English | 中文 | 日本語

knowledge-base-builder

A meta-skill that transforms your books and documents into a fully functional domain-specific skill — just like game-dev-knowledge serves all game development work, you can generate the equivalent for any domain.

---

What It Does

Your books/documents  →  knowledge-base-builder  →  Domain Expert Skill
  (PDF, MD, TXT)            (generate skill)          (ready to use)

The generated skill works like game-dev-knowledge: it understands domain context, provides best practices, helps make decisions, and serves as an expert reference for any task in that domain.

---

Quick Start

1. Place your books in the book/ folder (PDF, MD, TXT, DOCX)
2. Type the trigger: generate skill
3. Answer 2 questions when prompted
4. Done — your domain skill is created in .agent/skills/<domain>-knowledge/

---

Trigger Keyword

Keyword	Action
generate skill	Starts the full 6-step generation workflow

---

The 6-Step Workflow

Step	Name	What Happens
1	Environment Scan	Scans book/, asks 2 questions, runs scaffold.ps1 to build the folder tree
2	Knowledge Extraction (Map)	Reads books one by one, saving intermediate draft-*.md files to .temp/
3	Plan Structure	Synthesizes drafts into knowledge domains and a decision tree
4	Generate Skill Files (Reduce)	Fills in templates to create the final SKILL.md and references/*.md files
5	Sync Markdown Copy	Copies all files to Markdown/<domain>-knowledge/
6	Verify	Runs validate.ps1 to ensure all fields are filled and files exist. Fixes if failed.

---

Supported Input Formats

Format	Tool Required	Notes
.pdf	pdf-official or firecrawl skill	Most feature-rich
.md	None	Recommended for notes and summaries
.txt	None	Always readable
.docx	docx skill	Word documents

If no PDF tool is available, the system prompts you to paste text content directly.

---

Output Structure

After running generate skill, you get:

.agent/skills/<domain>-knowledge/
├── SKILL.md            ← Full domain expert skill with knowledge domains + decision tree
├── README.md           ← Book list and usage guide
└── references/
    ├── <domain1>.md    ← Detailed content for knowledge domain 1
    ├── <domain2>.md    ← Detailed content for knowledge domain 2
    └── ...

Markdown/<domain>-knowledge/   ← Synchronized copy

The generated skill is modeled on game-dev-knowledge and includes:

• Knowledge Domains — organized topic areas with core concepts and best practices
• Key Insights — author-attributed insights from each book
• Decision Tree — "What are you working on? → Go to Domain X"
• Source Books Table — complete reference for deeper reading

---

Example Session

User: generate skill

AI: Scanning book/ folder...
    Found: physics-textbook.pdf, chemistry-notes.md, biology-guide.txt

    No pdf-official skill detected.
    Can read: chemistry-notes.md, biology-guide.txt
    Cannot read: physics-textbook.pdf → will ask for summary

    Questions:
    1. Domain name? → "Life Sciences"
    2. Work scenarios? → "Research review, experiment design, lab troubleshooting"

AI: Extracting knowledge from 2 readable files...
    [Identified 3 domains: Biology, Chemistry, Lab Methods]
    [User provides physics summary → merged into extraction]

    Planning skill structure: life-sciences-knowledge
    Generating files...
    ✓ .agent/skills/life-sciences-knowledge/SKILL.md
    ✓ .agent/skills/life-sciences-knowledge/references/biology.md
    ✓ .agent/skills/life-sciences-knowledge/references/chemistry.md
    ✓ .agent/skills/life-sciences-knowledge/references/lab-methods.md
    ✓ .agent/skills/life-sciences-knowledge/README.md
    Syncing to Markdown/life-sciences-knowledge/ ✓

    Verifying: "How do I design a control experiment?" → ✓ answered from new skill
    Generation complete!

---

File Reference

File	Purpose
SKILL.md	Main skill entry — full 6-step workflow
scripts/scaffold.ps1	Setup script that generates the target folder structure
scripts/validate.ps1	CI/CD style integrity checker to enforce AI completion
book/README.md	How to place books in the book/ folder
references/pdf-tools.md	PDF tool detection and fallback strategy
references/skill-template.md	Standard template for the generated skill
references/knowledge-extraction.md	Knowledge extraction methodology

---

Design Principles

• Map-Reduce state management — processes books individually into .temp/ files to overcome context limits
• CI/CD style validation — uses validate.ps1 to prevent the AI from skipping steps
• Modeled on game-dev-knowledge — the gold standard for domain skill structure
• Progressive disclosure — SKILL.md provides summaries, references/ holds details
• Actionable knowledge — extracts specific, usable practices, not generic advice
• Decision-tree navigation — helps users find the right approach for their specific task