type/scope prefixes, issue numbers, and file/function names exactly as they are.

Output

• 1 polished commit message (matching the format of the original — if there was only a subject, only a subject).
• In ambiguous cases, up to 1-2 alternative expressions may be listed alongside.
• A one-line rationale for what was changed and why (e.g., "'~을 수행함' → '~함' — an unnecessarily formal verb in a commit message").

Notes

• This skill's substitution table is a derivative specialized for the commit message register from the prescriptions of skills/humanize-korean/references/ai-tell-taxonomy.md's A-7, A-8, A-9, A-12 (translationese constructions) and F-4 (Sino-Korean nominalization -성/-적/-화) (for the full ID list, see the top of references/commit-lexicon.md ). When the taxonomy is updated, also check references/commit-lexicon.md— python3 scripts/check_commit_lexicon_ids.pyThis allows checking header-body ID consistency and taxonomy existence.
• When polishing paragraph-level prose (reports, columns, blogs, etc.), do not use this skill; instead, use humanize-korean.