Skip to content

korean humanizer 1.1.0

Latest

Choose a tag to compare

@dotoricode dotoricode released this 10 Oct 03:06

korean-humanizer v1.1.0

This release adds six guidelines for trimming redundant wording while preserving meaning, along with the improvements merged in PR #5, #6, and #8. Expressions such as 있다, 것, 수, and conjunctions are candidates for review, rather than banned words. Edits must preserve ongoing actions, possibility, ability, permission, cause and effect, and the writer's voice. Sentences and closing remarks are shortened only when their information and purpose remain intact.

Personal settings are loaded through an explicit personal=path/to/file argument. The catalog headers are now 검토할 표현 / 치환 후보. Fixed limits such as editing 20% of sentences, making three edits per paragraph, or retaining 90% of the original length have been replaced with checks for information preservation and natural phrasing. These changes differ from the earlier 1.x compatibility promise, so check your settings and parsers in a separate project before upgrading. v1.0.1 is a previously documented version; its tag is not available in the remote repository. See the upgrade notes and changelog for details.

Validation passed: 22 regression tests (15 existing, 4 comparison, and 3 packaging tests), 25 fixtures, and the pattern, cross-file, and example checks. The same-input comparison records 16 responses across 8 inputs. The previous instructions also preserved the main meaning in this sample, so these results do not establish a general quality improvement or guarantee success on every model run. Fresh installation and first-invocation checks of the v1.1.0 ZIP in both CLIs were not performed.

Use the attached korean-humanizer-v1.1.0.zip to update the listing on agensi.io. The package is free and MIT-licensed, with 17 text files at the archive root, including SKILL.md. To install it, extract the archive into an empty korean-humanizer directory. Use this package rather than GitHub's automatically generated source-code ZIP. The manifest records the source commit and per-file SHA-256 hashes; SHA256SUMS records the hashes of the attached ZIP and manifest. The package has not been uploaded directly to agensi.io.