Skip to content

v0.4.0

Choose a tag to compare

View all tags
@zk-xyz zk-xyz released this 10 Apr 08:26
v0.4.0
926bdec

Added

  • Multimodal visual extraction (I8). brand_extract_visual captures a 2x DPR screenshot via headless Chrome, extracts computed styles from semantic elements plus :root CSS custom properties, and returns the screenshot as an MCP image block for agent-side vision analysis.
  • Deep site extraction. brand_extract_site discovers representative pages, captures desktop+mobile screenshots, samples multiple components per page, and persists .brand/extraction-evidence.json.
  • Design synthesis + DESIGN.md. brand_generate_designmd produces .brand/design-synthesis.json (radius, shadow, spacing, layout, motion, component, personality signals) and .brand/DESIGN.md (portable agent-facing design brief).

Improved

  • brand_start auto-mode visual fallback. When static CSS extraction scores LOW or finds fewer than 2 colors, auto-mode escalates to visual extraction, then deep site extraction when Chrome is available.
  • Richer compiled tokens. brand_compile and brand_start(auto) now compile synthesis-driven radius, shadow, layout, spacing, and motion groups into tokens.json.
  • 34 tools (up from 31). New: brand_extract_visual, brand_extract_site, brand_generate_designmd.

This is the extraction precision release. JS-rendered sites (Basecamp, Elementor, Squarespace) that returned zero colors from static CSS parsing now get computed-style extraction via headless Chrome.

Assets 2
Loading