capture-translate

Scrape an article from localhost, then produce a translation alongside the original capture.

Flow

1. Scrape first. Invoke scrape-article (or scrape-authenticated if the user indicates the page is behind auth). Use the saved markdown + frontmatter as the source — do not re-fetch just to translate.
2. Detect or confirm source language. Default assumption: source is Hebrew. If frontmatter language disagrees or the user specified otherwise, use that.
3. Pick target language. Default target is English (en). User can override (--to es, "translate to French", etc.).
4. Translate.
5. Save translation to <capture-root>/translations/YYYY/MM/YYYY-MM-DD--HHMM--<slug>--<src>-<tgt>.md. Resolve <capture-root> per reference/save-location.md (project-local preferred; global fallback).

Translator choice

Default: Claude (inline). For the user's typical volume (one article at a time), Claude translating directly in-conversation gives the best quality with zero extra deps. This is the right choice for most calls.