← Plugin catalog
Finance

Cashback Card

Harsha vardhan Gaddam v2.0.1

Which cards should I use here?

Create a free account to save this plugin.

From the developer

Declared capabilities

  • Web research
  • Reward calculation
  • Local wallet storage

Not specified in the available declarations: limitations, pricing & account requirements.

Onboarding example

Which of my cards earns the most at Costco?
Sources & dates

Developer declarations; not independently tested. Listing last checked Nov 19, 2026. Package observed Nov 19, 2026.

View package source →

Description

From the marketplace listing

Name a store and get the card in your wallet that nets the most on that purchase. Accounts for how the merchant actually codes (Target is a discount store, not a supermarket), spending caps you have partly used, categories you have not activated, what your points are really worth, and foreign transaction or fuel surcharge fees. Verifies rates against issuer terms rather than quoting them from memory. Covers the US, India and the UK.

Language: English · Automatically detected from descriptions.

What changed

Similar plugins

Finding similar plugins… Results will appear here automatically.

Browse plugins →

Keywords

Files & skills

File archives

Plugin package17 files · 157 KBBrowse files →
Skill instructions
find-best-cashback8.44 KB

View saved version →

---
name: find-best-cashback
description: Work out which credit card in someone's wallet earns the most on a purchase at a named store - Shell, Target, Walmart, Costco, Amazon, Swiggy, Tesco, a restaurant, a supermarket. Use when the user names a merchant and asks which card to use, what earns the most cashback or rewards there, how a store is likely to code, why a card did not earn a bonus rate, or which new card would be best for a merchant they shop at often. Covers the US, India and the UK.
---

# Find Best Cashback

Answer one question precisely: **for this purchase, at this merchant, in this country, which card in this wallet nets the most money?**

The hard part is not arithmetic. It is that a card's headline rate frequently does not apply — because the merchant codes as a superstore, because a cap is already used up, because the category was never activated, or because the points are worth less than a cent. This skill exists to catch those cases.

## The one rule that matters

**Never state a reward rate from memory.** Every rate you quote comes from one of three places, and you must say which:

| Source | How to label it |
| --- | --- |
| Live issuer page you just read | "verified today at [link]" |
| Bundled catalog in `data/` | "catalog snapshot from *as_of*, worth confirming" |
| The user told you | "per the terms you gave me" |

If a number decides the recommendation and you could not verify it, say so in the answer. Do not launder a stale rate into a confident sentence.

## Step 1 - Get the four inputs

You need: **merchant**, **country**, **cards held**, and — only when it changes the answer — **amount** and **channel**.

- Country: infer from context if the signal is strong (₹ or a named Indian merchant → India). Otherwise ask.
- Cards: check for a saved wallet first (see *Wallet* below). If none exists, ask for **card product names only**.
- Amount: ask only when a cap, a fixed-value offer, a minimum spend, or a fee could change the winner. For "which card at Shell?" a representative amount is fine — say which you assumed.
- Channel: ask when the merchant codes differently by channel and the difference flips the winner. `--explain-merchant` tells you when that is true.

Ask at most one consolidated question. Never request a card number, CVV, expiry, PIN, login, or statement. If the user pastes one, tell them to rotate nothing but redact it, and continue from the product name alone.

## Step 2 - Resolve how the merchant codes

```bash
python scripts/rank_cards.py --explain-merchant walmart --country US
```

This returns the spend categories, likely MCCs, a confidence level, and the known traps for that merchant. Read the caveats — they are the substance of a good answer.

If the merchant is not in the table, research it or ask the user how similar purchases have posted before. Do not guess a category from the brand name alone.

Merchant category codes are set by the merchant's acquiring bank, not by the card issuer. A franchise location, a separate fuel forecourt, an in-app payment, or a third-party delivery order can all code differently from the flagship store. Say "likely to code as", never "codes as".

## Step 3 - Score the wallet

Build a request and run the engine. Do not do this arithmetic in your head — caps, slabs, point values and fees interact in ways that are easy to get wrong.

```bash
python scripts/rank_cards.py request.json
```

```json
{
  "purchase": {
    "merchant": "shell", "channel": "pump", "amount": 60,
    "country": "US", "currency": "USD"
  },
  "wallet": [
    "amex-bcp",
    {"id": "chase-freedom-flex", "activated": true,
     "cap_used": {"chase-rotating": 1200}},
    {"id": "discover-it", "offers": [
      {"label": "10% back up to $15", "type": "percent", "value": 10, "cap": 15}]}
  ]
}
```

Run `--help` for the full schema and `--list-cards US` for the available ids. A card the user holds that is not in the ca

Preview of the first 4,000 characters. Read full instructions →

Referenced files: 9 — file references are metadata, not package contents.

manage-wallet3.04 KB

View saved version →

---
name: manage-wallet
description: Save, update, list or clear the set of credit cards a user holds, so card recommendations do not have to ask every time. Use when the user says which cards they have, asks to remember or forget a card, wants to see their saved cards, or reports that a spending cap or rotating category has changed.
---

# Manage Wallet

Keep a small local file of which cards the user holds, so `find-best-cashback` can answer "which card at Shell?" without re-asking.

## What goes in the file

Product identities and reward-relevant state. Nothing else.

Store: card ids, the user's country, chosen categories, activation status, and cap usage they have told you about.

**Never store**: card numbers, CVV, expiry dates, PINs, names as they appear on the card, credit limits, balances, statement PDFs, or login credentials. This file is plain text on disk. If the user offers a card number, tell them you do not need it and that the product name is enough.

## Location

`~/.cashback-wallet.json` — expand `~` yourself; the engine does not.

## Format

```json
{
  "version": 1,
  "country": "US",
  "updated": "2026-07-25",
  "cards": [
    {"id": "amex-bcp", "nickname": "the blue one"},
    {"id": "chase-freedom-flex", "activated": true,
     "cap_used": {"chase-rotating": 1200},
     "note": "Q3 rotating category is gas stations"},
    {"id": "boa-customized-cash", "chosen_category": "online_retail"}
  ]
}
```

Card ids come from the catalog:

```bash
python ../find-best-cashback/scripts/rank_cards.py --list-cards US
```

If the user names a card that is not in the catalog, still save it — record what they told you so it is not lost:

```json
{"id": "local-credit-union-visa", "in_catalog": false,
 "user_terms": "2% everywhere, no annual fee, told to me 2026-07-25"}
```

`find-best-cashback` will pass that through as an inline card definition rather than dropping it.

## Reading it

Read the file before asking the user which cards they hold. If it does not exist, ask once and offer to save the answer — do not save without being asked or agreeing first.

## Writing it

Rewrite the whole file; do not append. Always set `updated` to today.

Before overwriting an existing wallet, read it and confirm what changes: adding a card is safe to just do, but removing or replacing the list should be echoed back ("dropping the Discover it, keeping the other three — right?").

## Cap and activation state goes stale

`cap_used` and `activated` are true on the day they were written and drift immediately afterwards. When a saved value is more than a few weeks old, or when the calendar has crossed into a new quarter, treat it as unknown and re-ask rather than computing on it. A confidently wrong cap number produces a confidently wrong recommendation.

Rotating categories reset quarterly. A saved `activated: true` from last quarter means nothing this quarter.

## Clearing

If the user asks you to forget their cards, delete the file rather than emptying it, and confirm afterwards that it is gone.

Referenced files: 1 — file references are metadata, not package contents.

Package details

Publisher declarations from the archived package. These are separate from our research and the live service's terms.

Package license
MIT
Package author
Harsha Vardhan Gaddam
Keywords
See keywords

Declared capabilities

  • Web research
  • Reward calculation
  • Local wallet storage

Package observed Nov 19, 2026.

Technical details
First seen
Sep 30, 2026 · 22:02 UTC
Last seen
Nov 19, 2026 · 18:00 UTC
Collection status
Collected

plugins_6a657351f3108191a509c5efe534fc5f

Download plugin data (JSON)

Plugin details: served from cache. Prepared . Refreshed when plugin data changes or after 6 hours. Your account and Follow status are live.