claude-json-extraction

Claude JSON Extraction Fallback

patterns claude json api

Claude wraps output in ```json markdown fences ~30% of the time even when the system prompt says not to. Need a robust extraction strategy.

3-Tier Fallback

1. Try raw json.loads() on the response
2. Strip markdown fences via regex, try again
3. Extract first {...} block from the response, try again

Context

Used in site_generator.py _extract_json_from_response(). The SiteContent pydantic model is large (hero, about, services, testimonials, FAQ, SEO fields, color tokens, typography) but Claude Sonnet handles it without truncation.

Related

• site-builder-session-3 — where this was implemented
• persistent-browser-context — another site-builder pattern

---

Source: raw/example-meeting-notes.md | Ingested: 2026-04-13