YIAN6557/eian-article-art

Article Cover and Illustrations (Blackboard Chalk Style)

This is the style set by the author Eian. Before starting, open and look at examples/final/cover_bw3.png and examples/final/fig6.png first, then draw. For English articles, also look at examples/final/en/.

If the user requests a different style or provides other reference images, follow the user's instructions and do not apply this style.

Below, <skill> refers to the folder where this skill is located (when installed in Codex, it's usually ~/.agents/skills/eian-article-art; when installed in Claude Code, it's usually ~/.claude/skills/eian-article-art).

Strict Style Rules

• Background color: Pure black #000, no texture, gradient, or borders.
• Lines: Thin white chalk scribble lines forming shapes. Dense lines on bright surfaces, sparse lines on dark surfaces, with a few stray lines at the edges. Do not use clean thick white outlines, gray fills, or glow effects.
• Stick figures: Chubby, smooth, with a large round head, no face, no clothes.
• Two reference images, each must be included: Image 1 assets/refs/figure_ref.png controls the figure; Image 2 controls the line texture: use examples/raw/cover_slot_bw_a.png for covers, examples/raw/fig2_cost_a.png for in-text figures. If only Image 1 is included, the drawing becomes too smooth, felt-like, with white halos at the edges.
• Colors: Only two places may have color: tool icons (retain their official colors); measurement annotations and numbers (orange #D97757, #E8896A). Everything else is black and white.
• Text: Font Hannotate SC, white. AI does not draw text; all text is written with HTML, ensuring the text is always correct. Text on images should use the article's language.
• Content: Each image conveys only one thing, easy to understand at a glance. Do not add corner markers, page numbers, watermarks, or decorations.

Dimensions and Font Sizes

Image	Dimensions	Chinese Font Size	English Font Size
Cover	2000×800, image left text right	Title 92, subtitle 50	Title 92, subtitle 44 (can be split into two lines)
In-text figure	1672×941, image top text bottom	Label 58, explanation 45, main title 72; minimum 45	Label 52, explanation 40–45, main title 52–54; minimum 40

English text is much longer than Chinese: if a line cannot fit, split into two lines, or reduce the font size to the minimum value in the table, ensuring text does not overflow the image or overlap with the drawn elements.

Full Article Workflow

1. Read the entire article, select positions: Follow references/pick_spots.md to calculate the number of images and select positions. Create a checklist table (which paragraph to place after, what to illustrate, why it needs an image, the scene, text on the image).
2. Create a folder: Create <article_name>_images/ next to the article, containing prompts/, raw/, logos/, final/ subfolders.
3. Pause first: Present the checklist table and one sample image (usually the cover) to the user. After user approval, draw all remaining images in one go. Make changes the user requests, do not change anything else.
4. Write prompts: First read references/pitfalls.md. Modify from templates/prompt_cover.txt or templates/prompt_fig.txt and save to prompts/. Clearly specify the quantity, position percentages, and blank space reserved for text. Describe character actions as body poses, not emotional words. examples/prompts/ has 7 approved complete examples.
5. Draw images (one at a time, always include both reference images):
   • If you are Codex: Use your built-in image generation tool, pass the paths of both reference images to it (referenced_image_paths, Image 1 first, Image 2 second), and use the full prompt from prompts/. After drawing, the tool will tell you where the image is saved (~/.codex/generated_images/...), then cp it to raw/<name>.png.
   • If you are not Codex (e.g., Claude Code): Call Codex locally using a script. Run it in <article_name>_images/ (for in-text figures, replace the last reference image with fig2_cost_a.png):

     bash <skill>/scripts/codex_image2.sh prompts/x.txt raw/x.png <skill>/assets/refs/figure_ref.png <skill>/examples/raw/cover_slot_bw_a.png
     

6. Add text: Copy templates/cover.html or fig.html to <article_name>_images/, copy needed icons to logos/, modify the image and text (first read references/layout.md), then take a screenshot:

   bash <skill>/scripts/shot.sh fig1.html final/fig1.png
   

   Screenshots must use Chrome. Chrome cannot be opened in the Codex sandbox; the script will report FAILED. In that case, request permission to run the same command outside the sandbox (escalate privileges), and inform the user that this step is simply using Chrome to capture the webpage as an image. If escalation is not possible (e.g., codex exec), write an export script with all screenshot commands and tell the user to run it outside the sandbox.
7. Check (open each image after it is captured; if unqualified, redraw or modify the HTML; see references/pitfalls.md and references/layout.md for how to fix):
   • No stray letters, no garbled text, correct number of objects, no cropping.
   • The stick figure has a round head and no face, is the same person as in the reference images, not stiff, not alien-like.
   • Objects are drawn with scribble lines, with light and shadow. Do not use fine thin-line frames like a ruler.
   • Text uses handwritten font (Hannotate SC). If it appears as regular sans-serif, the font has not been downloaded; download it from Font Book first.
   • Text does not overlap with drawn elements, font size is not smaller than the table above, text does not overflow the image.
   • Text is at least 70 pixels from the four edges of the canvas; adjacent labels are at least 40 pixels apart; labels in the same row have the same line count.
   • The information conveyed by the text should also be visually evident in the image itself; do not rely solely on the text.
   • Compare all images with the sample images in examples/final/; the style must be consistent.
8. Deliver to the user:
   • All PNGs in final/.
   • <article_name>_illustrated.md: The original article text unchanged, with ![](<article_name>_images/final/fig1.png) inserted after the paragraphs specified in the checklist. Place the cover at the top.
   • The checklist table (final version).
   • To embed into X articles or WeChat public account drafts, refer to references/publish_drafts.md (optional, requires an agent capable of operating a browser).

If only one image is needed (not a full article), skip step 1 and start from step 4, still showing the user this single image first.

Cover Rules

• Left side: one stick figure + one large object, using a metaphor to convey the article's core. Example: for "draw fewer cards", draw a figure pulling a slot machine lever, with a pile of film strips on the floor.
• Use a 3:2 aspect ratio for the cover drawing (templates/cover.html places the image at 3:2). If the drawing is wider, crop it from the left to 3:2 before inserting.
• The drawing should be large: fill the full height of the left half, same size as cover_bw3.png. If the drawing is too small, enlarge it in the HTML (see references/layout.md).
• The image should not be too close to the edge: the leftmost point of the drawing should be at least 72 pixels from the left edge of the canvas (after enlarging #art, re-check).
• Right side, top to bottom: a row of icons (only for tools mentioned in the article; if none are mentioned, do not place icons) → a large title → a subtitle.
• The title must fit: Chinese font size × number of characters does not exceed 1200. Default size 92 supports at most 13 characters; if longer, reduce the font size (minimum 76); if it still doesn't fit, split into two lines. If the title is short and there are no icons, increase to 110. For English, see examples/cover_en.html.
• Text and drawing should be at least 40 pixels apart. If the drawing exceeds the left half, first reduce #art slightly to prevent text from touching the drawing.
• For icons, keep only the central pattern; remove the rounded background and borders. assets/logos/ contains Claude, Codex, Seedance logos. For other tools, find their official icons and similarly remove the background.

Requirements

• macOS: The font Hannotate SC is built-in on Mac (on first use, click download in Font Book).
• Google Chrome (or Chromium, or Edge), for screenshots.
• For drawing: Codex can draw itself. Other agents need Codex installed locally (codex command or ChatGPT desktop app); codex_image2.sh will find it automatically.