plan-rivalPrIntel

Prompt file imported from traianBaciu/rivalpr-intel (.github/prompts/plan-rivalPrIntel.prompt.md). Copyright stays with the author.

Plan: RivalPR Intel — Full-Stack Implementation

Build the RivalPR Intel PR Agency CRM from your Architecture.md. 4 mandatory layers (Next.js, Go/Gin, PostgreSQL, Docker) plus Anthropic Claude AI for pitch generation. Structured into 4 phases aligned with the hackathon timeline, with custom Copilot agent modes per layer.

Phase 0 — Project Scaffolding & Agent Setup (Start of Week 2)

1. Initialize Git repo with proper .gitignore (Go, Node, Docker, .env)
2. Create project folder structure:
   • backend/ — Go module with cmd/server/, internal/{models,handlers,middleware,services,worker,config}/
   • frontend/ — Next.js App Router with src/{app,components,lib}/
   • .github/agents/ — 4 Copilot agent modes
   • .github/copilot-instructions.md — project-wide conventions
3. Create 4 custom Copilot agent modes:
   • backend.agent.md — Go/Gin specialist (tools: read, edit, search, execute). Knows Gin routing, GORM, JWT, Go project layout. Restricted to backend/ only.
   • frontend.agent.md — Next.js/shadcn/ui specialist (tools: read, edit, search, execute). Knows App Router, Tailwind, TypeScript. Restricted to frontend/ only.
   • database.agent.md — PostgreSQL/GORM specialist (tools: read, edit, search). Knows model definitions, migrations, relationships, soft-delete, UUID PKs. Restricted to models/migration code.
   • devops.agent.md — Docker/compose/CI specialist (tools: read, edit, search, execute). Knows multi-stage Dockerfiles, compose orchestration, GitHub Actions. Restricted to infra files.
4. Create workspace-wide copilot-instructions.md with project conventions (naming, error handling, auth patterns)

Phase 1 — Database & Backend Core (Week 2: Apr 8–14)

5. Docker Compose baseline (parallel with step 6) — docker-compose.yml with frontend, backend, db services. PostgreSQL alpine with persistent volume. Go backend with air hot-reload. .env.example with all 6 env vars.
6. GORM models & auto-migration (parallel with step 5) — All 8 tables: User, Outlet, Client, Journalist, CrmRelationship, Campaign, Pitch, PitchVersion. UUID PKs, gorm.Model for soft-delete tables, CHECK constraint on pitches.status, composite unique indexes.
7. Auth system (depends on 6) — POST /auth/register (bcrypt cost ≥ 12), POST /auth/login (returns JWT), JWT middleware extracting user ID into Gin context.
8. CRUD API endpoints (depends on 7) — Full REST for Clients, Outlets, Journalists (cursor-paginated), CRM Relationships, Campaigns, Pitches — all user-scoped where applicable. 6 resource groups.
9. Rate limiting middleware (parallel with 8) — Token bucket per user: 10/min auth, 20/min AI, 120/min other.
10. Postman/Bruno collection (depends on 8) — Exported collection with all endpoints + auth flow examples.

Week 2 Deliverable: Working API with core endpoints + API collection

Phase 2 — Frontend & Integration (Week 3: Apr 15–21)

11. Next.js project setup — App Router, TypeScript, Tailwind, shadcn/ui. API client utility with JWT token management. Auth context/provider.
12. Auth pages (depends on 11) — Login (/login), Register (/register), auth guard middleware.
13. Core views (5 pages) (depends on 12):
    • Dashboard (/) — client count, active campaigns, recent pitches
    • Clients (/clients) — list, create, edit, soft-delete
    • Journalists (/journalists) — paginated list, outlet selection, CRM score slider
    • Campaigns (/campaigns) — list, create with press release text
    • Pitches (/pitches) — create, generate AI version, version picker with side-by-side comparison
14. Docker integration (parallel with 13) — Frontend Dockerfile (multi-stage), verify docker compose up runs all 3 services.
15. Responsive design pass (depends on 13) — Mobile viewport support, consistent component usage.

Week 3 Deliverable: FE connected to BE + docker-compose running locally

Phase 3 — AI Integration & Polish (Week 4: Apr 22–30)

16. Anthropic Claude integration — Async goroutine worker + job channel. Context injection: press release, journalist niche, outlet name, CRM score. POST to Claude API. Store pitch_versions with prompt_snapshot.
17. Optimistic UI (depends on 16) — Local status updates on "Mark as Sent", loading states with polling for AI generation.
18. README.md (parallel with 16) — Project description, prerequisites, setup, architecture overview, env var docs, docker compose up instructions.
19. Unit tests (bonus +5 pts, parallel with 16) — Backend tests for auth service and CRUD handlers.
20. CI/CD pipeline (bonus +5 pts, parallel with 19) — GitHub Actions: lint, test, build Docker images.
21. Demo prep (depends on all) — Seed DB with realistic PR data. Practice demo flow: register → create client → add journalist → create campaign → generate AI pitch → compare versions → send. Prepare architecture defense.

Week 4 Deliverable: Live demo + code walkthrough + architecture defense

Relevant Files

Verification

1. docker compose up — all 3 containers start; frontend at :3000, backend at :8080, DB at :5432
2. Register user → login → receive JWT → use for authenticated requests
3. CRUD all resources (clients, outlets, journalists, campaigns, pitches) via API
4. Trigger AI pitch generation → verify pitch_versions created with body + prompt snapshot
5. Navigate all 5 frontend views — real backend data, no hardcoded mocks
6. Full demo flow: register → client → journalist with outlet → campaign → generate pitch → compare versions → send
7. Rate limiting: confirm 429 responses when exceeding limits
8. Soft delete: delete a client, verify it disappears but historical pitches remain

Decisions

• Go backend — aligns with company's V3 rewrite evaluation; demonstrates concurrency via goroutine worker pool
• PostgreSQL — relational integrity for PR domain (pitches → clients, campaigns, journalists)
• shadcn/ui — maximizes frontend velocity for solo developer
• Async AI (202 + polling) — prevents request timeouts on Claude API calls
• Scope boundary — Core CRM + AI pitch generation. Excluded: real email sending, analytics dashboards, multi-tenancy

Further Considerations

1. Anthropic API key: Obtain before Week 4. During Weeks 2-3, stub the AI worker with mock responses so the full flow is testable without the key.
2. Live deployment (+5 bonus): Docker images make this straightforward via Railway or Fly.io. Decision deferred — plan supports it.
3. Test coverage (+5 bonus): Focus on backend unit tests for auth and pitch generation — highest ROI for a solo build.