Refund Radar

**Judgement:** Refund Radar is a strong, privacy-focused skill for analyzing transaction data, surfacing recurring and suspicious charges, and drafting refund/dispute messages. It is well-suited for personal finance audits and subscription cleanup workflows, especially when users are comfortable exporting CSVs from their banks. **Strengths:** - Runs entirely locally with no network calls, which is ideal for sensitive financial data. - Robust parsing and normalization of varied CSV and text formats. - Clear, rule-based detection of recurring charges, duplicates, spikes, fees, and currency anomalies, with severity tagging. - Produces interactive HTML reports plus ready-to-copy dispute and refund templates in multiple tones. **Key Risks & Limitations:** - Heuristic classification: false positives and false negatives are likely, so flagged charges must not be treated as definitive evidence of fraud or entitlement to a refund. - Legal and financial context is not embedded; outputs should not be considered professional advice, and incorrect dispute framing could affect customer–merchant or customer–bank interactions. - Parsing may fail or misinterpret edge-case bank formats or nonstandard text exports, requiring manual verification of results. - Users still expose sensitive financial data to whatever environment the skill runs in; local-only design reduces but does not eliminate privacy risk if the host system is compromised. **Recommended Scenarios:** - Monthly or quarterly self-audits of bank and card statements to find forgotten subscriptions and unexpected fees. - Assisting users who suspect a small set of problematic charges and want structured, copy-pasteable refund or dispute text. - Generating clear, categorized HTML reports for budgeting, financial coaching, or preparing to talk with a bank or accountant. - Power users or tools that can orchestrate CSV export and review steps, while keeping a human in the loop to confirm any disputed transactions.