A developer utility web app: paste JavaScript/PHP code (or link a GitHub file), and get AI-generated structural improvement notes, security vulnerability detection, and refactoring suggestions — with results exportable as Markdown or PDF.

• Frontend: React + Vite, CodeMirror 6 (syntax-highlighted editor)
• Backend: Node.js + Express, @anthropic-ai/sdk for the LLM call
• Database: SQLite (via better-sqlite3) — zero-config, single file
• Export: Markdown (raw text) and PDF (via pdfkit)

ai-code-review-assistant/
├── backend/                 Express API
│   ├── server.js            Entry point
│   ├── src/
│   │   ├── config/db.js     SQLite connection + schema bootstrap
│   │   ├── db/schema.sql    Table definitions (mirrors /database/schema.sql)
│   │   ├── routes/          analyze.js, history.js, reports.js
│   │   ├── services/        llmService, githubService, exportService, promptTemplates
│   │   └── middleware/      errorHandler.js
│   └── .env.example
├── frontend/                React app
│   └── src/
│       ├── components/      CodeEditor, AnalysisReport, FindingCard, ExportButtons, etc.
│       ├── api/client.js    Axios wrapper for the backend API
│       └── App.jsx
├── database/
│   ├── schema.sql           Canonical schema reference
│   └── seed.sql             Optional sample data
└── README.md

1. Input: paste a JS/PHP snippet into the CodeMirror editor, or paste a GitHub file URL (https://github.com/{owner}/{repo}/blob/{branch}/{path}) — the backend fetches the raw file.
2. Analyze: clicking "Analyze Code" sends the snippet (with line numbers) to the backend, which calls the Anthropic API using a system prompt tuned for security vulnerability detection (SQLi, XSS, insecure deserialization, hardcoded secrets, etc. — mapped to CWE IDs where applicable), structural/code-quality issues, and refactor suggestions. The model returns strict JSON, which is persisted as structured findings rows.
3. Review: results are shown as a scored summary (structural + security scores out of 100) plus a severity-sorted list of findings, each with a description, line reference, and a concrete suggested fix.
4. Export: "Export Markdown" / "Export PDF" generate a downloadable report file containing the summary, scores, all findings, and the original code.
5. History: past submissions/analyses are listed in a sidebar and can be reopened at any time.

cd backend
cp .env.example .env
# edit .env and set ANTHROPIC_API_KEY (get one at https://console.anthropic.com/)
npm install
npm run db:init      # creates the SQLite file + applies schema.sql
npm run dev           # starts on http://localhost:4000

cd frontend
cp .env.example .env   # optional — defaults to same-origin /api via the Vite dev proxy
npm install
npm run dev             # starts on http://localhost:5173

Open http://localhost:5173 in your browser.

sqlite3 backend/data/code_review.db < database/seed.sql

• The ANTHROPIC_API_KEY must be supplied by whoever runs the app — it is never hardcoded or shipped with this project.
• express-rate-limit caps /api/analyze at 10 requests/minute per client to protect against runaway LLM API costs; adjust in backend/server.js as needed.
• The GitHub URL fetch supports public files via raw.githubusercontent.com; set GITHUB_TOKEN in backend/.env to raise rate limits or access private repos you have access to.
• To add more languages, extend the language CHECK constraint in schema.sql, add a CodeMirror language extension on the frontend, and update the prompt in promptTemplates.js.