What problem does it solve?
Reading long articles or entire books and distilling their core arguments, evidence, and logical structure is time-consuming, and casual summaries often miss the author's actual claims or fabricate evidence. This Skill performs disciplined deep reading and produces structured analytical reports with explicit evidence citations and confidence levels.
Core Features & Use Cases
- Five Reading Modes: quick (key points), deep (full argument analysis), map (claim-evidence-data knowledge map), feynman (11-step Feynman learning loop), and book (whole-book chapter-by-chapter synthesis).
- Multi-Source Input: Reads local files (.txt/.md/.html/.pdf, including raw PDF stream parsing with ToUnicode CMap decoding), fetches web pages such as WeChat articles, or analyzes pasted text, chunking content over ~9000 characters.
- Evidence Discipline: Every claim is tagged with evidence, source location, and a four-level confidence rating (author's intent / stated fact / reasonable inference / unverifiable), with explicit "no evidence provided" markers instead of fabrication.
- Optional Exports: On request, writes Markdown reports, FreeMind mind maps (.mm), or standalone HTML reports to a deepread-output directory.
- Use Case: Paste a long WeChat article and ask for a knowledge map to receive a claim-evidence table, key data table, Mermaid mind map, XMind outline, and active-recall questions.
Quick Start
Use the dsh-deepread skill to deep-read this article and extract its core claims, argument structure, and supporting evidence.
