6756d11b5f
Full Chromium serves both headed and headless sessions. The separate shell duplicates the browser payload and is not needed for either mode. Remove the shell from PM and Docker. Select the managed Chromium executable for agent-browser and the full Chromium channel for direct Playwright callers. Route setup through PM and remove retired packages from cached bundle stores without changing the user's tool store. Update signing, architecture checks, launch probes and install guidance. Leave llama packages and Docker archive cleanup unchanged. Verification: - Real agent-browser navigation, clicks, DOM reads and screenshots pass in headed and headless modes with the same Chromium executable. - The direct Playwright doctor probe passes. - Focused Python and desktop packaging tests pass, as do six Docker checks and both real-browser task-scroll tests. - The built linux/amd64 image is 1.393 GB compressed, 223.6 MB smaller. - The broader PM suite and two unrelated setup tests still fail. Those failures reproduce on unchanged HEAD. - Five updated eval scripts parse; their full scenarios were not run.
Thread reading-intent probe
This renders the production Thread, real assistant-ui external-store runtime,
and production styles in Chromium, with deterministic synthetic transcript
records. No backend or provider is started. It is not a full native-app or
reporter-transcript reproduction.
From the repository root, copy fixture.tsx to
apps/desktop/src/scroll-campaign-probe.tsx and fixture.html to
apps/desktop/scroll-campaign-probe.html. Both are temporary files, not production
entries. Start Vite from apps/desktop with its normal config and a free loopback
port. Set a private Vite cache under a node_modules directory; a cache outside
node_modules can be transformed again by the React compiler.
Use isolated HOME and HERMES_HOME for the server and probe. Set
PLAYWRIGHT_BROWSERS_PATH explicitly if HOME changes Chromium's cache lookup.
Both root and apps/desktop dependencies must be available in a worktree.
Run under the campaign's shared test lock:
THREAD_SCROLL_OUTPUT=/path/to/artifacts \
THREAD_SCROLL_URL=http://127.0.0.1:18480/scroll-campaign-probe.html?thread \
PROBE_TAG=before \
node evals/desktop_bug_campaign/thread-scroll/probe.mjs
The probe records real scroll geometry for fresh load, wheel escape, A→B→A,
and a delayed empty→loaded refresh. It asserts preservation of distance from
bottom and a fresh session's tail position. Run against unchanged main first,
then the candidate in a fresh browser context. Each run writes <PROBE_TAG>.json.
The optional ?twins fixture mounts two independent production Threads with
different runtime IDs; it is useful for diagnosis but the automated probe above
expects one viewport. It does not substitute for a tiled pane-shell reproduction.
ownership-probe.mjs extends the same fixture (?ownership) with real
Thread remounts after changing the production profile and connection-scope
stores, a real document reload, and a partial transcript whose remaining
history is released with startTransition. The namespace controls use 12
messages to exclude virtualization estimation from that ownership assertion;
the hydration release expands to the original 200-message history. No DOM
geometry is overridden. Wheel input comes from Playwright's native input path.
The probe retains strict document-reload assertions. It also exercises the 200-message full remount/reload and live response growth after restoration. A continuation now observes late Markdown/intrinsic-row resizes until reader input or a live run takes over: 12-message reload drift changed from 17px to 0px, and the current 200-message reload control changed from 72px to 0px. Earlier historical long-remount drift was 323px; the current control retains its distance through remount, but this is not native-app or issue acceptance.
Remove the two temporary Desktop entry files after stopping the server.