Improvement backlog 25-Aug-26 (v0.8.0)

[neusse]

Konnect improvement backlog — 25-Aug-2026 (v0.8.0)

Disclosure. I posted this Codex-assisted evaluation after reviewing the
Konnect v0.8.0 release and source, the current roadmap and contributor rules,
every open issue, every open pull request, relevant closed/merged work,
maintainer responses, project discussions, and our actual end-to-end KiCad
use with the version-matched konnect-codex plugin. The priorities below are
recommendations, not maintainer assignments. Before implementing an item, I
will claim it on the issue, agree on the design, and follow the roadmap and
contribution process with one focused PR.

Snapshot

• Release reviewed: v0.8.0,
  tag commit dee8a27.
• Current upstream main: c68c745,
  one metadata-only PCM packaging commit ahead of the tag.
• Release delta: 98 commits and 61 changed files since v0.7.0.
• Live inventory at this snapshot: 32 open issues and 13 open pull requests.
• Tool surface: 19 toolsets, 204 registered tools, 210 total.
• Companion reviewed and installed: konnect-codex v0.8.0 companion revision 1.

Executive assessment

v0.8.0 is a major reliability release. It closes 13 items from the 21-August
backlog, repairs legacy footprint corruption, fixes repeated board-outline
corruption, replaces move_connected false success with an honest refusal,
adds catalogue-wide schema parameter checks, and moves several PCB reads and
zone creation onto live IPC.

The top risk has changed. #326
now belongs at P0 because create_netclass can omit wire_width from the
Default netclass, causing Eeschema to suppress or strip junction dots and
potentially change project-wide connectivity. The v0.8.0 shared connectivity
index also remains bus-blind (#328), and add_bus_entry reports its two
endpoints backwards (#329). These are small-looking defects with outsized
agent-workflow consequences.

Released progress since v0.7.0

• #294 adds a dry-run-first,
  revision-gated repair_corrupted_footprints tool, proven against a real
  68-footprint board with 848 phantom pads.
• #314 fixes
  set_board_size appending overlapping Edge.Cuts on every call.
• #315 stops
  move_connected from claiming it preserved wires when it did not. The real
  connected-move capability remains open and depends on Component and label moves don't add or prune junction dots #120.
• #285 lands the Tools accept parameters they never apply and report them as honoured #251
  parameter-honesty series and a catalogue-wide CI guard.
• #307–#313 and
  #317 fix idempotent edits,
  sheet movement, duplicated UUIDs, text justification, project library-table
  targeting, generated-pin positions, SVG return, and invalid sheet-pin syntax.
• #323 gives schematic
  connectivity tools one shared index; fix(sch): resolve power symbols in the schematic net graph #262 resolves power symbols and fix(schematic): respect intentional unconnected pins #267
  respects intentional no-connects.
• #316 creates/refills zones
  through live IPC when possible and reports explicit live/file provenance.
• #207 reads board state and
  pads from the live board; fix(pcb): report a User paper size's dimensions in get_board_info #324 adds custom User paper dimensions.
• #327 uses the local JLCPCB
  catalogue Datasheet column before network lookup.
• #306 ships the approved
  developer architecture, tool, integration, testing, and release guides.

P0 — correctness and non-destructive behavior

1. Preserve the complete Default netclass — #326

create_netclass omits wire_width,
and an incomplete Default netclass can disable or strip schematic junctions.

Recommended implementation:

• serialize every KiCad Default netclass field, including wire_width;
• use KiCad defaults when optional caller values are absent;
• round-trip a real project through KiCad and prove the full field set survives;
• verify T-junction connectivity and ERC before/after the mutation;
• add a regression test that fails if any required Default field disappears.

Until fixed, workflows should not create or overwrite Default without explicit
round-trip evidence.

2. Make schematic moves connectivity-safe — #120 and #315

#120 is the underlying
junction/connectivity problem; #315
tracks the missing wire-carrying move. v0.8.0 correctly refuses instead of
lying. PR #330 is a useful,
green junction-on-move prerequisite, not a complete move_connected solution.

Land this as staged, testable work: reconcile affected junctions first, then
implement explicit wire stretching/shrinking, rotation and delete behavior,
with KiCad ERC and serialized connectivity fixtures at every stage.

3. Refuse stale-file mutation after live IPC loss — #240/#241

#240 remains the central
wrong-state hazard: after a tool has observed an open document, IPC loss must
not silently fall back to editing a stale file. #241
should supply the reusable document-answering IPC test harness.

Every mutator should bind to the requested document identity, report its
source, and fail closed if a formerly live document becomes unavailable.

4. Finish unit-aware mutation — #182

The shared connectivity index improves reads, but mutation is still incomplete.
Rebase PR #273 over #323 and
prove multi-unit placement, movement, rotation, deletion, annotation, pin lookup,
and connectivity without duplicate references or cross-unit corruption.

5. Bound project-root and library discovery — #189

#189 can still select an
unrelated ancestor project or library table. Search must stop at an explicit
project boundary, expose candidates, and refuse ambiguity rather than choosing
the first match.

6. Verify every reported artifact — #252

#252 remains open even after
the schema-honesty work. Every snapshot, export and manufacturing response must
verify the requested output path, existence, nonzero size, format signature,
board/revision identity, and per-artifact failure state.

P1 — high-value workflow reliability

Bus and response correctness — #328/#329

• #328: make the shared
  connectivity index understand buses, bus entries and bus-attached labels.
  Until then, KiCad ERC is authoritative for bus sheets; agent guidance must not
  “repair” correct geometry from a Konnect-only orphan result.
• #329: return bus_side and
  wire_side from the endpoints actually written. Add all-orientation tests and
  require response geometry to equal serialized geometry.

Lifecycle and client reachability — #103/#233/#242/#325

• #103: own server children,
  track multiple instances, and remove orphaned processes safely.
• #242: MCP startup must not
  reinstall guidance after an explicit uninstall. Use an explicit-init model or
  a durable tombstone distinct from “never installed.”
• #233 and
  #325: support clients that
  cap tools or do not refetch after dynamic loading. The roadmap's compact
  surface plus tool-directory resource is the right direction. Codex remains
  functional through eager toolsets and does not need a proxy today.

Board identity, platform and mutation completeness

• #256: opening a requested
  PCB must prove that exact document became active.
• #257: remove KiCad 11 SWIG
  assumptions and move necessary DSN/SES/editor operations toward supported IPC.
• #231: provide a live,
  verified Update Footprints from Library path; rebase feat(pcb): update placed footprints from libraries #232 and repeat real-KiCad evidence.
• #254: discover supported
  per-user Windows KiCad installations without a hard-coded Program Files path.
• #258: support explicit
  custom-field upsert in batch schematic edits.
• #291: honor or reject the
  requested schematic SVG filename instead of silently changing it.
• #119: bound, categorize and
  preserve complete DRC reporting.
• #221: make live-CI claims
  match actual coverage and fix the rotation read-back race.

Freerouting bridge

Konnect v0.8.0 intentionally has no autoroute tool. The roadmap now adopts the
standalone-JAR DSN/export-route-SES/import bridge we recommended: distinguish
engine detection from bridge availability, bind the exact board, import
atomically, and verify placement/inventory/unrouted/short/DRC results. The
version-matched konnect-codex plugin already supplies a non-overwriting bridge
and remains an evidence source, not a Rust implementation dependency.

P2/P3 — bounded enhancements and maintenance

• #84 — finish structural
  replacement of indentation-sensitive schematic scans. P1 because it
  underpins correctness, but stage it behind active P0 mutations.
• #118 — real layer-aware 2-D board plot.
• #131 — sign/notarize both macOS slices and final artifact. P1 release work.
• #154 — Homebrew after stable signed artifacts.
• #181 — preserve lock-name compatibility before sha2 bump.
• #210 — reduce whole-sheet serialization diff churn.
• #225 — select footprint graphics by identity, not only layer.
• #226 — resolve placed metadata fidelity from measured impact.
• #241 — reusable open-document refusal mock; P2 support for P0 Closed-board fallback cannot tell "KiCAD was never running" from "KiCAD just died holding this board" #240.
• #296 — advanced symbol/footprint generation as agreed focused PRs.
• #305 — add or deliberately route placed-footprint 3-D model editing.

Completed since the 21-August backlog

Issue	Released resolution
#219	User paper dimensions through #324.
#234	Required nested fields and missing-coordinate refusal through #268/#318.
#251	Parameter-honesty consolidation and CI guard through #285.
#255	Local-catalog datasheet lookup through #327.
#286	Text justification through #308.
#287	Sheet captions and pins move with the sheet through #309.
#288	Fresh nested UUIDs on duplication through #310.
#289	Correct project library-table targeting through #311.
#290	Actual SVG result returned through #313.
#292	Idempotent edits through #307.
#293	Resolved generated-pin geometry through #312.
#303	Valid sheet-pin rotation through #317.
#304	Pre-6.0 footprint upgrade hint through #319.

Open pull-request assessment

“Conflicting” means the branch needs reconciliation; it is not a quality
verdict. Green status is recorded at this snapshot.

PR	State	Backlog assessment
#176 reload server	Conflicting, stale	Supersede or redesign for #103/#242 rather than reviving broad old behavior.
#232 live footprint update	Conflicting	Rebase and repeat real-KiCad preservation evidence.
#243 setup-python v7	Mergeable, green	Small CI dependency update.
#264 placement batch	Conflicting	Still useful after design confirmation and rebase.
#265 pad-read hardening	Draft, conflicting	Rebase only its remaining delta over merged #207.
#268 nested validation	Conflicting	Reduce to malformed-input coverage not already landed in #318/#285.
#270 artifact verification	Conflicting	Rebase and narrow to unresolved #252 behavior.
#273 multi-unit mutation	Conflicting	Rebase over #323; remains the main #182 implementation path.
#275 symbol bounds	Conflicting	Rebase after unit-aware work.
#320 viewer UUID update	Mergeable, green	Bounded dependency update.
#321 workspace UUID update	Mergeable, green	Bounded dependency update.
#322 structural library fields	Mergeable, Windows failing	Useful read path, not ready until Windows is green.
#330 junction reconciliation	Mergeable, green	High-priority partial #120 prerequisite; remove stray artifacts and preserve its explicit partial scope.

Complete open-issue disposition

Issue	Priority	Current disposition
#84	P1	Finish structural schematic parsing in staged conversions.
#103	P1	Server ownership and multi-instance lifecycle.
#118	P2	Layer-aware 2-D board plot.
#119	P1	Bounded, complete DRC output.
#120	P0	Junction-safe move/delete; #330 is a partial prerequisite.
#131	P1	macOS signing/notarization.
#154	P2	Homebrew after signed artifacts stabilize.
#181	P3	Preserve lock-name compatibility before dependency bump.
#182	P0	Rebase and finish multi-unit mutation.
#189	P0	Bound root/library discovery and refuse ambiguity.
#210	P2	Reduce serialization diff churn.
#221	P1	Fix live-CI claims and rotation race.
#225	P2	Select footprint graphics by identity.
#226	P3	Resolve placed metadata fidelity with evidence.
#231	P1	Live Update Footprints from Library.
#233	P1	Client tool-list refresh/cap compatibility.
#240	P0	Refuse stale fallback after observed-live IPC loss.
#241	P2	Shared document-answering IPC test harness.
#242	P1	Remove startup guidance mutation.
#252	P0	Verify every reported artifact.
#254	P1	Per-user Windows KiCad discovery.
#256	P1	Open and prove the requested PCB document.
#257	P1	KiCad 11 IPC/SWIG transition.
#258	P1	Explicit custom-field upsert.
#291	P1	Honor or reject requested SVG filename.
#296	P2	Focused advanced symbol/footprint generation.
#305	P2	Route placed-footprint 3-D model editing.
#315	P1	Implement real connected move after #120.
#325	P1	Compact surface/resource directory for capped clients.
#326	P0	Preserve Default netclass and schematic junctions.
#328	P1	Make connectivity bus-aware.
#329	P1	Correct bus-entry response endpoints.

Recommended execution order

1. Claim and fix create_netclass omits wire_width, which silently disables junction dots across the whole project #326 with a full KiCad Default-netclass round trip.
2. Review/land the bounded part of fix(sch): re-judge junction dots after a move #330, then continue Component and label moves don't add or prune junction dots #120/move_connected leaves every wire behind and reports success #315 in focused PRs.
3. Implement Closed-board fallback cannot tell "KiCAD was never running" from "KiCAD just died holding this board" #240 with reusable refuse_if_board_open_in_kicad's refusal branch is untested: no mock can answer GetOpenDocuments #241 IPC document fixtures.
4. Rebase fix(schematic): update every placed component unit #273 for sch_batch, sch_export, sch_analysis and design_review still resolve pins against unit 1 #182, then address project_root_for's ancestor walk is still unbounded when the file's directory has no table and no project #189 and [BUG] snapshot_project returns artifact paths it never verifies #252.
5. Fix The shared connectivity model is bus-blind: wire ends on bus entries and labels on buses are reported floating #328/add_bus_entry's response swaps bus_side and wire_side #329 before trusting agent-driven bus cleanup.
6. Implement the roadmap Freerouting bridge with exact-board and import evidence.
7. Address Plugin leaks orphan konnect server processes: PID file only tracks the last one #103/[BUG] MCP startup silently reinstalls guidance after uninstall #242 and compact client discovery (Linux OS issue with loading the sch_components toolset via load_toolset #233/Issue: VS Code Copilot only exposes a subset of Konnect tools as callable #325).
8. Rebase, narrow or close stale PRs before adding more overlapping work.

Release and benchmark follow-through

The version-matched konnect-codex v0.8.0 plugin has been reviewed against all
17 upstream guidance files. It carries the three upstream skill changes plus
temporary safety gates for #315/#326/#328/#329, preserves the companion
Freerouting bridge, and removes the nonexistent Konnect autoroute hook.

The remaining evidence step is a v0.8.0 end-to-end benchmark that includes:

• structured schematic SVG return and visual readability inspection;
• a Default-netclass/junction probe for create_netclass omits wire_width, which silently disables junction dots across the whole project #326;
• a bus sheet that exposes The shared connectivity model is bus-blind: wire ends on bus entries and labels on buses are reported floating #328/add_bus_entry's response swaps bus_side and wire_side #329 until fixed;
• exact-board live reads and live zone creation;
• local-catalog datasheet resolution;
• footprint-corruption repair dry-run/apply evidence;
• transfer inventory, placement image, Freerouting provenance, route-import
  inventory, direct ERC/DRC, 3-D review, and manufacturing artifacts.

This backlog should be refreshed after that benchmark or whenever a new minor
release changes the issue/PR inventory.

[mixelpixx]

This backlog revision is, frankly, a co-maintainer's document, and it's now load-bearing: ROADMAP.md has been rewritten around it — seven themes in dependency order, crediting this discussion and your benchmark work directly. Corrections welcome where I've misread your intent.

Status against your execution order, as of tonight (a lot happened today):

Landed since your snapshot: #303 (via #317), #288 (#310), #292 (#307), #286 (#308), #287 (#309), #289 (#311), #293 (#312), #290 (#313), #304 (#319), #219 (#324), #255 (#327), your #306 docs set, the #251 consolidation through #285 exactly as you prescribed ("avoid an uncoordinated cascade" — it caught its first real bug within an hour of merging, in my own code), #267, #316 (add_zone over IPC, live-verified), #318's half of #234, #207, and #323's shared connectivity index (which supersedes the #272 half of #182; #273 rebases onto it next). Also #294 merged after the verification you'd want: I built the actual v0.6.1 binary from its tag, used it to corrupt a 68-footprint board live (848 phantom pads, zero graphics — the true #244 damage, not a fixture), then ran the repair against it: all 68 detected, repaired in one undo commit, and the file KiCad saved matches the pristine original per reference.

Two more from your P0 list went down today outside the PR queue: #314 (set_board_size had appended a second outline on every call since the first release — both paths now replace, non-segment outlines refuse) and #315 (move_connected was a stub reporting success; it now refuses honestly until the #120-dependent implementation exists).

Your pick. You said you'd take one approved item, design-first. Everything on your P0 list is now either merged or claimed except: the Freerouting bridge (your own #253 design is written into the roadmap as the agreed direction), #120 (connectivity on move — which also unblocks the real move_connected), #240/#241 (stale-board evidence + the document-answering mock), and #189 (bounded root walk). All four are yours for the choosing; if you want my honest preference, the Freerouting bridge is the one nobody else can do as well as you, and #295's skill-layer work unblocks the moment v0.8.0 tags. But it's your pick, and design-first on the issue works for me exactly as you proposed.

And the re-run: v0.8.0 tags shortly with all of the above. A konnect-codex benchmark against it would be the most valuable QA event this project gets — your last two runs found what 900+ tests couldn't.