Konnect (J3Techs) errors
8 recurring problems engineers report in Konnect (J3Techs), ranked by impact.
#54 by errors trackedECAD
Recurring Konnect (J3Techs) errors, by impact
- 1A Konnect user's agent built a schematic sheet as rows of parts with a net label on every pin and no wires. It netted up, passed ERC, and nobody could read it. The rule against exactly that already existed in a bundled skill, but the agent never saw it, because the skills were installed into one config directory and the session ran from another. The fix moves the rule into the server's own initialize instructions and flags label-only two-pin parts.ECAD · Agent failure · 1 report74
- 2KiCad 10's IPC Pad message has no field for pad fabrication type, so heatsink, castellated, BGA, fiducial, test point and mechanical all go missing, and any live update that sends a whole footprint back re-creates its pads without that property. Konnect had a guard for this, but three tools sent whole footprints without running it, including the footprint repair tool.ECAD · Agent failure · 1 report74
- 3PreparedFootprintPlacement sent pads, drawings, fields and library data, but not the library footprint's child zones; place_component_array, duplicate_component and update_pcb_from_schematic lost them the same way. It was confirmed live: a Tag-Connect footprint placed through Konnect saved with no keepout. KiCad's API can create the zone, but the rule area message has no field for connect_pads, min_thickness or fill, so the zone gets the API's defaults and kicad-cli pcb drc then reports lib_footprint_mismatch. The fix creates the zones it can describe and refuses the placement for any it cannot.ECAD · Agent failure · 1 report74
- 4A user reported apply_route_plan refusing a plan the native router had just written. Reading the code showed three gaps: the optimizer accepted bends down to 89 degrees while apply linted at 90, the optimizer read a joint with the board's tracks as context while apply read the plan file alone, and with optimize off, raw routed copper reached the plan with no lint at all. The fix puts one limit and one verdict function in both places, with a test that fails if they diverge.ECAD · Agent failure · 1 report66
- 5After one routing job on a dense board, the remaining nets had to be routed by hand, because later jobs left connections unreachable however they were set up. The router treated every existing track as a hard obstacle, could not tell a locked track from an unlocked one, and treated filled outer-layer pours as walls.ECAD · Agent failure · 1 report66
- 6An earlier Konnect left footprint child zones away from their footprint. update_footprints_from_library and repair_corrupted_footprints only work with KiCad running, and the first refuses any library footprint with a zone. The PR adds a dry run that reports each zone as in place, misplaced or not in the library, then rewrites only the misplaced outline and checks every other byte is unchanged.ECAD · Agent failure · 1 report66
- 7Flips, footprint metadata, net classes and design rules only work on a closed board, while other operations need it open, so KiCad tools that need both meant asking the user to save, close and reopen for each step. KiCad's IPC has no close request, and the quit call either does nothing or quits all of KiCad. The new tool saves over IPC, posts a window close, runs the closed-board edit, and relaunches the editor.ECAD · Agent failure · 1 report66
- 8The plan_plane_vias routine rejected nearly every via site for a net that pours on several layers, and the only way through was to delete the outer pours, plan, and add them back. The router treated every filled pour as an obstacle, so connections under an outer pour failed, and with fills missing it routed GND and power as ordinary tracks. Pours had to be deleted before routing in PR #39 too.ECAD · Agent failure · 1 report66
How it fails
Sourced from public community reports. Costs and delays on individual entries are modeled estimates, not measurements.