Skip to content

Open-source AI browser agent for Chrome and Firefox (monorepo) 🧠

Ask eval: record prompt variants from the trace export; note the service-worker version trap

Claudebaf2954ff8e5
diff --git a/scripts/check-webgpu-ask.mjs b/scripts/check-webgpu-ask.mjs
index e02e50cbc..18e4fe9b9 100644
--- a/scripts/check-webgpu-ask.mjs
+++ b/scripts/check-webgpu-ask.mjs
@@ -14,6 +14,9 @@
// --extension DIR unpacked extension to load (default src/chrome). The
// extension ID, and so the model cache, follows the path:
// reuse one directory and profile to compare two builds.
+// Chrome keeps the registered service worker while the
+// manifest version is unchanged, so give the second build a
+// different version (for example append ".1").
// --model ID WebGPU preset (default Bonsai 2 2B Vision)
// --download download the model if it is not cached yet
// --tasks a,b run only these tasks
@@ -114,7 +117,9 @@ try {
const message = (action, data = {}) => harness.evaluate(({ action, data }) => chrome.runtime.sendMessage({ target: 'background', action, ...data }), { action, data });
await harness.evaluate(() => chrome.storage.local.set({
onboardingComplete: true, wbLocale: 'en', helpImproveWebBrain: false, agentAllowLocalNetwork: true,
+ tracingEnabled: true,
}));
+ result.extensionVersion = await harness.evaluate(() => chrome.runtime.getManifest().version);
result.adapter = await harness.evaluate(async () => {
const adapter = await navigator.gpu?.requestAdapter({ powerPreference: 'high-performance' });
return adapter ? { vendor: adapter.info.vendor, architecture: adapter.info.architecture } : null;
@@ -177,7 +182,14 @@ try {
.map(update => update.data.step)).size;
const answer = String(reply?.content || '');
const passed = !error && task.expect.test(answer) && !(task.reject && task.reject.test(answer.trim()));
- const entry = { name: task.name, seconds, steps, passed, answer, ...(error ? { error } : {}) };
+ // The trace export names each model request's prompt variant, which
+ // shows which profile actually answered (for example tiny_ask or ask).
+ let promptVariants = [];
+ try {
+ const exported = await message('export_traces', { tabId });
+ promptVariants = [...String(exported?.markdown || '').matchAll(/\bprompt ([a-z_]+)/g)].map(match => match[1]);
+ } catch { /* Tracing is diagnostic only. */ }
+ const entry = { name: task.name, seconds, steps, passed, promptVariants, answer, ...(error ? { error } : {}) };
result.tasks.push(entry);
console.log(JSON.stringify({ ...entry, answer: answer.slice(0, 160) }));
}