| name | rich-document-creator |
| description | Use this skill whenever the user wants to create rich, visually compelling documents, presentations, or PDFs — especially when they mention images, charts, design quality, or need professional output. Triggers include: "create a presentation", "make slides", "build a deck", "write a report with images", "generate a PDF brochure", "make a Word document with visuals", "create a pitch deck", "make a professional presentation on X topic", "generate slides about Y", "create a document with good design", "make a pptx", "make a docx", "make a pdf". This skill is essential for multi-format document creation (PPTX, DOCX, PDF) with real images sourced from the web. It is also the go-to skill for any presentation with more than 5 slides, any document requiring images or charts, and any request explicitly mentioning visual quality. Always prefer this skill over individual pptx/docx/pdf skills when images, design quality, or multi-format output is involved.
|
Rich Document Creator
Creates professional, image-rich documents across PPTX, DOCX, and PDF formats.
Quick Decision Guide
| What they want | Format | Guide |
|---|
| Slides / presentation / deck | PPTX | pptx-guide.md |
| Report / letter / memo / Word doc | DOCX | docx-guide.md |
| PDF brochure / handout / polished doc | PDF | pdf-guide.md |
| Multiple formats at once | All | Generate each in sequence |
When to skip this skill. If the user just wants a plain document — markdown notes, a text-only report, a basic table-only doc — daemora's built-in create_document tool is faster (one call, no install). Use this skill only when the document needs real images, custom layouts, or visual design that markdown can't express.
Step 1: Plan Content First
Before writing any code, plan the document structure:
- Identify the topic — extract key subject matter, goals, and audience from the user's request
- Decide slide/section count — use the table below as a baseline
- List key themes — what major points need to be made?
- Pick a visual style — color palette, font, image style (see design guide in each format file)
Slide / Section Count Guide
| Document Type | Recommended Count |
|---|
| Short presentation (intro/overview) | 8–12 slides |
| Standard presentation | 12–18 slides |
| Comprehensive presentation | 18–25 slides |
| Short report / memo (DOCX/PDF) | 3–6 sections |
| Full report | 6–12 sections |
Always err on the side of MORE content, not less. A thin 5-slide deck looks unfinished.
Step 2: Source Images
This is the key differentiator of this skill. Every document must include real images.
All downloads land under the project's data directory. Use ./data/images/ (create it once with mkdir -p ./data/images) so the artifacts survive the run and are easy to find.
Image Sourcing Workflow
There is no dedicated image-search tool — use one of these three paths, in this order:
Path 1 — Direct image URL (fastest, no API key). Some hosts return a real image straight from a keyword URL. Try first:
mkdir -p ./data/images
curl -L --max-time 20 --user-agent "Mozilla/5.0" \
-o ./data/images/hero.jpg \
"https://source.unsplash.com/1600x900/?renewable,energy,solar"
curl -L --max-time 20 -o ./data/images/bg1.jpg \
"https://picsum.photos/1600/900?random=1"
file ./data/images/hero.jpg
ls -lh ./data/images/hero.jpg
Path 2 — web_search then download. Use this when you want a specific image (a known landmark, branded product, news photo, etc.). web_search is text-only — it returns page URLs and descriptions, not images directly. So:
web_search({
query: "<subject> <visual style> site:unsplash.com OR site:commons.wikimedia.org",
count: 8
})
Good query patterns:
"neural network visualization site:unsplash.com""solar farm aerial photo site:commons.wikimedia.org""open office team collaboration stock photo""data dashboard screenshot site:pexels.com"
Then web_fetch the most promising result page, extract the actual image URL (look for og:image, the largest <img> src, or a download link), and curl it the same way as Path 1.
Path 3 — Generated placeholder (fallback). If both paths above fail or the image is purely decorative, generate one with Pillow via execute_command:
from PIL import Image, ImageDraw
def create_placeholder(width, height, text, color1, color2, output_path):
img = Image.new('RGB', (width, height), color1)
draw = ImageDraw.Draw(img)
for y in range(height):
ratio = y / height
r = int(int(color1[1:3], 16) * (1-ratio) + int(color2[1:3], 16) * ratio)
g = int(int(color1[3:5], 16) * (1-ratio) + int(color2[3:5], 16) * ratio)
b = int(int(color1[5:7], 16) * (1-ratio) + int(color2[5:7], 16) * ratio)
draw.line([(0, y), (width, y)], fill=(r, g, b))
draw.text((width//2, height//2), text, fill="white", anchor="mm")
img.save(output_path)
return output_path
create_placeholder(1280, 720, "Section Visual", "#1a3a5c", "#0d7377", "./data/images/hero.jpg")
Image Rules
- Always try Path 1 first — it's a single curl, no extra search/fetch steps.
- Each major slide/section needs at least one image — never ship a deck with no visuals.
- Match image content to slide topic — keywords in the URL or query should reflect the slide's actual subject.
- Always verify with
file ./data/images/<name>.jpg before embedding — a 200-byte HTML error page silently breaks pptxgenjs/docx/reportlab.
- Download before embedding — every renderer below loads from a local path, not a URL.
- Resize if needed using Pillow before embedding (target ≤ 1600 px on the long edge).
Step 3: Generate the Document
Read the format-specific guide:
Step 4: QA and Polish
After generating, render each page to a JPEG and eyeball it. Use soffice (LibreOffice's CLI) directly — no helper script needed:
soffice --headless --convert-to pdf ./data/output.pptx --outdir ./data/
pdftoppm -jpeg -r 120 ./data/output.pdf ./data/slide
ls ./data/slide-*.jpg
soffice --headless --convert-to pdf ./data/output.docx --outdir ./data/
pdftoppm -jpeg -r 120 ./data/output.pdf ./data/page
pdftoppm -jpeg -r 120 ./data/output.pdf ./data/page
If soffice isn't on PATH, try libreoffice (same flags) or install via the package manager (brew install --cask libreoffice on macOS, apt install libreoffice on Debian/Ubuntu). On macOS the binary may also live at /Applications/LibreOffice.app/Contents/MacOS/soffice.
View the resulting images and check for:
- Text overflow or cut-off
- Images not loading (404 / broken / placeholder boxes where a photo should be)
- Visual hierarchy issues
- Consistent styling across pages
Design Principles (Apply to ALL formats)
Color Strategy
- Choose ONE dominant color (60–70% of visual space)
- ONE supporting color (20–30%)
- ONE accent color for highlights (10%)
- Dark backgrounds for title/cover pages, light for content
Typography
- Title text: 36–44pt, bold
- Subheadings: 20–28pt
- Body text: 12–16pt
- Never mix more than 2 font families
Layouts to Use
- Hero layout: Full-bleed image with text overlay (great for title slides)
- Two-column: Text on left, image on right (or vice versa)
- Card grid: 2×2 or 3×2 grid of content blocks
- Full-bleed image: Image fills the slide, sparse text overlay
- Stat callout: Large numbers (60–80pt) with small label below
Things to NEVER Do
- Text-only slides with no visual element
- Default blue/white color scheme without thinking
- All slides with the same layout
- Bullet-point dumps without hierarchy
- Missing images (always add something visual per section)
Dependencies
Install once per machine. All three formats share these:
npm install -g pptxgenjs docx
pip install Pillow reportlab WeasyPrint pypdf pdfplumber --break-system-packages
brew install --cask libreoffice
brew install poppler
Run each install via execute_command. If a package is already present npm / pip / brew are no-ops, so it's safe to run on every fresh session.
