Writing Expert Team: Multi-Role Writing Pipeline

Problem

General-purpose writing assistants often mix title brainstorming, source gathering, structuring, drafting, fact-checking, compliance review, and formatting into one pass. For regulated topics such as law, tax, or medicine, this can miss statute currency, source attribution, and publishing details like Chinese quotation marks or HTML entities. For personal voice, the output can still sound generic. This skill treats content production as a routed pipeline, switching roles inside a single session and moving from a vague idea to an exportable draft.

How It Works

The skill uses S.流程调度员 and M.元数据桥接师 as control surfaces: S invokes roles in order, pauses at interaction points, and handles retries; M normalizes outputs into a shared YAML shape so downstream roles can consume them and trace changes. The default pipeline includes style profiling, topic planning, research, knowledge organization, drafting, fact-checking, editorial review, legal review, and formatting/export, with optional 公众号发布. Named roles include 刘风格, 赵选题, 张素材, 王整理, 李文章, 吴查查, 周审稿, 刘律师, and 陈排版. Two core assets are references/viral_formula.md, covering title, opening, structure, and ending patterns, and references/redlines.md, enforcing gates for Chinese quotes, HTML entities, glyph coverage, and legal formatting. scripts/redline_check.py can scan .md/.txt files with --fix.

Boundaries

It fits self-media writing, legal/tax/medical explainers, style cloning, and single-role requests such as topic planning, structuring, review, or layout. Keep in mind that sources must be cited, red fact-check items must be fixed before continuing, and expired statutes in regulated domains must be corrected. WeChat draft push is disabled by default and requires explicit confirmation before publishing. If knowledge-base or layout scripts are not configured, the skill may skip those calls or output a layout instruction set rather than directly generating every platform asset.