Cangjie Distillation Pipeline — Changhong Edition

Problem

Turning a book or course into capabilities an AI Agent can call is hard: the result often degenerates into a summary, notes, or generic templates. Directly publishing to a skill platform can also run into review or rate-limit issues caused by topic, quotation, and wording. cangjie-distill-changhong treats this as an engineering pipeline: extract reusable decision frameworks from source text, then precheck content, publish gradually, and revise based on feedback.

How It Works

It runs a distillation pipeline: confirm source, title, goal, and publishing needs, then perform overview, parallel extraction, validation, RIA++ construction, linking, and stress tests. For Chinese sources, it adapts quoted material by category instead of copying risky passages into the skill.
- Topic triage: classify content into caution, general, and safe tiers to decide rewriting or consultation.
- Staged publishing: release the safest item first, then continue after platform feedback.
- Rate backoff: pace batch publishing to reduce trial-and-error cost.
- Knowledge cards: write cards, topic indexes, and cross-book comparison markers for retrieval and evolution.

Limits

Use it for methodologies, frameworks, checklists, and principles, not simple summaries, translations, or memory-based extraction without text. Platform publishing must keep prechecks and probe-style flow; if the topic itself needs rewriting, softening wording alone is not sufficient.