- Simple 2-variable config: OPENAI_API_KEY + DEFAULT_MODEL
- Claude API auto-detected via model name "claude*"
- No code changes needed, just drop key and model name
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

- Add Claude API support via requests library (bypass Cloudflare blocks)
- Implement manual JSON parsing for Claude models (no structured output support)
- Fix system prompt extraction and injection in AnthropicWrapper
- Update classifier/stylist prompts with JSON format requirements
- Fix graph.py to handle both dict and object outputs from LLM
- Add debug logging for raw API responses
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

- Removed orphan folder 08-lead-search/ (replaced by 08a/08b/08c/08d)
- Removed 8 orphan HTML/MD files from lead folders
- Fixed duplicate section titles (JS h2 + MD heading)
- Progress counter now auto-counts from registry.json
- Bumped cookbook.js to v9

- Journey Shopping (5 Modes): SVG funnel + 6 sections (Overview, Modes, Stage Transition, Upsell, Personalization, Troubleshooting)
- Dual-Agent Flow: SVG architecture + 6 sections (Overview, Classifier, Stylist, State, Tracing, Troubleshooting)
- Split Query & Cascading: SVG 7-tier + 6 sections (Overview, Literal, Inferred, Cascading, Enrichment, Troubleshooting)
- Modification Guide: SVG file map + 6 sections (FileMap, Tags, Tone, Upsell, NewTool, Checklist)
- Fixed mermaid.run() API call in cookbook.js (nodes param instead of querySelectorAll string)
- Bumped cookbook.js version to v8

fix(agent): make Stylist proactive — stop asking questions, auto-build combos from suggest_items + ai_matches

feat(agent): add COMBO rules + examples to Stylist prompt — use suggest_items and ai_matches to build full outfits

feat(agent): add 6 few-shot examples to Classifier prompt for combo, correction, recall, and early exit scenarios

fix(agent): prevent classifier from inheriting narrow product_line (like underwear) when user asks for a combo/set

fix(agent): update RECALL rules to correctly route queries with gender/product corrections to search tool

- config.py += FREE_LLM_API_KEY, LANGFUSE_PUBLIC_KEY, LANGFUSE_SECRET_KEY
- sqlite_db.py: use SQLITE_DB_PATH from constants
- n8n_desc.py: import FREE_LLM_API_KEY from config
- faq_route.py: import FREE_LLM_API_KEY from config
- live_monitor_route.py: import Langfuse keys from config
- prompt_optimizer_route.py: update docstrings (remove hardcoded IPs)
- NEW: tests/conftest.py with centralized BASE_URL

- NEW: backend/common/constants.py (SQLITE_DB_PATH, SQLITE_GENDER_MAP, KID_SAFE_KEYWORDS, normalize_gender)
- NEW: config.py += CHATBOT_API_BASE, LANGFUSE_SELF_HOSTED_URL, N8N_BASE_URL, LITELLM_BASE_URL, STARROCKS_INGEST_HOST
- REFACTORED: stylist_engine.py (removed duplicate maps, inline imports, dead _get_fallback_mappings)
- REFACTORED: lead_search_tool.py, outfit_matcher_worker.py, sqlite_mock.py, outfit_db.py (use SQLITE_DB_PATH)
- REFACTORED: live_monitor_route, prompt_optimizer, user_simulator, simulation_runner (use CHATBOT_API_BASE)
- REFACTORED: faq_route.py, n8n_desc.py (use LITELLM_BASE_URL)
- ARCHIVED: seed_rules.py, migrate_outfit_matches.py, check_*, inspect_db.py, recover_links.py to _archive/
- ZERO hardcoded IPs remaining in runtime code

- experiment_log_route: self-contained _get_pool/_now (remove notes_route dep)
- roadmap_flow_route: self-contained _get_pool/_now (remove notes_route dep)
- check_history_route: query langgraph_chat_histories via async pool
  (was incorrectly querying dashboard_canifa.chat_history with wrong columns)
- real_user_history/check_history_route: same fix
- Verified: identity 468707 returns full chat history with products