v0.5.3 — Extraction Reliability

Four extraction reliability improvements from S004.

M1: Parallel Extraction with Timeout

Playwright visual extraction now runs with a 30-second timeout cap. If it times out, static CSS results (already computed in parallel) stand alone. Previously, a Playwright timeout could block the entire pipeline for 60+ seconds and return nothing. Static CSS is always fast (1-2s). Visual extraction enriches when it finishes.

M2: Firecrawl Integration

Set FIRECRAWL_API_KEY=fc-... to use Firecrawl for HTML retrieval. Handles JS rendering, anti-bot protection, and proxy management. Fallback chain: Firecrawl (if key set) → Playwright (if Chrome available) → static CSS (always). Wired into both brand_start auto mode and brand_extract_web.

M9: Benchmark CI

New GitHub Actions workflow runs the 10-brand extraction audit on every release. Installs Playwright in CI. Results uploaded as artifacts for trend tracking.

M10: Figma Dogfood Plan

Documented test plan for brand_extract_figma against real Figma files. Requires connected Figma MCP for manual testing.

Numbers

36 tools. 337 tests. 3 extraction fallback layers. Firecrawl → Playwright → CSS.