PROMPT ENGINEERING

# Prompt Engineering Specification / プロンプトエンジニアリング仕様 / 提示词工程规范

> Brain → Router → Canvas role design + prompt meta-principles for the company document generator

[EN](#english) · [日本語](#japanese) · [中文](#chinese)

---

## English {#english}

> makedocskill's core methodology is the Brain → Router → Canvas pipeline, achieving high-quality company documents (formal Word/PDF board materials, IR / CFO proposals, bilingual JP/ZH documents, and 16:9 HTML/PDF decks) through carefully designed prompt structures.

### Overview

makedocskill is not a traditional "give AI a paragraph and let it generate" workflow. Through the three-stage pipeline, Brain graded depth, Canvas role definition, and other prompt engineering techniques, it transforms document generation from "one-shot generation" to "structured reasoning + step-by-step verification" — so a board pack is convincing because the material was read correctly and rebuilt, not because a color theme was applied.

### 1. The Three-Stage Pipeline

#### Design Philosophy

The core insight: **Depth comes not from more steps, but from better prompt quality.**

The traditional approach lets AI complete all work at once (understand materials → organize structure → render output), producing shallow, template-like documents that round numbers, miss board questions, and bury the conclusion.

The pipeline decomposes this into three independent stages, each with a clear role: