I run this every morning. Over 6 weeks it produced 17 drafts; my honest conversion is still low (think single-digit clicks, not "月10万"), but the machinery works and the failure modes are interesting. This is the build log, not a get-rich post.
The architecture: one Python file, Claude for prose, a hard validation gate
So the contract is:
Claude (claude-opus-4-8 via the Anthropic SDK): topic → structured JSON {title, sections[], keywords_used[]}.
Python: takes that JSON, matches section text against a curated link table, and inserts at most one link per 400 words.
