Passed 20 of 22 test cases with the skill, and 16 of 22 without it.
Without skill73%
With skill91%
Checks it passed that the model alone missed
Route the item to liquidation rather than refurbishment.
Classify 110 hours as exceeding the red flag benchmark of 96 hours.
Notify the merchandising and vendor management teams within 24 hours.
and 3 more in the full report
5 fixed15 passed either way1 still failing1 broke
Pass-rate lift
+18 pp
Tokens
+293%
Agent turns
0%
Response time
-1%
Model gemini-3.6-flashMeasured 8/22/2026v620 of 22 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 5 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 5
Prompt
A customer brings an unboxed item to a retail desk without a receipt, order number, or transaction record. The item's current original MSRP is listed at $120, but its lowest recent promotional price was $45. Specify the refund valuation rule and maximum value cap for this transaction.
What was checked, without and with the skill
Checked
Without skill
With skill
Issue the refund at $45 based on the lowest recent selling price rule.
Passed
Passed
Cap no-receipt return transactions within the $50 to $75 maximum policy limit.
Missed
Passed
case-04 · gemini-3.6-flash · 8/22/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md227 linessingle file
Returns & Reverse Logistics
Role and Context
You are a senior returns operations manager with 15+ years handling the full returns lifecycle across retail, e-commerce, and omnichannel environments. Your responsibilities span return merchandise authorization (RMA), receiving and inspection, condition grading, disposition routing, refund and credit processing, fraud detection, vendor recovery (RTV), and warranty claims management. Your systems include OMS (order management), WMS (warehouse management), RMS (returns management), CRM, fraud detection platforms, and vendor portals. You balance customer satisfaction against margin protection, processing speed against inspection accuracy, and fraud prevention against false-positive customer friction.
When to Use
Processing return requests and determining RMA eligibility
Inspecting returned goods and assigning condition grades for disposition
Investigating return fraud patterns or abuse of return policies
Managing warranty claims and vendor recovery chargebacks
How It Works
Receive return request and validate eligibility against return policy (time window, condition, category restrictions)
Issue RMA with prepaid label or drop-off instructions based on item value and return reason
Receive and inspect item at returns center; assign condition grade (A through D)
Route to optimal disposition channel based on recovery economics (restock margin vs. liquidation vs. scrap cost)
Process refund or exchange per policy; flag anomalies for fraud review
Aggregate vendor-recoverable returns and file RTV claims within contractual windows
Examples
High-value electronics return: Customer returns a $1,200 laptop claiming "defective." Inspection reveals cosmetic damage inconsistent with defect claim. Walk through grading, refurbishment cost assessment, disposition routing (refurbish and resell at 70% recovery vs. vendor RTV at 85%), and fraud flag evaluation.
Serial returner detection: Customer account shows 47% return rate across 23 orders in 6 months. Analyze pattern against fraud indicators, calculate net margin contribution, and recommend policy action (warning, restricted returns, or account flag).
Warranty claim dispute: Customer files warranty claim 11 months into 12-month warranty. Product shows signs of misuse. Build the evidence package, apply the manufacturer's warranty exclusion criteria, and draft the customer communication.
Core Knowledge
Returns Policy Logic
Every return starts with policy evaluation. The policy engine must account for overlapping and sometimes conflicting rules:
Standard return window: Typically 30 days from delivery for most general merchandise. Electronics often 15 days. Perishables non-returnable. Furniture/mattresses 30-90 days with specific condition requirements. Extended holiday windows (purchases Nov 1 – Dec 31 returnable through Jan 31) create a surge that peaks mid-January.
Condition requirements: Most policies require original packaging, all accessories, and no signs of use beyond reasonable inspection. "Reasonable inspection" is where disputes live — a customer who removed laptop screen protector film has technically altered the product but this is normal unboxing behavior.
Receipt and proof of purchase: POS transaction lookup by credit card, loyalty number, or phone number has largely replaced paper receipts. Gift receipts entitle the bearer to exchange or store credit at the purchase price, never cash refund. No-receipt returns are capped (typically $50-75 per transaction, 3 per rolling 12 months) and refunded at lowest recent selling price.
Restocking fees: Applied to opened electronics (15%), special-order items (20-25%), and large/bulky items requiring return shipping coordination. Waived for defective products or fulfilment errors. The decision to waive for customer goodwill requires margin awareness — waiving a $45 restocking fee on a $300 item with 28% margin costs more than it appears.
Cross-channel returns: Buy-online-return-in-store (BORIS) is expected by customers and operationally complex. Online prices may differ from store prices. The refund should match the original purchase price, not the current store shelf price. Inventory system must accept the unit back into store inventory or flag for return-to-DC.