Skip to content

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

Keep Cloud continuations tracked and report actual completion outcomes

Barack Sokullu8e32033ce193
diff --git a/package.json b/package.json
index d7ac613dc..f847e7148 100644
--- a/package.json
+++ b/package.json
@@ -16,7 +16,7 @@
"test:attachment-drop": "node test/attachment-drop.mjs",
"test:workflow-editor": "node --test test/workflow-editor.mjs",
"test:security": "node test/security/injection-corpus.mjs",
- "test:cloud-bridge-approval": "node --test test/cloud-bridge-approval.mjs test/cloud-conversation-isolation.mjs",
+ "test:cloud-bridge-approval": "node --test test/cloud-bridge-approval.mjs test/cloud-conversation-isolation.mjs test/cloud-run-outcomes.mjs",
"test:discord": "node --test test/discord-guard.mjs",
"test:toolbar-guard": "node test/rich-text-toolbar-guard.mjs",
"test:pdf-read": "node test/pdf-read.mjs",
diff --git a/src/chrome/src/agent/agent.js b/src/chrome/src/agent/agent.js
index a36ff7e79..c48f0833f 100644
--- a/src/chrome/src/agent/agent.js
+++ b/src/chrome/src/agent/agent.js
@@ -30997,7 +30997,8 @@ If the user has already named or confirmed this exact recipient, do NOT ask agai
}
async _scheduleAutoProgressResume(tabId, onUpdate = () => {}) {
- if (!this.scheduler) return null;
+ const cloudResume = this.cloudRunContexts.get(tabId)?.deferResume;
+ if (!this.scheduler && !cloudResume) return null;
const mode = this._effectiveRunMode(tabId);
if (!this._isActionMode(mode)) return null;
if (!this._shouldBlockDoneForProgress(tabId)) return null;
@@ -31005,16 +31006,17 @@ If the user has already named or confirmed this exact recipient, do NOT ask agai
const { tabUrl, tabTitle } = await this._getTabUrlTitle(tabId);
let result;
try {
- result = await this.scheduler.createResumeJob({
+ const args = {
+ after_seconds: 90,
+ reason: 'The active progress-ledger task hit consecutive stalled model outputs before finishing.',
+ resume_instruction: this._buildAutoProgressResumeInstruction(tabId),
+ };
+ result = cloudResume ? await cloudResume(args) : await this.scheduler.createResumeJob({
tabId,
conversationId: this.conversationIds.get(tabId) || null,
resumeTaskId: this._resumeTaskId(tabId, { create: true }),
mode,
- args: {
- after_seconds: 90,
- reason: 'The active progress-ledger task hit consecutive stalled model outputs before finishing.',
- resume_instruction: this._buildAutoProgressResumeInstruction(tabId),
- },
+ args,
currentUrl: tabUrl,
currentTitle: tabTitle,
});
@@ -34263,6 +34265,8 @@ If the user has already named or confirmed this exact recipient, do NOT ask agai
return await this._inspectDevEventListeners(tabId, args || {});
}
if (name === 'schedule_resume') {
+ const cloudResume = this.cloudRunContexts.get(tabId)?.deferResume;
+ if (cloudResume) return cloudResume(this._resumeArgsWithProgressGuard(tabId, args || {}));
if (!this.scheduler) {
return {
success: false,
@@ -42211,7 +42215,7 @@ If the user has already named or confirmed this exact recipient, do NOT ask agai
else this._runProviderOverrides.delete(tabId);
const previousCloudContext = this.cloudRunContexts.get(tabId);
if (runOptions.cloudRun) {
- this.cloudRunContexts.set(tabId, { outputSchema: runOptions.outputSchema ?? null, schemaRepairUsed: false });
+ this.cloudRunContexts.set(tabId, { outputSchema: runOptions.outputSchema ?? null, schemaRepairUsed: false, deferResume: runOptions.deferResume });
}
try {
if ((mode === 'act' || mode === 'dev') && !this._isStandaloneChatRun(runOptions)) {
@@ -43771,6 +43775,7 @@ If the user has already named or confirmed this exact recipient, do NOT ask agai
shareRequest = messages;
}
} catch {}
+ runOptions.onRunFinished?.(_traceStatus);
await this._endTraceRun(tabId, runId, _traceStatus, finalResponse, { provider, messages, mode, shareRequest, shareResponse: shareRawResponse, hadProviderCompletion: shareHadProviderCompletion });
}
}
@@ -43825,7 +43830,7 @@ If the user has already named or confirmed this exact recipient, do NOT ask agai
else this._runProviderOverrides.delete(tabId);
const previousCloudContext = this.cloudRunContexts.get(tabId);
if (runOptions.cloudRun) {
- this.cloudRunContexts.set(tabId, { outputSchema: runOptions.outputSchema ?? null, schemaRepairUsed: false });
+ this.cloudRunContexts.set(tabId, { outputSchema: runOptions.outputSchema ?? null, schemaRepairUsed: false, deferResume: runOptions.deferResume });
}
try {
if ((mode === 'act' || mode === 'dev') && !this._isStandaloneChatRun(runOptions)) {
@@ -44864,6 +44869,7 @@ If the user has already named or confirmed this exact recipient, do NOT ask agai
shareRequest = this._pruneOldImages(modelMessagesForRun(), provider);
}
} catch {}
+ runOptions.onRunFinished?.(_traceStatus);
await this._endTraceRun(tabId, runId, _traceStatus, finalResponse, { provider, messages, mode, shareRequest, shareResponse: shareRawResponse, hadProviderCompletion: shareHadProviderCompletion });
}
}
diff --git a/src/chrome/src/cloud-runs.js b/src/chrome/src/cloud-runs.js
index 9f39e1525..afa75a63c 100644
--- a/src/chrome/src/cloud-runs.js
+++ b/src/chrome/src/cloud-runs.js
@@ -4,6 +4,7 @@ import {
normalizeSavedWorkflow,
} from './agent/workflows.js';
import { isCredentialField } from './agent/credential-fields.js';
+import { validateResumeArgs } from './agent/scheduler.js';
const DEFAULT_CLOUD_BRIDGE_URL = 'ws://127.0.0.1:17374/extension';
const CLOUD_RUN_STORAGE_KEY = 'webbrainCloudRunSnapshots';
@@ -694,6 +695,7 @@ export function createCloudRunController({
workflowTrace = null,
now = () => new Date(),
makeRunId = () => `run_${globalThis.crypto.randomUUID()}`,
+ waitForResume = ms => new Promise(resolve => setTimeout(resolve, ms)),
} = {}) {
const api = chromeApi;
const runs = new Map();
@@ -883,12 +885,10 @@ export function createCloudRunController({
return redactWorkflowRuntimeValues(withoutSecrets, known, '[redacted strict URL]');
}
- // Terminal prose reaches the caller by the same two routes as an update row,
- // so it takes both secret and trace-only URL redaction. A structured
- // `run.result` is handled separately below and takes only credential-value
- // redaction so ordinary schema-valid URL components remain usable.
+ // Answers retain ordinary article links. Credentials remain value-redacted;
+ // trace updates still withhold page URLs and arbitrary tool payloads.
function redactStrictTerminal(run, value, strictSecretMode) {
- return redactStrictTraceValues(run, value, strictSecretMode);
+ return redactStrictSecretValues(run, value, strictSecretMode);
}
function pushUpdate(run, type, data, runtimeValues = []) {
@@ -935,7 +935,7 @@ export function createCloudRunController({
// with every string and number replaced by a placeholder, which satisfies
// strict mode by making the run useless.
const publicSummary = strictSecretMode
- ? redactStrictTraceValues(run, result.summary, true)
+ ? redactStrictSecretValues(run, result.summary, true)
: safeResult.summary;
if (result.cloudFailed) {
run.status = 'failed';
@@ -948,6 +948,11 @@ export function createCloudRunController({
run.summary = publicSummary || run.summary;
}
}
+ if (type === 'tool_result' && ['done', 'done_json'].includes(data?.name)
+ && ['partial', 'failed', 'blocked'].includes(data?.result?.outcome)) {
+ run.agentOutcome = data.result.outcome;
+ }
+ if (type === 'run_status' && data?.status) run.agentOutcome = data.status;
if (type === 'captcha_gate') {
// Keep the latest sanitized frame/vendor snapshot at run level so it
// survives the rolling 200-update window in exported cloud traces.
@@ -1066,7 +1071,7 @@ export function createCloudRunController({
: String(msg.task || msg.text || '').trim();
if (!task) throw new Error('cloud_run requires `task`.');
if (startingTabs.has(tabId) || agent.isRunning(tabId)) {
- throw new Error(`Tab ${tabId} already has an active WebBrain run.`);
+ throw cloudRunError('This browser tab is busy with another run. Wait for it to finish.', 409);
}
const apiMutationsAllowed = msg.apiMutationsAllowed === true || msg.api_mutations_allowed === true;
@@ -1140,7 +1145,7 @@ export function createCloudRunController({
// agent's active-run map cannot identify the parent here. The cloud
// run record keeps the actual trace id; resolve its session from that
// trace instead of attaching the tab's potentially unrelated session.
- const parentTraceRunId = parentRun?.traceRunId || null;
+ let parentTraceRunId = parentRun?.traceRunId || null;
let parentTraceSessionId = null;
if (parentTraceRunId && typeof workflowTrace?.getRun === 'function') {
try {
@@ -1212,18 +1217,62 @@ export function createCloudRunController({
);
}
} else {
- content = await agent.processMessage(tabId, task, publishUpdate, mode, [], {
- cloudRun: true,
- independentRun: true,
- apiMutationsDenied: mode === 'ask',
- outputSchema,
- onTraceStarted(traceRunId) {
- run.traceRunId = traceRunId;
- schedulePersist();
- },
- parentRunId: parentTraceRunId,
- parentSessionId: parentTraceSessionId,
- });
+ let prompt = task;
+ let continuation = false;
+ let resumeCount = 0;
+ do {
+ let requestedResume = null;
+ run.agentOutcome = null;
+ content = await agent.processMessage(tabId, prompt, publishUpdate, mode, [], {
+ cloudRun: true,
+ independentRun: !continuation,
+ trustedContinuation: continuation,
+ apiMutationsDenied: mode === 'ask',
+ outputSchema,
+ deferResume(args) {
+ if (resumeCount >= 3) return { success: false, dispatched: false, noDispatch: true,
+ error: 'Cloud continuation limit reached. Return an honest partial or failed outcome.' };
+ const parsed = validateResumeArgs(args, now().getTime());
+ if (!parsed.ok) return { success: false, dispatched: false, noDispatch: true, error: parsed.error };
+ // No browser-native alarm: it would run outside this Cloud handle,
+ // releasing the workspace lock and allowing the profile to pause.
+ requestedResume = parsed;
+ return { success: true, scheduled: true, done: true, jobId: run.runId,
+ scheduledAt: parsed.scheduledAt, summary: `Scheduled a resume for ${parsed.scheduledAt}.` };
+ },
+ onRunFinished(status) {
+ // An explicit failed/partial done outcome overrides generic 'done'.
+ if (!['partial', 'failed', 'blocked'].includes(run.agentOutcome)) run.agentOutcome = status;
+ },
+ onTraceStarted(traceRunId) {
+ run.traceRunId = traceRunId;
+ schedulePersist();
+ },
+ parentRunId: parentTraceRunId,
+ parentSessionId: parentTraceSessionId,
+ });
+ if (!requestedResume || ['failed', 'aborting', 'aborted'].includes(run.status)) break;
+ resumeCount++;
+ pushUpdate(run, 'cloud_resume_wait', { status: 'waiting', scheduledAt: requestedResume.scheduledAt, attempt: resumeCount });
+ // Keep the same run, tab reservation, permission scope and secret
+ // registries until completion or confirmed cancellation. Polls keep
+ // the worker alive; a worker restart fails safely instead of replaying.
+ await persist();
+ while (run.status !== 'aborting' && now().getTime() < requestedResume.scheduledAtMs) {
+ await waitForResume(Math.min(1000, requestedResume.scheduledAtMs - now().getTime()));
+ }
+ if (run.status === 'aborting') break;
+ parentTraceRunId = run.traceRunId;
+ parentTraceSessionId = null;
+ if (parentTraceRunId && typeof workflowTrace?.getRun === 'function') {
+ try {
+ parentTraceSessionId = (await workflowTrace.getRun(parentTraceRunId))?.conversationId || null;
+ } catch { /* lineage lookup must not fail a continuation */ }
+ }
+ prompt = requestedResume.resumeInstruction;
+ continuation = true;
+ pushUpdate(run, 'cloud_resume_wait', { status: 'resuming', attempt: resumeCount });
+ } while (true);
}
run.pendingInput = null;
// Terminal fields are published over the bridge and persisted just like
@@ -1241,7 +1290,11 @@ export function createCloudRunController({
run.status = 'aborted';
run.error = run.error || 'Aborted by cloud_abort.';
} else if (run.status !== 'failed') {
- if (structured && run.result === undefined) {
+ if (run.agentOutcome && run.agentOutcome !== 'done') {
+ run.status = run.agentOutcome === 'cancelled' ? 'aborted' : 'failed';
+ run.error = `Browser task did not finish successfully (${run.agentOutcome}). Review the returned result before retrying.`;
+ if (!structured) run.result = run.content;
+ } else if (structured && run.result === undefined) {
run.status = 'failed';
run.error = 'Structured cloud run finished without a valid done_json result.';
} else {
diff --git a/src/firefox/src/agent/agent.js b/src/firefox/src/agent/agent.js
index 668b911b0..35f6ef9fd 100644
--- a/src/firefox/src/agent/agent.js
+++ b/src/firefox/src/agent/agent.js
@@ -28616,7 +28616,8 @@ Rules: no prose intro, no conclusion, no "this screenshot shows...", no layout d
}
async _scheduleAutoProgressResume(tabId, onUpdate = () => {}) {
- if (!this.scheduler) return null;
+ const cloudResume = this.cloudRunContexts.get(tabId)?.deferResume;
+ if (!this.scheduler && !cloudResume) return null;
const mode = this._effectiveRunMode(tabId);
if (!this._isActionMode(mode)) return null;
if (!this._shouldBlockDoneForProgress(tabId)) return null;
@@ -28624,16 +28625,17 @@ Rules: no prose intro, no conclusion, no "this screenshot shows...", no layout d
const { tabUrl, tabTitle } = await this._getTabUrlTitle(tabId);
let result;
try {
- result = await this.scheduler.createResumeJob({
+ const args = {
+ after_seconds: 90,
+ reason: 'The active progress-ledger task hit consecutive stalled model outputs before finishing.',
+ resume_instruction: this._buildAutoProgressResumeInstruction(tabId),
+ };
+ result = cloudResume ? await cloudResume(args) : await this.scheduler.createResumeJob({
tabId,
conversationId: this.conversationIds.get(tabId) || null,
resumeTaskId: this._resumeTaskId(tabId, { create: true }),
mode,
- args: {
- after_seconds: 90,
- reason: 'The active progress-ledger task hit consecutive stalled model outputs before finishing.',
- resume_instruction: this._buildAutoProgressResumeInstruction(tabId),
- },
+ args,
currentUrl: tabUrl,
currentTitle: tabTitle,
});
@@ -31029,6 +31031,8 @@ Rules: no prose intro, no conclusion, no "this screenshot shows...", no layout d
return await this._resizeWindow(tabId, args || {});
}
if (name === 'schedule_resume') {
+ const cloudResume = this.cloudRunContexts.get(tabId)?.deferResume;
+ if (cloudResume) return cloudResume(this._resumeArgsWithProgressGuard(tabId, args || {}));
if (!this.scheduler) {
return {
success: false,
@@ -35236,7 +35240,7 @@ Rules: no prose intro, no conclusion, no "this screenshot shows...", no layout d
this._runModeOverrides.set(tabId, mode);
const previousCloudContext = this.cloudRunContexts.get(tabId);
if (runOptions.cloudRun) {
- this.cloudRunContexts.set(tabId, { outputSchema: runOptions.outputSchema ?? null, schemaRepairUsed: false });
+ this.cloudRunContexts.set(tabId, { outputSchema: runOptions.outputSchema ?? null, schemaRepairUsed: false, deferResume: runOptions.deferResume });
}
try {
if ((mode === 'act' || mode === 'dev') && !this._isStandaloneChatRun(runOptions)) await firefoxBidi.startRun(tabId, this._runAbortSignal(tabId));
@@ -36623,6 +36627,7 @@ Rules: no prose intro, no conclusion, no "this screenshot shows...", no layout d
shareRequest = messages;
}
} catch {}
+ runOptions.onRunFinished?.(_traceStatus);
await this._endTraceRun(tabId, runId, _traceStatus, finalResponse, { provider, messages, mode, shareRequest, shareResponse: shareRawResponse, hadProviderCompletion: shareHadProviderCompletion });
}
}
@@ -36668,7 +36673,7 @@ Rules: no prose intro, no conclusion, no "this screenshot shows...", no layout d
this._runModeOverrides.set(tabId, mode);
const previousCloudContext = this.cloudRunContexts.get(tabId);
if (runOptions.cloudRun) {
- this.cloudRunContexts.set(tabId, { outputSchema: runOptions.outputSchema ?? null, schemaRepairUsed: false });
+ this.cloudRunContexts.set(tabId, { outputSchema: runOptions.outputSchema ?? null, schemaRepairUsed: false, deferResume: runOptions.deferResume });
}
try {
if ((mode === 'act' || mode === 'dev') && !this._isStandaloneChatRun(runOptions)) await firefoxBidi.startRun(tabId, this._runAbortSignal(tabId));
@@ -37556,6 +37561,7 @@ Rules: no prose intro, no conclusion, no "this screenshot shows...", no layout d
shareRequest = this._pruneOldImages(modelMessagesForRun(), provider);
}
} catch {}
+ runOptions.onRunFinished?.(_traceStatus);
await this._endTraceRun(tabId, runId, _traceStatus, finalResponse, { provider, messages, mode, shareRequest, shareResponse: shareRawResponse, hadProviderCompletion: shareHadProviderCompletion });
}
}
diff --git a/src/firefox/src/cloud-runs.js b/src/firefox/src/cloud-runs.js
index a76e48e82..f240e100d 100644
--- a/src/firefox/src/cloud-runs.js
+++ b/src/firefox/src/cloud-runs.js
@@ -4,6 +4,7 @@ import {
normalizeSavedWorkflow,
} from './agent/workflows.js';
import { isCredentialField } from './agent/credential-fields.js';
+import { validateResumeArgs } from './agent/scheduler.js';
const DEFAULT_CLOUD_BRIDGE_URL = 'ws://127.0.0.1:17374/extension';
const CLOUD_RUN_STORAGE_KEY = 'webbrainCloudRunSnapshots';
@@ -696,6 +697,7 @@ export function createCloudRunController({
workflowTrace = null,
now = () => new Date(),
makeRunId = () => `run_${globalThis.crypto.randomUUID()}`,
+ waitForResume = ms => new Promise(resolve => setTimeout(resolve, ms)),
} = {}) {
const api = chromeApi;
const runs = new Map();
@@ -885,12 +887,10 @@ export function createCloudRunController({
return redactWorkflowRuntimeValues(withoutSecrets, known, '[redacted strict URL]');
}
- // Terminal prose reaches the caller by the same two routes as an update row,
- // so it takes both secret and trace-only URL redaction. A structured
- // `run.result` is handled separately below and takes only credential-value
- // redaction so ordinary schema-valid URL components remain usable.
+ // Answers retain ordinary article links. Credentials remain value-redacted;
+ // trace updates still withhold page URLs and arbitrary tool payloads.
function redactStrictTerminal(run, value, strictSecretMode) {
- return redactStrictTraceValues(run, value, strictSecretMode);
+ return redactStrictSecretValues(run, value, strictSecretMode);
}
function pushUpdate(run, type, data, runtimeValues = []) {
@@ -937,7 +937,7 @@ export function createCloudRunController({
// with every string and number replaced by a placeholder, which satisfies
// strict mode by making the run useless.
const publicSummary = strictSecretMode
- ? redactStrictTraceValues(run, result.summary, true)
+ ? redactStrictSecretValues(run, result.summary, true)
: safeResult.summary;
if (result.cloudFailed) {
run.status = 'failed';
@@ -950,6 +950,11 @@ export function createCloudRunController({
run.summary = publicSummary || run.summary;
}
}
+ if (type === 'tool_result' && ['done', 'done_json'].includes(data?.name)
+ && ['partial', 'failed', 'blocked'].includes(data?.result?.outcome)) {
+ run.agentOutcome = data.result.outcome;
+ }
+ if (type === 'run_status' && data?.status) run.agentOutcome = data.status;
if (type === 'captcha_gate') {
// Keep the latest sanitized frame/vendor snapshot at run level so it
// survives the rolling 200-update window in exported cloud traces.
@@ -1068,7 +1073,7 @@ export function createCloudRunController({
: String(msg.task || msg.text || '').trim();
if (!task) throw new Error('cloud_run requires `task`.');
if (startingTabs.has(tabId) || agent.isRunning(tabId)) {
- throw new Error(`Tab ${tabId} already has an active WebBrain run.`);
+ throw cloudRunError('This browser tab is busy with another run. Wait for it to finish.', 409);
}
const apiMutationsAllowed = msg.apiMutationsAllowed === true || msg.api_mutations_allowed === true;
@@ -1142,7 +1147,7 @@ export function createCloudRunController({
// agent's active-run map cannot identify the parent here. The cloud
// run record keeps the actual trace id; resolve its session from that
// trace instead of attaching the tab's potentially unrelated session.
- const parentTraceRunId = parentRun?.traceRunId || null;
+ let parentTraceRunId = parentRun?.traceRunId || null;
let parentTraceSessionId = null;
if (parentTraceRunId && typeof workflowTrace?.getRun === 'function') {
try {
@@ -1214,18 +1219,62 @@ export function createCloudRunController({
);
}
} else {
- content = await agent.processMessage(tabId, task, publishUpdate, mode, [], {
- cloudRun: true,
- independentRun: true,
- apiMutationsDenied: mode === 'ask',
- outputSchema,
- onTraceStarted(traceRunId) {
- run.traceRunId = traceRunId;
- schedulePersist();
- },
- parentRunId: parentTraceRunId,
- parentSessionId: parentTraceSessionId,
- });
+ let prompt = task;
+ let continuation = false;
+ let resumeCount = 0;
+ do {
+ let requestedResume = null;
+ run.agentOutcome = null;
+ content = await agent.processMessage(tabId, prompt, publishUpdate, mode, [], {
+ cloudRun: true,
+ independentRun: !continuation,
+ trustedContinuation: continuation,
+ apiMutationsDenied: mode === 'ask',
+ outputSchema,
+ deferResume(args) {
+ if (resumeCount >= 3) return { success: false, dispatched: false, noDispatch: true,
+ error: 'Cloud continuation limit reached. Return an honest partial or failed outcome.' };
+ const parsed = validateResumeArgs(args, now().getTime());
+ if (!parsed.ok) return { success: false, dispatched: false, noDispatch: true, error: parsed.error };
+ // No browser-native alarm: it would run outside this Cloud handle,
+ // releasing the workspace lock and allowing the profile to pause.
+ requestedResume = parsed;
+ return { success: true, scheduled: true, done: true, jobId: run.runId,
+ scheduledAt: parsed.scheduledAt, summary: `Scheduled a resume for ${parsed.scheduledAt}.` };
+ },
+ onRunFinished(status) {
+ // An explicit failed/partial done outcome overrides generic 'done'.
+ if (!['partial', 'failed', 'blocked'].includes(run.agentOutcome)) run.agentOutcome = status;
+ },
+ onTraceStarted(traceRunId) {
+ run.traceRunId = traceRunId;
+ schedulePersist();
+ },
+ parentRunId: parentTraceRunId,
+ parentSessionId: parentTraceSessionId,
+ });
+ if (!requestedResume || ['failed', 'aborting', 'aborted'].includes(run.status)) break;
+ resumeCount++;
+ pushUpdate(run, 'cloud_resume_wait', { status: 'waiting', scheduledAt: requestedResume.scheduledAt, attempt: resumeCount });
+ // Keep the same run, tab reservation, permission scope and secret
+ // registries until completion or confirmed cancellation. Polls keep
+ // the worker alive; a worker restart fails safely instead of replaying.
+ await persist();
+ while (run.status !== 'aborting' && now().getTime() < requestedResume.scheduledAtMs) {
+ await waitForResume(Math.min(1000, requestedResume.scheduledAtMs - now().getTime()));
+ }
+ if (run.status === 'aborting') break;
+ parentTraceRunId = run.traceRunId;
+ parentTraceSessionId = null;
+ if (parentTraceRunId && typeof workflowTrace?.getRun === 'function') {
+ try {
+ parentTraceSessionId = (await workflowTrace.getRun(parentTraceRunId))?.conversationId || null;
+ } catch { /* lineage lookup must not fail a continuation */ }
+ }
+ prompt = requestedResume.resumeInstruction;
+ continuation = true;
+ pushUpdate(run, 'cloud_resume_wait', { status: 'resuming', attempt: resumeCount });
+ } while (true);
}
run.pendingInput = null;
// Terminal fields are published over the bridge and persisted just like
@@ -1243,7 +1292,11 @@ export function createCloudRunController({
run.status = 'aborted';
run.error = run.error || 'Aborted by cloud_abort.';
} else if (run.status !== 'failed') {
- if (structured && run.result === undefined) {
+ if (run.agentOutcome && run.agentOutcome !== 'done') {
+ run.status = run.agentOutcome === 'cancelled' ? 'aborted' : 'failed';
+ run.error = `Browser task did not finish successfully (${run.agentOutcome}). Review the returned result before retrying.`;
+ if (!structured) run.result = run.content;
+ } else if (structured && run.result === undefined) {
run.status = 'failed';
run.error = 'Structured cloud run finished without a valid done_json result.';
} else {
diff --git a/test/cloud-conversation-isolation.mjs b/test/cloud-conversation-isolation.mjs
index 360c42f34..599623722 100644
--- a/test/cloud-conversation-isolation.mjs
+++ b/test/cloud-conversation-isolation.mjs
@@ -46,7 +46,7 @@ for (const browser of ['chrome', 'firefox']) {
const racing = await Promise.allSettled([c.startRun({ conversationKey: key1, task: 'held' }), c.startRun({ conversationKey: key1, task: 'race' })]);
assert.equal(racing.filter(r => r.status === 'fulfilled').length, 1);
const clearCount = clears.length;
- await assert.rejects(c.startRun({ conversationKey: key1, task: 'busy' }), /active|starting/);
+ await assert.rejects(c.startRun({ conversationKey: key1, task: 'busy' }), error => error.status === 409 && /busy|starting/.test(error.message));
assert.equal(clears.length, clearCount);
release(); hold = null; await finished(racing.find(r => r.status === 'fulfilled').value);
await assert.rejects(c.startRun({ conversationKey: key1, tabId: 7, task: 'conflict' }), /independent/);
diff --git a/test/cloud-run-outcomes.mjs b/test/cloud-run-outcomes.mjs
new file mode 100644
index 000000000..6ab466f5b
--- /dev/null
+++ b/test/cloud-run-outcomes.mjs
@@ -0,0 +1,126 @@
+import test from 'node:test';
+import assert from 'node:assert/strict';
+import fs from 'node:fs';
+import vm from 'node:vm';
+
+const tick = () => new Promise(resolve => setTimeout(resolve, 5));
+for (const browser of ['chrome', 'firefox']) {
+ const { createCloudRunController } = await import(`../src/${browser}/src/cloud-runs.js`);
+ function harness(processMessage, extra = {}) {
+ const stored = {};
+ let seq = 0;
+ const agent = { isRunning: () => false, processMessage, abort() {}, strictSecretMode: true };
+ const api = { storage: { session: { get: async key => ({ [key]: stored[key] }), set: async data => Object.assign(stored, data) } },
+ tabs: { get: async () => ({ id: 7, url: 'https://example.com' }), query: async () => [{ id: 7, url: 'https://example.com' }], update: async () => {} } };
+ const c = createCloudRunController({ chromeApi: api, agent, makeRunId: () => `run_${++seq}`, ensureOffscreen: async () => {}, ...extra });
+ return { c, stored, agent, api };
+ }
+ async function finished(c, run) {
+ for (let i = 0; i < 100 && ['running', 'aborting'].includes(c.runs.get(run.runId).status); i++) await tick();
+ return c.status({ runId: run.runId });
+ }
+ test(`${browser}: tracked continuation keeps its run and tab lock, evidence and secrets`, async () => {
+ let clock = Date.parse('2026-10-05T10:00:00Z'), releaseWait;
+ const waiting = new Promise(resolve => { releaseWait = resolve; });
+ const observed = [];
+ const { c } = harness(async (tab, prompt, update, mode, _attachments, options) => {
+ observed.push({ tab, prompt, mode, options });
+ options.onTraceStarted(`trace_${observed.length}`);
+ if (observed.length === 1) {
+ update('tool_call', { name: 'set_field', args: { text: 'private-login-secret' } });
+ const resume = options.deferResume({ after_seconds: 30, reason: 'Wait for page', resume_instruction: 'Continue the same task' });
+ assert.equal(resume.success, true);
+ options.onRunFinished('scheduled_resume');
+ return resume.summary;
+ }
+ update('tool_result', { name: 'read_page', result: { url: 'https://news.ycombinator.com/item?id=12345678' } });
+ options.onRunFinished('done');
+ return 'Article https://news.ycombinator.com/item?id=12345678; private-login-secret';
+ }, { now: () => new Date(clock), waitForResume: async () => { await waiting; clock += 30000; }, workflowTrace: { getRun: async () => ({ conversationId: 'same-conversation' }) } });
+ const run = await c.startRun({ task: 'Read a public article' });
+ await tick();
+ assert.equal((await c.status({ runId: run.runId })).status, 'running');
+ assert.equal(c.runs.get(run.runId).completedAt, null);
+ await assert.rejects(c.startRun({ task: 'Other task' }), e => e.status === 409);
+ releaseWait();
+ const done = await finished(c, run);
+ assert.equal(done.status, 'completed');
+ assert.equal(observed.length, 2);
+ assert.equal(observed[1].options.trustedContinuation, true);
+ assert.equal(observed[1].options.independentRun, false);
+ assert.equal(observed[1].options.parentRunId, 'trace_1');
+ assert.equal(observed[1].options.parentSessionId, 'same-conversation');
+ assert.equal(observed[1].prompt, 'Continue the same task');
+ assert.match(done.result, /item\?id=12345678/);
+ assert.equal(JSON.stringify(done).includes('private-login-secret'), false);
+ assert.equal(JSON.stringify(done.updates).includes('item?id=12345678'), false);
+ });
+ test(`${browser}: cancellation during a deferred resume never dispatches continuation`, async () => {
+ let release, calls = 0;
+ const waiting = new Promise(resolve => { release = resolve; });
+ const { c } = harness(async (_tab, _task, _update, _mode, _attachments, options) => {
+ calls++;
+ const r = options.deferResume({ after_seconds: 30, reason: 'Waiting', resume_instruction: 'Continue' });
+ options.onRunFinished('scheduled_resume');
+ return r.summary;
+ }, { waitForResume: () => waiting });
+ const run = await c.startRun({ task: 'Read' }); await tick();
+ assert.equal((await c.abort({ runId: run.runId })).status, 'aborting');
+ release();
+ assert.equal((await finished(c, run)).status, 'aborted');
+ assert.equal(calls, 1);
+ });
+ test(`${browser}: failed, partial and recovery-limit outcomes preserve output without reporting success`, async () => {
+ for (const outcome of ['partial', 'failed', 'empty_output', 'max_steps', 'required_tool_missing']) {
+ const { c } = harness(async (_tab, _task, update, _mode, _attachments, options) => {
+ update('tool_result', { name: 'done', result: { outcome: outcome === 'failed' ? 'failed' : undefined } });
+ options.onRunFinished(outcome === 'failed' ? 'done' : outcome);
+ return 'Only a draft, not posted';
+ });
+ const run = await c.startRun({ task: 'Comment on an article' });
+ const done = await finished(c, run);
+ assert.equal(done.status, 'failed');
+ assert.match(done.error, /did not finish successfully/);
+ assert.equal(done.result, 'Only a draft, not posted');
+ }
+ });
+ test(`${browser}: continuation budget and restart stop safely instead of replaying page actions`, async () => {
+ let clock = Date.now(), calls = 0;
+ const { c, stored, agent, api } = harness(async (_tab, _task, _update, _mode, _attachments, options) => {
+ calls++;
+ const resume = options.deferResume({ after_seconds: 30, reason: 'Wait', resume_instruction: 'Continue' });
+ options.onRunFinished(resume.success ? 'scheduled_resume' : 'partial');
+ return resume.summary || resume.error;
+ }, { now: () => new Date(clock), waitForResume: async ms => { clock += ms; } });
+ const run = await c.startRun({ task: 'Read' });
+ assert.equal((await finished(c, run)).status, 'failed');
+ assert.equal(calls, 4);
+ const row = structuredClone(stored.webbrainCloudRunSnapshots[0]);
+ row.status = 'running'; row.completedAt = null;
+ stored.webbrainCloudRunSnapshots = [row];
+ const restarted = createCloudRunController({ chromeApi: api, agent, ensureOffscreen: async () => {} });
+ assert.equal((await restarted.status({ runId: row.runId })).status, 'failed');
+ assert.equal(calls, 4, 'restart never replays a possible consequential action');
+ });
+ test(`${browser}: real auto-resume helper routes Cloud work to the controller, native scheduling stays unchanged`, async () => {
+ const source = fs.readFileSync(new URL(`../src/${browser}/src/agent/agent.js`, import.meta.url), 'utf8');
+ const method = source.slice(source.indexOf(' async _scheduleAutoProgressResume('), source.indexOf(' _plainFinalProgressBlock('));
+ const helper = vm.runInNewContext(`({${method}})._scheduleAutoProgressResume`);
+ let native = 0, cloud = 0;
+ const fake = { scheduler: { createResumeJob: async () => { native++; return { success: true }; } },
+ cloudRunContexts: new Map([[7, { deferResume: async args => { cloud++; assert.equal(args.after_seconds, 90); return { success: true }; } }]]),
+ _effectiveRunMode: () => 'act', _isActionMode: () => true, _shouldBlockDoneForProgress: () => true,
+ _getTabUrlTitle: async () => ({}), _buildAutoProgressResumeInstruction: () => 'Continue',
+ conversationIds: new Map(), _resumeTaskId: () => 'task' };
+ await helper.call(fake, 7);
+ assert.equal(cloud, 1); assert.equal(native, 0);
+ fake.cloudRunContexts.clear(); await helper.call(fake, 7);
+ assert.equal(native, 1);
+ const start = source.indexOf(" if (name === 'schedule_resume') {");
+ const route = source.slice(start, source.indexOf(' if (!this.scheduler)', start));
+ const toolRoute = vm.runInNewContext(`(async function(name,tabId,args){${route}\n}})`);
+ fake.cloudRunContexts.set(7, { deferResume: async () => ({ success: true, cloud: true }) });
+ fake._resumeArgsWithProgressGuard = (_tab, args) => args;
+ assert.equal((await toolRoute.call(fake, 'schedule_resume', 7, {})).cloud, true);
+ });
+}
diff --git a/test/run.js b/test/run.js
index 055c5e14d..c020c9fb9 100644
--- a/test/run.js
+++ b/test/run.js
@@ -26044,11 +26044,11 @@ test('trace lineage: _startTraceRun and replay plumb parent ids in both builds',
assert.match(agentSource, /parentRunId: replayParentRunId,[\s\S]*?parentSessionId: replayParentSessionId,/, `${browser}: replay does not pass captured lineage to tracing`);
}
const chromeCloudRuns = fs.readFileSync(path.join(ROOT, 'src/chrome/src/cloud-runs.js'), 'utf8');
- assert.match(chromeCloudRuns, /const parentTraceRunId = parentRun\?\.traceRunId \|\| null;/, 'cloud-runs does not use the completed parent trace');
+ assert.match(chromeCloudRuns, /(?:const|let) parentTraceRunId = parentRun\?\.traceRunId \|\| null;/, 'cloud-runs does not use the completed parent trace');
assert.match(chromeCloudRuns, /workflowTrace\.getRun\(parentTraceRunId\)/, 'cloud-runs does not resolve the parent trace session');
assert.match(chromeCloudRuns, /parentRunId: parentTraceRunId,[\s\S]*?parentSessionId: parentTraceSessionId,/, 'cloud-runs does not thread resolved parent lineage');
const firefoxCloudRuns = fs.readFileSync(path.join(ROOT, 'src/firefox/src/cloud-runs.js'), 'utf8');
- assert.match(firefoxCloudRuns, /const parentTraceRunId = parentRun\?\.traceRunId \|\| null;/, 'Firefox cloud-runs does not use the completed parent trace');
+ assert.match(firefoxCloudRuns, /(?:const|let) parentTraceRunId = parentRun\?\.traceRunId \|\| null;/, 'Firefox cloud-runs does not use the completed parent trace');
assert.match(firefoxCloudRuns, /parentRunId: parentTraceRunId,[\s\S]*?parentSessionId: parentTraceSessionId,/, 'Firefox cloud-runs does not thread resolved parent lineage');
});