蜻蜓文档 TTWriter

A WordCraft distribution for writing in Chinese
National-standard layout for Chinese official documents, theses and legal papers, plus Chinese word counting and East Asian font / language marking.

English · 简体中文

---

What this is

蜻蜓文档 TTWriter is a Chinese edition of WordCraft: a complete source repository that builds, runs and can be changed, with the layer that writing in Chinese actually needs added on top.

WordCraft is an open-source word processor written from scratch in Rust, aiming at Microsoft Word parity while being faster, more open, and drivable by other programs. Upstream already ships a Simplified Chinese interface (zh-hans), CJK glyphs and CJK line breaking. What we add is the writing-standard layer: how a national-standard document is laid out, how characters are counted, and how fonts are written into the file.

Full upstream source	Not a plugin, not a skin. crates/ is WordCraft. One clone, and it builds.
Chinese word counting	A Chinese "word count" counts characters, not space-separated tokens. Counts match Word's Chinese word count (G2).
East Asian fonts and language	w:rFonts/@w:eastAsia and w:lang/@w:eastAsia round-trip correctly, so a Chinese document keeps its fonts and its language marking when Word opens it.
Four layout skills	Chinese official documents (GB/T 9704), theses (GB/T 7713 / 7714), legal papers, and a general skill that documents every WordCraft command.
Four Chinese presets	General, official document, thesis, legal — applied in one step.
Offline	No network, no upload. Your writing stays on your machine.

---

Install and use

Prebuilt binaries

Download the build for your platform from Releases and run wordcraft.

Build from source

git clone https://github.com/matou1118/TTWriter
cd TTWriter
cargo build --release -p wordcraft

Needs Rust 1.95+ (rust-version in Cargo.toml). The full recipe for Windows (GNU toolchain + MinGW) is build.ps1; test-all.ps1 runs build, tests, formatting, skill validation and an end-to-end layout smoke test in one command.

Chinese fonts are not committed (font binaries never are). The official Chinese edition embeds four faces as an optional build input, so its binary still shows Chinese — in the right styles — on a machine with no Chinese fonts installed:

craft-fonts/fonts/manifest.txt   # plus the font files; all gitignored
./build.ps1                      # picks them up automatically

What is embedded: 阿里巴巴普惠体 (Alibaba PuHuiTi, a sans face), 朱雀仿宋 (Zhuque Fangsong) and 霞鹜文楷 GB (LXGW WenKai GB), the last two under the SIL Open Font License. All three are fallbacks only: a machine that has the font the document asks for uses that font, exactly as before (see substitutes() in crates/fonts/src/word.rs; 仿宋_GB2312 and 方正小标宋体 resolve here too). 宋体 is not bundled at all — Windows, macOS and Linux all ship one — which keeps 28 MB out of the binary. A plain cargo build gives upstream behaviour: Chinese comes from the system.

Command line and automation

Everything is a command. The ribbon, keyboard shortcuts, the CLI, the control channel and MCP all dispatch through the same command registry.

wordcraft-cli convert draft.docx draft.pdf          # convert formats
wordcraft-cli render draft.docx page1.png --page 1  # render to an image
wordcraft-cli pdf merge all.pdf in1.pdf in2.pdf     # merge PDFs; also info / rotate / delete / extract
wordcraft-cli commands                              # list every command
wordcraft-cli mcp                                   # run as an MCP server for AI agents

---

The Chinese layout layer

Skills

Under skills/, these are operating manuals for AI assistants: which command to run when, what the standard requires, and where the traps are. Each ships runnable Python scripts.

Skill	What it does	Standard
skills/wordcraft	How to use every WordCraft command, how to verify availability, plus an evidence ledger	—
skills/gongwen	Chinese official documents: notices, requests, reports, letters, minutes, replies	GB/T 9704—2012
skills/thesis	Theses: title page, running head, table of contents, figure/table numbering, references	GB/T 7713.1, GB/T 7714—2015
skills/legal-docs	Civil complaints, lawyer's letters, legal opinions, contract review reports	—

All four are layout only: the content must already be final, and a skill will never write, invent or edit a word of it. Writing the content is a job for a writing skill.

Presets

Four sets of Chinese page setup and default styles under presets/, each a runnable script:

wordcraft-cli run --file draft.docx --script presets/zh-gongwen.json --save out.docx

Preset	For
zh-general.json	General Chinese documents
zh-gongwen.json	Official documents (type area, sizes, line spacing)
zh-thesis.json	Theses
zh-legal.json	Legal papers

Pages

Official document	Thesis	Civil complaint

---

What changed relative to upstream

The baseline is WordCraft v0.9.0. Changes fall into two groups, both described in OPEN-SOURCE-STRATEGY.md.

Offered upstream (prepared as PRs; patches in upstream-prs/)

1. Count Chinese characters as words — a run of Chinese used to be split on spaces and counted, which badly under-counted. It now follows Word's Chinese word count (G2).
2. Read and write w:lang/@w:eastAsia — Chinese paragraphs used to lose their East Asian language marking, so Word would run its spell check in English.

This edition only

3. Four layout skills and four presets — see above.
4. Fixed the font names used by the official-document layout — 仿宋_GB2312 and 楷体_GB2312 are not installed family names on modern Windows (the family in simfang.ttf is FangSong / 仿宋), so asking for them made the renderer silently fall back to a sans face and lose the FangSong/Kai distinction that a 公文 depends on. The names are now resolved in order against what is actually installed.
5. Fixed the command order in the thesis script — inserting a header moves the caret into the header story, after which main-text-only commands fail. The table of contents and the bibliography now come before the footer and header.

Known limits, documented in the skills rather than hidden:

• No Chinese spelling or grammar proofing. WordCraft proofs English, Russian, Polish, Portuguese and Italian only; this edition does not pretend otherwise.
• The GB/T 7714 entry template has no {issue} slot, so a journal issue number must be added by hand.
• The bibliography heading defaults to the English References; pass title to references.bibliography to get 参考文献.
• 方正小标宋简体 (the 2-point 小标宋 a 公文 title requires) is not bundled with Windows; without it the script falls back to a sans title and says so.

---

Relationship to upstream

This is a thin distribution: divergence from upstream is kept small, and generic improvements go upstream as PRs.

• The repository holds the complete upstream source plus our patches. Merging upstream is easy: there are only two patches, both inside small functions, and both git apply --check clean against v0.9.0.
• Internal identifiers are unchanged — crate and binary names, config directories, the bundle id ai.storyteller.wordcraft. Only the displayed name differs: the window title, the About box, the .exe's properties, the installer entries — about twenty-five lines in all. That keeps syncing with upstream cheap; the generic patches are offered upstream as PRs.
• We use no ArtCraft marks and do not represent ArtCraft. Under upstream's trademark terms this repository says, in plain text, that it is based on WordCraft by the ArtCraft team. All brand art has been replaced with our own (ATTRIBUTION.md).
• Upstream code stays MIT OR Apache-2.0. What this repository adds (skills, presets, tooling, docs) is Apache-2.0.

Issues and PRs are welcome. Upstream has an excellent development guide in AGENTS.md and this repository follows it; see CONTRIBUTING.md for the workflow and SECURITY.md to report a vulnerability.

---

Homepage and provenance

• Homepage / portfolio: www.ttread.cn (蜻蜓伴读 · 码头哥)
• Upstream: storytold/wordcraft — WordCraft is developed by the ArtCraft team, MIT OR Apache-2.0

Credits

• The ArtCraft team and every WordCraft contributor. This Rust word processor written from scratch is the ground everything here stands on; upstream built the architecture and the OOXML compatibility solidly enough that we only had to add one layer.
• The Simplified Chinese interface (zh-hans) comes from upstream, not from this edition.
• The open fonts (Inter, JetBrains Mono, Source Sans/Serif), hyphenation and spelling lists we use are credited file by file, with licences, in ATTRIBUTION.md.
• Upstream's public sources include ECMA-376 / ISO 29500 (OOXML).

Release-by-release notes and thanks are in CHANGELOG.md.

Licence

• Code in this repository: MIT OR Apache-2.0 (LICENSE-MIT, LICENSE-APACHE), the same as upstream.
• What this repository adds (skills/, presets/, tools/, docs): Apache-2.0.
• Assets and third-party material: see ATTRIBUTION.md.