Repository navigation
Releases
v0.6.0 — Văn phong mới và gói cài cho chatbot
Bản này bổ sung cách dùng cho người không viết code và mở rộng hướng dẫn biên tập theo loại bài.
- Văn phong: văn viết rõ ràng, trao đổi công việc, kỹ thuật, hành chính, bài đăng mạng xã hội và marketing. Có lựa chọn giữ loại bài gốc; kênh, giọng, xưng hô và mức sửa vẫn độc lập.
- Dùng trên chatbot: trang /start, gói Claude/Gemini, yêu cầu sao chép và tài liệu Markdown tự chứa. Gói plugin ChatGPT được chuẩn bị để nộp, chưa có listing công khai.
- Biên tập: giải thích đúng lúc, rà lối viết theo khuôn theo ngữ cảnh và giữ giọng cá nhân. Bổ sung nguồn đọc tiếng Việt, nghiên cứu quốc tế, ví dụ nên sửa và nên giữ.
- Kiểm chứng: 44 pattern, 103 example, 124 benchmark case. Case mới do agent tự viết; chưa có baseline được người bản ngữ độc lập đánh giá. Không tuyên bố tăng doanh số, tương tác hay phát hiện nguồn gốc văn bản.
Tải và cập nhật
Người dùng chatbot tải vietnamese-writing-skills-user-packs.zip, giải nén rồi mở BAT-DAU.txt. Người dùng Python cập nhật bằng:
python -m pip install --upgrade vietnamese-writing-skills==0.6.0Chi tiết: changelog tiếng Việt · nguồn cho ba profile mới.
v0.5.0 — Natural Vietnamese writing and rewrite controls
Vietnamese Writing Skills 0.5.0 adds guidance for more natural Vietnamese and reader/style context for rewrite requests.
Added
- Naturalness guidance on information flow, collocation, ellipsis, repetition, and pragmatic nuance, with attributed reading notes and separate chat/spoken-script guidance.
- Optional rewrite context and style profile, curated skill references, and structured output modes that keep review comments separate from replacement text.
- Thirty agent-authored long benchmark cases in 15 matched pairs; benchmark total is 70. No human-reviewed quality baseline is claimed.
Changed
- Revised six corpus examples to avoid superficial synonym swaps, clarify permission to reduce hype, and request author review for unsupported promotional claims.
- Rewrite consumers must inspect
output_mode: review comments and author decisions now return an emptyrewritepluscomments. The web app handles all four modes.
Fixed
- Updated Next.js and frontend test tooling to patched release lines, and refreshed vulnerable transitive dependencies while retaining the CI audit gate.
- The high/critical audit gate passes; moderate
minimatchadvisories remain in development lint tooling.
Validation: CI and Web CI passed before tagging; frontend tests, lint, type checking, production build, and the high/critical audit gate passed. The production dependency audit reports zero vulnerabilities.
Quality scope: The new benchmark cases are agent-authored; no independent human-reviewed quality baseline is claimed.
Full Changelog: v0.4.6...v0.5.0