SKILL.md — Claude Code Skills for RealtorDesk AI

This file defines domain-specific skills that Claude Code agents can use when working in this repo.

Base Skills

1. Frontend Component Builder

• Agent: agents/frontend/ (sub-task)
• Trigger: New UI component, page, or layout work
• Rules: PascalCase, co-located tests, shadcn/ui primitives, Tailwind only, absolute imports via @/

2. Supabase Backend Agent

• Agent: agents/backend/ (sub-task)
• Trigger: Edge Functions, RLS policies, database queries, auth flows
• Rules: Every table needs RLS, use supabase.auth, sequential migrations, never expose service role key client-side

3. Stripe Integration Agent

• Agent: agents/backend/ (sub-task)
• Trigger: Checkout, webhooks, subscription management, billing portal
• Rules: Verify webhook signatures, CAD currency, link checkout to Supabase user via metadata, never store card data

4. Testing Agent

• Agent: agents/testing/ (sub-task)
• Trigger: Unit tests, integration tests, E2E tests
• Rules: Co-locate with components, mock Supabase client for unit tests, Playwright for E2E

5. SEO & Content Agent

• Agent: agents/content/ (sub-task)
• Trigger: Meta tags, OG images, landing page copy, blog content
• Rules: Bilingual EN/FR, unique meta per page, structured data where applicable

6. DevOps Agent

• Agent: agents/devops/ (sub-task)
• Trigger: CI/CD, Vercel config, GitHub Actions, preview deployments
• Rules: Never deploy to production directly, preview deployments for PRs

---

Product-Specific Skills

Skill: Real Estate Data Agent

• Agent: agents/backend/ (sub-task)
• Trigger: CREA DDF integration, property data sync, MLS listing management
• Context Files: src/features/properties/, supabase/functions/crea-ddf-sync/

Capabilities:

• Implement and maintain CREA DDF API integration (OAuth 2.0 token refresh)
• Sync property listings to Supabase on schedule (Edge Function cron)
• Parse and normalize RETS/DDF property data into the app schema
• Handle listing photos, virtual tour URLs (Matterport / iGuide), and open house data
• Build property search with filters (city, price, bedrooms, property type)

Rules:

1. DDF sync runs via Edge Function — never from the client
2. Refresh OAuth tokens automatically — store encrypted in Supabase secrets
3. Property images are referenced by URL — do not store binary in Supabase
4. All DDF data synced with source: 'crea_ddf' marker for auditability
5. Respect CREA DDF rate limits — implement exponential backoff

Output:

• Edge Function: supabase/functions/crea-ddf-sync/index.ts
• DB migration: supabase/migrations/YYYYMMDD_properties.sql
• Types: src/types/property.ts

Skill: Lead Scoring Agent

• Agent: agents/backend/ (sub-task)
• Trigger: Lead scoring logic, AI chatbot routing, lead enrichment
• Context Files: src/features/leads/, supabase/functions/score-lead/, src/lib/openai.ts

Capabilities:

• Design and implement AI lead scoring using OpenAI API
• Score leads on: engagement history, property search behavior, response time, budget signals
• Trigger Lovable AI chatbot handoff for high-score leads
• Enrich lead profiles via Apify (LinkedIn, social data)
• Build drip campaign triggers based on score thresholds

Rules:

1. All OpenAI calls happen in Edge Functions — API key never touches the client
2. Lead scores are 0–100; store with scored_at timestamp for tracking
3. Score updates are async — never block UI on scoring
4. Log scoring input/output for model evaluation and debugging
5. Never store raw OpenAI response beyond what's needed — respect PIPEDA

Output:

• Edge Function: supabase/functions/score-lead/index.ts
• Hook: src/hooks/useLeadScore.ts
• Types: src/types/lead.ts