Merge branch 'main' into validation/web-transcript-selection

This commit is contained in:
2026-11-05 16:35:06 +00:00
16 changed files with 396 additions and 28 deletions
+22
View File
@@ -34,3 +34,25 @@ The initial capture attempt used waitfor on a static heading; the kit does not
index that static label, although the saved PNG visibly showed the warning.
Final captures use screenshots and a real Copy Input interaction instead.
No physical device or Safari claim is made here. Nothing was deployed.
## Summary request ordering follow-up
The integrated application test exposed a real ordering dependency: the shared
PipelinesClient serializes each request until its reply. Definition validation
was enqueued immediately while its widget mounted, before the schedule section
had enqueued `list_schedules`, so a delayed or unanswered optional validation
held up primary Summary content.
Validation now starts in a microtask after synchronous Summary mounting. This
lets the schedule request enter the existing transport queue first, without
adding a timer or changing the client's serialization contract. The selected
source and gateway are captured before deferral, so a later selection cannot
change a queued request's source. The application fixture answers the new
validation request and asserts that schedules and launch controls are available
while validation is still unanswered. No selection, backend or transport code
was changed. Focused application/definition-warning tests and analysis are
recorded in the follow-up handoff; no full suite or deployment was run here.
Follow-up result: 149 application/definition-warning tests pass through the
memory watchdog (`/tmp/memwatch/out.35VDe8`, peak 2105 MB, 49 s). Focused analysis
is clean (`flutter-analyze.NVN15m`).
@@ -0,0 +1,26 @@
# Artifact browser interaction verification — 2026-11-05
Verified existing HTML artifact behavior in an isolated Chrome fixture built
from the warning-display branch, using `tool/capture/`. No production chat/run
was changed and no artifact-viewer source changes were needed.
At 1100×900: opened run 7 → Artifacts → release-dashboard.html, entered Full
screen, scrolled to Section 8 and the bottom heading, returned to the top, then
panned horizontally to the table's RIGHT EDGE column. Inspected all PNGs under
`/private/tmp/artifact-fullscreen-check/` (inline, fullscreen, bottom, right edge).
At 560×1000: entered Full screen, clicked Zoom in, asserted the sandboxed frame
has a scale transform, clicked Actual size, asserted that transform is reset,
and exited fullscreen. Inspected all PNGs under
`/private/tmp/artifact-zoom-check/`. Both capture scripts completed and cleaned up
their browser/server processes. The listing capture in
`/private/tmp/artifact-interaction-check/` was also inspected.
This proves desktop Chrome wheel scrolling/panning and the viewer's zoom
controls, including a narrow viewport. It does not prove physical touch/pinch,
Safari link/clipboard policy, or native-device behavior. Web artifact zoom uses
the existing buttons: the sandboxed document does not receive native pinch zoom.
No Android devices were attached (`adb devices -l` returned an empty list).
A real Safari fixture rendered, but further native automation failed with
`cgWindowNotFound`; Safari interaction acceptance remains unverified. The owned
loopback fixture server was terminated and awaited; no personal Fold was used.