lieflat-gongwen

A Gongwen writing Skill grounded in 1.02 million Chinese characters.

lieflat
gongwen

Writing grounded in 1.02 million Chinese characters.

Lieflat
Gongwen

Built from a corpus of 1.02 million Chinese characters, this Gongwen writing Skill works with Codex, Claude Code and other agents compatible with agents.md. It turns writing style into measurable, reviewable parameters so an agent can better reproduce seven document genres: research reports, leadership speeches, work opinions, experience cases, work plans, experience summaries and Party-building materials.

The Skill uses corpus statistics to distill the style of strong Gongwen writing. It focuses on genres without a fixed template: two document families and seven document types in all.

Six official-document types: research reports, leadership speeches, work opinions, experience cases, work plans and experience summaries.

One Party-building type: a roughly 1,000-character Party-building or grassroots experience case.

The Skill does not cover fixed-format statutory documents such as notices, requests, replies and official letters. Their body structure is already prescribed, so it does not need to be inferred through corpus statistics.

Seven genres. Distinct writing DNA.

Genre	Writing focus	Family
Research report	Analysis and findings, with headings that carry an argument.	Official document
Leadership speech	A speaking structure shaped by audience and occasion.	Official document
Work opinion	Policy direction and actionable priorities.	Official document
Experience case	A concrete practice and why it worked.	Official document
Work plan	Goals, tasks, responsibilities, and sequence.	Official document
Experience summary	Review practice and extract reusable lessons.	Official document
Party-building case	A shorter, narrative-led grassroots case.	Party-building

The study covers six official-document types and one Party-building type. Fixed-format statutory documents, such as notices, requests and replies, are outside this Skill.

What the corpus actually shows

Observed feature	Corpus result	Implication
Opening formula	Only 6–8% open with formulae such as “according to” or “in order to”.	No single opening dominates.
Ending	75–87% end naturally.	A formulaic order is not required.
Percentages	60–72% use no percentages.	Never invent data to fill a page.
Analytical headings	Average first-level heading: about 17 Chinese characters.	A heading can carry the point, not just a topic.
Sentence rhythm	Average sentence: about 53–56 Chinese characters.	Parallel detail often structures longer sentences.

Figures are approximate and depend on genre and sample. Some genres have small samples. The original corpus is not included because of copyright, so these measurements cannot be independently recomputed from the public package.