สกิลสำหรับแปลง Open Source repository → ตำราเรียนภาษาไทย
ให้ AI (OpenCode, Claude Code, Codex) อ่าน Open Source repository แล้วเขียนเป็นบทเรียนภาษาไทย — จากไม่รู้ → เข้าใจ → วิเคราะห์ได้ → ลงมือทำได้
- คนที่อยากเรียนสถาปัตยกรรมจาก Open Source จริง
- คนที่อยากให้ AI สอนแทนการสรุปเฉยๆ
- นักพัฒนาไทยที่อยากอ่านตำราภาษาไทย
# ใน OpenCode
/skills add aetox-skills/opensource-textbookหรือเพิ่มใน opencode.json:
{
"skills": [
"https://github.com/aetox-skills/opensource-textbook"
]
}หรือวางในโฟลเดอร์ skills โดยตรง:
.opencode/skills/opensource-textbook/SKILL.md
.claude/skills/opensource-textbook/SKILL.md
.agents/skills/opensource-textbook/SKILL.md
| Pass Level | โมเดลขั้นต่ำ | แนะนำ |
|---|---|---|
| Quick Note | GPT-4o, Claude 3.5 Sonnet, DeepSeek V3 | — |
| Standard Lesson | Claude 3.5 Sonnet, DeepSeek R1/V4 | GPT-4o ขึ้นไป |
| Full Textbook | Claude Opus 4+, GPT-5+, DeepSeek V4 Pro | Opus 4.5+ หรือ GPT 5.5 Pro |
💡 Skill นี้ถูกออกแบบให้ Full Textbook ต้องใช้โมเดลที่ reasoning แข็งพอจะทำ 15 rules + 8 layers + comparison + deep terminology ได้ใน pass เดียว — ถ้าใช้โมเดลต่ำกว่านี้ อาจต้อง split เป็นหลาย pass
ใช้ opensource-textbook skill
ทำ Architecture Chapter จาก https://github.com/anomalyco/opencode
ภาษาไทย, Full Textbook
opensource-textbook/
├── SKILL.md ← contract สำหรับ AI (15 rules, 8-layer density, comparison framework, deep terminology)
├── README.md ← คู่มือมนุษย์ (ไฟล์นี้)
├── templates/ ← 11 templates บทเรียน (concept, architecture, code-reading, implementation, comparison, exercise, book-plan, index, glossary, source-map, capstone)
└── checklists/ ← 4 checklists (repo-inspection, evidence-check, textbook-quality, evaluation-rubric)
SKILL.md— 15 Operating Rules, Knowledge Sources, Book-Level Planning, 8-Level Knowledge Density (tiered), Architecture Comparison Framework, Deep Terminology Protocol, Rubric, Reporttemplates/book-plan.md— แผนทั้งเล่มtemplates/index.md— หน้าประตูเข้าหนังสือtemplates/concept-chapter.md— สอนแนวคิดtemplates/architecture-chapter.md— สอนภาพรวมระบบ + Architecture Comparison tablestemplates/code-reading-chapter.md— พาอ่านไฟล์จริง (🆕 Code Walkthrough layer)templates/implementation-chapter.md— สร้าง mini versiontemplates/comparison-chapter.md— เทียบ design choicetemplates/exercise-chapter.md— แบบฝึกหัดtemplates/glossary.md— คำศัพท์ (Standard + Deep format)templates/source-map.md— แนวคิด↔ไฟล์↔บทtemplates/capstone-project.md— โปรเจคจบเล่มchecklists/repo-inspection.md— 7 ขั้น Code Reading Pipelinechecklists/evidence-check.md— Evidence auditchecklists/textbook-quality.md— 9 ส่วน Teaching Formatchecklists/evaluation-rubric.md— Scoring /100
| ความสามารถ | รายละเอียด |
|---|---|
| 15 Operating Rules | ห้ามเขียนก่อนตรวจสอบ, ห้ามเดา, แยก evidence, สอนอย่าบอก |
| 6 Evidence Labels | [✓][→][?][!][M][W] — แยก repo/model/web |
| Knowledge Sources | Repo, Model Knowledge, Web — architecture=repo only |
| Code Reading Pipeline | 7 ขั้น — ห้ามข้าม, ห้ามเขียน arch จาก README |
| Book-Level Planning | book-plan.md — target reader, chapter order, dependency graph |
| 8-Level Knowledge Density | Overview→Concept→Mechanism→Code Walkthrough→Evidence→Design Reason→Transfer→Pitfall — tiered by pass level |
| Architecture Comparison Framework | ทุก component หลักเทียบกับทางเลือก — Performance/Complexity/Maintainability/Ecosystem fit |
| Deep Terminology Protocol | core terms 4 ชั้น: คำจำกัดความ→กลไก→ความสำคัญ→ตัวอย่าง [✓] |
| 7 Chapter Types | Concept, Architecture, Code Reading, Implementation, Comparison, Exercise, Book Planning |
| Pass Levels | Quick Note / Standard / Full + Artifact Budget |
| Evaluation Rubric | /100 — < 85 revise |
| Textbook Worthiness | Score /100 — ไม่ใช้ stars อย่างเดียว |
| Source Snapshot | Commit SHA, branch, date — ทุกบทต้อง pin |
| Citation Format | path/file.ts#L10-L45 @ <sha> |
- Architecture claims ต้องมี
[✓]จาก repo เท่านั้น - ต้องให้ AI อ่าน repo จริงก่อน (ผ่าน Code Reading Pipeline 7 ขั้น)
- ภาษาไทยเท่านั้น
ใช้ scripts/validate-output.js ตรวจ output แบบ deterministic — เช็ค Source Snapshot, evidence counts, citations, required sections, และ self-score:
node scripts/validate-output.js ./my-textbook-output --pass fullPass levels: quick, standard, full — แต่ละ level มี min evidence ต่างกัน
MIT — ใช้ได้อิสระ