Card Wallet

Card Wallet

Return a compact wallet audit for a set of cards the user holds.

When To Use

When the user asks to evaluate their card lineup, check for overlap, or optimize their wallet. Trigger phrases: "card-wallet", "wallet audit", "my cards", "do I have overlap", "which cards should I keep".

Input Format

The user provides a comma-separated list of card names:

• card-wallet Chase Sapphire Preferred, Amex Gold, Citi Double Cash

Workflow

1. Parse card list from comma-separated input.
2. Resolve each card — normalize and match to exact variants. If any card is ambiguous, return a numbered choice list for that card and stop.
3. Search — use WebSearch by default per card. If BRAVE_API_KEY is available and curl exists, you may use one Brave Search API call per card instead. Classify results as issuer or secondary by domain.
4. Fetch pages — fetch issuer and approved secondary pages before deciding whether more searches are needed.
5. Pace batch searches — when multiple cards require searches, serialize or batch them gently instead of firing a large burst.
6. Collect — for each card: annual fee, top earning categories, statement credits, key benefits.
7. Analyze — identify overlapping earn categories, uncovered categories, redundant benefits, total fee burden.
8. Confidence — flag uncertain claims.

Step 1: Card Identity Resolution

Common Abbreviations

Only shorthands and ambiguous names need entries here. Cards with full, unambiguous names (e.g., "Chase Marriott Bonvoy Boundless", "Chase United Explorer", "American Express Hilton Honors Aspire") are resolved via search — no table entry needed.

Supported Issuers

American Express, Bank of America, Barclays, Bilt, Capital One, Chase, Citi, Discover, Robinhood, U.S. Bank, Wells Fargo.

Step 2: Search (Per Card)

Use the platform's WebSearch and WebFetch tools by default. If BRAVE_API_KEY is available and the runtime also provides curl, you may use Brave Search API instead for faster and more repeatable search results.

Optional Brave template:

curl -sS "https://api.search.brave.com/res/v1/web/search?q=CARD+NAME+benefits+credits&count=10" \
  -H "X-Subscription-Token: $BRAVE_API_KEY"

Do not assume any search provider tolerates a large burst of parallel searches.

Search Budget Rule

Treat search as scarce and paced. Built-in web search is the default path; if Brave mode is used, it may rate-limit after only a few closely spaced requests.

• Start with the most important cards first.
• Fetch issuer and approved secondary pages before deciding whether more searches are needed.
• When multiple cards require searches, serialize them in small batches or add short waits of about 2 to 5 seconds between bursts.
• If Brave returns 429, wait about 8 to 15 seconds and retry once for the still-missing search.
• If Brave is unavailable, continue with WebSearch + WebFetch.
• If it still fails, continue with the best evidence already gathered and note the limitation in ## 📋 Confidence Notes.

Classify results by domain: issuer pages (use Issuer Domains table below) vs approved secondary sources. Optionally use 1 secondary source (prefer thepointsguy.com) for cross-checking.

Fetch Pages

For each card, fetch the top issuer URL from search results. Optionally fetch 1 secondary URL with WebFetch.

An approved secondary page means a URL whose hostname matches the preferred secondary domains named in this skill. Do not fetch or cite secondary pages from any other domain.

URL Safety Rules

• Prefer WebFetch for page retrieval. Use curl only for the optional Brave Search API calls above, not for arbitrary result URLs.
• Never execute a shell command that interpolates a raw URL taken directly from search results.
• Only fetch URLs when all of the following are true:
  1. scheme is https
  2. hostname matches a supported issuer domain or an approved secondary domain from this skill
  3. the URL is being passed to WebFetch, not inserted into a shell pipeline
• If a result URL fails those checks, skip it and use the next valid result.

Run fetches in parallel. Search snippets alone miss detailed credit and rate info. Combine fetched page content + search snippets + training knowledge.

Issuer Domains (for classifying results, not constraining searches)

Required Output Sections

## 💰 Annual Cost

Total annual fees across all cards, listed per card.

## 📈 Earning Map

Table showing the best card for each major spend category (dining, travel, groceries, gas, streaming, other).

## 🏷️ Credits Stack

Statement credits, cash-back rebates, and complimentary subscriptions only — not enhanced earn rates or point multipliers. All credits across the wallet with total annual value.

## 🔁 Overlap

Numbered list of redundant earn categories or duplicate benefits.

## 🕳️ Gaps

Numbered list of common spend categories not covered at a bonus rate by any card.

## 📋 Confidence Notes

Flag any uncertain, unconfirmed, or conflicting claims.

## 🔗 Sources

Numbered list of URLs fetched, as markdown hyperlinks with short "Site - Topic" labels.

Output Rules

• Use one emoji per section heading and numbered lists for overlap/gaps.
• When listing credits, fees, or any monetary amounts, sort from highest to lowest dollar value.
• Keep content to condensed facts — no prose padding.
• Omit the Card Identity section when all matches are confident.
• Do not show YAML blocks in output.
• End every report with a ## 🔗 Sources section listing each URL fetched during research as a markdown hyperlink with a short "Site - Topic" label, e.g. [Chase - Sapphire Preferred](https://...).

Confidence Definitions

• confirmed: supported by issuer terms or multiple approved sources
• unconfirmed: plausible but not fully resolved
• conflicting: sources disagree on a material fact