Lobster Edition WeChat Article Creator

Problem it addresses

This skill turns WeChat article creation into a reviewable four-step pipeline for cases where direct generation is hard to control: messy source extraction, unclear title direction, drafts that sound AI-generated, and mid-process changes to material, style, or structure. Instead of producing a final article in one pass, it exposes intermediate artifacts: extracted content, candidate titles, first draft, and humanized final copy. That makes it easier to inspect where a result goes wrong.

How it works

• Extraction: prioritizes mp.weixin.qq.com links, extracts the article title and body, and stores the result in 01-extract.json.
• Title selection: generates five candidate styles based on the extracted content: pain-point, numeric, suspense, counterintuitive, and story-driven; the chosen title is injected as a theme signal for the next step.
• Drafting: creates 03-draft.md from the selected title, extracted content, and writing requirements, with room to revise style, key points, or structure.
• Humanizing: removes AI-like writing patterns and writes 04-final.md.
• State management: uses checkpoint.json to track current_node and completed steps, reducing state drift from fixed file names; each completed step updates the checkpoint so the workflow can recover from a failure point.
• Optional images: when OPENAI_API_KEY is configured and image generation is available, it can create a cover image from the title or one or two article images for key scenes.

Boundaries and caveats

It is a workflow controller for writing, not a publishing automation. Source extraction still needs human confirmation, titles and drafts may need rework, and changed page structures may require supplemental text. Final publishing, copyright checks, and fact verification remain the user's responsibility.