kangarooking/cangjie-skill transforms static knowledge from books, videos, and podcasts into structured, callable AI agent skills using the RIA-TV++ pipeline. The pipeline applies a seven-stage distillation process with triple verification and stress testing to produce independently executable agent skills.
Project overview
It targets the gap between passive content consumption and active agent execution by generating testable, composable skills rather than summary documents.
Project type
AI Agent · Workflow · Prompt Engineering · Learning Resources
Use cases
Learning & Education
Deployment
Refer to project documentation
License
MIT
Best for
Developers looking to convert long-form passive text content into structured, independently testable agent skills for coding and workflow environments.
Key capabilities
Processes raw text through a seven-stage sequence that includes extraction, triple verification, and stress testing to produce structured skills.
Outputs a repository of multiple independent skills accompanied by an index, overview, digest, glossary, and test prompts rather than a single summary file.
Installs tested skills directly into Claude Code or Cursor skills directories to make them callable.
Limitations and risks
Video distillation operations require a separate video-downloader skill to provide transcription.
The triple verification system typically yields a pass rate of 25-50% during the skill extraction process.
Getting started
Setup difficulty levels, reasons, and the initial path to first success are not documented.
Alternatives and comparisons
Provides a continuously updated collection of over 100 standardized persona skills and multi-agent frameworks for diverse scenarios, compared to generating skills from specific source content.
A TypeScript-first framework for building LLM applications with typed signatures and structured generation, rather than a pipeline for distilling books or videos into callable skills.
A multi-model chat interface for red teaming and cognition research, supporting local models and input perturbation, rather than distilling content into structured agent skills.