Draft a tech blog post (velog, work-docs)

Write a tight, visuals-first Korean tech blog post for velog, saved as a draft in the public work-docs repo. velog does not render Mermaid, so every diagram becomes an image here (you upload those images to velog when publishing).

Step 0: Prereqs

• work-docs cloned at ${WORK_DOCS_REPO:-$HOME/Devel/work-docs} (shared with note/worklog).
• curl (for Kroki diagram rendering). For data charts, the report skill's venv + figures.py. Optional: an Excalidraw skill for hand-drawn diagrams.

Step 1: Scaffold the post folder

bash ${CLAUDE_SKILL_DIR}/assets/new_blog.sh <slug>

Creates blog/<YYYY-MM-DD>-<slug>/index.md + an assets/ folder, and prints the draft's GitHub URL.

Step 2: Ground it (do not write from memory)

Verify facts before writing: cubrid-manual, cmt-manual, Understand-Anything (if installed), and the real code. A tech post must be correct. Cite sources in 참고.

Step 3: Write the post (minimal text, visuals-first)

Korean. Structure: 제목 → 요약(1~2줄) → 들어가며(왜 읽나, 무슨 문제) → 핵심 섹션들 → 정리 → 참고.

• 핵심 텍스트만: 짧은 문단, 개조식.
• 개념 설명은 글보다 그림으로 (다음 단계에서 이미지로).
• 표는 비교·수치, 코드블록은 코드·로그.
• em-dash(—)는 쓰지 않는다: 쉼표·콜론·괄호·마침표로 대체.

Step 4: Make visuals as IMAGES (velog cannot render Mermaid)

Put each source in assets/, render it to an image, and embed with ![설명](assets/…).

• Diagram-as-code (Mermaid / Graphviz / D2 / PlantUML): write assets/diagram-N.mmd, then render to PNG for velog:

  bash ${CLAUDE_SKILL_DIR}/assets/render.sh assets/diagram-N.mmd assets/diagram-N.png

  Keep the .mmd source so the diagram stays editable and reproducible. (velog accepts PNG uploads; SVG upload is often blocked, so render PNG for velog.)
• Data charts: build a chart JSON and render it with the report skill's figures.py to assets/chart-N.png (see the report skill for the figures.py <block.json> <out.png> command).
• Hand-drawn / architecture: if an Excalidraw skill is installed, use it to produce assets/*.png.
• Keep diagrams simple (roughly 10-15 nodes); dense graphs render poorly.
• Prefer Mermaid for structured visuals (flow/sequence/state/ER/class): nodes auto-size to their text, so labels never clip.

Verify every image (render → look → fix → repeat): after rendering, Read the PNG and check it: is every label fully inside its node, nothing clipped or overlapping, the whole figure readable at blog width? If a diagram is wrong, fix the .mmd (switch LR↔TB, shorten labels, or split it) and re-render, then look again. Repeat until each image is clean. Never embed a diagram you have not looked at.

Note: Kroki (https://kroki.io) is a public service, so the diagram source is sent there. That is fine for public blog content. Point KROKI_URL at a self-hosted instance if you prefer.

Step 5: Commit (never push)

REPO="${WORK_DOCS_REPO:-$HOME/Devel/work-docs}"
git -C "$REPO" add "blog/<YYYY-MM-DD>-<slug>/"
git -C "$REPO" commit -m "blog: <제목>"

Give the user the draft path and GitHub URL. Pushing is the user's responsibility.

Step 6: Publish on velog (manual)

velog 에디터에 index.md 본문을 붙이고, assets/의 이미지를 드래그해 업로드하면 velog가 CDN URL로 바꿔 삽입합니다. 다이어그램이 이미지라 velog에서 정상 표시됩니다.