| name | kdp-frontmatter-pages |
| description | Replace a KDP coloring book's plain text title / copyright / thank-you pages with illustrated front & back matter (and add a "This Book Belongs To" page). Write personalized image prompts, generate the art with the built-in ChatGPT image_gen tool, then merge into interior.pdf. USE WHEN user says 'tao trang bia trong sach', 'frontmatter pages', 'thay trang thank you', 'title page dep', 'this book belongs to page', 'trang dau cuoi sach', 'decorative interior pages', 'lam trang title va thank you', 'ghep frontmatter', 'illustrated title page', 'personalize thank you page'. |
KDP Frontmatter Pages
Turn the ugly plain-text title / copyright / thank-you pages into themed
illustrations that match the book's interior. Always includes a
"This Book Belongs To" page for ALL books (adults and kids alike). Fully automated flow:
Claude writes the prompts → Claude generates the art with the built-in
ChatGPT image_gen tool → Claude merges them into interior.pdf.
Image generation uses the same built-in image_gen tool as the
kdp-chatgpt-cover-creator skill — no external renderer, no manual hand-off.
(If image_gen is unavailable in the current runtime, fall back to writing
the .txt prompts and asking the user to generate 1.png/2.png/3.png
by hand, then resume at Phase 2.)
Phase Routing
| Phase | Trigger | What Claude does |
|---|
| 1 — Prompts + Generate | "generate frontmatter", "lam trang title/thank you", book has no frontmatter/*.png yet | Read plan.json, write 3 personalized prompts to output/<theme>/frontmatter/{1,2,3}.txt, then call the built-in image_gen tool to produce 1.png/2.png/3.png |
| 2 — Merge | "ghep lai", "ghep frontmatter vao", frontmatter/1.png & 3.png already exist | Inspect images, run the assembly script, run QC |
Phase 1 normally flows straight into Phase 2 in a single run — generate, then
merge. Stop between phases only if image_gen is unavailable (hand-off mode).
Always confirm the theme_key (folder under output/). If unclear, list
output/*/plan.json and ask which book.
Phase 1 — Write the prompts and generate the art
- Read
output/<theme>/plan.json. Note: audience, page_size, author,
title, and skim several page_prompts + the cover_prompt to capture the
exact visual style and any recurring mascots/objects (e.g. a kawaii
teddy bear, balloons, a smiling cake).
- Load
references/prompt-templates.md and fill the bracketed slots so the 3
pages clearly belong to THIS book — same style, same characters, real objects
from the book mentioned in the thank-you message.
- Personalize the copy:
- Title (1): short punchy hero title (not the full 60-char KDP title) +
genre sub-banner + 3-line subtitle + copyright line.
- Belongs To (2): ALWAYS generate for every book — adults and kids alike.
For adults adapt the wording (e.g. "This Book Belongs To" + name line, no
age field); for kids include the "AGE: ____" line.
- Thank You (3): warm, theme-specific message + an Amazon-review CTA.
- Write the three prompts to
output/<theme>/frontmatter/{1,2,3}.txt
(create the folder) as a record of what was generated.
- Generate the images with the built-in
image_gen tool — one call per
prompt. This is the same ChatGPT image-generation path the
kdp-chatgpt-cover-creator skill uses; do not ask the user to make the
art by hand and do not route through image_providers.py.
- Request square output for
8.5x8.5 books, portrait 3:4 for
8.5x11 (match plan.json.page_size).
- Spell every word of the baked-in text EXACTLY (title, subtitle, author such
as
BoBoArt, thank-you copy) — misspellings are the top reject reason.
- Save the chosen renders into the book folder:
output/<theme>/frontmatter/1.png — Titleoutput/<theme>/frontmatter/2.png — This Book Belongs To (always required)output/<theme>/frontmatter/3.png — Thank You
- If a render comes back garbled or off-style, re-call
image_gen for just
that page before moving on.
- Proceed directly to Phase 2 — Merge (no user hand-off needed). Only stop
and ask the user to generate the PNGs manually if
image_gen is unavailable
in this runtime.
Style is non-negotiable: square for 8.5x8.5 / portrait 3:4 for 8.5x11;
grayscale (black line + soft gray) on white; text baked in with EXACT wording;
no full-page border/frame. See references/prompt-templates.md.
Phase 2 — Merge into the interior
- Inspect
frontmatter/1.png (and 2.png, 3.png) with the Read tool.
Check: text spelled correctly, grayscale, square/portrait matches trim, art
not touching edges. If small print is garbled, re-call image_gen for that
page (or, in hand-off mode, ask the user to regenerate it). The script's plain
copyright page still carries the legal text, so minor garbling on the
title/thank-you art is only cosmetic.
- Run the assembly script (it backs up the old PDF, upscales the art to
300 DPI, keeps coloring pages on right-hand pages with blank backs, forces an
even page count):
python .claude/skills/kdp-frontmatter-pages/scripts/assemble_frontmatter.py <theme_key>
Useful flags: --size 8.5x11|8.5x8.5, --author "Name", --age 3-7,
--no-copyright.
- QC the result:
python scripts/pdf_qc.py --pdf output/<theme>/interior.pdf --trim <size> --require-even-pages
- Visually verify by rendering a few pages (title, copyright, belongs-to,
first coloring page, last/thank-you) and Reading them. Report GO/NO-GO.
Page order produced
Title(1.png) → Copyright(text) → This Book Belongs To(2.png) →
(blank pad if needed) → coloring pages (right-hand, blank backs) →
Thank You(3.png) as the even last page.
Image conventions
| File | Page | Required |
|---|
frontmatter/1.png | Title | ✅ |
frontmatter/2.png | This Book Belongs To | ✅ (always — adults + kids) |
frontmatter/3.png | Thank You | ✅ |
Generate at ≥2550px square (or 2550×3300 portrait). The script grayscales +
upscales to the trim's pixel target so print is ≥300 DPI.
Quality criteria (GO checklist)
Scripts
scripts/assemble_frontmatter.py — merge frontmatter images + coloring pages
into interior.pdf. Reads trim/author/age from plan.json. Run from repo root.
References
references/prompt-templates.md — the 3 prompt blueprints + style rules
(load during Phase 1).
