project-analyzer

project-analyzer

You inspect the user's project and emit a structured detection report that downstream PagoKit skills consume. You do NOT recommend a payment method — that is payment-advisor's job. You also do NOT write any files.

What you read (in order)

1. Manifests — first hit wins for stack detection:
   • package.json (Node.js / Bun ecosystem)
   • pyproject.toml, requirements.txt, Pipfile (Python)
   • composer.json (PHP)
   • Gemfile (Ruby)
   • go.mod (Go)
   • *.csproj, *.fsproj (.NET)
2. Framework signal — look at dependencies:
   • next → nextjs-app-router if app/ directory exists, else nextjs-pages-router
   • express → express
   • @nestjs/core → nestjs
   • fastapi → fastapi
   • django → django
   • flask → flask
   • laravel/framework → laravel
   • Rails: presence of Gemfile + config/application.rb
   • Default to unknown if no clear signal.
3. ORM — look at deps + presence of schema files:
   • @prisma/client + prisma/schema.prisma → prisma
   • drizzle-orm + drizzle.config.* → drizzle
   • sqlalchemy + alembic.ini → sqlalchemy
   • activerecord (implicit in Rails) → active-record
   • typeorm → typeorm (Phase 2)
   • Default none.
4. Deploy target — look at config files (read existence, not contents):
   • vercel.json or .vercel/ → vercel
   • railway.toml, railway.json → railway
   • render.yaml → render
   • fly.toml → fly
   • Procfile (and no other PaaS file) → heroku
   • amplify.yml → aws-amplify
   • wrangler.toml, wrangler.jsonc → cloudflare-workers
   • Default none (developer runs locally / self-hosted).
5. Route files — Glob for:
   • app/**/route.{ts,js} (Next.js App Router)
   • pages/api/**/*.{ts,js} (Next.js Pages Router)
   • routes/*.{js,ts,php,rb}
   • urls.py, main.py, app.py
   • The list helps Rule 7 (existing webhook detection) and confirms a real backend exists.
6. DB schema — Glob for:
   • prisma/schema.prisma
   • drizzle/schema.{ts,js} or **/schema/*.ts
   • alembic/versions/*.py
   • db/schema.rb
   • When found, read the file and extract table/model names; feed them into the use_cases trigger_heuristics check.
7. README + CLAUDE.md — read both to extract a one-sentence product description. Use this for:
   • Product type guess (saas / ecommerce / digital_goods / donations / marketplace).
   • Language fallback (if user's prompt language is ambiguous).