What problem does it solve? Manually reverse-engineering a customer's brand style from their decks and documents is slow and inconsistent. This Skill analyzes existing PPTX, DOCX, and web artifacts and extracts a reusable style palette (colors, fonts, layout distribution, visual patterns) so future document generation matches the source brand without hand-curating design tokens. ## Core Features & Use Cases - Multi-format extraction: Parses PPTX (via python-pptx or pptx-genjs), DOCX (via python-docx or docx-templater), and HTML/CSS (via :root variables and computed styles) with auto-detection by file extension. - Structured palette output: Classifies colors into primary/secondary/accent/text/background roles using the 60-30-10 rule, captures heading and body fonts, layout distribution, and visual patterns like logo position and bullet style. - Dual output artifacts: Writes a machine-readable palette JSON plus a human-readable STYLE.md summary to ~/.lintel/brand/palettes/ for downstream generation runs. - Use Case: A customer sends their brand deck. Run the skill on the deck to produce a 'customer-A' palette, then pass --palette customer-A to future document generation so reports match the customer's branding. ## Quick Start Ask the agent to extract a style palette from your brand deck by running generate-style-learn on deck1.pptx with the palette name customer-A.