dsh-read-url

DeepSeek Harness agents can often search the web, but they still get stuck at the next step: turning a URL into clean, usable body text. A webpage is not a search snippet; navigation, ads, sidebars, and repeated scripts can quickly consume context, while Chinese pages may also bring GBK/GB2312, Big5, or Shift-JIS encoding problems. JSON APIs and RSS/Atom feeds also deserve better treatment than ordinary HTML scraping. dsh-read-url fills that gap.

Instead of dumping the whole page into the model, it focuses on what the agent actually needs. It can fetch webpages, JSON APIs, and RSS/Atom feeds, detect common encodings automatically, extract clean main content from article containers, and auto-join paginated articles. The default output is compact text with the title, useful metadata, and body; Markdown is available as an option. Cached reads, offset continuation, and batch reading help reduce repeated network requests and keep context usage under control.

This plugin is useful for anyone who needs DSH agents to read sources, check news, crawl forums or blogs, research multiple pages, or map a site's structure. It is especially helpful for workflows involving Chinese or legacy-encoded websites, JSON/RSS data sources, and token-sensitive agent loops. If your goal is to make the model read the original text first and then reason over it, dsh-read-url provides a lighter, more practical layer than whole-page fetching.