What each of its 17 skills does for a garment factory, where it stops, and six margin checks you can run this week with Claude.

Claude for Manufacturing is Anthropic's plugin for factories, launched November 12, 2026. Its 17 skills draft quotes, quality reports, supplier requests and a morning production report from your ERP and exports. It wasn't built for apparel: none of its 17 skills knows a size run, a standard minute or a chargeback. Here's what each one does, where it stops, and six margin checks you can run this week.
The problem it doesn't solve. Claude for Manufacturing drafts documents from your data. It calculates from the numbers your systems hold; it doesn't check whether they're right, and it doesn't tell you what an order is worth before you accept it. In apparel, that's where margin is made or lost: a quote can look profitable on the costing sheet and lose money once the real minutes, fabric and chargebacks land. Below: what each skill does, six places factories lose money, a prompt to find each one, and why finding it once isn't enough.
Short on time? Book a 30-minute margin check and we run one of the six checks below with you, on your data.
Book a margin checkIt's a plugin for Claude Team and Enterprise plans with 17 skills: 12 single skills and 5 chains that run several in a row. It connects to Microsoft 365, Oracle NetSuite, Infor M3 and Autodesk Fusion. You don't need a connector to start: 11 of the 12 skills work from a folder of regular exports. A person reviews every draft before it reaches a customer or supplier.
The “Set us up” step. Tests your connected systems, finds your export folder, learns your shift patterns and plant vocabulary, and lists what's missing.
Teach it your words on day one: cut ticket, bundle, SAM, size run, colorway. It learns vocabulary; it doesn't know apparel's.
Reads your exports and connected systems, joins them on shared IDs, and gives back one workbook plus a list of records that didn't match.
The most useful skill for apparel. Joining orders, cut tickets, payroll and shipments is half of every margin question. The unmatched list is the first honest look at your data quality.
Connects a system the plugin doesn't support, trying seven options from best to simplest and stopping at the first that works.
Most apparel plants run a local ERP, a payroll system and spreadsheets, not NetSuite or Infor. This is how those get in.
Reads a request for quotation in any format, records where every field came from, checks it against what your plant can build, and recommends bid or no-bid.
Works on a buyer's tech pack and PO request. “What your plant can build” means machines and capabilities; it won't know your capacity is booked for the next six weeks unless you give it the plan.
Prices a job with the working shown: material cost, labor and machine time, setup spread across each quantity break, and the margin.
Built around machine time and setup. Map your operation bulletin as its routing and SAM as the time standard, and add fabric consumption per size and marker efficiency. Then it shows the margin, but only on the costs you gave it.
Writes a nonconformance report that cites the exact requirement broken, works out which other parts may be affected, and checks earlier reports.
A failed buyer inspection against an AQL level or a spec sheet is exactly this. Saves the QA manager real hours.
Builds a timeline across your data, ranks likely causes with evidence for and against, and drafts most of the 8D corrective-action report.
Big brands ask for 8D reports on repeat defects like shading, broken stitches or measurement failures. It drafts most of the report; the fix and the sign-off are still yours.
Builds the request package for suppliers, drafts the emails, tracks replies and compares the bids. It never picks the winner; your buyer does.
Fabric, trims and packaging RFQs. Comparing bids on price, lead time and minimums is daily work for a sourcing team.
Ranks the likely causes of a machine failure from sensor data, repair history and the manufacturer's guide, with the cheapest test to tell them apart.
Built for machines with sensor data and troubleshooting guides. Most sewing machines have neither, and a mechanic fixes them in minutes. The exception: automatic cutters, embroidery heads, printing and heat-press lines.
Builds the morning report: output against plan, equipment effectiveness (OEE), downtime ranked by cost, scrap and downtime in dollars, and the decisions waiting.
Sewing lines are measured in efficiency against standard minutes, not OEE. Map it, and “downtime and scrap in dollars” every morning is worth having, as long as the cost rates you feed it are right.
Generates variations of a part in Autodesk Fusion by changing measurements on the open model. Needs Fusion running; there's no fallback.
Fusion is mechanical CAD. Garment pattern and 3D tools aren't supported.
Turns a filled-in decision matrix into a ranked comparison with sources, and tests how sensitive the result is. It never invents a score for an empty cell.
Works for any comparison you can score: two fabric mills, three new machines, near-shore against offshore. You fill in the scores; it ranks them and never invents one.
Runs without Fusion at a lower level, and says so. Same costing caveat as quote build.
The strongest fit: failed inspection to draft 8D in one go.
The only chain with no approval step; it writes to its own dashboard. You have to schedule it in Cowork or Claude Code; the plugin can't schedule itself.
Puts the cost of a breakdown into the next morning report. Inherits breakdown diagnosis.
A bad fabric lot becomes an RFQ to other mills.
Not sure which of these fit your plant? On the same 30-minute call we map the 17 skills to the systems and exports you run today.
Map my skillsWhat it writes back. Anthropic's manufacturing page says the plugin doesn't “write to your systems of record”. The plugin's own documentation lists the NetSuite connector as read and write, Infor M3 and Microsoft 365 as read only. In practice, every skill ends with a person approving the draft. It also doesn't operate equipment.
Its status. The plugin's GitHub page calls it a reference implementation, “not actively maintained”, provided as is. Treat it as a strong starting kit, not a supported product.
Anthropic says it fits best where each job gets its own quote, drawing and routing. Apparel works that way, but its money moves through things a generic manufacturing skill doesn't model.
The plugin can read data about all of these if it's in your systems. It doesn't know what to look for. Anthropic's launch cites a 2025 Manufacturing Leadership Council survey: 79% of manufacturers enter data by hand. The same survey found only 39% have a process to check it. A fast draft from an unchecked costing sheet is still wrong.
These come from eight years running our own garment factory in Santiago, Dominican Republic, and from the plants we work with now. Each check takes one or two reports you already have and one prompt. The thresholds are our starting points, not industry standards: change them to fit your business.
Save the files listed under “Files to attach” as Excel or CSV, attach them to a Claude chat and paste the prompt. Column names don't have to match ours. Missing a column? Run it anyway: every prompt tells Claude to list what's missing instead of guessing.
Can't pull these files in an hour? That's the finding: your margin depends on data nobody connects. Send us what you have, and we'll run the check with you.
One example of why they matter. Grupo Beta is a vertical apparel manufacturer in Tegucigalpa, Honduras, that produces two to three million garments a month. HGI's case study reports that the CreateOne platform there identified “more than $425,000 in supplier chargebacks that had gone unnoticed” and “$200K+ in recurring monthly savings through SKU optimization and supplier price analysis.”
Source: HGI, “Why Ask AI to Just Find a Problem When It Can Fix It Too?”
Discounts get given on the phone and never approved. The biggest customers usually get the biggest unapproved ones.
Files to attachSales order lines from your ERP, last 90 days: order, customer, style, quantity, list price, invoiced price, approved discount.
Red flagInvoiced more than 3% below list, after approved discounts.
PromptThis finds discounts you already gave. That money left with the invoice. Computing value means the quote shows the real margin of the order at the price you're about to accept, before anyone says yes on the phone.
The costing sheet was right the day it was quoted. Fabric consumption, minutes and trims drift after that, and nobody re-checks a style that's already selling.
Files to attachThe costing sheet for your highest-volume style, and its last 3 production orders: fabric issued, garments cut and shipped, minutes booked, trims used.
Red flagActual cost more than 5% above the costing sheet, on any line item.
PromptThis shows the drift on 3 orders you already shipped. That money is gone. Computing value means every quote uses the real minutes and fabric of the style, not the costing sheet from the day it was first priced.
Lines wait for cut work early in the week and pay overtime to catch up at the end. Payroll sees the hours, production sees the idle time, nobody puts them side by side.
Files to attachOvertime by line by week from payroll, and idle time by line by week from production tracking or the supervisors' log. Last 8 weeks.
Red flagAny line with overtime in a week it also logged more than 4 idle hours.
PromptThis shows overtime you already paid. Computing value means the plan sees that a line will wait for cut work before the week starts, and what that wait will cost.
Customers are ranked by revenue. Ranked by margin after chargebacks, freight and rework, the order often flips, and your second-biggest customer can be your least profitable.
Files to attachRevenue, cost, deductions, freight and returns by customer, last 12 months, from your ERP or accounting system.
Red flagAny customer in your top 10 by revenue that's in your bottom 3 by margin.
PromptThis needs 12 months of results, and it can't split the costs no single customer owns. Computing value means knowing what a customer's next order is worth, with every cost assigned, before you take it.
Customers deduct from payments and suppliers owe you for late or defective materials. Both sit in remittances and emails nobody reconciles. This is where Grupo Beta's $425,000 was.
Files to attachCustomer deductions from accounts receivable, and late or rejected supplier deliveries from the receiving log. Last 6 months.
Red flagAny deduction older than 30 days with no dispute, and any late or rejected supplier delivery with no claim.
PromptThis recovers some of what you're owed. It doesn't stop the next deduction. Computing value means each order carries its chargeback risk from the start: the labels, ship dates and compliance rules it can fail.
Cut bundles that stall become cash you've spent and can't ship. After a month, most of it is either a forgotten order or a quality problem nobody raised.
Files to attachYour open WIP report: cut ticket or bundle, style, customer, quantity, start date, current operation, cost so far.
Red flagAnything open more than 30 days.
PromptThis finds work that's already stuck. Computing value means knowing what each bundle costs you for every day it waits, while there's still time to move it.
Bring one export: customer deductions, open WIP, overtime by line or your last 90 days of order lines. We run the matching check with you on a call, trace every flagged number to its source row, and show what a model of your business would add. You keep the results. Mutual NDA on request.
With Claude for Manufacturing: run the /data-consolidate skill to build one workbook with the data listed under “Files to attach”, from your connected systems and export folder. Then paste the prompt.
With any Claude plan, or a spreadsheet: attach the files with the prompt. Every check is also a few columns of formulas if you'd rather not use AI at all.
Check your plan's data policy before you upload customer prices or payroll.
Check one answer by hand. Pick one flagged row and trace it to the source document. If it holds, trust the rest more. If it doesn't, you've found a data problem worth fixing first.
Start with check 5. It's the one most likely to find money you can recover this month.
Install it if you already run NetSuite or Infor, pay for Claude Team or Enterprise, and your team spends hours a week writing quotes, quality reports and morning summaries.
Run the six checks first if most of your costs live in spreadsheets nobody checks. The plugin will draft fast and confidently from wrong numbers.
Don't wait for it to find your margin. It drafts documents. Knowing which orders, customers and styles make money, and acting on it before you say yes to the next order, needs a model of how your business makes money. That model is what CreateOne builds for apparel factories and brands: connected to the systems you already run, with every number traced to its source. Book a margin check to see it on your data
No. It's a general manufacturing plugin; Anthropic says it fits best where each job gets its own quote, drawing and routing. Of its 17 skills, 6 work for apparel today (data, connectors, quality, supplier and trade-study skills), 4 need apparel terms mapped in (onboarding, inquiry intake, quoting, production reporting), and 2 don't apply (Fusion design and breakdown diagnosis).
Yes. It connects to Oracle NetSuite or Infor M3 (a plant connects one), plus Microsoft 365 and Autodesk Fusion. Other systems work through scheduled exports or the /custom-connector skill, and 11 of the 12 skills run on exports alone.
No. It drafts from what your systems hold and cites where each figure came from, but it doesn't verify the figures. The /data-consolidate skill lists records that don't match across systems, which is a good first look at your data quality.
Anthropic publishes the plugin as a reference implementation: not actively maintained, provided as is. Treat it as a starting kit.
It's a plugin for Claude Team and Enterprise plans. Anthropic's announcement lists no separate manufacturing price.
Claude for Manufacturing has no textile-specific skills, as it has none for apparel. The general skills (data consolidate, NCR write-up, root cause, supplier quote request, trade study) work for fabric and textile plants the same way. A mill swaps the apparel terms in this guide, like size runs and standard minutes, for its own, and checks 1, 4, 5 and 6 apply as written.
On the Claude side, check your plan's data policy before you upload customer prices or payroll. With CreateOne, a mutual NDA is available on request, we connect only the data the agreed questions need (scoped, logged, never a full data dump), data is encrypted in transit, and your team approves every action.
Yes. In Claude for Manufacturing, every skill ends with a person approving the draft. In CreateOne, your team approves every action and every number traces to its source, so anyone can check where it came from. At Grupo Beta, HGI reports 110 active users, from the shop floor to the CEO's desk, asking the system questions directly.
No tool makes a wrong number right on its own. AI makes messy data more visible: the /data-consolidate skill lists the records that don't match across systems. CreateOne connects to your systems as they are and traces every number to its source, so you can see where a figure came from before you act on it.
According to HGI's case study, SKU optimization recovered more than $100,000 a month in lost revenue, and tracing supplier price changes back to individual purchase orders generated another $100,000 a month. Consolidating and analyzing events across the business also surfaced more than $425,000 in supplier chargebacks that had gone unnoticed.
Yes. Send one export and we run the check with you on a 30-minute call, with every flagged number traced to its source row. Book a margin check.