Status + a re-scope of the backfill
Audited every board/molecule page on the wiki. The backfill as written in this issue is not executable — the premise that "--render-board works off the wiki source — no design machine needed" only holds for pages that actually have EDA source published, and most of them don't.
What's actually on those ~18 pages
| page |
render/ |
EDA source on page |
renderable? |
adom/bme690-molecule |
yes |
.kicad_pcb + .kicad_sch |
✅ done |
adom/bmi270-molecule |
yes |
.kicad_pcb + .kicad_sch |
✅ done |
adom/bmv080-molecule |
yes |
.kicad_pcb + .kicad_sch |
✅ done |
adom/bma580-molecule |
yes |
.kicad_pcb + .kicad_sch |
✅ done |
adom/usb3-to-ethernet-dataonly-molecule |
yes |
full KiCad project |
✅ done |
cosmiic/502535-lipo-molecule |
yes |
none (render pushed from elsewhere) |
n/a |
cosmiic/adipose-molecule |
yes |
none |
n/a |
cosmiic/pm-coil-molecule |
yes |
none |
n/a |
cosmiic/cosmiic-biopotential-amplifier-eval-board-465000 |
no |
Altium .zip |
❌ Altium not supported |
cosmiic/cosmiic-pulse-generator-eval-board-545000 |
no |
Altium .zip |
❌ Altium not supported |
cosmiic/cosmiic-power-module-eval-board-357000 |
no |
KiCad project .zip |
⚠️ needs unpack + re-push |
adom/7in-display-moleculte |
no |
gerbers + PNGs only |
❌ no source |
adom/bmv080-i2c-mpin-molecule |
no |
gerbers + PNGs only |
❌ no source |
adom/capacitor-molecule-panelized |
no |
gerbers + PNGs only |
❌ no source |
adom/capacitormolecule2 |
no |
gerbers + PNGs only |
❌ no source |
adom/capacitormoleculepanelized2 |
no |
gerbers + PNGs only |
❌ no source |
adom/coral-breakout-stand-molecule |
no |
gerbers + PNGs only |
❌ no source |
adom/ina219-breakout-board |
no |
gerbers + PNGs only |
❌ no source |
adom/molecule-rp2350b-pican2b-v1-318000 |
no |
gerbers + PNGs only |
❌ no source |
adom/moleculentx15r |
no |
gerbers + PNGs only |
❌ no source |
adom/naneye2d-control-molecule |
no |
gerbers + PNGs only |
❌ no source |
adom/pico2-2w-molecule |
no |
gerbers + PNGs only |
❌ no source |
adom/rp2040-devboard-molecule |
no |
gerbers + PNGs only |
❌ no source |
adom/vl53l4cdv-molecule |
no |
gerbers + PNGs only |
❌ no source |
adom/coldfire-xilinx-dev-board |
no |
gerbers .tar.gz + GLB |
❌ no source |
adom/rp2040-test-board |
no |
GLB only |
❌ no source |
So: ~14 of these are Fusion-360 exports that only ever published gerbers/PNG/BOM/CPL, never a .kicad_pcb/.kicad_sch. Two more are Altium-only. There is nothing for --render-board to consume. This is a missing-source problem, not a pipeline problem — it needs the original designs re-exported from a design machine (Kyle for the Fusion set, COSMIIC for the Altium set), which is the opposite of this issue's "no design machine needed" assumption.
Split out as a separate issue so it doesn't block the deprecation half.
Backfill actually done
Re-ran --render-board on all 5 pages that do have KiCad source. Two real defects fixed beyond just "has a render/ folder":
bmi270-molecule was serving a stale render — manifest.current was 1.0.2 while the published package.json was 1.0.3. Now 1.0.3 with 2 archived versions.- 4 of 5 were still on flat v1 manifests (
bme690, bmv080, bma580, usb3-…), so APM's version selector had nothing to fall back on but single-current. All 5 are now v2 with current == published version and a populated versions[].
Worth noting the v1→v2 upgrade only happens when something re-runs --render-board; pages rendered before APM 0.4.0 stay flat until touched. Any page not in the renderable list above is still flat.
Deprecation — prepped, not pushed
molecule-publish 2.1.2 already covers everything publish-board did, so it's ready to retire. One dangling thread: molecule-publish Stage 2.5 still points at skills/publish-board/kicad_board_bundle.py on the adom-project-manager page — i.e. the live skill sends you into the skill we're deprecating. That needs to move before publish-board goes away.
I have both edits ready but haven't pushed to adom/molecule-publish since it's your page, @drew:
publish-board/SKILL.md → deprecated: true, user-invocable: false, superseded-by: adom/molecule-publish, description rewritten, and a warning banner at the top with a port table. Body kept for reference.kicad_board_bundle.py → copied into molecule-publish as skills/kicad_board_bundle.py, with Stage 2.5 rewritten to call it in-place (plus the --normalize invocation, which Stage 2.5 never actually spelled out) and a note that publish-board is deprecated.
Say the word and I'll push #1 to the APM page and hand you #2 — or take the diff and land it yourself with the next molecule-publish cut.
The three sanity-check decisions
All three are already documented in 2.1.2 — confirming so they can be closed out:
- EAGLE render skip — Stage 3.5: EAGLE/Fusion
.brd/.sch report "nothing rendered" and skip gracefully, APM shows a static version label. ✅ agreed.
- deps install side-effect — Stage 2.5: deps are real registry deps, so
pkg install <board> pulls the component pages too. ✅ agreed, correct tradeoff.
- render refresh cadence — Stage 3.5: re-run on every content update, after
pkg publish. ✅ agreed. The bmi270 stale-manifest case above is exactly what skipping it looks like.
Acceptance, restated