Stripe transaction review agent

Why you need it

‍
Fraud often shows up as small patterns across many transactions, not just one obvious bad charge. Manually scanning Stripe exports for strange velocity, odd merchants, or location mismatches is slow and easy to miss. This agent pulls recent transactions from Stripe, checks them with rules and LLM pattern detection, and flags anything that looks off. It sends a simple Slack summary with evidence for each case and produces JSON that your fraud or ops team can plug into dashboards or internal tools.

What you need in Vellum

• Stripe integration with access to recent transactions
• An input or config for lookback window, for example last 24 hours or last 7 days
• Logic or tools to apply rule based checks such as velocity, high value, country or IP mismatch
• An LLM step to look for unusual patterns that simple rules might miss
• A formatter that creates structured JSON with flagged cases, reasons, and scores
• Slack integration with a field for the target channel name or ID
• A summary template for posting flagged cases to Slack with short explanations
• A trigger such as a schedule that runs every hour, day, or week