Verified Agent Skill record

Cangjie Skill

cangjie-skill is a structured editorial workflow for turning long-form source material into a connected collection of reusable skill documents. It covers source understanding, candidate extraction, screening, skill construction, cross-linking, prompt-based testing, digest creation, and delivery while preserving traceable intermediate materials.

Tasks

Distill a book, long-video transcript, podcast transcript, course, interview, long article, or document collection into atomic, reusable skills based on its methods, frameworks, principles, checklists, and concepts.
Create a source overview, extract candidate methodology units, screen those candidates, and turn accepted units into structured SKILL.md files.
Link related skills, compile a shared glossary and index, create test prompts, record test results, and produce a reader-facing digest.
Resume interrupted work from books/<slug>/PIPELINE_STATE.md rather than restarting the process.

Inputs

The accessible source text, supplied as a PDF, EPUB, TXT file, subtitle file, transcript path, or plain text.
For a book: title, author, and publication year.
For a video, podcast, or course: title, creator or speaker, and publication date.
User confirmation of the initial content overview and the screened candidate list.
The user's preferred installation location during delivery.

Outputs

books/<book-slug>/PIPELINE_STATE.md
books/<book-slug>/BOOK_OVERVIEW.md
books/<book-slug>/verified.md
books/<book-slug>/INDEX.md
books/<book-slug>/GLOSSARY.md
books/<book-slug>/DIGEST.md
Candidate extraction files under books/<book-slug>/candidates/
Rejected units and rejection reasons under books/<book-slug>/rejected/

Limitations and checks

This is a documented methodology and template-driven workflow; the source does not document a standalone executable automation tool or installation command.
It is not intended for simple summaries, book reviews, excerpt collections, or role-playing as the author.
The process cannot proceed responsibly without accessible source text.
Parallel extraction depends on an environment that supports sub-agents; the documented fallback is to run the same five extractor prompts serially.
Non-book sources require chapter references to be mapped to traceable alternatives such as timestamps, parts, episodes, or lessons.
Verify that every proposed skill passed all three documented screening questions: support from at least two independent source passages, ability to address a new question, and distinctiveness beyond generic advice.
Verify that every SKILL.md contains all six required sections: Reading, Interpretation, Past Application, Future Trigger, Execution, and Boundary.
Verify that quoted passages remain within the stated limits of 150 Chinese characters or 100 English words per passage.