What the agent does
A public launch puts the sign-up and login of the production site in front of bots as well as people. Review how authentication works end to end and fix the gaps attackers try first, without changing how the flows look to a real user.
1. Map the auth flows
List sign-up, login, logout, password reset, email change, social or magic-link sign-in, and session handling, with the files for each. Say whether auth is a library or provider (Auth.js, Clerk, Supabase, Devise, Laravel Breeze...) or custom code: with a provider, the job is configuration, not rewriting it.
2. Passwords and sessions
- Passwords hashed with a slow algorithm (argon2id, bcrypt or scrypt), never stored or logged in plain text.
- Session cookies
HttpOnly,Secure,SameSite=Lax(orStrict), with a sensible lifetime; sessions rotated on login and invalidated on logout and password change. - State-changing requests protected against CSRF (the framework's built-in protection, or SameSite plus an origin check).
3. Abuse
Rate limit login, sign-up, password reset and verification emails per IP and per account, with a clear 429. Make login and reset answers the same whether or not the email exists, so accounts cannot be enumerated. If sign-up is open to bots, propose a light challenge (Turnstile, hCaptcha) for me to decide on.
4. Reset and verification tokens
Tokens must be random, single-use, short-lived (an hour or less for resets) and stored hashed. A used or expired token gives a clear message and a way to request a new one.
Done means
- Passwords use a slow hash and never appear in logs.
- Session cookies are HttpOnly, Secure and SameSite, and sessions rotate on login.
- Login, sign-up and reset are rate limited and do not reveal whether an email exists.
- Reset and verification tokens are random, single-use, short-lived and stored hashed.
- The sign-up and login flows look and work the same for real users.
Questions
I use Clerk, Auth0 or Supabase. Is this still useful?
Yes, but the agent works on configuration: session lifetime, rate limits, email verification and how your app uses the session, rather than rewriting what the provider does well.
Why make login answers the same for unknown emails?
Different answers let anyone check which emails have accounts, which helps targeted phishing and credential stuffing.
Do I need a captcha?
Not always. Rate limits stop most abuse. If bots still create accounts, a light challenge such as Turnstile is the next step, and the prompt asks you before adding one.
Does this prompt add links to Aback Launch to my site?
No. Nothing in the prompt mentions Aback Launch or asks your agent to link to us. The only exception is an optional badge step, which stays off unless you tick it. Some prompt libraries slip their own link into your codebase by default; we think that should always be your choice.
Which coding agents does it work with?
Claude Code, Cursor, Codex, GitHub Copilot, Windsurf, Cline, and any agent that can read your project and edit files. You can copy it as a plain prompt, save it as a Claude Code skill or a Cursor rule, or add it to AGENTS.md.
Is it safe to run on my real codebase?
The prompt makes the agent read first, show a plan and wait for your approval, and it forbids touching secrets, deploying and deleting data. Still review the changes and run your tests before you deploy, as with any change.
Changelog
- v1.0First version.