jamro / verify-jlcpcb-bom

Verify JLCPCB BOM

Report only. Do not edit the schematic, PCB, BOM, or footprints.

hardware/boards/<board>/production/ is gitignored. Read bom.csv from disk. Do not infer nets from schematic geometry.

Run

Execute this script (do not reimplement parsing or the JLCPCB request):

python3 .cursor/skills/verify-jlcpcb-bom/scripts/verify_bom.py \
  --board main-control-board

--board accepts main-control-board or hardware/boards/main-control-board.

If the user did not name a board, run it for the only hardware/boards/*/production/bom.csv. If several exist, ask which board.

Missing production/bom.csv: tell the user to generate it with KiCad's Fabrication Toolkit. Do not invent a BOM.

How to read the report

The script prints FAIL lines, then WARN lines, then one PASS N line. N is BOM rows with no fail. Exit code 1 means at least one fail.

• Fail — wrong or missing LCSC id, BOM row disagrees with the schematic (designator, value, LCSC, footprint), schematic footprint was not pushed to the board, JLCPCB package is not in the footprint name, or a passive value disagrees with the catalog Resistance / Capacitance / Inductance.
• Warn — DNP part still in the BOM, JLCPCB stock is 0, part is extended (expand), schematic value is a label that does not contain the manufacturer part number, or a project-library footprint pad count disagrees with a catalog pin count.
• A lookup failure is a fail. It means package and value were not checked. Do not fill those in from memory.

Quote the script output. Do not dump the JSON response. Do not re-fetch each part unless the script could not reach JLCPCB.

After the quote, interpret every FAIL and WARN. Group lines that share one cause. Do not stop at the raw script text.

Interpret

Verdict for each line: real, false positive, or info.

A lookup failure is neither a design bug nor a false positive. For refs whose lookup failed, say the catalog was not reached and skip package and value claims. Continue interpreting findings for all other refs.

Package line is a name search

package … not in footprint … means JLCPCB componentSpecificationEn was not found as a substring of the footprint name. The script lowercases, then strips spaces, _, -, and =. It skips only these generic tokens: smd, tht, plugin, dip, radial, axial, chip, connector, throughhole, through-hole. It splits on ASCII , ; / only. A fullwidth comma ， glues the surrounding words into one token, so Surface Mount，Right Angle never matches a KiCad name.

Surface Mount is not in the skip list. It fails against footprints that say nothing, or only SMD.

Pitch P=1mm does not match the text P1.00mm (p1mm vs p1.00mm). Body 3x3 does not match L3.0-W3.0. Those are format misses, not evidence the land pattern differs.

False positive

Call the package line a false positive when every stated dimension agrees, even if the words differ:

Also a false positive when the footprint name contains the manufacturer part number (or the standard equivalent, such as 53261-0271 for a PicoBlade-style 1.25 mm header) and pitch, pin count, and orientation agree.

could not compare value and package … not comparable are inconclusive. Say what was not checked. Do not call them false positives.

Real package mismatch

Keep real when any of these disagree:

• Pitch (P1.25 vs P2.00, P0.50 vs P0.65)
• Pin count or row shape (1x02 vs 1x04, TQFN-16 vs TQFN-20)
• Body size that is not a zero/format difference (0603 vs 0805, 3x3 vs 4x4)
• Orientation (Horizontal / Right Angle vs Vertical / top-entry / straight)
• Mount (SMD vs THT / pin-header vertical)
• Connector family whose land pattern differs (PicoBlade vs PH, JST SH vs GH)

If the line is still ambiguous, read that ref's schematic footprint, value, and LCSC fields, and the footprint name on the board. Do not infer nets from coordinates. If it stays ambiguous, verdict uncertain and name the dimension that is missing. Do not downgrade that to a false positive.

Other lines

Recommend a fix

Recommend a fix only for real and for info the user would act on (stock 0, extended cost). False positives: no change.

Name the field and the direction. Do not invent a replacement LCSC id. If a different part is required, say to search JLCPCB for the same footprint, pitch, pin count, and value.

• Wrong passive value: set the schematic value to the catalog value, or replace the LCSC with the part whose Resistance / Capacitance / Inductance matches the schematic. Regenerate production/bom.csv.
• Wrong or missing LCSC: set LCSC Part # on the schematic symbol to the intended in-stock id, then regenerate the BOM.
• Footprint text or pad count disagrees with the ordered package: assign the footprint that matches that package, update the PCB from the schematic, regenerate the BOM.
• Stale BOM vs schematic: regenerate with Fabrication Toolkit. Do not hand-edit bom.csv.
• Extended or zero stock, and the land pattern is already right: optional search for a basic-library equivalent. No schematic edit until that part is chosen.
• Function label vs MPN: no rename.

Reply

Use this shape. Quote the script output unchanged. One block per cause, not per ref, when the cause is the same.

<script output>

## Reading
- **J1, J3 — false positive.** JLCPCB says right-angle SMD at 1.25 mm / 1.00 mm. Footprints say `Horizontal` and `P1.25mm` / `P1.00mm`. `Surface Mount，Right Angle` is one token because of the fullwidth comma, so the name search fails. Pitch and orientation agree. No change.
- **U3 — false positive.** `TQFN-16-EP(3x3)` is the same body as `TQFN-16_L3.0-W3.0-…-EP`. No change.
- **R1 — real.** Schematic 10k, catalog resistance 4.7k. Set the schematic value to 4.7k or pick the LCSC whose Resistance is 10k. Do not invent the id. Regenerate the BOM.
- **J1, U1 — info.** Extended library. No schematic change. Optional: search a basic-library part with the same footprint.

## Do next
Only real design or BOM changes. Omit this section when every line is a false positive or a no-change info.