← Back to library

Supervise a browsing agent through cancel flow and support chat

ChrisGPT describes watching a Codex-powered browser agent cancel an Amazon subscription, open chat with an associate, and argue for a refund in real time. We treat it as a supervised pattern: you watch, you take over at identity checks, and you decide whether any refund offer is acceptable.

  • OutcomeSupervise a browsing agent while it ends a subscription in the vendor UI and chats with support about money back, taking over at login and payment steps.
  • SetupMinutes
  • LevelIntermediate
  • CreditChrisGPT (@ChrisGPT)
View original
Works with
ChatGPT agentClaudeClaude CodeChatGPT agent

Published Oct 23, 2026Updated Oct 23, 2026AI dry run on sample input: passed Oct 23, 2026Source verified Oct 23, 2026

When should you use it?

  • A subscription UI is confusing and chat support is the real path to a refund
  • You can sit and watch the session
  • You already have cancel-by-letter playbooks for locked-in contracts

Not for: Handing over passwords for the agent to type; Disputing charges after a refund window without checking the terms

What do you need, and what do you get?

You need

  • The vendor account open on your machine
  • What you want canceled and why you believe a refund is fair
  • Screenshots or order numbers

You get

  • Cancellation confirmation
  • Chat transcript with support
  • Refund decision you accepted or rejected

How does it work, step by step?

  1. State the askTell the agent the plan name, the refund rationale and that you will take over at login.
  2. Watch the clicksStay on the session while it navigates settings and help.
  3. Take over for identityEnter passwords, MFA and payment confirmations yourself.
  4. Approve the closeRead the final offer before accepting any refund or credit.

The prompt

Our own prompt, inspired by ChrisGPT (@ChrisGPT) and written to work in ChatGPT agent, Claude, Claude Code and others. Values in lime are placeholders.

Fill in the blanks (optional; the copied prompt uses your values)

Prompt3 variables
Cancel {subscription} on {vendor} and request a refund because {reason}.
Rules: I will take over at every login, MFA and payment screen. Do not accept a credit if I asked for a refund unless I say so in chat. Narrate each step. If support asks for a phone call, stop and hand control to me.
When done, paste confirmation numbers and the final offer.

Inspired by ChrisGPT (@ChrisGPT): Browser agent cancels Amazon subscription and chats for a refund. The prompt above is written by playbooks.bot in our own words; nothing from the original is quoted.

What can go wrong?

  • Agents can accept a credit when you wanted cash back. Say which outcomes are OK up front.
  • Distinct from drafting a cancellation letter: this is live UI and chat.

Our take

Live demos of agents beating support chat are entertaining; the durable lesson is supervision. Cancellation and refund chats mix UI navigation with negotiation, so the human should own credentials and the accept button. We do not promise refund outcomes. In our dry run on a fictional SaaS cancel flow, the agent reached chat, stated the billing-period argument from our brief, and paused when asked for the last four card digits. Say up front whether a credit is acceptable or only a cash refund. Take over the second chat asks for identity. Save the transcript yourself; do not trust the agent to archive it. If support pushes a phone call, stop the session. This is supervised navigation, not a guarantee you will get money back. Say up front whether a credit is acceptable or only a cash refund. Take over the second chat asks for identity. Save the transcript yourself; do not trust the agent to archive it. If support pushes a phone call, stop the session. This is supervised navigation, not a guarantee you will get money back.

Where the time goes back: You avoid hunting through nested cancel menus while still keeping control of identity and the final yes.

FAQ

Is this the same as cancel-subscription-fee-waiver?
No. That playbook drafts a written waiver request. This one supervises a browser agent inside the vendor's site and chat.
What if chat asks for a phone call?
Take over. Do not let the agent speak on a voice line for you.

Source and credit

Inspired by ChrisGPT (@ChrisGPT). We describe the idea in our own words, link to the original, and wrote the prompt ourselves; nothing is quoted.

Inspired by
Browser agent cancels Amazon subscription and chats for a refund
Creator
ChrisGPT (@ChrisGPT)
Published on
X
Original date
Apr 29, 2026
License
Not openly licensed (the poster keeps all rights), so we quote nothing and link to the original
Originally for
Claude Code
AI dry run
Oct 23, 2026: passed. Dry-run tested by our AI test agent on sample input. PASS: Supervised; identity handoff.
AI dry run on sample data; not run with live connected accounts. The playbooks.bot test agent filled the prompt with synthetic inputs, produced the output and checked it against the page's stated outputs and guardrails.
Verified
Oct 23, 2026: source link resolves to the original, the credited creator published it, our excerpt or prompt matches it, and the license allows our use.

Are you the creator? Claim or correct this page.

More in Money & Admin

All 82 Money & Admin playbooks →

Related playbooks

Money & AdminPrompt✓ AI dry run

Audit internet and phone bills before you negotiate

Compare your current internet and phone plans to verified offers at your address, then get a negotiation draft you approve before anyone is contacted.

Credit: Michael Fenech (@Michael_Fenech_)

Under an hour
Money & AdminPrompt✓ AI dry run

Audit subscriptions from statements before you cancel

From 12 months of bank statements and billing emails, get a ranked cancel list with renewal dates, fees and evidence — then approve every cancel yourself.

Credit: Michael Fenech (@Michael_Fenech_)

Minutes
Money & AdminPrompt✓ AI dry run

Practice on errands before high-stakes customer work

Run a weekly low-stakes errand bot with compare-and-wait rules so you learn failure modes before granting customer or money permissions.

Credit: Carnage (@0xCarnagee)

Under an hour
Money & AdminPrompt✓ AI dry run

Recover money from unwanted charges with evidence

Find duplicate charges, post-cancel billings and refund-eligible purchases, prepare evidence-backed requests, and track requested versus actually received refunds after you approve each send.

Credit: Michael Fenech (@Michael_Fenech_)

Under an hour