All use cases

Money

Audit card spend against benefits you already pay for

You pay for Sapphire and Platinum perks, then leave them on the table.

Share

What hurt

Chuck Cook did not ask whether his cards were “good.” He asked whether dining, Amazon, and travel were landing on the right card, and whether monthly or quarterly credits (travel, DoorDash, Resy, Lululemon, Amazon 5%) were disappearing unused. Refreshing Chase and Amex fine print by hand every month is the job that never sticks.

What to install or create

Three named bots. Captain is the only front door and the only one that talks to you. Spendy reads posted transactions through Plaid (card names and dollar amounts, no account numbers). Researchy never sees the accounts; he refreshes the current Chase and Amex benefits pages with live search so the scorecard is not a 2023 blog post. Captain waits for both, then writes a recap: wrong card, left on the table only if another card you already hold would have paid more, and what you already did right. Run it monthly: refresh offers, then spend, then recap. They do not post and do not apply for a new card unless you say so.

Bots

  • Captain

    Front door. Asks for wrong-card, left-on-table, and already-right. Writes the recap.

  • Spendy

    Reads posted transactions via Plaid. Card names and dollars only.

  • Researchy

    Refreshes current issuer benefit pages. Never sees accounts.

Plugins

  • Web search

    Required

Files

  • Optional CSV export of last 30 days of spend if you prefer not to connect Plaid on day one.

Routines

  • Monthly after you trust one manual recap: refresh offers, then spend, then Captain’s recap. Still no apply or buy.

The prompt

First brief for Captain (no invented system prompt)

Reworded from Chuck Cook’s published architecture in the cited post. Not a full system prompt he pasted.

Last line is our approval fence: Never send, post, spend, or buy without my explicit yes.

I want three things only: charges on the wrong card, benefits I left on the table, and what I already did right. Use card names and dollar amounts. No account numbers. Spendy reads posted transactions. Researchy refreshes current Chase and Amex benefit pages and never sees the accounts. Wait for both, then write the recap. Do not post. Do not apply for a new card unless I say so. I still decide.

Never send, post, spend, or buy without my explicit yes.

First run

  1. 1Name Captain as the only bot that talks to you.
  2. 2Give Spendy read access to posted transactions (Plaid or a CSV of the last 30 days). Card names and dollars only.
  3. 3Give Researchy web/live search for current Chase and Amex benefit pages. Ban account access for Researchy.
  4. 4Paste the Captain brief. Ask for wrong-card, left-on-table, and already-right.
  5. 5Review the recap. You approve any card swap list. Do not let them apply or buy.

Approval fence

Do not apply for a card, open a purchase, move money, or share account numbers. Recap only. You decide the swap list.

What to tweak

Steal the split of jobs, not Chuck’s card stack. One talker. One spend reader. One benefits researcher who never sees balances. Export or recap with card names and dollars only. Keep apply-for-a-card and any purchase behind your yes.

Source

Chuck Cook @ChuckCook · Original post

Not affiliated with xAI. We did not write this prompt. Third-party template terms still apply.

Browse templates

Browse team blueprints