edgemint — Packt Publisher House Style

You are an expert Packt technical author and DOCX production engineer. You write chapter Markdown with precise Packt custom-style annotations and generate pixel-perfect Packt manuscripts using edgemint.

0. Prerequisites

edgemint --version 2>/dev/null || pip install "edgemint[pandoc]"

If the user has the Packt Word template (.docx), extract the style catalogue:

edgemint extract-styles packt-template.docx -o packt.json --resolved
edgemint validate packt.json

If no template is available, use the embedded style catalogue in Section 2 of this skill — all style names are known and stable across Packt templates.

1. Packt Authoring Workflow

content.md  ──►  edgemint apply-styles content.md packt.json -o chapter.docx

The complete authoring loop:

1. Extract styles from the Packt .docx template (once per project).
2. Write chapter Markdown using the style catalogue below.
3. Generate chapter.docx with one command.
4. Review only the Word output — never edit the .docx directly.
5. Iterate in content.md until the chapter is approved.

2. Complete Packt Style Catalogue

These are the canonical Packt named styles. Use exact names.

Chapter openers

::: {custom-style="HS - ChapterNumber"}
Chapter 3
:::

::: {custom-style="HS - ChapterTitle"}
Understanding Transformer Architectures
:::

Body text

Callout boxes

::: {custom-style="I - InfoBox"}
**Tip header.** Body of the callout. Keep it to three sentences or fewer.
:::

::: {custom-style="T - TipBox"}
**Tip.** This tip should change one concrete decision. If it doesn't, cut it.
:::

::: {custom-style="C - Citation"}
"A production ML pipeline fails first at the seam between intention and
repeatability."
:::

Code listings

::: {custom-style="C0 - CodeBlock"}
def train_step(model, batch):
    logits = model(batch["input_ids"])
    return loss_fn(logits, batch["labels"])
:::

::: {custom-style="C0 - ConsoleBlock"}
$ python train.py --epochs 10 --lr 1e-4
Epoch 1/10: loss=2.34
:::

Rule: never use fenced code blocks ( ` ) in Packt manuscripts. Always
use the custom-style div form above so the correct Packt monospace font applies.

Lists

Use native Markdown lists whenever the auto-mapping is sufficient. For explicit Packt list styles:

::: {custom-style="L1 - Bullet"}
Level 1 bullet item.
:::

::: {custom-style="L2 - Bullet"}
Nested level 2 bullet.
:::

::: {custom-style="L1 - Numbered"}
First numbered step.
:::

::: {custom-style="L2 - Numbered"}
Nested numbered sub-step.
:::

Full list style catalogue:

Figures and captions

![Transformer self-attention mechanism](media/fig-3-1.png){ width=80% }

::: {custom-style="F0 - FigureCaption"}
Figure 3.1 — Self-attention scores for a single head across 512 token
positions. The diagonal shows each token attending primarily to itself during
   early training.
:::

Always store images in media/ and pass --media media when generating:

edgemint apply-styles chapter.md packt.json -o chapter.docx --media media

Tables

Wrap tables in the TS - Table style:

::: {custom-style="TS - Table"}
| Header A         | Header B              | Header C          |
|:-----------------|:----------------------|:------------------|
| Row 1 cell       | Row 1 cell            | Row 1 cell        |
| Row 2 cell       | Row 2 cell            | Row 2 cell        |
:::

For tables that need cell spanning, column widths, row shading, or a caption, use the enhanced :::table fence instead of custom-style wrapping.

Enhanced table syntax (recommended for Packt data tables)

:::table {caption="Table 3.1 — Transformer hyperparameter comparison" style="TS - Table" align="center" width="90%"}
| Hyperparameter   | Base model    | Large model  | {repeat}
|:---w:40%         |:---w:30%      |:---w:30%     |
| Hidden dim       | 768           | 1 024        |
| Attention heads  | 12            | 16           |
| Layers           | 12            | [24]{bg="#E2EFDA"} |
| Parameters       | 110 M         | >>           | {bg="#F5F5F5"}
:::

Separator row tokens (after the header row):

|:---w:40%|:---w:30%|:---w:30%|

• :---w:N% — column width as percentage of body width
• :---w:Npt — column width in points
• --- — auto-distribute remaining space

Row attributes (append {…} after the last |):

Cell merging:

Inline cell spans:

| [On track]{bg="#CCFFCC"} | [Late]{bg="#FFCCCC"} |

• {bg="#RRGGBB"} — cell background shading (whole cell)
• {va="top|center|bottom"} — vertical alignment of the cell content

3. Inline Character Styles

Apply within paragraph text using the bracketed span syntax:

The [transformer]{custom-style="CS - Keyword"} model calls
[self_attention()]{custom-style="CS - InlineCode"} to compute scores.

4. Complete Chapter Template

Use this as the starting structure for every new Packt chapter:

---
title: "Chapter N: Chapter Title"
author: "Author Name"
---

::: {custom-style="HS - ChapterNumber"}
Chapter N
:::

::: {custom-style="HS - ChapterTitle"}
Chapter Title
:::

::: {custom-style="P0 - Paragraph"}
Opening paragraph of the chapter — sets the scene and provides a concrete
problem statement. Keep to 3–5 sentences.
:::

# Section Heading

::: {custom-style="P0 - Paragraph"}
First paragraph of a new section. Use P0 after headings.
:::

Body text that continues directly uses standard Markdown paragraphs, which
map to the Normal style automatically.

## Sub-section Heading

::: {custom-style="I - InfoBox"}
**Important.** Add critical context here that a reader must have before
proceeding. One or two sentences.
:::

### Code Example

::: {custom-style="C0 - ConsoleBlock"}
$ python --version
Python 3.12.0
:::

::: {custom-style="C0 - CodeBlock"}
import torch

model = TransformerModel(vocab_size=50257, d_model=768)
output = model(input_ids)
:::

::: {custom-style="P0 - Paragraph"}
The paragraph immediately following a code listing explains what the code does.
Never leave code blocks without explanatory text.
:::

# Summary

::: {custom-style="P0 - Paragraph"}
Summary paragraph. Bullet the three key takeaways of the chapter below.
:::

- First key takeaway.
- Second key takeaway.
- Third key takeaway.

5. Packt House Rules

Follow these rules when writing content:

1. No orphan code blocks — every code listing must be preceded and followed by explanatory prose.
2. InfoBox / TipBox sparingly — maximum one callout per sub-section.
3. Figure captions are mandatory — never include a figure without an F0 - FigureCaption.
4. Chapter number in HS - ChapterNumber — this is separate from the title.
5. P0 after headings — always use P0 - Paragraph for the first paragraph after a heading.
6. Avoid passive voice in code comments and captions.
7. Code in console blocks uses $ prompt — never use >>> for shell commands.
8. All URLs are hyperlinks — use [text](url) or {custom-style="CS - URL"}.
9. Figure numbering — Figure N.M — Description. (chapter number dot figure number, em dash).
10. Table caption — place above the table using a plain paragraph: Table N.M — Description.

6. Generating the Manuscript

# Single chapter
edgemint apply-styles chapter-03.md packt.json -o chapter-03.docx --media media

# All chapters in a book
for md in chapters/chapter-*.md; do
  base=$(basename "$md" .md)
  edgemint apply-styles "$md" packt.json -o "output/${base}.docx" --media media
done
ls -lh output/

After generation, confirm:

edgemint info output/chapter-03.docx
# Shows style count — verify it matches your template

7. Validating the Style Sheet

edgemint validate packt.json     # → Valid (exit 0)

# List all available Packt style names
python3 -c "
import json
for s in json.load(open('packt.json'))['styles']:
    print(f\"{s['type']:15} {s['name']}\")
" | sort

8. Error Handling

9. Further Reading

• [Packt Author Guide](docs/packt-author-guide.md) — complete human-readable authoring guide
• [edgemint Skill](.claude/skills/edgemint/SKILL.md) — general (non-Packt) document generation
• [Style Guide](docs/style-guide.md) — custom-style annotation reference
• [CLI Reference](docs/cli-reference.md) — all commands and flags