external-blog-repost-publisher

Provide the article URL, local Markdown, or local HTML file, and the skill extracts the source, normalizes it to canonical Markdown, translates if needed, and publishes via sf-cli write-article. Verification checks confirm field-level correctness after publishing.

The translation flow classifies the article category, voice, and audience first, then translates while preserving facts, examples, and narrative framing. It writes both content and content_en fields for Chinese imports of English sources.

The skill retries extraction at least three times with backoff, downgrades through extractor paths, and tries reader mirrors such as r.jina.ai before refusing. If still blocked, it produces a source_extraction_blocker.md with the causal chain and dominant cause.

The db upsert-article command expects the ArticleRecord schema, not the API response schema. Normalize detailed_summary as a JSON string, include created_at and updated_at timestamps, or use db update-rows for narrow field patches.

A reprint notice with a clickable source URL is added to each updated language field, and the source date stays in the notice text. The articles.date field defaults to the local import date unless the user requests otherwise.