dsh-deepread: Evidence-first Deep Reading Plugin for DeepSeek Harness
xiehuan123/dsh-deepread
DeepRead is a DSH plugin that enables evidence-first deep reading for AI agents, turning articles, books, and PDFs into traceable claims, evidence, confidence levels, and knowledge maps.
dsh-deepread is a plugin for DeepSeek Harness that provides evidence-first deep reading for AI agents, turning long articles, books, PDFs, and document sets into traceable claims, evidence, confidence levels, knowledge maps, and review questions. It solves the problem of AI reading that cannot distinguish facts from inferences or fills missing support with invented evidence, ensuring every important claim is backed by source text and marking unverifiable claims explicitly. Core capabilities include five reading modes, knowledge-map mode, batch comparison, cost preview, and multiple export formats, helping users efficiently and credibly digest complex documents.
How to Install
dsh plugin --profile web add dsh-deepread- Category
- Data & Knowledge
- Platform
- DSH Plugin
- Author
- xiehuan123
- Distribution
- Plugin
dsh-deepread Key Features
dsh-deepread Repo Summary
What Does It Do?
DeepRead is a DSH plugin for DeepSeek Harness that provides evidence-first deep reading for AI agents, turning long articles, books, PDFs, and document sets into traceable claims, evidence, confidence levels, knowledge maps, and review questions. It solves the problem of AI reading that cannot distinguish facts from inferences or fills missing support with invented evidence, ensuring every important claim is backed by source text and marking unverifiable claims explicitly. Core capabilities include five reading modes, knowledge-map mode, batch comparison, cost preview, and multiple export formats, helping users efficiently and credibly digest complex documents.
Core Features
- Five reading modes: quick key takeaways, deep in-depth reading, map knowledge map, feynman Feynman technique (11-step loop + spaced repetition), and book whole-book reading.
- Knowledge-map mode: categorizes content into ten types (conclusion, sub-claim, mechanism, fact, data, case, hidden premise, objection, limitation, actionable advice), pairs every claim with evidence, and marks four confidence levels (author intent, original facts & data, reasonable inference, unverifiable).
- Three input methods: WeChat article URLs, files (.txt/.md/.html/.pdf, with a built-in pure-JS PDF extractor supporting Chinese ToUnicode mapping), and pasted text.
- Batch compare: pass 2-10 documents to get per-document summaries plus a cross-document report with comparison matrix, conflicts, complementarity, and synthesis.
- Citations: reports carry page/paragraph provenance, with a citation table locating claims back to 【第N页】 markers in the source.
- Cost preview: use estimate: true to preview token spend, model-call count, and expected time per mode without calling the model.
How to Use This Plugin?
After enabling it in DSH, you can call the deepread tool in chat or use the 📖 reading panel in the browser UI. Enter a link, file path, or pasted text, select a mode (such as map knowledge-map mode) and export format (md, mm, html, or all), then click start. DeepRead will generate a report with evidence and confidence levels, writing output to the deepread-output/ directory in the workspace. To adjust behavior, modify reading focus or export options in settings.
This page is an independent rewrite of the plugin's official README — for authoritative documentation and the latest changes, refer to the source: xiehuan123/dsh-deepread. The plugin is third-party code that runs on your machine once installed; inclusion does not imply endorsement — please review the source before installing.