> Illustrative sample produced with the Cheek Priority Agent Roster skill. Mastercard is not affiliated with and has not endorsed this material. Owners are fictional names; priority scores are sample values in the AIDE Opportunities Audit's format, not real audit output.

# Mastercard Agent Roster

The one-page view for leadership review before anything gets created. Source: the AIDE Opportunities Audit's opportunity matrix (illustrative sample data); the four highest-priority agentic opportunities below were worked through the full Agent Role Definition worksheet, one creation prompt per agent. Waves map from the audit's bands: Pilot Now agents are Wave 1, Build Readiness agents are Wave 2. No agent in this roster ships without a named owner, at least one never, numeric guardrails, and both a quality and a safety metric.

## Wave 1: Pilot Now

### 1. Chargeback Dispute Triage Agent

- **Owner:** Dana Okafor, EVP Network Operations (fictional)
- **Goal:** every chargeback case opened in the past 24 hours is classified, evidence-checked, and routed to the correct resolution queue with a triage note before analysts start the morning.
- **Source opportunity:** Dispute triage and routing, priority score 87, band Pilot Now, passed deep assessment.
- **Creation prompt:** dispute-triage-agent-creation-prompt.md
- **Top guardrail:** autonomous routing only for disputes at or under $500; everything above, and all fraud-coded cases, routes to the human review queue.
- **Never:** never moves money and never decides a dispute outcome; disposition stays human.

### 2. Services Report Drafting Agent

- **Owner:** Elena Marsh, VP Advisors Delivery (fictional)
- **Goal:** every Advisors engagement with a scheduled deliverable has a first-draft report assembled from that engagement's own data room, in the practice template, waiting in the engagement manager's review folder two business days before the client date.
- **Source opportunity:** Consulting deliverable first drafts, priority score 81, band Pilot Now, passed deep assessment.
- **Creation prompt:** services-report-drafting-agent-creation-prompt.md
- **Top guardrail:** 20 drafts per week maximum; every draft carries a DRAFT watermark and cites only documents from its own engagement's data room, verified by the citation check in the log.
- **Never:** never sends anything to a client; drafts land in the engagement manager's review folder only.

## Wave 2: Build Readiness

### 3. Merchant Onboarding Verification Agent

- **Owner:** Tomas Aguilar, SVP Customer Onboarding (fictional)
- **Goal:** every new merchant application has the documented verification checklist completed, discrepancies flagged with evidence, and a recommendation note attached within 4 business hours of submission.
- **Source opportunity:** Merchant onboarding document verification, priority score 78, band Build Readiness.
- **Creation prompt:** merchant-onboarding-verification-agent-creation-prompt.md
- **Top guardrail:** any application with a beneficial ownership discrepancy or a composite risk score above 60 routes untouched to the named compliance officer; the agent completes checks only from the documented checklist, nothing improvised.
- **Never:** never approves or declines a merchant; verification findings only.

### 4. Fraud Rule Tuning Agent

- **Owner:** Grace Lin, VP Fraud Strategy (fictional)
- **Goal:** each week, the fraud rules committee receives a ranked set of proposed rule adjustments, each simulated against the trailing 30 days of transactions with the fraud-catch and false-positive deltas quantified.
- **Source opportunity:** Fraud rule tuning proposals, priority score 74, band Build Readiness.
- **Creation prompt:** fraud-rule-tuning-agent-creation-prompt.md
- **Top guardrail:** a proposal is eligible for committee review only if its simulated false-positive delta is under 0.5 percent; maximum 10 proposals per week.
- **Never:** never pushes a rule change to production; proposals exist only in the committee sandbox until a human commits them.

## Before anything runs

Every creation prompt above goes through the Agent Definition Review (/cheek-agent-definition-review) for the operator's grading, weakest layer named, before its agent is created in the Claude Console. Wave 2 does not start until both Wave 1 agents have run for four weeks with zero guardrail breaches in their action logs.

IMPORTANT: download this file and upload it to your Claude project, or keep it in this working folder if you are in Claude Code, so the other skills in this collection can find it and build on your work.