m-master

Context-aware marketing platform.

Core Rules

• 개발 절대규칙: DEVELOPMENT_RULES.md
• 구현 규칙: IMPLEMENTATION_RULES.md
• 검증 규칙: VERIFICATION_RULES.md
• 위키 작업 시스템: wiki/README.md

Deployment workflow

• production branch: all routine development and deployment happen here

GitHub Actions deployment

• pushes to production deploy to the Vercel production target and rebind m-master.vercel.app
• Vercel Git auto-deploy should stay disabled to avoid duplicate deployments

Required repository secret

• VERCEL_TOKEN: token that can deploy and manage aliases for the watchers0930s-projects/m-master Vercel project

Vercel strategy

• GitHub Actions is the deployment entry point
• local CLI direct deploys should be avoided except for emergency maintenance
• m-master.vercel.app is the fixed production domain

Promotion checklist

1. Work only on production
2. Run local verification: at minimum npm run build
3. Push production
4. Wait for the GitHub Actions production deploy
5. Verify m-master.vercel.app

If production promotion is not clean

• Do not force-push production
• If a direct production push is blocked by branch protection or branch divergence, create a temporary alignment branch from the latest origin/production
• Apply the verified content onto that branch
• Open a clean PR from the alignment branch into production
• If the production deploy workflow does not auto-start after merge, run deploy.yml with workflow_dispatch on production

Database

• Separate Neon database for m-master
• Prisma uses:
  • DATABASE_URL for application queries
  • DIRECT_URL for schema operations

Image generation

• OPENAI_API_KEY: enables GPT image generation for the image studio
• OPENAI_TEXT_MODEL: optional override for GPT draft generation, defaults to gpt-5.4-mini
• OPENAI_IMAGE_MODEL: optional override, defaults to gpt-image-1-mini
• cost-optimized default: low quality and 1 variant per request
• if the API key is missing or the OpenAI request fails, the app falls back to the built-in SVG generator
• channel draft generation also uses the same OPENAI_API_KEY; if the request fails, it falls back to the built-in rule-based draft generator