dsh-rss-reader

For many readers, RSS remains the backbone of deep, curated content intake, yet traditional readers suffer from fragmentation: feed syncing, article archiving, note-taking, and reading stats live in separate tools and rarely form a closed loop with AI conversation. dsh-rss-reader pulls the entire workflow—incremental feed fetching, article materialization into Markdown with YAML front matter, notes and highlights, an activity timeline, and a three-column reading interface with per-day navigation—into a single DeepSeek Harness plugin tab. Everything runs as local-first services inside the dsh runtime, with no external server or cloud dependency.

Reading and thinking happen on the same surface. Opening an article silently injects it into the current session's agent context, so the model immediately knows what you are reading. A one-click action—summarize, translate, or extract key points—or a free-form question pushes the article into the conversation flow, and the view switches back to the Chat tab to display the reply; selecting a passage first attaches it as a highlighted excerpt. Each sync run leaves a compact status card in the chat stream, making feed health visible at a glance. The model can also invoke tools such as rss_sync, rss_search, rss_read_article, and rss_save_note, complemented by chat commands like /feeds, /book, and /export-review, completing a browse-to-review-export loop.

The plugin is well suited for heavy RSS readers, knowledge workers, and anyone building a personal "read – annotate – ask – review" pipeline inside the DeepSeek Harness ecosystem. All data is persisted locally via SQLite and Markdown files; article IDs are SHA-256 stable across syncs for idempotency, image assets are deduplicated locally, and feed URLs pass SSRF checks, balancing offline usability with security.