Reorder Policy

<!-- Copyright 2026 Anthropic PBC -->

<!-- SPDX-License-Identifier: Apache-2.0 -->

Reorder Policy

This skill encodes StockPilot's reorder rules. Use it whenever you need to decide whether to reorder a SKU and how many units to order.

Inputs you need

| Value | Where to get it |

|---|---|

| on_hand | Latest row for the SKU in /mnt/user/data/stock_levels.csv (sum across warehouses unless the task is warehouse-specific) |

| reorder_point | /mnt/user/data/products.csv |

| avg_daily_sales | Mean of units_sold over the last 14 days in /mnt/user/data/sales_history.csv |

| lead_time_days | From the chosen supplier (see the supplier-selection skill) |

| forecast_qty, confidence, flags | From the forecasting skill, only if the task looks forward more than 14 days or mentions a promo/season |

Decision rules

• Do we reorder at all? Reorder if on_hand < reorder_point. If on_hand ≥ reorder_point, the answer is "no reorder needed" — stop here.

• How much? The target order quantity is 30 days of cover plus safety stock, minus what's already on hand:

   safety_stock = 1.5 × avg_daily_sales × lead_time_days
   order_qty    = (avg_daily_sales × 30) + safety_stock − on_hand

Round up to the supplier's min_order_qty multiple.

• Confidence guard. If you obtained a forecast and confidence < 0.6, do not place a PO automatically. Instead:
• Write an escalation to /mnt/user/sinks/outbox.jsonl using the notify-templates skill (escalation template), including the flags from the forecast.
• In your final answer, state the recommended quantity and that it requires human review, with the reason.

• Expedite? If on_hand / avg_daily_sales < lead_time_days (you'll stock out before the order arrives), pick the supplier with the shortest lead time even if it's not the cheapest, and note "expedited" in the PO.

Worked example

SKU-0057: on_hand = 38, reorder_point = 120, avg_daily_sales = 18.2, chosen supplier lead_time_days = 7, min_order_qty = 50.

• 38 < 120 → reorder.
• safety_stock = 1.5 × 18.2 × 7 = 191.1
• order_qty = (18.2 × 30) + 191.1 − 38 = 546 + 191.1 − 38 = 699.1 → round up to next 50 → 700
• 38 / 18.2 = 2.1 days of cover, lead time is 7 → expedite.

Prioritization (when many SKUs are at risk)

Sort by tier, then by days-of-cover ascending within a tier:

• Stockouts — on_hand = 0 at any warehouse. Always first.
• Will stock out before PO arrives — days_of_cover < lead_time_days. Expedite or transfer.
• High-velocity below reorder — top-100 sellers below reorder point.
• Routine replenishment — everything else below reorder point.
• Trending toward reorder — note only, no action.

If you can't action every SKU in one pass, say how many you handled, how many

remain, and list the remaining SKU IDs so the next run picks them up.

Transfer vs reorder

When one warehouse is low and another has surplus:

• Transfer when the surplus warehouse has >30 days cover for itself, the

3–5 day transfer beats the best supplier lead, and the qty needed is ≲200

units. WH-CENTRAL is the default source.

• Reorder when no warehouse has surplus, the qty is large, or supplier lead

is comparable to transfer lead anyway.

• Both when the shortage is urgent and large — transfer a bridge qty now,

PO the remainder.

State which path you chose and why.

Compliance

• Any single PO over $10,000 needs a one-line rationale included with the

PO so ops can copy it into the ERP.

• If you place more than five POs in one task, end with a one-line total

committed spend.

• Do not place a second PO for a SKU that already has an open PO covering the

need.

Output

When you act, append to /mnt/user/sinks/purchase_orders.jsonl:

{"sku": "SKU-0057", "supplier_id": "SUP-03", "qty": 700, "expedite": true, "reason": "below reorder point; 2.1d cover"}

When you only recommend (no side-effect requested), return a structured ReorderDecision:

{"sku": "...", "reorder": true, "qty": 700, "supplier_id": "SUP-03", "expedite": true, "confidence": 0.85, "notes": "..."}