What problem does it solve? Facing hundreds to tens of thousands of Markdown articles or notes, users cannot read them one by one and have no map of what is worth deep reading. This Skill compresses the entire corpus through a five-layer funnel and outputs a structured series of thematic deep-reading cards instead of a single summary. ## Core Features & Use Cases - Layered Compression Funnel: Surveys the corpus with a shell script, clusters filenames by domain, deep-reads only benchmark files, then cross-maps concepts across domains to surface axioms versus noise. - Thematic Card Series Output: Produces a master topic table (00 主题总表.md) plus individual topic cards with fixed structure: cold-start case, core axiom, three pillar books, counterexample, cross-disciplinary mapping, and a curated reading list of real in-corpus files. - Parallel Sub-Agent Dispatch: Ships reference templates for dispatching Explore sub-agents across batches of roughly 500 filenames each, with a ten-domain taxonomy and compact reporting format. - Use Case: A user with 5,000 Markdown book notes asks to distill the library; the Skill batches the filenames, dispatches parallel drill-down agents, synthesizes 12-20 cross-domain themes, and generates one deep-reading card per theme with verifiable in-corpus sources. ## Quick Start Ask the agent to distill your Markdown notes directory into a thematic deep-reading card series using the corpus-distiller workflow.
