Distilling books into agent-ready skill packs

What it does cangjie-skill is a structured pipeline—dubbed RIA-TV++—for converting a book’s methodology into a repository of discrete, callable AI skills. It runs text through six stages: whole-book analysis, parallel extraction by five specialist extractors, triple verification (typically filtering out half to three-quarters of candidates), structured RIA++ formatting, Zettelkasten-style cross-linking, and adversarial stress testing with bait questions. The output is not a summary but a multi-file skill pack: a book overview, an index map, individual SKILL.md modules, and validation prompts.

The interesting bit The rigor is the hook. Most “book to AI” projects stop at summarization; this one borrows Mortimer Adler’s analytical reading and Zhao Zhou’s RIA note-taking, then adds execution steps and boundary conditions specifically so an agent knows when not to apply a skill. It is essentially a quality-control layer for knowledge distillation, treating books as raw material that must survive cross-domain verification and predictiveness tests before graduating to skill status.

Key highlights

• Six-stage RIA-TV++ pipeline from raw text to tested skill repository
• Triple verification filters candidates for uniqueness, cross-domain support, and predictive power; typical pass rate is 25–50%
• Outputs include BOOK_OVERVIEW.md, INDEX.md, linked SKILL.md modules, and test-prompts.json for adversarial validation
• Already applied to over a dozen sources, from Buffett’s letters to the Huangdi Neijing and Mao’s Selected Works
• Designed to plug into the broader “nuwa/cangjie/darwin” skill ecosystem: distill people, distill books, then evolve both

Caveats

• The README describes the methodology and shows generated results, but the actual automation layer (how the six stages are orchestrated) is not fully visible in the provided sources
• Most demonstrated outputs live in separate repositories; this repo contains the prompts, templates, and methodology docs rather than a turnkey engine

Verdict Worth a look if you’re building agent workflows and want book-derived methodologies that are filtered, linked, and tested rather than summarized. Skip it if you’re hoping for a one-click CLI that ingests PDFs and spits out skills without human steering.