Skip to content
This repository was archived by the owner on Nov 2, 2026. It is now read-only.

v0.10.1

Choose a tag to compare

@github-actions github-actions released this 29 Aug 00:01
· 527 commits to main since this release

A patch for two things v0.10.0 shipped wrong

Both were found by @neusse, who audited the
released v0.10.0 package rather than the source tree. Neither is a code
defect — both are things Konnect told you or your agent, and both were
wrong.

The bundled guidance wired an LED backwards

konnect init installs skills that Claude reads while designing. One of them,
wiring-patterns.md, gave a "known-good" LED pattern that reversed the part:
it fed the current-limiting resistor into "anode/pin 1" and grounded
"cathode/pin 2".

KiCad's standard Device:LED is pin 1 = K (cathode), pin 2 = A (anode) —
the opposite. An agent copying that pattern placed the LED backwards, and the
LED simply never lights.

Corrected, and the reference now states the real pin order and tells the agent
to confirm polarised parts with get_symbol_info instead of assuming an
order.

Eleven library identifiers did not exist

The quick-reference list of common KiCad symbols carried identifiers that do
not resolve in KiCad 10:

  • the four transistors were listed under Device: — they live in
    Transistor_BJT and Transistor_FET
  • Device:Ferrite_Bead has never existed; it is Device:FerriteBead
  • the MCP1700 and CP2102N entries used pre-KiCad-10 family names
    (MCP1700x-330xxTT is the SOT-23 part)
  • the duplicated table inside kicad-schematic/SKILL.md had additionally
    drifted from its own reference, disagreeing on the BJT suffix

An agent following one of these could not place the part at all.

Pin-numbering guidance is no longer stated as universal

The library skill previously gave a table of pin-order "conventions" —
IC counter-clockwise, diode pin 1 = anode, BJT 1=B/2=C/3=E. Those are not safe
general rules: physical lead numbers depend on the exact manufacturer part,
package suffix and the drawing's view direction, and KiCad's own LED symbol
contradicts the diode rule outright.

That table is now a short list of genuinely generic cases plus an explicit
verification contract: confirm the manufacturer part and package, note the
drawing view, reconcile datasheet leads against symbol pins against footprint
pads, read the result back, and refuse to use the part while any mirror or
ambiguity is unresolved.

.gitattributes (contributors only)

v0.10.0 added byte-preserving rules for KiCad-authored test fixtures but
replaced the repository's existing attributes instead of extending them,
dropping * text=auto and the explicit binary markings for PNG/ICO/ICNS.
Restored in #354. Fixture
byte-preservation is unaffected — the more-specific rules still win, and every
fixture blob hash is unchanged.

If you installed v0.10.0

Guidance is installed by konnect init, so it is on your machine until you run
it again. Run konnect init to pick up the corrected skills.

Then, if you built anything with agent assistance on v0.10.0:

  • Check any LED you did not place by hand. Confirm the cathode (pin 1, the
    K pin) goes to the lower-potential side. This is the one with physical
    consequences — a reversed LED does not light and is not caught by ERC.
  • Anywhere a transistor, MOSFET, ferrite bead, MCP1700 or CP2102N failed to
    place, the identifier was wrong, not your project.

Nothing was written incorrectly to your files by Konnect itself. No tool,
schema, or response shape changed in this release.

Guard

Both classes are now mechanically checked. Every Lib:Symbol identifier in
bundled guidance must resolve against the installed KiCad libraries, and the
existing sibling guard already required every tool name in prose to exist in
the registry. The new check is neuter-verified: reintroducing
Device:Q_NPN_BEC fails the build naming it.

Still open from the same audit

neusse's #357 and
#358 cover the larger half
— all nine bundled reference files are currently unreachable by progressive
disclosure (Claude is never told they exist), and the PreToolUse hook's
output does not reach Claude's context under the current hook contract. Both
are confirmed and open.


What's Changed

  • fix(repo): restore text and binary attributes by @neusse in #354

Full Changelog: v0.10.0...v0.10.1