Most high-value content (books, long videos, podcasts) contains repeatable methods, but that knowledge rarely becomes actionable in real-world workflows. This project focuses on extracting method-oriented units from long-form content and packaging them as agent-callable, testable skills so an AI agent can execute, combine, and pressure-test those methods instead of merely summarizing them.
What Sets It Apart
- Structured RIA‑TV++ pipeline: a seven-stage flow (Adler-style overview, five parallel extractors, triple verification, RIA++ structuring, Zettelkasten linking, pressure testing, and delivery) that prioritizes verifiability and executionability over plain compression.
- Triple verification and pressure tests: candidate skills must have cross-domain evidence, predictive power, and non-triviality; each skill is validated with adversarial test prompts and mixed-skill confusion cases before being accepted.
- Agent-first outputs: instead of a single summary, the repo emits a skill pack (BOOK_OVERVIEW.md, INDEX.md, DIGEST.md, SKILL.md modules, test-prompts.json) and provides integration points to install skills into agent platforms such as Claude Code and Cursor.
Who It's For and Trade-offs
Great fit if you regularly consume long-form content and need the actionable parts turned into reusable agent tools — knowledge workers, instructional designers, AI product teams, and researchers building agent toolchains. Look elsewhere if your source materials are purely narrative or personal memoirs with little transferable methodology, or if you need an out-of-the-box, zero‑effort summary: the approach requires clean transcripts/subtitles, and rigorous human-in-the-loop checks for high-stakes use.
Where It Fits
Compared with person-centric "skill" distillation projects, this repository targets systematized method extraction (books, courses, long interviews) rather than mimicking an individual's style. It complements nuwa-skill (distilling human personas) and darwin-skill (skill evolution) by focusing on method → skill conversion and validation.
