Hardware Manufacturing

Move hardware from CAD to mass production — mechanical design, PCB schematic, layout and DFM review, prototyping, EVT/DVT/PVT builds, factory selection, quality control, certification, and logistics. Use when the user asks about manufacturing a physical product, PCBs, tooling, or production runs.

Skill metadata

Source	Bundled (installed by default)
Path	skills/venture-studio/hardware-manufacturing
Version	1.0.0
Author	Fabric
License	MIT
Platforms	linux, macos, windows
Tags	manufacturing, cad, pcb, dfm, evt-dvt-pvt, production
Related skills	d2c-smart-products

Reference: full SKILL.md

info

The following is the complete skill definition that Fabric loads when this skill is triggered. This is what the agent sees as instructions when the skill is active.

Hardware Manufacturing

Use this skill when the user needs to take a physical product from an idea or CAD file to repeatable mass production: enclosure design, PCB design and fabrication, prototype iteration, EVT/DVT/PVT builds, factory selection, quality plans, certification, and freight. You cannot mill aluminum from a terminal, but you can do most of the leverage work: interrogate requirements, run design-rule checks, audit BOMs against live distributor stock, draft RFQ packages, compute tooling break-evens, and write the gate checklists that keep a build honest.

Do NOT use this skill for firmware or app work on the device (ordinary software workflow), for go-to-market, channel, and subscription questions on a connected product (load d2c-smart-products with skill_view), or for a quick "is this sensor even viable" feasibility hack — that is a spike, not a production program.

Workflow

1. Scope the program. Ask the user before designing anything: target retail price and unit-cost ceiling, lifetime and first-year volume, launch markets (drives certification), power source (mains, battery, USB), environment (indoor/outdoor, IP rating), and hard deadline. Write the answers into a one-page product requirements doc in the repo — every later trade-off resolves against it.
2. Design mechanical for the real process. Pick the production process from the volume table below first, then CAD to its constraints. A part designed for 3D printing and later "converted" to injection molding is a redesign, not a conversion.
3. Design the electronics. Schematic capture, then layout, then run ERC/DRC, then a DFM pass against the chosen fab's published capabilities. Lock the BOM with availability checks before layout is final.
4. Prototype in tight loops. Order boards and printed enclosures in small batches; bring up, test, log defects, revise. Each loop should answer named questions, not just "see how it looks."
5. Run the stage gates. EVT proves the design works, DVT proves it survives, PVT proves the factory can build it. Do not skip gates to save calendar time — each gate exists because the next phase is more expensive to fix things in.
6. Select the factory and cut tools. RFQ at least three suppliers, audit the winner, sign off golden samples, and only then release tooling payment milestones.
7. Certify. Engage an accredited test lab during EVT for pre-scans; formal FCC/CE/UL testing runs on DVT units, and certificates must exist before PVT units ship to customers.
8. Package and ship. Design packaging that survives ISTA drop testing, pick Incoterms deliberately, and decide air vs. ocean per shipment based on margin and urgency.

Mechanical CAD phase

Tool	Best when	Notes for the agent
Fusion 360	Startup default; parametric solids plus built-in CAM and simulation	Free for personal/non-commercial use only — verify current Autodesk licensing before recommending for a business; closed format, export STEP for portability
Onshape	Team collaboration, browser-only environments	Free plan makes documents public; version control is built in
FreeCAD	Open source, scriptable, headless-friendly	You can drive it from Python for batch STEP/STL export and geometry checks
CadQuery / build123d	Code-defined parametric parts	Fully agent-operable: generate, mutate, and export models in a script

Design to the process you will actually ship on, chosen by lifetime volume:

Volume (units)	Process	Piece price	Up-front cost	Design constraints to respect
1–50	3D print (FDM/SLA/MJF)	High	~none	Orientation, overhangs, layer strength anisotropy
50–500	CNC machining	Medium-high	Fixturing only	Internal corner radii, tool access, stock sizes
100–1,000	Urethane casting	Medium	Soft silicone tool, low thousands	Tool life ~20–25 shots; near-molding geometry
1,000+	Injection molding	Low (often under a dollar)	Tooling: low tens of thousands typical	Draft angles, uniform walls, ribs not slabs, avoid undercuts

Injection-molding rules that save redesigns: 1–2 degrees of draft on every face parallel to pull direction; keep walls uniform (thick sections sink and warp); use ribs at 50–60% of wall thickness for stiffness; every undercut adds a side action and real money to the tool; put the parting line and ejector marks where cosmetics allow. Get a DFM review from the molder before the tool is cut — reputable shops provide it free with a quote.

PCB phase

Default to KiCad unless the user has an existing toolchain. It is free, has no seat limits for contractors, and kicad-cli lets you run ERC, DRC, and gerber/BOM export from the terminal — meaning you can verify a design without opening a GUI:

kicad-cli sch erc project.kicad_sch
kicad-cli pcb drc project.kicad_pcb
kicad-cli pcb export gerbers project.kicad_pcb -o fab/

Sequence: schematic capture → ERC clean → footprint assignment → layout → DRC clean against the fab's rule set (import their trace/space, drill, and annular-ring limits; do not trust defaults) → DFM review → panelize if the assembly house wants arrays (add rails, fiducials, and mouse bites or v-scores per their spec).

Fab and assembly. For prototypes and small runs, turnkey fab+assembly houses (JLCPCB, PCBWay, Seeed, Aisler, or a domestic quick-turn shop for schedule-critical spins) are cheaper than splitting fab and assembly. For production, keep assembly close to the enclosure factory to shorten the box-build chain.

BOM management. Before layout is final, check every line item on a parts aggregator (Octopart or distributor APIs) for: live stock at two or more distributors, lifecycle status (reject EOL and NRND parts for new designs), and price breaks at your target volume. Nominate a second source for every critical IC or note it as single-source risk in the gate checklist. Store the BOM as CSV in the repo with columns for MPN, second-source MPN, reference designators, and per-unit cost at target volume — you can then re-audit availability with a script on every build.

Prototype iteration loop

Run each spin as a small, named experiment rather than a vibes-based reorder:

1. State the questions. Before ordering, list what this revision must answer (e.g. "does the antenna meet range spec inside the enclosure", "does the snap fit survive 50 cycles"). If a spin answers nothing new, skip it.
2. Order small. 5–10 boards assembled, 2–3 enclosures printed. Speed beats quantity until DVT.
3. Bring up methodically. Power rails first with current limiting, then clocks, then peripherals. Log every deviation in a bring-up notes file in the repo, keyed to board serial number.
4. Disposition every defect. Each issue gets a root cause and one of: fix in next rev, tolerate with rationale, or needs-more-data. Track them in a defects table per revision.
5. Bump the revision everywhere at once. Board silkscreen, CAD file names, BOM, and the release folder all move together; mixed-revision confusion at the factory is expensive and entirely self-inflicted.

Three to four board spins before EVT is normal for a product with a radio; budget calendar time for it rather than pretending rev A will be final.

Stage gates: EVT / DVT / PVT

Gate	Builds what	Typical qty	Proves	Exit criteria
EVT (Engineering Validation)	Works-like units, often prototype processes	20–100	The design functions	All features demoed; major bugs dispositioned; BOM risks listed; pre-scan EMC results reviewed
DVT (Design Validation)	Looks-like/works-like units on production tooling	50–500	The design survives	Reliability, drop, thermal, and battery tests passed; certification testing complete or in flight; cosmetic spec signed
PVT (Production Validation)	Sellable units on the production line at rate	500+ or 5–10% of first order	The factory can build it	Line yield at target (typically >95%); cycle time measured; QC plan executed; units are sellable

Treat exit criteria as blocking. A "conditional pass" with an untracked list of exceptions is how programs slip a quarter. Record every waiver with an owner and a close-by date.

Tooling cost and MOQ math

Amortize tooling into unit economics before committing:

effective unit cost = piece price + (tooling cost / expected lifetime volume)

Worked example: a $30,000 steel mold over 10,000 units adds $3.00/unit to a $1.20 molded part ($4.20 effective). The same part CNC'd at $18 needs no tooling. Break-even is tooling ÷ (CNC price − piece price) = 30,000 ÷ 16.80 ≈ 1,786 units — above that, cut the tool. Aluminum "soft" tools cost a third to a half as much but are typically good for 10k–100k shots; hardened steel multi-cavity tools cost multiples more and run to the millions. Factory MOQs work the same way: a stated MOQ is usually a price point, not a wall — ask for the price at MOQ, at half, and at 2x, and negotiate from the curve.

Factory selection and communication

Shortlist 3–5 candidates (sourcing platforms, referrals, or the fab house's assembly partners). Send each an identical RFQ package: STEP files, 2D drawings with tolerances and critical dimensions flagged, gerbers and centroid files, BOM, cosmetic spec, target volumes, and required certifications. Compare quotes line by line — a low headline price with vague tooling terms is worse than a clear expensive one.

Before mass production: audit the factory (in person or via a third-party audit firm), agree on a golden sample — a signed-off reference unit both sides keep — and write a QC plan covering incoming material inspection, in-process checks, and final inspection. Use AQL sampling (general inspection level II is the norm) with typical thresholds of 0 for critical defects, 1.0–2.5 for major, 4.0 for minor. Book third-party pre-shipment inspection for at least the first three shipments; it costs a few hundred dollars and catches container-level disasters.

Certification pass

• FCC (US): every electronic product needs unintentional-radiator testing; anything with a radio needs intentional-radiator certification. Using a pre-certified radio module with its modular approval intact dramatically shrinks scope — do not route antennas differently than the module's grant allows.
• CE (EU): mostly self-declaration against harmonized standards (EMC Directive, LVD if mains-powered, RED if it has a radio); keep a technical file. UKCA parallels CE for Britain.
• UL/ETL (safety): not legally mandatory in the US but required by most retailers and insurers for mains-powered products. Batteries need UN 38.3 to ship lithium cells at all, and often IEC 62133.
• Engage an accredited test lab at EVT for informal pre-scans (a day of chamber time finds EMC problems while the board can still change); run formal testing on DVT hardware; budget several weeks and five figures for a wireless consumer product across US+EU.

Packaging and logistics

Design the retail box, inserts, and master carton together; validate with ISTA 3A (parcel) or 1A drop testing at DVT. Print unit and carton barcodes the 3PL can scan. On shipping terms, know the Incoterm you sign:

Incoterm	Seller responsibility ends	Use when
EXW	At the factory door	You have your own freight forwarder and want control
FOB	Loaded on vessel at origin port	The common default; you own ocean freight and import
DDP	Delivered, duties paid, at your door	You want zero logistics work and accept the markup

Air freight costs roughly 4–8x ocean per kg but arrives in days, not 4–6 weeks port-to-port plus customs. Standard play: air the first small batch to hit launch, ocean the rest. Duties, tariffs, and customs brokerage belong in landed cost from day one — compute landed cost per unit, not factory cost.

Build-readiness checklist template

Draft one of these in the repo before each gate and review it with the user; the gate is passed when every box is checked or waived with an owner.

# [Product] — [EVT|DVT|PVT] Build Readiness — [date]

## Design
- [ ] CAD frozen at rev ___; STEP + drawings exported to /releases/[rev]/
- [ ] Schematic/layout frozen; ERC and DRC clean (attach reports)
- [ ] Open issues from previous build dispositioned (list waivers below)

## Supply
- [ ] BOM availability re-checked within 7 days; no EOL/NRND lines
- [ ] Long-lead parts (>8 weeks) ordered or in stock at CM
- [ ] Single-source risks listed with mitigation

## Build
- [ ] Quantity ___ agreed; PO issued; build dates confirmed
- [ ] Test plan written: what each unit must pass before it counts
- [ ] Fixtures/test jigs ready (DVT+); golden sample signed (PVT)

## Compliance & downstream
- [ ] Cert status: pre-scan / formal testing / certificates on file
- [ ] Packaging rev ___ ready (DVT+); ISTA drop test passed (PVT)
- [ ] QC plan + AQL levels agreed with factory (PVT)

## Waivers
| Item | Risk | Owner | Close by |
|---|---|---|---|

Common failure modes

• Designing for the prototype process. A beautiful 3D-printed enclosure with zero draft, undercuts everywhere, and 4 mm walls is not "almost ready for molding." Pick the production process in step 2 and CAD to it from the start.
• Freezing layout before the BOM is buyable. A single out-of-stock IC with a 40-week lead time stalls the whole program. Audit availability before layout is final and again before every build.
• Skipping EVT because "the prototype already works." One working bench unit proves nothing about twenty; EVT exists to find the failures that only show up in a population.
• Treating MOQ as a wall instead of a price curve. Ask for quotes at multiple volumes; almost everything is negotiable against the amortization math.
• No golden sample. Without a signed reference unit, every cosmetic dispute with the factory becomes your word against theirs — and you lose, because they already built the units.
• Certifying after tooling. An EMC failure found at PVT means board respins against a frozen enclosure. Pre-scan at EVT while everything can still move.
• Quoting factory cost as unit cost. Landed cost includes tooling amortization, freight, duty, inspection, and yield loss. Products die on the difference.
• Answering channel and pricing questions from here. Retail strategy, subscriptions, and D2C funnels for a connected product belong to d2c-smart-products — load it with skill_view rather than improvising.