Official Document Writing Assistant

What It Addresses

This skill is aimed at drafting formal documents for state-owned enterprises and government agencies, covering 38 document types such as requests, reports, summaries, speeches, meeting minutes, notices, letters, opinions, approvals, work statements, research reports, and party lectures. It targets concrete writing problems: unclear document structure, weak openings and endings, mismatch between the audience and tone, awkward headings or sentences, formatting that does not follow GB/T 9704, and missing data such as figures, document numbers, names, or dates that could otherwise be filled in casually.

How It Works

The skill does not bundle a local knowledge base or writing method. During a writing task, it retrieves the relevant fill-in template, knowledge skeleton, and execution logic chain through an online API. The user usually provides the document type, topic, and key points; the skill then guides the user through the returned template, covering audience profile, optional examples, core purpose, and supplemental details. Generation follows the logic_chain constraints, using an outline-first flow, audience matching, retrieval decisions, evidence binding, and pre-delivery checks. When required details are absent, the skill marks them as 〔To be filled: ××〕 or 〔To verify: ××〕 instead of inventing them. For formatting, it structures the title, addressee, body, and signature according to GB/T 9704 and reports the format check result before delivery.

Limits and Notes

It is best for drafting new official documents from requirements and generating first drafts by document type. It is not intended for proofreading, polishing, or rewriting existing documents, nor for non-official content such as social posts, fiction, or marketing copy. The skill depends on network access and an online service; initial use requires email registration, and some document types are subject to free usage limits. Retrieved skeletons are online-protected content and should not be exported, copied, cached, or treated as an offline knowledge base.