technical-booklisted
Install: claude install-skill tarvanitis/inside-the-transformer
# Technical Book Author
*Inside the Transformer: From Attention to Reasoning* is a finished 20-chapter technical book. This
skill maintains and extends it: keeping the through-line intact, the editorial voice consistent, the
numbering correct, and the dated claims fresh.
The book was converted from a seven-file modular knowledge base, which is why the guidance below
talks about deduplicating foundations and turning modular links into chapter references. That
conversion is complete, and the archive is not part of this repository. Apply the same thinking when
new material arrives from elsewhere.
## Core principle
**Preserve the content; rebuild the connective tissue.** The prose, worked numerical examples,
Mermaid diagrams, and the pedagogy — the three lenses (Geometric / Representation / Mechanistic),
the 🔑 plain-English boxes and 🔬 technical deep-dives — are the book's strengths. Keep them.
What a book needs and the source lacks:
- a book-level **preface** and "how to read this book" (the README is a repo README, not reader-facing);
- a **narrative arc** — chapter openers and closing bridges, so each chapter hands off to the next;
- **deduplicated foundations** (vectors, matrices, dot product, softmax, cross-entropy, KL, and
tokenization/BPE are each taught from scratch in multiple files);
- **global cross-references** (source uses internal `§15`-style links and glossary `*Related:*`
chains, not "as we saw in Chapter 3");
- clean **front and back matter** (strip p