tougu-writer-factory

Advisor Writing Skill Factory

A complete meta-skill for training an exclusive advisor writing Skill from 3-5 satisfactory articles

Core Positioning

This Skill is a "factory" — it receives 3-5 satisfactory articles submitted by the advisor and outputs a ready-to-use exclusive writing Skill.

Core Capabilities

Capability	Description
8-dimension deep extraction	Title/paragraph/sentence pattern/vocabulary/analysis logic/data standards/fixed expressions/data requirements
Fixed modules	Market data acquisition + compliance red lines (general modules, no extraction needed)
Self-evaluation loop	Generate → compare → fix → loop, ensuring the Skill meets the standard
Complete output	SKILL.md + fixed modules + references/ + scripts/

Quick Start

Trigger Methods

Say to the assistant:

"Help me build a closing commentary skill"
"Train an exclusive writing template"
"Extract my writing style and build a skill"
"This is my writing style, help me package it into a skill"

Operation Flow

Phase 1: 文章采集存档（需用户确认样本）
    ↓
Phase 2: 8维深度提取
    ↓
Phase 3: 对比整合 → 风格画像（需用户确认风格）
    ↓
Phase 4: Skill组装
    ↓
Phase 5: 自测评循环（自动执行，无需用户介入）
    ↓
✅ 完成 → 触发安装检查

Output Directory

output/{文章类型}-writer/(Relative path, cross-platform compatible with Windows/Mac/Linux)

Task Objective

The objective of this Skill factory is:

Receive 3-5 satisfactory articles submitted by the advisor
Extract writing features through the 8-dimension deep extraction analysis method
Generate a ready-to-use exclusive writing Skill
Ensure generation quality through a self-evaluation loop

Complete Workflow

Phase 1: Article Collection and Archiving ✋

Input: 3-5 satisfactory articles of the same type

Requirements：

Same type of articles (closing commentary/pre-market outlook/industry interpretation, etc.)
Markdown or plain text, preserving complete paragraph structure
Indicate the author's name and the reason for satisfaction

Operations：

Save the articles to a temporary directory
Verify whether the number of articles is sufficient
Record the basic information of the articles

Output: Archiving confirmation report

Detailed Guide: See references/phase1.md

Phase 2: 8-Dimension Deep Extraction

The core extraction stage, extracting features in 8 dimensions from the sample articles

No.	Dimension	Extraction Content	Weight
1	Title Format	Main title structure, subtitle format, column name	20%
2	Paragraph Structure	Number of paragraphs, functional positioning of each paragraph, inter-paragraph transition logic	20%
3	Sentence Pattern Features	Sentence length, conditional judgment sentences, probability expressions	15%
4	Vocabulary Features	Professional terminology bank, distinctive metaphorical words, cautious words	15%
5	Analysis Logic	Data → event → logic → conclusion → recommendation framework	20%
6	Data Standards	Data types, precision requirements, comparative expressions	Reference
7	Fixed expressions	Fixed opening phrases for paragraphs, fixed ending format	10%
8	Data requirements	Required data, optional data, information needs	Reference

Detailed guide: See references/phase2.md

Phase 3: Comparative integration → Style profile ✋

Cross-compare the 8-dimension extraction results of 3-5 articles:

Consistency rate	Handling method
≥80%	Core pattern, include in style profile
50-79%	Standard pattern, mark as optional variant
<50%	Personal preference, treat as personalization option

Output：

{风格名}创作风格.md - Style guide (including paragraph examples + complete cases)
{风格名}创作风格_微观特征.md - Detailed description of micro-level features

Detailed guide: See references/phase3.md

Phase 4: Skill assembly

Assemble the generated resources into a complete Skill:

output/{文章类型}-writer/
├── SKILL.md                    # 可直接使用的写作Skill
├── 固化模块/
│   ├── compliance.md           # 合规红线
│   └── market_data.py         # 市场数据获取脚本
├── scripts/
│   ├── fetch_market_data.py   # 可执行脚本
│   └── market_data.md         # LLM数据获取驱动指南
└── references/
    ├── {风格名}创作风格.md
    ├── {风格名}创作风格_微观特征.md
    ├── {风格名}html海报模板.html
    ├── {风格名}合规检查.md
    ├── step1.md ~ step4.md
    └── self_eval.md

Fixed modules(general, copy directly):

固化模块/compliance.md → Compliance red lines
固化模块/market_data.py → Market data acquisition script

Detailed guide: See references/phase4.md

Phase 5: Self-evaluation loop

Purpose: Ensure the generated Skill truly meets the case standards

Process：

初始生成
    ↓
随机选题测试（自动）
    ↓
Skill执行：生成测试文章
    ↓
格式校验（skill-creator规范或内置校验）
    ↓
7维对比分析
    ↓
差距识别
    ↓
修复优化（如需要）
    ↓
循环直到达标（≥80%）或达3轮上限

Passing standards：

Overall match ≥ 80% → Pass
Single-dimension match ≥ 60% → Acceptable
Any dimension < 50% → Must be fixed

Detailed guide: See references/phase5.md

Quality check checklist

After generation is complete, you must verify:

File integrity check

SKILL.md contains the complete workflow
Fixed modules/ contains compliance.md and market_data.py
scripts/ contains fetch_market_data.py
references/ contains step1-4.md and self_eval.md

Content integrity check

Creative style.md contains writing examples for all paragraph types
Creative style.md contains at least 2 complete case excerpts
Micro-level features.md contains sentence features, vocabulary features, and a polishing checklist

Style consistency check

Title format is consistent with the samples
Number of paragraphs is consistent with the samples
All fixed opening phrases are extracted
All distinctive vocabulary is included
The analytical logic framework is correct

Technical dependencies

Component	Purpose
Python 3	Runtime environment
akshare	Market data acquisition
pandas	Data processing
markdown	Text processing
skill-creator	Format validation (optional, use built-in validation when not installed)

Notes

Fixed modules are general - Copy directly, no need to extract from samples
SKILL.md must be detailed - Include complete operational steps, checklists, and notes
Style documents must be substantial - Creative style.md and micro-level features.md must be detailed and complete
Self-evaluation requires generating articles - First call the generated Skill to generate a test article, then compare
Sample quality determines Skill quality - Submit the most satisfactory version, 3-5 articles of the same type

Reference document index

File	Content
`references/phase1.md`	Phase 1 Detailed Guide: Article Collection and Archiving
`references/phase2.md`	Phase 2 Detailed Guide: 8-Dimensional Deep Extraction
`references/phase3.md`	Phase 3 Detailed Guide: Comparison and Integration
`references/phase4.md`	Phase 4 Detailed Guide: Skill Assembly
`references/phase5.md`	Phase 5 Detailed Guide: Self-Assessment Loop
`references/self_eval.md`	Self-Assessment Checklist (LLM-driven)
`固化模块/compliance.md`	Compliance Red Lines (Fixed)
`固化模块/market_data.py`	Market Data Retrieval Script (Fixed)