KiCad - the KiCad Bridge

Motion gate: GetForegroundWindow can lag AD's z-order right after a raise, which aborted the symbol beat's motion at its first gesture on John's laptop while AD had already measured the window on top; the gate now waits up to 1.5 s for the OS to agree. Panel record keeps the callback URL across stage-advance shows.

Tour verification (John: 'tool your bridge code to definitively know that the correct thing is showing'): every beat now MEASURES, after its raise, that its own window is the top of AD's z-order, that the live title carries the expected document, that the GL canvas is rendered (kicad_ui canvas probe) and that two captures 0.8 s apart are identical (settled). The result rides on the step and on progress as ; motion runs only on a verified window, otherwise the step says exactly which fact failed and which window was actually on top. Zoom discipline: 2D beats start and end on extents with two wheel clicks in and back out; 3D beats zoom one menu step in and out and end on Zoom to Fit while staying isometric.

Tour 3D motion is now deterministic: KiCad's own View menu commands (Rotate X/Z by 10 degrees, Zoom In/Out, Zoom to Fit) resolved from the live menu bar and posted as WM_COMMAND, so the isometric and the orbit are the same on every DPI and need no mouse or foreground; mouse drags remain only as the fallback (they flipped the chip viewer and did nothing in the board viewer on 2026-09-03). Each step in the result now carries the window handle and title it drove.

Tour window targeting (John's laptop: fourteen KiCad windows from days of demos; the 2D beat showed an empty untitled PCB Editor and the 3D beat the old sample cube): every beat now targets the tour's own document titles (esc-g431 schematic and board, the Adom STM32 symbol, the LQFP footprint), reuses an existing window instead of opening a duplicate, and opens each 3D view from that exact editor (open_3d_viewer accepts pcbHwnd/fpHwnd). The clean slate at start also closes old adom-demo documents and an empty untitled PCB Editor.

Tour identity fix (John's dashboard run, 2026-09-03): ab refused the demo panel, every caption, every raise and every window close from the tour thread with 'carries no caller identity' (toast fallback, no captions, nothing in front). Every loopback call the bridge makes now carries identity: the X-Adom-Caller headers of the request being served, or the bridge's own self caller otherwise; the tour thread and ab's callback thread adopt the identity of the request that started the tour. Also: a failed panel show no longer demotes later tours to the toast (per-tour reset, one retry), and every 3D viewer plus the Footprint Editor are closed before the beats so the footprint beat opens a fresh, re-indexed editor.

Tour sync: the verb-side control channel (kicad_demo {control}) is cleared at tour start too, so a pause/next/exit queued against a previous tour never steers a new one; completes the stale-press drain of 0.9.302.

Tour sync fixes (John's run on 0.9.301): stale panel presses from the previous show are drained before a tour starts (his earlier NEXT presses made the new tour jump to step 3 at once); the narration mute is reset from the call's muted flag at thread start and pushed to the panel on the first show; STOP from the remote sets stopping on progress immediately so the dashboard can close its own pill; progress carries stopping.

Taskbar hygiene (John: frozen progress bars on every KiCad window, no Adom badge): the boot sweep that clears leftover bars ran once at +6 s while ab was still restarting and failed silently; it now retries across the first two minutes until a window answers, and a five-minute idle reaper clears progress and badge on every KiCad window whenever the bridge is driving nothing. Filed adom/adom-bridge#138 for a progress TTL on ab's side.

Tour 3D motion recalibrated: the orbit drags totalled about a third of the viewer width and turned the board nearly edge-on (KiCad's trackball is roughly 0.1 degree per pixel). Now ~30 degrees of tilt and ~40 of yaw for the isometric, then three gentle orbits with zoom, slower and smoother for the recording.

Tour board-in-3D fix: KiCad's 3D viewers are unowned top-level windows, so picking 'the viewer owned by the PCB Editor' never matched and the chip's viewer was never closed (0.9.298 test-box run: board3d failed with the viewer actually open). The beats now take the viewer that APPEARS after their open, the chip's viewer is remembered and closed by handle before the board's opens, with the topmost viewer as fallback.

Fix: the 0.9.297 tour job died on a NameError (_EXIT undefined) before its first progress update, so Play the demo showed the dashboard's transport pill and nothing else (John, lost, 2026-09-03). _EXIT is defined and the job's reset sits inside its try so any early failure lands on progress as an error instead of a silent stall.

Tour: motion on every beat for the recording (fit, wheel zoom in/out, middle-drag pans in the 2D editors; Home, an isometric left-drag and three orbits with zoom in the 3D viewers), sent only while the beat's window is confirmed in front and aborted at the next gesture on STOP or when the user switches away. The board-in-3D beat now closes the chip's 3D viewer first and picks the viewer owned by the PCB Editor (the last beat was showing the chip's viewer by title match). STOP from the remote's X now clears the caption and ends the tour within the beat, audio already off. Per-beat motion reports ride on the result.

Tour foreground fix (John: 'why do I not see you foregrounding any kicad window'): the beat raise called SetForegroundWindow from the bridge process and reported true regardless; Windows refuses that from a background process while the user works elsewhere. The bridge now lifts its no-activate mark, asks ab's desktop_bring_to_front to raise the beat's window once, and reports foregrounded only when the z-order afterwards shows that window on top.

The tour now walks a real Adom board, adom/esc-g431 (149 footprints, 227 symbols): its STM32G431 symbol and LQFP-48 footprint are lifted from the wiki files at tour time and installed into the Adom library, then symbol, footprint, chip in 3D, schematic, board 2D, board 3D in that order. Captions were never painting because ab requires reason on desktop_caption; fixed, medium size, one line per beat. Narration re-rendered to match. Wording no longer claims the background: each beat's window is brought to the user once. Dead local 3D model paths in the board are dropped.

Demo panel callback URL fix: the bridge read its bound port from a second copy of the server module and sent no eventCallbackUrl to ab (panel presses reached the tour only via poll). The port now travels through the environment, so every press is POSTed to the bridge the instant it is clicked. Verified on the test box: a press to the loopback route paused within the beat, muted at once, and resumed on play.

kicad_demo progress now reports panel (last show reply, stage, callback URL), muted, muteSource, lastPress and paused; each step in the final result carries foregrounded (whether the beat's window was brought forward once).

Demo panel: reason is sent on every show (ab marks it required on show; a stage-advance show without it was refused live and the remote stayed at 1/7), and every show's reply is recorded on kicad_demo progress (panel: shown, stage, callbackUrl, error) so a refused show is visible instead of silent.

kicad_demo adopts the ab 2.1.17 demo-panel contract: the first show passes eventCallbackUrl (loopback POST /demo/event on the bridge's own port) so a press lands the instant it is clicked; mute/unmute silence or replay the beat's narration at once; pause/stop silence at once and apply at the beat boundary; stage-advance shows send only the stage so a user's pause or mute is never undone; the poll stays as backstop and skips presses already delivered. The tour now foregrounds each beat's window exactly once (John: the user must see which window is being demoed, but keeps full control between beats); no persistent focus guardian during the tour. One short ab caption per beat stays.

kicad_demo drives ab's desktop_demo_panel (Adom Bridge 2.1.0+, wiki adom/adom-bridge#116): the floaty play/pause/next/back/stop remote is shown at stage 0 with the seven stops, polled between beats, stage and state updated per beat, hidden on done/exit/error; the toast transport remains only as the fallback for an older ab. Every beat paints one short ab caption (KiCad tour n/7: ...) for about the beat's length instead of a paragraph for a minute (John, 2026-09-03).

John, 2026-09-03: no caption on a Symbol Editor launch or rescan reopen (the 0.9.285 one was large, shown too long, and wrong on his laptop: it promised 'behind your window' while KiCad raised itself in front while loading). Launch is silent. _foreground now also reports editorInFrontAtEnd, measured from AD's z-order at return time, so a response never claims background about a window the user can see on top. Dev skill records the rule: one early look, no fighting the user for the foreground, captions only on the opt-in allowFocusSteal path and short.