Spaces:
Sleeping
Sleeping
Commit History
-
docs: update all docs for v3.2 + add Part 2 component generation research
- Rewrite Medium article for v3.2: color classifier, naming authority chain, DTCG compliance, 8-source extraction, component generation teaser - Fix AURORA prompt contradiction in llm_agents.py: align SYSTEM_PROMPT (advisory only) with PROMPT_TEMPLATE (optional naming_map) - Update LinkedIn post, image guide, and context doc for v3.2 - Add PART2_COMPONENT_GENERATION.md: 30+ tool research, custom plugin decision, MVP scope (5 components, 86 variants), architecture plan - Update CLAUDE.md Phase 5 with research findings and decision Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>f0ceb42
-
feat: W3C DTCG v1 compliance + single naming authority (v3.2)
Fix 3 — DTCG strict compliance: - _to_dtcg_token() now supports $extensions with namespaced metadata (com.design-system-extractor: {frequency, confidence, category, evidence}) - Color, radius, shadow exports include rich metadata - Spec-compliant: $type, $value, $description, $extensions Fix 4 — Resolve naming authority contradiction: - Color classifier is PRIMARY naming authority (deterministic) - AURORA is SECONDARY: can only assign semantic roles (brand/text/bg/border/feedback), cannot override palette names - _get_semantic_color_overrides() rewritten with clear authority chain - filter_aurora_naming_map() added to llm_agents.py - _generate_color_name_from_hex() deprecated to thin wrapper - semantic_analyzer.py marked deprecated (absorbed elsewhere) - CLAUDE.md updated to v3.2 with current status and future roadmap All 113 tests pass. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>6b43e51
-
v3.2: Transparent logging — full reasoning visibility
- Remove all LLM reasoning truncation (90→unlimited, 60→unlimited) - Add step-by-step reasoning to every rule engine check: • Type scale: show ratio calculations between consecutive sizes • Accessibility: explain WCAG criteria and pass/fail methodology • Spacing: show GCD detection and grid alignment counts • Color stats: explain dedup process and palette size assessment • Radius: show base-4/base-8 alignment and strategy explanation • Shadow: show elevation hierarchy analysis - Add shadow insufficiency warnings with industry benchmarks (Material 6 levels, Tailwind 6, Polaris 5, Carbon 4, Chakra 6) and specific blur value suggestions for missing levels - Add score breakdown in summary (component scores out of max) - Increase log capacity from 100→500 entries - Expand log textbox from 20→30 lines - Show all failing colors with both white/black contrast ratios - Expand post-validation display from 5→10 issues Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>24adae3
-
v3.1: Rule-based color classifier — kill LLM naming
- NEW core/color_classifier.py: 100% deterministic color classification - Aggressive dedup (RGB dist < 30 within same category) - Capped categories: brand(3), text(3), bg(3), border(3), feedback(4) - Supports semantic/tailwind/material naming conventions - Every decision logged with evidence - app.py: Convention picker dropdown + preview button in Stage 3 - Export functions now use classifier instead of LLM naming_map - Removed _get_semantic_color_overrides() / _consolidate_colors() dependency - AURORA demoted to advisory-only (no more naming_map output) - CLAUDE.md updated with v3.1 architecture Fixes: "nonary/decenary" Latin ordinal naming, raw hex keys, mixed conventions, 3-way naming merge chaos Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>d0d3d7e -
v3: ReAct multi-agent architecture with parallel execution
- AURORA (ReAct): Names ALL colors with numeric shades, analyzes all 5 token types - ATLAS (ReAct): Benchmarks against industry design systems - SENTINEL (ReAct): Grounded scoring with rule-engine cross-validation - NEXUS (Tree of Thought): 2-perspective synthesis, picks best - All 3 specialist agents run in parallel via asyncio.gather - Python-based critics validate each agent output (no LLM critics) - Font family detection via document.fonts API - Radius grid analysis + shadow elevation analysis in rule engine - AURORA naming_map as single export authority (no 3-way merge) - 113 tests passing Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>abab3e7