---
title: "A self-improvement loop for agents that build sign-up · Proxy Mail QA"
description: "How a coding agent improves its own sign-up, login and reset work: it tests with a real inbox, reads what the email did, fixes, runs again and keeps what it learnt."
url: https://proxymail.qa/solve/agent-self-improvement-loop
---

1. [Home](https://proxymail.qa/)
2. [What we solve](https://proxymail.qa/solve)

# A self-improvement loop for agents that build sign-up

A coding agent gets better at a task when it can see whether its own work did what it was meant to, and try again. For most code, tests and a browser are enough. Sign-up, login codes and password resets end in an email, which the agent can't see, so the loop stops there. Proxy Mail QA is the tool that closes it.

Updated 12 November 2026

## What a self-improvement loop needs

An agent improves its own work when four things are true:

1. **It can act**: write the code, run the app, fill in the form.
2. **It can observe the result**, not only its own output. “The code compiled” isn’t “the user got the email”.
3. **It can compare** this attempt with the last one, so it knows whether a change helped.
4. **It can keep what it learnt**, so the next attempt, or the next session, starts further on.

Agents already do the first well. The other three depend on tools.

## Where the loop breaks for sign-up

- **The result is in an inbox.** The agent submits the form and sees “Check your email”. Whether anything was sent, where its link points and whether the code works are all out of its reach.
- **So it reports success too early.** “Sign-up is implemented” usually means the form submits. The agent has no evidence either way, and neither do you.
- **A made-up address can’t finish the flow**, and your own address works once.
- **Nothing carries over.** Without a record of the last run, the agent can’t tell a fix from a coincidence.

## The loop with Proxy Mail QA

The agent does every step. Proxy Mail QA is what it checks with.

| Step    | The agent                                                      | The tool it uses                          |
| ------- | -------------------------------------------------------------- | ----------------------------------------- |
| Act     | Builds or changes the flow, opens the sign-up page             | Its own editor and browser                |
|         | Gets an address, a persona and a password for the form         | form\_values or create\_address           |
| Observe | Waits for the code or link after submitting                    | wait\_for\_code, wait\_for\_link          |
|         | Asks why nothing came                                          | check\_delivery                           |
|         | Checks the email’s sending, link, expiry and size              | check\_email, count\_emails               |
| Compare | Records the session and reads what changed since the last ones | start\_run, finish\_run                   |
| Keep    | Saves what it learnt about the form, with your agreement       | save\_custom\_field, save\_signup\_screen |

A first pass often goes like this:

1. The agent adds email verification and signs up with a new test address.
2. The wait finds nothing. The delivery check says the email provider only sends to verified addresses. The agent sees its work isn’t done, and why.
3. It fixes the provider’s setup and runs again. The email arrives in four seconds, but the check says its link points at `localhost`.
4. It fixes the site URL and runs again. The link opens staging, the account is verified, and one email came.
5. The run’s report shows all three attempts. The agent tells you what it changed and shows the evidence.

Without the inbox, the agent would have stopped after step 1 and said it was finished.

## Three things the agent improves

**Its work, in this session.** Each check answers with what failed and why, so the agent’s next change is aimed at something. This is the loop above.

**The project’s setup, for every later session.** When the agent meets a field it can’t fill, such as company size, it can ask you and save the answer as a custom field. When it finds the form runs over several screens, it can save them. The next agent, or the next person with the Chrome extension, gets the form right first time. The agent changes these only when you ask or agree.

**Its own instructions, for itself.** Ask the agent to end a session by writing what it learnt into the file it reads at the start of each one (`CLAUDE.md`, `AGENTS.md`, or a skill): “staging’s email provider needs the domain verified”, “the reset link expires in 10 minutes, so open it straight away”. This is the part closest to the word recursive: the agent uses the evidence from one run to change how it will work in the next.

Prompt

```text
After each change to sign-up, login or password reset, run the auth regression pack on
staging with Proxy Mail QA. If the report shows a new failure, fix it and run again.
When it passes, add one line to AGENTS.md for anything you learnt that the next session
should know, and show me the report.

```

## Knowing whether it got better

A loop needs a measure. The [auth regression pack](https://proxymail.qa/skills) makes the same checks every time: sign-up, a used link opened twice, a replaced reset link, an invitation. Its report ends with what changed since the last runs: slower emails, extra emails, new or intermittent failures, newly failing checks, a new sender or link site.

That gives the agent a before and an after, and gives you something to read that the agent didn’t write about itself.

When a person reports a bug instead, the same loop starts from their report: see [Triage sign-up and login bug reports with an agent](https://proxymail.qa/solve/bug-report-triage).

## What stays with you

- **Proxy Mail QA changes no model and no code.** The improving is the agent’s, and so are the mistakes. Read the report, not only the summary.
- **Test builds only.** The loop runs on development, staging and preview sites, never on your live site or real users’ accounts.
- **Lasting changes need your yes.** Custom fields, sites and sign-up screens affect everyone in the project.
- **The report holds no email text.** It keeps addresses, timings, checks and each link’s site, so it can go on a pull request.
- **Email only.** Codes sent by SMS aren’t covered.

## Automating this?

Run the loop after every change: the auth regression pack makes the same checks each time, and its report says what changed since the last runs.

[Node.js client](https://proxymail.qa/docs/node) [Python](https://proxymail.qa/docs/python) [Set up your coding agent](https://proxymail.qa/docs/api#agent-setup)

## Try it on your next sign-up test

Free plan, no credit card. Create an address and watch the code or link arrive.

[Create a free account](https://app.proxymail.qa/?mode=signup&from=site-solve-agent-self-improvement-loop)

## Questions

### Is this recursive self-improvement (RSI)?

Recursive self-improvement usually means an AI system that improves itself, and then uses the improved version to improve again. Proxy Mail QA doesn't do that and doesn't change any model. It is a tool an agent uses inside its own loop: the agent improves its work, its setup and its instructions, and Proxy Mail QA tells it what the emails did.

### What does the agent improve?

Three things, in order of how long they last: the code it just wrote, by fixing what the check found; the project's setup, by saving what it learnt about the sign-up form; and its own instructions, by writing down the lesson so the next session starts with it.

### Does Proxy Mail QA fix anything itself?

No. It creates the address, hands back the code or link, checks the email and keeps the report. It never opens a link, fills in a form or changes your code. The agent does those with its own tools.

### Does the agent need a person in the loop?

Not for building, testing and fixing on a development or staging site. It does for changes that outlast the session: the MCP server tells agents to change a project's custom fields, sites or sign-up screens only when the person asks or agrees.

### Which agents can do this?

Any agent that can use an MCP server or a REST API and drive a browser or a simulator: Claude Code, Codex, Cursor, GitHub Copilot's coding agent, Gemini in Antigravity and others.

## Related guides

- [Test sign-up and email verification](https://proxymail.qa/solve/email-verification-testing): A fresh, private address for every sign-up test, with the verification link or code ready the moment it lands.
- [Test OTP and one-time login codes](https://proxymail.qa/solve/otp-code-testing): Codes pulled out of the email with a Copy button and a countdown when the email says how long they last.
- [Test magic links and passwordless sign-in](https://proxymail.qa/solve/magic-link-testing): Single-use links that are never opened or prefetched on our side, so they still work when you click them.
- [Test password reset emails](https://proxymail.qa/solve/password-reset-testing): Request, receive and use reset links, and keep the new password saved next to the test account.
- [Stop sharing one Gmail inbox for QA](https://proxymail.qa/solve/shared-test-inbox): Private, disposable addresses for every tester instead of one shared mailbox and a pile of plus-addresses.
- [Stop using Gmail for test codes](https://proxymail.qa/solve/gmail-test-inbox): Why reading codes from Gmail over its API or IMAP keeps breaking your tests, and how one address per test removes the inbox from the problem.
- [Email limits in auth providers while you test](https://proxymail.qa/solve/auth-email-limits): Supabase's 2 an hour, Firebase's 5 email links a day, Auth0's 10 a minute and Clerk's 100 a month, and how to send your own.
- [Let test addresses through a disposable-email block](https://proxymail.qa/solve/disposable-email-blocks): Keep the block for real users and let your test domains through in development and staging, with one setting you can't ship by mistake.
- [CAPTCHA, Google and Apple sign-in in tests](https://proxymail.qa/solve/captcha-and-social-sign-in-tests): Published test keys for reCAPTCHA, hCaptcha and Turnstile, and what to test instead of automating Google and Apple sign-in.
- [Take QA notes by voice with Wispr Flow](https://proxymail.qa/solve/voice-notes-wispr-flow): Dictate what each test account is for, its state and what you found, straight into its notes, in seconds.
- [Test phone-first sign-up without real SMS](https://proxymail.qa/solve/phone-first-sign-up-testing): Test numbers with fixed codes, a test-mode code kept out of release builds, and why virtual numbers usually fail.
- [Test sign-up for people in China](https://proxymail.qa/solve/china-sign-up-testing): Real-name rules, a 姓名 field and province menus, a safe test code, and 身份证 numbers that can't be anyone's.
- [Test sign-up for people in Japan](https://proxymail.qa/solve/japan-sign-up-testing): Furigana fields, 全角 and 半角, postcode lookups and LINE sign-in, with test accounts that have Japanese names.
- [Made-up test accounts, not masked production data](https://proxymail.qa/solve/synthetic-test-data): A masked copy of production still holds real people. Test sign-up with made-up accounts that no one owns instead.
- [Test sign-up for people in India](https://proxymail.qa/solve/india-sign-up-testing): Test accounts with one name, PIN codes that agree with the city, phone-first sign-up, Hindi labels, ID fields and card mandates.
- [Test sign-up for people in Pakistan](https://proxymail.qa/solve/pakistan-sign-up-testing): Test accounts with Pakistani names in both scripts, CNIC fields that never hold a real number, and phone-first sign-up.
- [Fill phone and ID fields safely](https://proxymail.qa/solve/phone-id-and-tax-number-fields): Numbers set aside for fiction, your app's own test numbers, ID values that never match a real person, and KYC sandbox values, filled the same way by agents and tests.
- [Verification link points to localhost:3000](https://proxymail.qa/solve/localhost-links): Why emails from staging link to localhost, how to fix it in Supabase, Firebase and other stacks, and how we flag it.
- [Common Supabase sign-up and login problems](https://proxymail.qa/solve/supabase-auth-problems): Emails that never arrive, links that expire or sign nobody in, and sign-ups that fail, with the fix for each.
- [Common Firebase sign-up and login problems](https://proxymail.qa/solve/firebase-auth-problems): Missing verification and reset emails, verified users who still look unverified, failing links and sign-up limits, with the fix for each.
- [Sign-up and login problems in vibe-coded apps](https://proxymail.qa/solve/vibe-coded-login-problems): The login bugs AI-built apps usually ship with: emails only you receive, links to the preview, missing pages, and what to ask the builder.
- [Check that one user can't see another's data](https://proxymail.qa/solve/vibe-coded-app-data-leaks): A curl with the anon key for tables anyone can read, and two signed-up accounts for policies that let every user read every row.
- [Triage sign-up and login bug reports with an agent](https://proxymail.qa/solve/bug-report-triage): An agent signs up with a new test address, watches the inbox, and answers replicated, not replicated or inconclusive, with the evidence.
- [Test sign-up emails in Playwright](https://proxymail.qa/solve/playwright-email-tests): A complete Playwright test for sign-up codes and links, from a fresh inbox to the report on the pull request.
- [Test sign-ups over several screens](https://proxymail.qa/solve/multi-step-sign-up-testing): One test account across every screen of the sign-up, with your form's own questions filled the same way each time.
- [Test sign-up emails in Cypress](https://proxymail.qa/solve/cypress-email-tests): A complete Cypress test for sign-up links and codes, with the email waits running as Cypress tasks.
- [Test sign-up emails with pytest](https://proxymail.qa/solve/pytest-email-tests): A complete pytest test for sign-up codes and links, with Playwright for Python or Selenium and a one-file helper.
- [Test coupon codes and welcome offers](https://proxymail.qa/solve/coupon-code-testing): The coupon with its discount, end date and arrival time, what checkout did with it, and a failure when an expired or reused one is accepted.
- [Test unsubscribe links and one-click unsubscribe](https://proxymail.qa/solve/unsubscribe-testing): The unsubscribe headers and link checked on every marketing email, and a watch that flags any marketing after unsubscribing, with no time limit.
- [Test that a welcome offer can't be claimed twice](https://proxymail.qa/solve/welcome-offer-abuse): Versions of one address that reach the same inbox, and a check that fails when the same person got the offer more than once.
- [AI sign-up page review](https://proxymail.qa/solve/ai-sign-up-page-review): Let an agent check the sign-up through to the email: delivery, the link, how it was sent, and that only one arrives.
- [AI registration page review](https://proxymail.qa/solve/ai-registration-page-review): An agent tries every field with values made for it, then finishes the registration through the email.
- [Review a sign-up page by hand](https://proxymail.qa/solve/how-to-review-my-sign-up-page): A checklist for your browser and phone, in the order that finds the most problems.
- [Review a simple sign-up form](https://proxymail.qa/solve/review-a-simple-sign-up-form): 12 checks for a form with an email, a password and perhaps a name.
- [Review a multi-page sign-up form](https://proxymail.qa/solve/review-a-multi-page-sign-up-form): One test account across every screen, and the verification email wherever it comes.
- [Review a financial sign-up form](https://proxymail.qa/solve/review-a-financial-website-sign-up-form): The emails, age and country rules and awkward values in a financial sign-up, and what needs other tools.
- [An email address for testing registration](https://proxymail.qa/solve/email-for-checking-my-registration-form): A new, real address for every registration try, with the code or link ready to use.
- [An email address for testing sign-in](https://proxymail.qa/solve/email-for-checking-my-sign-in-form): Test accounts with saved passwords and an inbox for codes, magic links and resets.
- [How long should a password be?](https://proxymail.qa/solve/how-long-should-my-registration-form-password-be): NIST's numbers for password length and rules, and test passwords to check yours.
- [No error for a password that breaks the rules](https://proxymail.qa/solve/why-is-my-registration-form-not-showing-password-errors): The eight usual causes of a silent password field, and how to tell which one it is.
- [Help me check my website](https://proxymail.qa/solve/help-me-check-my-website): A whole-site check in an afternoon, in the order that finds the costliest problems first.
- [Tools to build and check a website](https://proxymail.qa/solve/tools-to-help-build-and-check-my-website): A toolkit by area, most of it free and none of it ours, except for account emails.
- [What tools do I need for a website?](https://proxymail.qa/solve/what-tools-do-i-need-when-building-a-website): The full list, then which parts a simple site can skip.

[Built your app with an AI tool? Pick yours](https://proxymail.qa/built-with)

Claude Code, Codex, Cursor, GitHub Copilot, Gemini and Antigravity are products of their owners. Proxy Mail QA is independent and isn’t affiliated with or endorsed by them.
