Page Capture and Parsing

Two Capture Paths

1. Extension Popup full-page capture

Entry: src/services/pageMarkdown.ts

Responsibilities:

• inject into the current tab
• identify candidate content roots
• filter ads, nav, and hidden elements
• convert structured DOM into Markdown

2. Content-script lightweight capture

Entry: src/inline-composer/pageContext.ts

Responsibilities:

• read the current selection
• fall back to a full-page text snapshot when nothing is selected
• extract image references related to the content root

How Multiple article/main Roots Are Handled

The implementation does not just take the first one.

Instead it:

1. collects multiple candidate content roots
2. scores them by text length, paragraph count, headings, images, and link density
3. keeps the strongest content root
4. merges additional strong roots when they look like part of the same main body

This reduces the chance of treating feed/list containers as the final article body.

How Image Resolution Works

The extractor now tries:

• currentSrc
• src
• srcset
• data-src
• data-original
• data-lazy-src
• other data-* image attributes

In full-page mode, images are appended as Markdown references rather than uploaded as binary attachments.