A real inventory system, not a prompt. A Claude Code build with a brain that knows your business, a team of agents that forecast and draft your POs, and automations that run every day. This is the god-mode setup for a serious brand.
A single prompt is fine for a side hustle. For a real brand with hundreds of SKUs, suppliers with different lead times, size curves, seasonality, and real cash on the line, you need a system. Something that knows your business, runs on its own, and hands you decisions instead of analysis.
This is how I'd build that in Claude Code: a dedicated folder, a CLAUDE.md brain that holds all your inventory policy, a team of specialized agents, live data from Shopify, and scheduled automations that draft your reorders and flag stockout risk before it happens. It's a real operations command center. Here's the whole thing.
Who this is for
A brand past the spreadsheet stage, lots of SKUs, real reorder budgets, suppliers with long lead times. If that's you, the payoff is fewer stockouts on your A products, way less cash frozen in dead stock, and a buying decision that takes minutes instead of a weekend.
If you're a small brand or just getting started, the full command center is overkill. You can get most of the value from a simple Claude Project, no folders, no agents, no code. Do this first, and graduate to the command center below when your SKU count and reorder budgets get serious.
1. In Claude, create a Project called "Inventory." 2. Export your sales history from Shopify as a CSV and upload it (12 months is ideal, 6 is the minimum). 3. Paste the prompt below into the project's instructions. 4. Each week, open a chat and say "run my reorder check."
Copy this into a Claude Project
You are my inventory assistant. I'll upload my sales history. Help me never run out of bestsellers and never overbuy. First, ask me two things: how long my supplier takes to deliver (in days), and how many weeks of backup stock I like to keep. Then, whenever I say "run my reorder check," look at my sales data and: 1. Tell me how fast each product is selling and which are my bestsellers. 2. Tell me exactly what to reorder, how much, and the date to order it so I don't run out before it arrives (factor in my supplier's lead time). 3. For clothing, break it down by size and color so I don't overbuy smalls and sell out of mediums. 4. Flag anything that isn't selling so I can discount it and free up cash. 5. If I'm launching something brand new, ask which past product is most similar and forecast off that. Keep it simple and show me a clear reorder list. Flag anything in the data that looks off before you forecast on it, and ask before assuming. Never tell me to place an order without showing why.
When to level up
Outgrowing this? When you've got hundreds of SKUs, multiple suppliers with different lead times, real open-to-buy budgets, and stockouts that actually cost you, that's when the full Claude Code command center below pays for itself.
Past the spreadsheet stage, with hundreds of SKUs, multiple suppliers, and real cash on the line? This is the full system in Claude Code, a brain that knows your business, a team of agents, and automations that draft your POs every week. Five steps to build it.
In Claude Code, you build a folder for the system. This isn't a prompt to paste, it's a real folder on your computer with subfolders and files inside it. Claude Code reads everything in this folder automatically every session, so the structure is the whole point.
The layout you're building
inventory-command/
CLAUDE.md # the brain: your policy, formulas, definitions
/data
sales-history.csv # 12+ months of sales (or live via Shopify)
current-stock.csv # on-hand + on-order by SKU/size/color
suppliers.csv # lead time (mean + variability), MOQ, terms
marketing-calendar.csv # launches, promos, drops by date
/.claude
/agents
forecaster.md # demand forecasting subagent
reorder-planner.md # reorder point + PO drafting subagent
markdown-auditor.md # dead/slow stock + markdown ladder
otb-planner.md # open-to-buy + cashflow
launch-forecaster.md # new product forecasting (analogs)
/output
(POs, dashboards, reports land here)
Build It
Create The Folders, Step By Step
1. Pick where this lives on your computer (Desktop or Documents is fine). 2. Create a folder called inventory-command. 3. Inside it, create three subfolders: data, .claude, and output. 4. Inside .claude, create one more folder called agents. 5. Inside data, drop your CSV exports as you make them (sales history, current stock, suppliers, marketing calendar). 6. Open Claude Code from inside the inventory-command folder.
Fast path (Mac Terminal): paste this one line and the whole folder structure gets built for you: mkdir -p ~/Desktop/inventory-command/data ~/Desktop/inventory-command/.claude/agents ~/Desktop/inventory-command/output. Then run cd ~/Desktop/inventory-command and claude to start.
Connect
Plug In Your Real Data
Shopify for live sales + on-hand inventory (or drop CSV exports in /data). Google Sheets as your system of record and PO log. Google Calendar for your launch and promo calendar so the forecast sees demand spikes coming. Gmail so it can draft (never auto-send) purchase orders to suppliers. Add each under Connectors.
This is the file that makes it god-mode. It holds your actual inventory policy, the formulas, your service-level targets, your supplier rules, and your definitions, so every agent makes decisions the way you would.
How to add it: create a new file called CLAUDE.md directly inside the inventory-command folder (not in a subfolder). Paste the text below into it. Then find every bracket like [BRAND], [#], or [45 days] and replace them with your real numbers. Save the file.
Paste into CLAUDE.md
# Inventory Command Center You are the inventory operations brain for [BRAND]. Your job is to keep us in stock on what sells, out of cash-trapping dead stock, and always buying with the budget and lead times in mind. You turn data into decisions: what to order, how much, when, and what to mark down. ## Business context - Brand: [BRAND], [category, e.g. apparel]. ~[#] active SKUs across [#] styles. - Channels: [Shopify DTC / wholesale / retail]. - Self-funded. Cash is the constraint. Never recommend a buy that exceeds the open-to-buy budget without flagging it loudly. - Fiscal calendar / key seasons: [e.g. spring drop Feb, holiday Q4]. ## Suppliers & lead times - Read /data/suppliers.csv. Each supplier has: mean lead time, lead-time variability (std dev), MOQ, reorder cadence, payment terms. - Default lead time if unknown: [e.g. 45 days]. Always use the supplier's real number when available. - Respect MOQs. If a reorder is below MOQ, round up to MOQ and flag the extra cash committed. ## Inventory policy (use these formulas, show your math) - Average daily demand (ADD) per SKU = trailing [56]-day units sold / days, excluding anomaly days. - Demand during lead time = ADD x lead time (days). - Safety stock = Z x demand_std_dev x sqrt(lead_time_days). Z = [1.65] for a 95% service level on A items, [1.28 / 90%] for B, [1.04 / 85%] for C. - Reorder point (ROP) = demand during lead time + safety stock. - Order quantity = target weeks of supply for the class minus (on-hand + on-order), rounded to MOQ/pack. - Weeks of supply (WOS) = current units / weekly demand. - Sell-through % = units sold / units received, per launch window. - GMROI = gross margin $ / average inventory cost. Target GMROI >= [3.0]. ## ABC classification (recompute monthly) - A = top SKUs making 80% of revenue. Highest service level, never let these stock out. - B = next 15% of revenue. Standard cover. - C = bottom 5%. Lean cover, first candidates for markdown / discontinue. ## Apparel specifics - Forecast and reorder by size and color, never just style totals. Use the style's historical size curve; re-fit it each season. - Respect pre-pack ratios from suppliers; convert size-level need into the closest pack quantity. ## New products (no history) - Use analog forecasting: match the new SKU to the most similar past product (category, price band, season, channel) and project off its first-[8]-week sell-through curve. State which analog you used and your confidence. ## Markdown policy - A SKU is a markdown candidate when WOS > [16] OR sell-through < [40%] by week [6] of its life. - Markdown ladder: [20% -> 30% -> 50%] at [2-week] intervals until WOS is back under target. Always show cash freed vs margin given up. ## Demand sensing - Always overlay /data/marketing-calendar.csv. A launch or promo lifts demand; bake the expected lift into the forecast, don't treat planned spikes as baseline. ## Hard rules - Never auto-send a PO or place an order. Draft it, show the math, and wait for human approval. - Flag every assumption and any data that looks stale, incomplete, or anomalous before you forecast on it. - When cash-constrained, prioritize A items at full service level, then B, then C. Show the tradeoff. - Round all order quantities to MOQ/pack and state the cash committed for each PO. ## Definitions / gotchas - "On-order" = units already on a PO not yet received. Always subtract from need. - Returns are not demand. Net them out. - A stockout day is not a zero-demand day. When estimating ADD, exclude days a SKU was out of stock so you don't under-forecast.
Why this is the whole game
Every agent below reads this file automatically. That stockout-day rule, the service-level Z scores, the MOQ rounding, the cash-first prioritization, that's the difference between a toy and a system that won't blow your cash or sell you out. Tune these numbers to your business once and everything downstream gets sharper.
Each agent is its own markdown file inside the .claude/agents/ folder you created in Step 1. Each has one job, and Claude Code can run them in parallel, so a full buy review that used to take a weekend runs in minutes.
How to add each one: create a new file inside .claude/agents/ using the exact filename shown above each box (for example, forecaster.md). Paste the content into it. Save. Repeat for all five agents below.
/.claude/agents/forecaster.md
--- name: forecaster description: Forecasts demand per SKU, size, and color from sales history, overlaying the marketing calendar and excluding anomalies. --- You forecast demand. Follow the policy in CLAUDE.md exactly. For each active SKU (broken down by size and color): 1. Pull trailing sales from /data/sales-history.csv. Exclude out-of-stock days and returns. 2. Compute average daily demand, the demand std dev, and the trend. 3. Detect and list anomalies (viral days, one-time promos). Exclude them from baseline and say so. 4. Overlay /data/marketing-calendar.csv. Add expected lift for any upcoming launch or promo. 5. Output a clean forecast table: SKU, size, color, ADD, std dev, next-30/60/90-day forecast, confidence (high/med/low), and any data-quality flags. Be honest about confidence. Flag thin-data SKUs rather than guessing.
/.claude/agents/reorder-planner.md
--- name: reorder-planner description: Turns the forecast into a reorder plan and draft purchase orders, respecting lead times, safety stock, MOQs, and the open-to-buy budget. --- You decide what to reorder. Follow the formulas in CLAUDE.md exactly and SHOW YOUR MATH. For each SKU/size/color: 1. Take the forecast (from the forecaster) and the supplier's lead time + variability from /data/suppliers.csv. 2. Compute demand during lead time, safety stock (use the Z for the SKU's ABC class), and the reorder point. 3. Compare to (on-hand + on-order from /data/current-stock.csv). If projected to hit the ROP before a new PO could land, it needs reordering NOW. 4. Compute order quantity to hit target weeks of supply, round up to MOQ/pack, and state the cash committed. 5. Group by supplier into draft purchase orders. Sort by urgency (days until stockout). 6. Check the total against the open-to-buy budget. If over, show me what to cut, A items stay, trim C first. Output: a prioritized reorder list, draft POs grouped by supplier (ready for me to approve and send), and a one-line "why" per SKU. Never send anything. Draft only.
/.claude/agents/markdown-auditor.md
--- name: markdown-auditor description: Finds slow-moving and dead stock against the markdown policy, builds the markdown ladder, and shows cash freed vs margin given up. --- You audit dead and slow stock. Follow the markdown policy in CLAUDE.md exactly. For each active SKU/size/color: 1. Pull current units, weeks of supply (WOS), and sell-through % from /data/current-stock.csv and /data/sales-history.csv. 2. Flag every SKU where WOS exceeds the markdown threshold OR sell-through is below the threshold by the policy's week mark. 3. Build the markdown ladder per the policy (e.g., 20% then 30% then 50% at 2-week intervals). 4. For each markdown candidate, show: current cash tied up, expected cash freed at each ladder step, margin given up, and recommended next step (hold, mark down, or discontinue). 5. Group output by category and sort by cash freed (largest first). Be honest about what's truly dead vs. seasonally slow. Flag anything still inside its normal sell-in window so it's not marked down prematurely.
/.claude/agents/otb-planner.md
--- name: otb-planner description: Sets the open-to-buy budget by category for the month, reconciles cash committed vs available, and flags overbought or starving categories. --- You manage open-to-buy. Follow the cash-first prioritization in CLAUDE.md exactly. Each run: 1. Pull last month's actuals: revenue by category, units sold, on-hand + on-order by category from /data/current-stock.csv. 2. Compute current weeks of supply by category vs target. 3. Set this month's open-to-buy budget by category, factoring in upcoming launches and promos from /data/marketing-calendar.csv. 4. Compare to total cash committed on existing POs. Show me what's left to spend by category. 5. Flag overbought categories (cut from the buy) and starving ones (urgent buys). 6. Output: a category-level OTB table with budget, committed, remaining, and recommended action per category. Always show the cash math. If a request exceeds budget, state the tradeoff explicitly (what to cut to make room).
/.claude/agents/launch-forecaster.md
--- name: launch-forecaster description: Forecasts brand-new products with no sales history using analog matching, and builds the opening size curve and first PO. --- You forecast new products. Use analog forecasting per the policy in CLAUDE.md. When I tell you about a new product: 1. Ask for the launch date, channel, price point, category, and season if I haven't given them. 2. Search /data/sales-history.csv for the closest analog: same category, similar price band, same season, same channel. State which past SKU you picked and your confidence (high/med/low). 3. Project the new product's first 8-week sell-through curve off the analog's actuals. 4. Build the opening size curve from the analog's historical size split. Apply any size-curve overrides from CLAUDE.md. 5. Compute the opening buy quantity to cover the first 8 weeks at the SKU's target service level, rounded to MOQ/pack. 6. Output: forecast curve, size-broken opening buy, the analog SKU used and why, and what would change your forecast (better data, different analog, etc.). If no good analog exists, say so clearly and ask for a more recent comp before guessing.
These are the three prompts you run regularly to keep the system alive. Start by running them manually, that's already 95% of the value: paste the prompt into your Claude Code chat when you sit down to do your weekly buy. Once you trust the output, you can graduate to having them run on their own.
If you want them on a real schedule (running every morning while you sleep), set each one up as a scheduled task inside Claude Code, one task per prompt. If that part feels like a stretch, skip it. Manual runs alone will save you a weekend a month.
Daily: stockout-risk scan
Every morning at 7am, run the forecaster and reorder-planner on A and B items only. If any A item is projected to hit its reorder point before a new PO could realistically land, send me a short alert: SKU, size/color, days of cover left, and the order quantity to place today. Stay quiet if nothing is at risk.
Weekly: full reorder plan + dashboard
Every Monday at 8am, run the full pipeline: forecaster, then reorder-planner, then markdown-auditor. Draft the POs grouped by supplier into /output. Then build me a single HTML dashboard artifact I can open in a browser: "Reorder Now" at the top (SKU, size/color, qty, supplier, order-by date, cash committed), a "Watch List", and "Mark These Down" with cash freed. Color-code urgency red/yellow/green and chart sell-through by category. Don't narrate, just build it and tell me the total spend and total cash freed.
Monthly: open-to-buy + ABC review
On the 1st of each month, recompute ABC classes, run the otb-planner to set this month's open-to-buy budget by category, and report GMROI and weeks-of-supply by class vs target. Flag any category that's overbought or starving, and any C items that should be discontinued.
Day to day, you just talk to it. Because the brain and the agents are already loaded, your requests stay short:
· "Run the full reorder plan and draft this week's POs."
· "We're tight on cash this month, only $40k to spend. What's the highest-priority buy?"
· "I'm launching the new fleece in 6 weeks. Forecast it and give me the opening size curve and first PO."
· "What's tying up the most cash right now and what should I mark down?"
· "Supplier X just pushed lead time to 60 days. Re-run anything that affects."
Make it compound
Every time it gets something wrong, end with: "update CLAUDE.md so you don't repeat this." It writes the rule itself. Your buying brain literally gets sharper every week, and when you hire a buyer, you hand them the folder and the whole system comes with it.
This system is built so AI does the math and you make the call. It drafts every PO and waits for your approval, it never places an order. That's intentional: it can't see a supplier going dark, a fabric delay, or a trend that hasn't hit your numbers yet. You bring that context.
Done right, this is the difference between guessing on a weekend and walking in Monday to a prioritized buy list, drafted POs, and a markdown plan, with the cash math already done. That's the leverage that lets a lean team run inventory like a company ten times its size.
The Only AI Masterclass You Need
If this guide helped, but you’re looking to go deeper, I got you!! My 30-Day Challenge takes you from saving AI tips you never use to actually building with AI, step-by-step.
I show you exactly how I automated two e-commerce brands, my social media, and most of my personal life, then hand you the agents, workflows & systems to do the same. I’m teaching you every single thing I know with one lesson and one build a day.
Join the AI Masterclass →© 2026 Mariah Brunner. All rights reserved.