What your coding agent forgets before you deploy.

The short answer

Before you deploy, ask your coding agent to go through this list. Most failures on a first deploy come from a handful of things that work on your machine and not on a server. The usual ones: an out-of-date lock file, no start script, a hardcoded port, localhost addresses, missing variables, tables that were never created, files saved to disk and timers inside the app.

The checklist

1. The lock file matches package.json

   Run npm install and commit package-lock.json. See the lock file error.

2. There is a start script, and the build passes

   npm run build should pass on your machine, and package.json needs a start script for a server app. See Express and Next.js.

3. The app reads PORT and listens on 0.0.0.0

   Never a fixed port, never 127.0.0.1. See never started listening.

4. There is no localhost in the code

   Search for localhost and 127.0.0.1. Use relative paths or an environment variable for the API address.

5. Every variable the code reads is set

   Names in your .env must exist online too. See environment variables.

6. The database comes from DATABASE_URL

   Not a local address and not SQLite. See add a database and SQLite resets.

7. Migrations create the tables

   Use a tool Antideploy detects, or run your migration before the app listens. See relation does not exist.

8. Uploads go to a bucket

   Nothing important is saved to the app's disk. See file uploads.

9. Timers are scheduled jobs

   Replace node-cron and setInterval with a route and a scheduled job. See scheduled jobs.

10. Secrets are not in the code

    No keys in files you commit, and none in NEXT_PUBLIC_ or VITE_ variables. Keep .env out of Git.

For Python projects

• Every import is in requirements.txt, including the server: gunicorn for Flask and Django, uvicorn for FastAPI
• The Python version is pinned in .python-version if your packages need a specific one
• Django lists your address in ALLOWED_HOSTS