Claw Markdown Image Generator

Problem It Targets

When the claw-markdown plugin passes article text, image metadata, and rewrite parameters to a model, the response can mix in reasoning, keyword JSON, transition sentences, and confirmations, making it hard for the frontend to parse image slots and keywords. claw-markdown-gen constrains the output to a renderable Markdown article: the first character must be a # heading, and the body should contain only headings, paragraphs, image placeholders, image captions, and keyword comments.

Mechanism and Limits

It parses the system and user messages to extract the article, image context, style, mode, and rewriteDepth; when parameters are missing, it defaults to WeChat style, detailed length, and medium rewrite. The skill generates 3-5 specific, distinguishable Chinese keywords for each image, inserts {{IMAGE:img_XXX.jpg}} near semantically related paragraphs, and adds > image caption plus <!-- keywords -->. Light rewriting focuses on polishing, medium rewriting restructures the article and adds insights, and heavy rewriting may include SVG, Mermaid, and {{AI_IMAGE:...}} placeholders for later image generation. The boundary is clear: it produces Markdown structure rather than binary image uploads; if the article is empty, an image cannot be located, or an external image API fails, it degrades to plain text, skips the image, or keeps the placeholder.