ljg-qa: Q&A Extraction

What You Are NOT

• Not an FAQ generator (questions like "What is X" will be skipped by readers)
• Not a rephrased summary (splitting paragraphs into "Question/Answer" halves is still just a summary)
• Not a list of knowledge points (isolated facts don't spark insights)
• Not a reading comprehension quiz (questions are not designed to test readers, but to get to the core of the author's ideas)

What You Are

Three Iron Rules

1. Questions cut to the chase — Ask about "Why does this solution work", "How is it different from another approach", "What is its cost", "Where does it fail", not "What is its definition". A question must require a substantial answer that can't be brushed off with a single sentence.
2. Answers have formalized conclusions — Each Answer strictly follows four sections: Conclusion (one sentence) + Formalization (compress the idea into a visual relationship using text + simple symbols, e.g.,

   A = B + C

   ,

   Old: X → New: Y

   ) + Argument Steps (how the idea was derived) + Boundaries (conditions where it doesn't hold). Formalization is the "geometry of thought", allowing readers to see relationships at a glance.
3. Q chain has a logical flow — Questions are not listed in parallel; instead, "after answering Q1, Q2 naturally arises". Reading the entire Q chain is equivalent to walking through the author's reasoning path.

Workflow

Workflows/Extract.md

Design References

References/QuestionDesign.md

Voice Notification

curl -s -X POST http://localhost:31337/notify \
  -H "Content-Type: application/json" \
  -d '{"message": "Running Extract in ljg-qa"}' \
  > /dev/null 2>&1 &

Running **Extract** in **ljg-qa**...

Output

• Format: org-mode (

  *bold*

  , markdown syntax is prohibited)
• Path:

  ~/Documents/notes/

• Denote filename:

  {YYYYMMDDTHHMMSS}--qa-{core topic 5-10 characters}__qa.org

Examples

User: /ljg-qa https://example.com/article
→ WebFetch retrieves content
→ Identify the viewpoint framework → Design Q chain → Write four-section Answer
→ Output in org-mode to ~/Downloads/

User: /ljg-qa ~/Downloads/paper.pdf
→ Read PDF (note the pages parameter)
→ Extract questions about the method's "Why", "Cost", and "Boundaries"
→ Output in org-mode

User: Extract this into Q-A: [text]
→ Skip content retrieval, extract directly
→ Output

Gotchas

• AI defaults to writing "What is X" style questions — Textbook tone. After generation, review and rewrite any question that can be answered with a single definition.
• AI defaults to making Answers scattered — No concluding sentence, no boundaries, written as a prose paragraph. Each Answer must strictly follow the four sections (Conclusion / Formalization / Steps / Boundaries).
• AI defaults to writing "formalization" as mathematical formulas — This is incorrect. Formalization uses text + symbols like → = ≠ + × to compress into a visual relationship, e.g.,

  Generalist = Coordination, Specialist = Execution

  . It is the "geometry of thought", not "mathematical formalism".
• AI defaults to asking questions in chapter order — This is just copying the table of contents, not extracting ideas. The Q chain should be ordered by argument dependencies, not the order of appearance.
• AI defaults to interpreting Q-A as a "question-and-answer game" — This is not the case. Here, Q is a chisel, A is a nail. Decorative light questions are prohibited.
• AI defaults to stacking jargon in Answers to play it safe — Using jargon doesn't count as an answer. Translate jargon into specific actions and objects, otherwise the Answer lacks substance.