Devit — AI-Powered Client Scoping Tool

An AI-powered scoping tool that interviews clients through a structured conversation, generates project specs, breaks them into tasks, and syncs with GitHub for tracking.

How It Works

1. Admin creates a project → generates a unique shareable link
2. Client opens the link → no login required → chats with an AI that runs a structured discovery interview
3. AI auto-transitions through phases: discovery → requirements (PRD) → implementation plan → task breakdown
4. Spec & tasks are generated → admin reviews and edits in the dashboard
5. Client sees progress → approved spec and task status on their link
6. GitHub sync → tasks become GitHub issues, status polls back automatically

Architecture

Monorepo using pnpm workspace:

• apps/user-application — TanStack Start frontend (admin dashboard + public client pages)
• apps/data-service — Hono API backend (Claude AI integration, GitHub API)
• packages/data-ops — Shared DB layer (schemas, queries, auth)

Stack: TanStack Start, Hono, Better Auth, Drizzle ORM, Cloudflare Workers, Neon Postgres, Claude API.

Key Entities

• Project — identified by unique slug for public access
• Message — conversation history (user + assistant) per project
• Spec — generated markdown specification
• Task — individual work items, optionally linked to GitHub issues

Auth Model

• Admin pages require Better Auth login
• Client pages are fully public (slug = access)

Setup

pnpm run setup

Installs all dependencies and builds data-ops package.

Development

pnpm run dev:user-application  # TanStack Start app (port 3000)
pnpm run dev:data-service      # Hono backend service (port 8788)

Database Migrations

From packages/data-ops/ directory:

pnpm run drizzle:dev:generate  # Generate migration
pnpm run drizzle:dev:migrate   # Apply to database

Replace dev with staging or production.

Environment Variables

• packages/data-ops/ — .env.dev, .env.staging, .env.production (see .env.example)
• apps/user-application/ — .env files per Vite mode
• apps/data-service/ — .dev.vars (local), Cloudflare dashboard (remote)

Testing

pnpm run test              # run all tests
pnpm run test:watch        # watch mode
pnpm run test:coverage     # with coverage report

Uses Vitest with workspace projects. Each package can also run tests independently via pnpm --filter <package> test.

Deployment

pnpm run deploy:staging:user-application
pnpm run deploy:staging:data-service
pnpm run deploy:production:user-application
pnpm run deploy:production:data-service

Secrets sync: bash apps/{app}/sync-secrets.sh {env}

Cloudflare Account Override

To deploy to a different CF account, copy .env.example to .env and fill in CLOUDFLARE_ACCOUNT_ID + CLOUDFLARE_API_TOKEN.

Package Docs

Each package has its own AGENTS.md with detailed structure, patterns, and workflows (CLAUDE.md symlinks to AGENTS.md).

Planning

• Implementation plan: plans/ai-scoping-tool.md
• Planning skills (brainstormer) are pre-configured via .claude/settings.json