pi-agent-browser-native 0.6.16 → 0.7.0

package/dist/extensions/agent-browser/lib/orchestration/browser-run/process-output.js

@@ -4,7 +4,7 @@ import { parseArgvDescriptor } from "../../argv-descriptor.js";
 import { isBrowserIndependentRead, needsManagedSession } from "../../command-policy.js";
 import { deleteIdentityKeysInNamespace, getAgentBrowserSessionIdentityKey, isAgentBrowserSessionIdentityKeyInNamespace } from "../../argv-grammar.js";
 import { batchHasSuccessfulCloseAll, getSuccessfulBatchCloseLifecycle } from "../../batch-lifecycle.js";
-import { isCloseAllCommand, isCloseCommand, isOpenNavigationCommand, isRecordPageTransitionCommand, isUnverifiedPageTransitionCommand, isWindowOrDiffPageTransitionCommand } from "../../command-taxonomy.js";
+import { isCloseAllCommand, isCloseCommand, isOpenNavigationCommand, isPageMutationCommand, isRecordPageTransitionCommand, isUnverifiedPageTransitionCommand, isWindowOrDiffPageTransitionCommand } from "../../command-taxonomy.js";
 import { OPEN_RESULT_TAB_CORRECTION_FLAGS } from "../../launch-scoped-flags.js";
 import { cleanupElectronLaunchResources, inspectElectronLaunchStatus } from "../../electron/cleanup.js";
 import { getResultingPageTargetState, commandRequiresLivePageVerification } from "../../page-target-validation.js";
@@ -147,6 +147,7 @@ function applyBatchNetworkRouteState(options) {
 }
 export async function processBrowserOutput(input) {
     const { ctx, cwd, electronPostCommandStatusSettleMs, implicitSessionCloseTimeoutMs, sessionPageStateUpdate, signal, state } = input;
+    const operationCwd = input.operationCwd ?? cwd;
     const { prepared, processResult } = input;
     const { electronChildProcesses, electronLaunchRecords, sessionPageState, traceOwners } = state;
     let artifactManifest = state.artifactManifest;
@@ -167,6 +168,7 @@ export async function processBrowserOutput(input) {
         observeNativeWebMcp(parsed.envelope?.data);
         let parseError = parsed.parseError;
         const recordingStopRecovery = await recoverRecordingStop({
+            modelVisible: input.modelVisible,
             artifactManifest, artifactRunStartedAtMs: input.artifactRunStartedAtMs, commandTokens: prepared.commandTokens, cwd,
             envelope: parsed.envelope, namespace: prepared.executionPlan.namespace, parseError, processResult,
             reservation: prepared.executionPlan.sessionName ? state.activeRecordingReservations?.get(getAgentBrowserSessionIdentityKey(prepared.executionPlan.sessionName, prepared.executionPlan.namespace)) : undefined,
@@ -240,6 +242,12 @@ export async function processBrowserOutput(input) {
         const envelopeSuccess = plainTextInspection ? true : presentationEnvelope?.success !== false;
         let succeeded = (processSucceeded && parseSucceeded && envelopeSuccess) || recordingStopRecovery?.recovery.healed === true;
         const inspectionText = plainTextInspection ? processResult.stdout.trim() : undefined;
+        const unobservedMutation = processResult.agentBrowserStarted && !plainTextInspection
+            && (processResult.aborted || processResult.timedOut || !parseSucceeded || presentationEnvelope === undefined)
+            && (prepared.executionPlan.commandInfo.command === "batch" ? batchCommandSteps : [prepared.commandTokens]).some(step => {
+                const action = step[0] === "find" ? step[step[1] === "nth" ? 4 : 3] ?? "click" : step[0];
+                return isPageMutationCommand(action, step[1]);
+            });
         const sessionStateKey = getSessionContextKey(prepared.executionPlan.sessionName, prepared.executionPlan.namespace);
         const closeAllApplied = nestedBatchClosesAll || (directCloseAllRequested && succeeded);
         if (sessionStateKey && processResult.agentBrowserStarted && sessionPageState.get(sessionStateKey).tabReopenPending === true) {
@@ -334,14 +342,14 @@ export async function processBrowserOutput(input) {
         const failedWebMcpSettlement = prepared.priorSessionTabTargetUnknown === true && ((!succeeded && isWebMcpSettlementCommand(prepared.executionPlan.commandInfo.command, prepared.executionPlan.commandInfo.subcommand))
             || (prepared.executionPlan.commandInfo.command === "batch" && batchHasFailedWebMcpSettlement(presentationEnvelope?.data)));
         const unsettledWebMcpMutation = pendingWebMcpMutation || failedWebMcpSettlement;
-        const observedSessionTabTarget = unsettledWebMcpMutation
+        const observedSessionTabTarget = unsettledWebMcpMutation || (unobservedMutation && !failedTransitionReverification)
             ? undefined
             : normalizeSessionTabTarget(navigationSummary)
                 ?? (trustsReportedPageTarget ? extractSessionTabTargetFromBatchResults(presentationEnvelope?.data) : undefined)
                 ?? (succeeded && trustsReportedPageTarget ? extractSessionTabTargetFromCommandData(prepared.commandTokens, presentationEnvelope?.data) : undefined);
         const safeObservedSessionTabTarget = observedSessionTabTarget;
         let currentSessionTabTarget = safeObservedSessionTabTarget;
-        if (!currentSessionTabTarget && nestedBatchClose === undefined) {
+        if (!currentSessionTabTarget && nestedBatchClose === undefined && !unobservedMutation) {
             // Window/diff responses do not report the final URL; URL2 is intent, not redirect evidence.
             currentSessionTabTarget = resultingPageState.pageTargetMayHaveChanged
                 ? succeeded && !destinationTransition ? normalizeSessionTabTarget({ url: resultingPageState.currentPageUrl }) : undefined
@@ -395,8 +403,8 @@ export async function processBrowserOutput(input) {
         let fillVerificationDiagnostic;
         let selectorTextVisibilityDiagnostics = [];
         let electronBroadGetTextScopeDiagnostics = [];
-        const timeoutPartialProgress = processResult.timedOut && !recordingStopRecovery && !prepared.readConfirmation ? await collectTimeoutPartialProgress({ commandTokens: prepared.commandTokens, compiledJob: prepared.compiledJob, cwd, namespace: prepared.executionPlan.namespace, sessionName: prepared.executionPlan.sessionName, stdin: prepared.runtimeToolStdin }) : undefined;
-        if (!currentSessionTabTarget && timeoutPartialProgress?.currentPage?.source === "live") {
+        const timeoutPartialProgress = processResult.timedOut && !recordingStopRecovery && !prepared.readConfirmation ? await collectTimeoutPartialProgress({ commandTokens: prepared.commandTokens, compiledJob: prepared.compiledJob, cwd, operationCwd: input.operationCwd, namespace: prepared.executionPlan.namespace, sessionName: prepared.executionPlan.sessionName, stdin: prepared.runtimeToolStdin }) : undefined;
+        if (!currentSessionTabTarget && timeoutPartialProgress?.liveUrlRecovered === true) {
             currentSessionTabTarget = normalizeSessionTabTarget(timeoutPartialProgress.currentPage);
         }
         if (succeeded) {
@@ -444,24 +452,26 @@ export async function processBrowserOutput(input) {
                 const plannedBatchTransitionInvalidation = !Array.isArray(presentationEnvelope?.data)
                     ? batchCommandSteps.map(getCommandRefSnapshotInvalidation).find((invalidation) => invalidation !== undefined)
                     : undefined;
-                const pageTransitionInvalidation = unsettledWebMcpMutation
-                    ? buildPageTransitionRefSnapshotInvalidation("A detached WebMCP invocation is still pending or failed to settle and can mutate, rerender, or navigate the page, so prior snapshot refs remain invalid after URL verification. Run webmcp result or cancel, then take a fresh snapshot before using page-scoped refs.")
-                    : processResult.agentBrowserStarted && (directTransitionInvalidation || plannedBatchTransitionInvalidation)
-                        ? directTransitionInvalidation ?? plannedBatchTransitionInvalidation
-                        : failedTransitionReverification
-                            ? buildPageTransitionRefSnapshotInvalidation("A failed eval/back/forward/reload/connect/state-load/tab command may still have changed the page, so the prior snapshot refs were invalidated. Run snapshot -i before using page-scoped refs.")
-                            : batchRefSnapshotState?.invalidation?.reason === "page-transition"
-                                ? batchRefSnapshotState.invalidation
-                                : undefined;
+                const pageTransitionInvalidation = unobservedMutation && !failedTransitionReverification
+                    ? buildPageTransitionRefSnapshotInvalidation("A dispatched mutation was interrupted without a final outcome. Verify the current URL and take a fresh snapshot before deciding whether to retry.")
+                    : unsettledWebMcpMutation
+                        ? buildPageTransitionRefSnapshotInvalidation("A detached WebMCP invocation is still pending or failed to settle and can mutate, rerender, or navigate the page, so prior snapshot refs remain invalid after URL verification. Run webmcp result or cancel, then take a fresh snapshot before using page-scoped refs.")
+                        : processResult.agentBrowserStarted && (directTransitionInvalidation || plannedBatchTransitionInvalidation)
+                            ? directTransitionInvalidation ?? plannedBatchTransitionInvalidation
+                            : failedTransitionReverification
+                                ? buildPageTransitionRefSnapshotInvalidation("A failed eval/back/forward/reload/connect/state-load/tab command may still have changed the page, so the prior snapshot refs were invalidated. Run snapshot -i before using page-scoped refs.")
+                                : batchRefSnapshotState?.invalidation?.reason === "page-transition"
+                                    ? batchRefSnapshotState.invalidation
+                                    : undefined;
                 if (currentSessionTabTarget) {
                     const tabUpdate = sessionPageState.applyTabTarget({ sessionName: sessionStateKey, target: currentSessionTabTarget, update: sessionPageStateUpdate });
                     if (!tabUpdate.applied && succeeded)
                         sessionPageState.markPinning(sessionStateKey, "drift");
                 }
-                else if (processResult.agentBrowserStarted && (resultingPageState.pageUrlUnknown || resultingPageState.pageTargetMayHaveChanged) && !(prepared.commandTokens[0] === "session" && prepared.commandTokens[1] === "info")) {
+                else if (processResult.agentBrowserStarted && (unobservedMutation || resultingPageState.pageUrlUnknown || resultingPageState.pageTargetMayHaveChanged) && !(prepared.commandTokens[0] === "session" && prepared.commandTokens[1] === "info")) {
                     sessionPageState.markTabTargetUnknown({ sessionName: sessionStateKey, update: sessionPageStateUpdate });
                 }
-                let refSnapshot = unsettledWebMcpMutation
+                let refSnapshot = unsettledWebMcpMutation || unobservedMutation
                     ? undefined
                     : prepared.executionPlan.commandInfo.command === "batch"
                         ? batchRefSnapshotState?.snapshot
@@ -472,7 +482,7 @@ export async function processBrowserOutput(input) {
                     ? prepared.commandTokens.filter((token) => token !== "--delta" && token !== "--full") : undefined);
                 // Native owns delta baselines. Read complete refs without advancing that baseline
                 // rather than maintaining a second revision/tree cache in the wrapper.
-                if (refreshArgs && !unsettledWebMcpMutation) {
+                if (refreshArgs && !unsettledWebMcpMutation && !unobservedMutation) {
                     refSnapshot = extractRefSnapshotFromData(await runSessionCommandData({ args: refreshArgs, cwd, namespace: prepared.executionPlan.namespace, sessionName: prepared.executionPlan.sessionName, signal }));
                     if (!refSnapshot)
                         sessionPageState.applyRefSnapshotInvalidation({ invalidation: buildPageTransitionRefSnapshotInvalidation("The native snapshot delta did not include complete refs and the full ref read failed. Run snapshot -i before using refs."), sessionName: sessionStateKey, update: sessionPageStateUpdate });
@@ -574,7 +584,8 @@ export async function processBrowserOutput(input) {
         }
         let managedSessionOutcome = buildManagedSessionOutcome({ activeAfter: managedSessionActive, activeBefore: priorManagedSessionActive, attemptedSessionName: managedCloseSessionName, command: commandClosesSession ? "close" : prepared.executionPlan.commandInfo.command, currentSessionName: managedSessionName, currentSessionNamespace: managedSessionNamespace, previousSessionName: priorManagedSessionName, replacedSessionName: replacedManagedSessionName, replacedSessionNamespace: priorManagedSessionNamespace, sessionMode: prepared.sessionMode, succeeded: managedTransitionSucceeded });
         if (prepared.executionPlan.managedSessionName && managedTransitionSucceeded && managedSessionActive) {
-            managedSessionCwd = cwd;
+            if (!priorManagedSessionActive || replacedManagedSessionName)
+                managedSessionCwd = cwd;
             managedSessionNamespace = prepared.executionPlan.namespace;
         }
         if (sessionStateKey && succeeded) {
@@ -681,7 +692,7 @@ export async function processBrowserOutput(input) {
         }
         if (plainTextUpgrade && errorText)
             presentationEnvelope = { ...presentationEnvelope, success: false, error: errorText };
-        let presentation = plainTextInspection ? { artifacts: undefined, batchFailure: undefined, batchSteps: undefined, content: [{ type: "text", text: inspectionText ?? "" }], data: undefined, fullOutputPath: undefined, fullOutputPaths: undefined, imagePath: undefined, imagePaths: undefined, savedFile: undefined, savedFilePath: undefined, summary: `${prepared.redactedArgs.join(" ")} completed` } : recordingStopRecovery && !recordingStopRecovery.batch ? recordingStopRecovery.presentation : await buildToolPresentation({ args: prepared.redactedProcessArgs, artifactManifest, artifactMaxUpdatedAtMs: Date.now(), artifactMinUpdatedAtMs: input.artifactRunStartedAtMs, artifactRequest: screenshotArtifactRequest, batchArtifactRequests: batchScreenshotArtifactRequests, commandInfo: prepared.executionPlan.commandInfo, compiledSemanticAction: prepared.compiledSemanticAction, cwd, envelope: presentationEnvelope, errorText, namespace: prepared.executionPlan.namespace, networkRouteDiagnostics, networkRoutes: activeNetworkRoutes, persistentArtifactStore, previousRecordingContactSheetPath: sessionStateKey ? state.activeRecordingReservations?.get(sessionStateKey)?.contactSheetPath : undefined, piCleanupOwnership: sessionStateKey && (state.ownedManagedSessions.has(sessionStateKey) || prepared.executionPlan.managedSessionName !== undefined) ? "wrapper-managed" : "caller-owned", sessionName: prepared.executionPlan.sessionName });
+        let presentation = plainTextInspection ? { artifacts: undefined, batchFailure: undefined, batchSteps: undefined, content: [{ type: "text", text: inspectionText ?? "" }], data: undefined, fullOutputPath: undefined, fullOutputPaths: undefined, imagePath: undefined, imagePaths: undefined, savedFile: undefined, savedFilePath: undefined, summary: `${prepared.redactedArgs.join(" ")} completed` } : recordingStopRecovery && !recordingStopRecovery.batch ? recordingStopRecovery.presentation : await buildToolPresentation({ modelVisible: input.modelVisible, args: prepared.redactedProcessArgs, artifactManifest, artifactMaxUpdatedAtMs: Date.now(), artifactMinUpdatedAtMs: input.artifactRunStartedAtMs, artifactRequest: screenshotArtifactRequest, batchArtifactRequests: batchScreenshotArtifactRequests, commandInfo: prepared.executionPlan.commandInfo, compiledSemanticAction: prepared.compiledSemanticAction, cwd: operationCwd, envelope: presentationEnvelope, errorText, namespace: prepared.executionPlan.namespace, networkRouteDiagnostics, networkRoutes: activeNetworkRoutes, persistentArtifactStore, previousRecordingContactSheetPath: sessionStateKey ? state.activeRecordingReservations?.get(sessionStateKey)?.contactSheetPath : undefined, piCleanupOwnership: sessionStateKey && (state.ownedManagedSessions.has(sessionStateKey) || prepared.executionPlan.managedSessionName !== undefined) ? "wrapper-managed" : "caller-owned", sessionName: prepared.executionPlan.sessionName });
         if (recordingStopRecovery)
             presentation = mergeRecordingRecoveryPresentation(presentation, recordingStopRecovery);
         const confirmation = readConfirmationEvent ?? prepared.readConfirmation;
@@ -701,7 +712,8 @@ export async function processBrowserOutput(input) {
             const errorContent = presentation.content[0];
             if (errorContent?.type === "text" && presentationEnvelope.data)
                 errorContent.text += `\n\n${presentationEnvelope.data}`;
-            presentation = await compactLargePresentationOutput({ artifactManifest, commandInfo: prepared.executionPlan.commandInfo, data: presentation.data, persistentArtifactStore, presentation });
+            if (input.modelVisible !== false)
+                presentation = await compactLargePresentationOutput({ artifactManifest, commandInfo: prepared.executionPlan.commandInfo, data: presentation.data, persistentArtifactStore, presentation });
         }
         if (electronHandoff?.error && electronHandoff.failureCategory)
             presentation.failureCategory = electronHandoff.failureCategory;
@@ -741,12 +753,17 @@ export async function processBrowserOutput(input) {
         const qaPreset = prepared.compiledQaPreset
             ? (processResult.timedOut ? analyzeQaPresetTimeout(prepared.compiledQaPreset) ?? analyzeQaPresetResults(presentationEnvelope?.data, prepared.compiledQaPreset) : analyzeQaPresetResults(presentationEnvelope?.data, prepared.compiledQaPreset))
             : undefined;
+        if (qaPreset?.passed && presentation.resultCategory === "failure") {
+            qaPreset.passed = false;
+            qaPreset.summary = presentation.summary;
+            qaPreset.failedChecks.push(presentation.summary);
+        }
         let qaAttachedTarget = prepared.compiledQaPreset?.checks.attached
             ? await collectQaAttachedTarget({ currentTarget: currentSessionTabTarget ?? prepared.priorSessionTabTarget, cwd, namespace: prepared.executionPlan.namespace, sessionName: prepared.executionPlan.sessionName, signal })
             : undefined;
         const sourceLookupElectronContext = prepared.compiledSourceLookup ? getSourceLookupElectronContext({ currentTarget: currentSessionTabTarget, electronLaunchRecords, namespace: prepared.executionPlan.namespace, priorTarget: prepared.priorSessionTabTarget, sessionName: prepared.executionPlan.sessionName }) : undefined;
-        const sourceLookup = prepared.compiledSourceLookup ? await analyzeSourceLookupResults(presentationEnvelope?.data, prepared.compiledSourceLookup, cwd, { electronContext: sourceLookupElectronContext, workspaceRoot: cwd }) : undefined;
-        const networkSourceLookup = prepared.compiledNetworkSourceLookup ? redactNetworkSourceLookupAnalysis(await analyzeNetworkSourceLookupResults(presentationEnvelope?.data, prepared.compiledNetworkSourceLookup, cwd)) : undefined;
+        const sourceLookup = prepared.compiledSourceLookup ? await analyzeSourceLookupResults(presentationEnvelope?.data, prepared.compiledSourceLookup, operationCwd, { electronContext: sourceLookupElectronContext, workspaceRoot: operationCwd }) : undefined;
+        const networkSourceLookup = prepared.compiledNetworkSourceLookup ? redactNetworkSourceLookupAnalysis(await analyzeNetworkSourceLookupResults(presentationEnvelope?.data, prepared.compiledNetworkSourceLookup, operationCwd)) : undefined;
         if (networkSourceLookup && presentation.content[0]?.type === "text")
             presentation.content[0] = { ...presentation.content[0], text: `${networkSourceLookup.summary}\n\n${presentation.content[0].text}` };
         else if (networkSourceLookup)
@@ -817,7 +834,7 @@ export async function processBrowserOutput(input) {
             : undefined;
         const sessionWarning = electronPostCommandHealth ? formatElectronPostCommandHealthText(electronPostCommandHealth) : electronSessionMismatch ? formatElectronSessionMismatchText(electronSessionMismatch) : aboutBlankSessionMismatch ? buildAboutBlankWarning(aboutBlankSessionMismatch) : undefined;
         const warningText = [sessionWarning, recordingPageWarning].filter(Boolean).join("\n\n") || undefined;
-        const redactedContent = buildRedactedPresentationContent({ exactSensitiveValues: prepared.exactSensitiveValues, plainTextInspection, presentation, presentationEnvelope, succeeded, userRequestedJson: prepared.userRequestedJson, warningText });
+        const redactedContent = input.modelVisible === false ? [] : buildRedactedPresentationContent({ exactSensitiveValues: prepared.exactSensitiveValues, plainTextInspection, presentation, presentationEnvelope, succeeded, userRequestedJson: prepared.userRequestedJson, warningText });
         const finalRecoveryState = await prepareFinalResultRecoveryState({ aboutBlankSessionMismatch, batchRefSnapshotState, commandTokens: prepared.commandTokens, compiledSemanticAction: prepared.compiledSemanticAction, currentRefSnapshot, currentRefSnapshotInvalidation, currentSessionTabTarget, cwd, electronPostCommandHealth, errorText, executionPlan: prepared.executionPlan, parseError, plainTextInspection, presentation, processResult, redactedProcessArgs: prepared.redactedProcessArgs, runtimeToolArgs: prepared.runtimeToolArgs, sessionPageState, sessionPageStateUpdate, sessionTabCorrection, signal, succeeded });
         currentRefSnapshot = finalRecoveryState.currentRefSnapshot;
         currentRefSnapshotInvalidation = finalRecoveryState.currentRefSnapshotInvalidation;
@@ -834,7 +851,7 @@ export async function processBrowserOutput(input) {
         const resultHeadedManagedAutosaveInterval = resultRetainsPreparedManagedSession && !prepared.ownedManagedSessionContext?.reuseOnly && !(commandClosesSession && succeeded)
             ? prepared.ownedManagedSessionContext?.headedManagedAutosaveInterval
             : undefined;
-        const result = buildFinalAgentBrowserToolResult({ aboutBlankSessionMismatch, artifactCleanup, categoryDetails: finalRecoveryState.categoryDetails, clickDispatchDiagnostic, commandTokens: prepared.commandTokens, comboboxFocusDiagnostic, compiledNetworkSourceLookup: prepared.compiledNetworkSourceLookup, compiledSemanticAction: prepared.compiledSemanticAction, compatibilityWorkaround: prepared.compatibilityWorkaround, currentRefSnapshot, currentRefSnapshotInvalidation, currentSessionTabTarget, currentSessionTabTargetUnknown, electronBroadGetTextScopeDiagnostics, electronFailedConnectCleanup, electronHandoff, electronLaunch: prepared.electronLaunch, electronLaunchRecord, electronLaunchRecords, electronPostCommandHealth, electronProfileIsolationDetails: input.electronProfileIsolationDetails, electronRefFreshnessDiagnostic, electronSessionMismatch, errorText, evalResultWarning, evalStdinHint, exactSensitiveValues: prepared.exactSensitiveValues, executionPlan: prepared.executionPlan, fillVerificationDiagnostic, headedLaunch: prepared.headedLaunch, inspectionText, preserveAttachedBrowserSession: input.preserveAttachedBrowserSession === true, providerLaunch: prepared.providerLaunch, managedSessionHeadedAutosaveDisabled: resultHeadedManagedAutosaveDisabled || undefined, managedSessionHeadedAutosaveInterval: resultHeadedManagedAutosaveInterval, managedSessionOutcome, managedSessionRestoreDisabled: state.managedSessionRestoreState.isDisabled(prepared.executionPlan.sessionName, prepared.executionPlan.namespace), navigationSummary, networkSourceLookup, noActivePageSnapshotFailure: finalRecoveryState.noActivePageSnapshotFailure, openResultTabCorrection, overlayBlockerDiagnostic, parseError, parseFailureOutput, parseSucceeded, plainTextInspection, presentation, presentationEnvelope, priorSessionTabTarget: prepared.priorSessionTabTarget, processResult, qaAttachedTarget, qaPreset, recordingDependencyWarning, redactedArgs: prepared.redactedArgs, redactedCompiledElectron: prepared.redactedCompiledElectron, redactedCompiledJob: prepared.redactedCompiledJob, redactedCompiledNetworkSourceLookup: prepared.redactedCompiledNetworkSourceLookup, redactedCompiledQaPreset: prepared.redactedCompiledQaPreset, redactedCompiledSemanticAction: prepared.redactedCompiledSemanticAction, redactedCompiledSourceLookup: prepared.redactedCompiledSourceLookup, redactedContent, redactedProcessArgs: prepared.redactedProcessArgs, redactedRecoveryHint: prepared.redactedRecoveryHint, resultArtifactManifest, richInputRecoveryDiagnostic: finalRecoveryState.richInputRecoveryDiagnostic, scrollNoopDiagnostic, selectorTextVisibilityDiagnostics, sessionMode: prepared.sessionMode, sessionTabCorrection, sourceLookup, succeeded, timeoutPartialProgress, unsettledWebMcpMutation, userRequestedJson: prepared.userRequestedJson, visibleRefFallbackDiagnostic: finalRecoveryState.visibleRefFallbackDiagnostic, visibleRefFallbackSessionName: finalRecoveryState.visibleRefFallbackSessionName });
+        const result = buildFinalAgentBrowserToolResult({ modelVisible: input.modelVisible, aboutBlankSessionMismatch, artifactCleanup, categoryDetails: finalRecoveryState.categoryDetails, clickDispatchDiagnostic, commandTokens: prepared.commandTokens, comboboxFocusDiagnostic, compiledNetworkSourceLookup: prepared.compiledNetworkSourceLookup, compiledSemanticAction: prepared.compiledSemanticAction, compatibilityWorkaround: prepared.compatibilityWorkaround, currentRefSnapshot, currentRefSnapshotInvalidation, currentSessionTabTarget, currentSessionTabTargetUnknown, electronBroadGetTextScopeDiagnostics, electronFailedConnectCleanup, electronHandoff, electronLaunch: prepared.electronLaunch, electronLaunchRecord, electronLaunchRecords, electronPostCommandHealth, electronProfileIsolationDetails: input.electronProfileIsolationDetails, electronRefFreshnessDiagnostic, electronSessionMismatch, errorText, evalResultWarning, evalStdinHint, exactSensitiveValues: prepared.exactSensitiveValues, executionPlan: prepared.executionPlan, fillVerificationDiagnostic, headedLaunch: prepared.headedLaunch, inspectionText, preserveAttachedBrowserSession: input.preserveAttachedBrowserSession === true, providerLaunch: prepared.providerLaunch, managedSessionHeadedAutosaveDisabled: resultHeadedManagedAutosaveDisabled || undefined, managedSessionHeadedAutosaveInterval: resultHeadedManagedAutosaveInterval, managedSessionOutcome, managedSessionRestoreDisabled: state.managedSessionRestoreState.isDisabled(prepared.executionPlan.sessionName, prepared.executionPlan.namespace), navigationSummary, networkSourceLookup, noActivePageSnapshotFailure: finalRecoveryState.noActivePageSnapshotFailure, openResultTabCorrection, overlayBlockerDiagnostic, parseError, parseFailureOutput, parseSucceeded, plainTextInspection, presentation, presentationEnvelope, priorSessionTabTarget: prepared.priorSessionTabTarget, processResult, qaAttachedTarget, qaPreset, recordingDependencyWarning, redactedArgs: prepared.redactedArgs, redactedCompiledElectron: prepared.redactedCompiledElectron, redactedCompiledJob: prepared.redactedCompiledJob, redactedCompiledNetworkSourceLookup: prepared.redactedCompiledNetworkSourceLookup, redactedCompiledQaPreset: prepared.redactedCompiledQaPreset, redactedCompiledSemanticAction: prepared.redactedCompiledSemanticAction, redactedCompiledSourceLookup: prepared.redactedCompiledSourceLookup, redactedContent, redactedProcessArgs: prepared.redactedProcessArgs, redactedRecoveryHint: prepared.redactedRecoveryHint, resultArtifactManifest, richInputRecoveryDiagnostic: finalRecoveryState.richInputRecoveryDiagnostic, scrollNoopDiagnostic, selectorTextVisibilityDiagnostics, sessionMode: prepared.sessionMode, sessionTabCorrection, sourceLookup, succeeded, timeoutPartialProgress, unsettledWebMcpMutation, userRequestedJson: prepared.userRequestedJson, visibleRefFallbackDiagnostic: finalRecoveryState.visibleRefFallbackDiagnostic, visibleRefFallbackSessionName: finalRecoveryState.visibleRefFallbackSessionName });
         const resultWithCloseAll = closeAllApplied
             ? { ...result, details: { ...(isRecord(result.details) ? result.details : {}), closeAllApplied: true } }
             : result;

package/dist/extensions/agent-browser/lib/orchestration/browser-run/recording-recovery.js

@@ -95,6 +95,7 @@ export async function recoverRecordingStop(options) {
         && !currentUsesSamePath;
     const data = matchedData ?? (expected ? { path: expected.absolutePath, recordingId: expected.recordingId, success: null } : undefined);
     const presentation = await buildToolPresentation({
+        modelVisible: options.modelVisible,
         artifactManifest: options.artifactManifest, artifactMinUpdatedAtMs: expected?.startedAtMs ?? options.artifactRunStartedAtMs, artifactMaxUpdatedAtMs: Date.now(),
         artifactRequest: expected ? { path: expected.path, absolutePath: expected.absolutePath, status: terminalMeasurements ? undefined : "unverified" } : undefined,
         commandInfo: { command: "record", subcommand: "stop" }, cwd: expected?.cwd ?? options.cwd,
@@ -139,7 +140,8 @@ export async function recoverRecordingStop(options) {
     presentation.nextActions = followups;
     presentation.recordingRecovery = recovery;
     const text = `Recording receipt recovery (${recovery.status}): ${recovery.reason}\nOriginal attempt: ${attemptError}`;
-    presentation.content.unshift({ type: "text", text });
+    if (options.modelVisible !== false)
+        presentation.content.unshift({ type: "text", text });
     return { batch, envelope, partialBatch: batch && !rows, stopIndex, presentation, recovery };
 }
 export function mergeRecordingRecoveryPresentation(base, result) {

package/dist/extensions/agent-browser/lib/orchestration/browser-run/screenshot-observation.js

@@ -0,0 +1,79 @@
+import { isRecord } from "../../parsing.js";
+import { getScreenshotPathTokenIndex, getScreenshotPositionalIndices } from "./artifact-paths.js";
+import { runSessionCommandData } from "./session-state.js";
+export function getScreenshotCapture(command) {
+    if (command[0] !== "screenshot")
+        return { kind: "unknown" };
+    if (command.includes("--full") || command.includes("-f"))
+        return { kind: "full-page" };
+    const positional = getScreenshotPositionalIndices(command);
+    const selectorIndex = positional.find(index => index !== getScreenshotPathTokenIndex(command));
+    return selectorIndex === undefined ? { kind: "viewport" } : { kind: "element", selector: command[selectorIndex] };
+}
+function finiteRecord(value, keys) {
+    return isRecord(value) && keys.every(key => typeof value[key] === "number" && Number.isFinite(value[key]));
+}
+export async function collectScreenshotSample(options) {
+    if (options.command[0] !== "screenshot" || options.signal?.aborted)
+        return undefined;
+    const { selector } = getScreenshotCapture(options.command);
+    const script = `(() => {
+		const root = document.documentElement, body = document.body, v = window.visualViewport;
+		let element;
+		try {
+			const selector = ${JSON.stringify(selector ?? null)};
+			const matches = selector ? document.querySelectorAll(selector) : [];
+			if (matches.length === 1) {
+				const rect = matches[0].getBoundingClientRect();
+				element = { x: rect.x, y: rect.y, width: rect.width, height: rect.height };
+			}
+		} catch {}
+		return { url: location.href, frame: window === window.top ? "main" : "child", childFrameCount: window.frames.length,
+			viewport: { width: innerWidth, height: innerHeight },
+			document: { width: Math.max(root.scrollWidth, body?.scrollWidth || 0, innerWidth), height: Math.max(root.scrollHeight, body?.scrollHeight || 0, innerHeight) },
+			scroll: { x: scrollX, y: scrollY }, dpr: devicePixelRatio,
+			visualViewport: { x: v?.offsetLeft ?? 0, y: v?.offsetTop ?? 0, scale: v?.scale ?? 1 }, element };
+	})()`;
+    try {
+        const data = await runSessionCommandData({ ...options, args: ["eval", "--stdin"], stdin: script, timeoutMs: 2_000 });
+        const sample = isRecord(data) ? data.result : undefined;
+        if (!isRecord(sample) || typeof sample.url !== "string" || !["main", "child"].includes(String(sample.frame))
+            || !finiteRecord(sample.viewport, ["width", "height"]) || !finiteRecord(sample.document, ["width", "height"])
+            || !finiteRecord(sample.scroll, ["x", "y"]) || !finiteRecord(sample.visualViewport, ["x", "y", "scale"])
+            || typeof sample.childFrameCount !== "number" || !Number.isInteger(sample.childFrameCount) || sample.childFrameCount < 0
+            || typeof sample.dpr !== "number" || !Number.isFinite(sample.dpr) || sample.dpr <= 0
+            || (sample.element !== undefined && !finiteRecord(sample.element, ["x", "y", "width", "height"])))
+            return undefined;
+        return sample;
+    }
+    catch {
+        return undefined;
+    }
+}
+export function buildScreenshotGeometry(options) {
+    const { before, after, capture, pixels } = options;
+    const unknown = (reason) => ({ status: "unknown", reason, before, after });
+    if (!before || !after)
+        return unknown("Capture was not bracketed by browser geometry samples; coordinates are unknown.");
+    if (JSON.stringify(before) !== JSON.stringify(after))
+        return unknown("Browser geometry changed across capture; no coordinate mapping is asserted.");
+    if (before.frame !== "main")
+        return unknown("Probe observed a child frame; its relation to the captured page is unknown.");
+    if (before.visualViewport.scale !== 1 || before.visualViewport.x !== 0 || before.visualViewport.y !== 0)
+        return unknown("Visual viewport is zoomed or offset; capture origin is unknown.");
+    let crop;
+    if (capture === "viewport")
+        crop = { ...before.scroll, ...before.viewport };
+    else if (capture === "full-page")
+        crop = { x: 0, y: 0, ...before.document };
+    else if (capture === "element" && before.element && before.childFrameCount === 0 && before.scroll.x === 0 && before.scroll.y === 0)
+        crop = before.element;
+    else
+        return unknown("Element crop/frame provenance is unverified (scrolled page, child frames, non-CSS selector, or ambiguous target).");
+    if (!pixels || crop.width <= 0 || crop.height <= 0 || crop.x < 0 || crop.y < 0
+        || !Object.values(crop).every(Number.isInteger)
+        || pixels.width !== crop.width * before.dpr || pixels.height !== crop.height * before.dpr)
+        return unknown("Image dimensions do not establish the sampled CSS crop at this DPR.");
+    return { status: "measured", before, after, crop, pixelsPerCssPixel: { x: pixels.width / crop.width, y: pixels.height / crop.height },
+        reason: "Matching pre/post samples, not an atomic capture guarantee. Pixels map to CSS document coordinates within crop; native mouse uses current viewport CSS coordinates. Recheck scroll/frame before input." };
+}

package/dist/extensions/agent-browser/lib/orchestration/electron-host/index.js

@@ -4,6 +4,7 @@ import { boundElectronProbeString } from "../../electron/cdp.js";
 import { buildOwnedManagedSessionRestoreContext, withOwnedManagedSessionContext, } from "../../managed-session-restore.js";
 import { getPageTargetValidationError } from "../../page-target-validation.js";
 import { isRecord } from "../../parsing.js";
+import { getBrowserResultMessage } from "../../browser-transcript.js";
 import { withAttachedBrowserSessionContext } from "../../process.js";
 import { buildAgentBrowserNextActions } from "../../results/action-recommendations.js";
 import { buildAgentBrowserResultCategoryDetails } from "../../results/categories.js";
@@ -125,10 +126,8 @@ function isElectronLaunchRecord(value) {
 export function restoreElectronLaunchRecordsFromBranch(branch) {
     const records = new Map();
     for (const entry of branch) {
-        if (!isRecord(entry) || entry.type !== "message")
-            continue;
-        const message = isRecord(entry.message) ? entry.message : undefined;
-        if (!message || message.toolName !== "agent_browser")
+        const message = getBrowserResultMessage(entry);
+        if (!message)
             continue;
         const details = isRecord(message.details) ? message.details : undefined;
         const electron = isRecord(details?.electron) ? details.electron : undefined;

package/dist/extensions/agent-browser/lib/orchestration/input-plan.js

@@ -3,10 +3,9 @@ import { normalizeUrlLessOpen } from "./batch-stdin.js";
 import { isPlainTextInspectionArgs, validateToolArgs, redactInvocationArgs, redactSensitiveText } from "../runtime.js";
 import { buildAgentBrowserResultCategoryDetails } from "../results/categories.js";
 import { compileAgentBrowserElectron } from "../input-modes/electron.js";
-import { compileAgentBrowserJob, compileAgentBrowserQaPreset } from "../input-modes/job.js";
+import { compileAgentBrowserQaPreset } from "../input-modes/job.js";
 import { compileAgentBrowserNetworkSourceLookup, compileAgentBrowserSourceLookup, redactNetworkSourceLookupArgs, redactNetworkSourceLookupUrl } from "../input-modes/lookups.js";
 import { compileAgentBrowserSemanticAction } from "../input-modes/semantic-action.js";
-import { AGENT_BROWSER_SCRIPT_MAX_TIMEOUT_MS, compileAgentBrowserScript } from "../input-modes/script.js";
 function redactCompiledElectron(compiled) {
     if (!compiled)
         return undefined;
@@ -68,46 +67,39 @@ function normalizeExplicitEvalStdinArgs(args, stdin) {
 export function resolveAgentBrowserInput(options) {
     const { getBatchPreflightValidationError, params } = options;
     const semanticActionResult = params.semanticAction === undefined ? {} : compileAgentBrowserSemanticAction(params.semanticAction);
-    const jobResult = params.job === undefined ? {} : compileAgentBrowserJob(params.job);
     const qaResult = params.qa === undefined ? {} : compileAgentBrowserQaPreset(params.qa);
     const sourceLookupResult = params.sourceLookup === undefined ? {} : compileAgentBrowserSourceLookup(params.sourceLookup);
     const networkSourceLookupResult = params.networkSourceLookup === undefined ? {} : compileAgentBrowserNetworkSourceLookup(params.networkSourceLookup);
     const electronResult = params.electron === undefined ? {} : compileAgentBrowserElectron(params.electron);
-    const scriptResult = params.script === undefined ? {} : compileAgentBrowserScript(params.script);
     const hasExplicitArgs = Array.isArray(params.args);
     const explicitInputModes = [
         hasExplicitArgs,
         Boolean(semanticActionResult.compiled),
-        Boolean(jobResult.compiled),
         Boolean(qaResult.compiled),
         Boolean(sourceLookupResult.compiled),
         Boolean(networkSourceLookupResult.compiled),
         Boolean(electronResult.compiled),
-        Boolean(scriptResult.compiled),
     ].filter(Boolean).length;
     const inputModeError = explicitInputModes !== 1
-        ? "Provide exactly one of script, args, semanticAction, job, qa, sourceLookup, networkSourceLookup, or electron."
+        ? "Provide exactly one of args, semanticAction, qa, sourceLookup, networkSourceLookup, or electron."
         : undefined;
     const compiledSemanticAction = semanticActionResult.compiled;
     const compiledQaPreset = qaResult.compiled;
     const compiledSourceLookup = sourceLookupResult.compiled;
     const compiledNetworkSourceLookup = networkSourceLookupResult.compiled;
     const compiledElectron = electronResult.compiled;
-    const compiledScript = scriptResult.compiled;
-    const compiledJob = jobResult.compiled ?? compiledQaPreset;
+    const compiledJob = compiledQaPreset;
     const compiledGeneratedBatch = compiledNetworkSourceLookup ?? compiledSourceLookup ?? compiledJob;
     const normalizedExplicitArgs = normalizeExplicitEvalStdinArgs(params.args ?? [], params.stdin);
-    const toolArgs = compiledElectron || compiledScript ? [] : compiledSemanticAction?.args ?? compiledGeneratedBatch?.args ?? normalizedExplicitArgs.args;
+    const toolArgs = compiledElectron ? [] : compiledSemanticAction?.args ?? compiledGeneratedBatch?.args ?? normalizedExplicitArgs.args;
     const toolStdin = compiledGeneratedBatch?.stdin ?? normalizedExplicitArgs.stdin;
     const redactedArgs = redactInvocationArgs(toolArgs);
     const generatedStdinError = params.stdin !== undefined
         ? compiledGeneratedBatch
-            ? "Do not provide stdin with job, qa, sourceLookup, or networkSourceLookup; those modes generate their own batch stdin."
+            ? "Do not provide stdin with qa, sourceLookup, or networkSourceLookup; those modes generate their own batch stdin."
             : compiledElectron
                 ? "Do not provide stdin with electron; electron mode is host-only or manages its own input."
-                : compiledScript
-                    ? "Do not provide stdin with script; browser call stdin belongs inside browser({ args, stdin })."
-                    : undefined
+                : undefined
         : undefined;
     const outputPathError = params.outputPath !== undefined && (typeof params.outputPath !== "string" || params.outputPath.trim().length === 0)
         ? "outputPath must be a non-empty string when provided."
@@ -118,52 +110,40 @@ export function resolveAgentBrowserInput(options) {
             ? compiledElectron.action === "list"
                 ? "electron.list has no configurable timeout; remove top-level timeoutMs."
                 : "Use electron.timeoutMs for this action; top-level timeoutMs applies only to browser CLI subprocess calls."
-            : compiledScript && params.timeoutMs !== undefined && params.timeoutMs > AGENT_BROWSER_SCRIPT_MAX_TIMEOUT_MS
-                ? `script timeoutMs must be ${AGENT_BROWSER_SCRIPT_MAX_TIMEOUT_MS} or less.`
-                : undefined;
-    const scriptSessionModeError = compiledScript && params.sessionMode !== undefined
-        ? "Do not provide sessionMode with script; script always uses its own isolated session."
-        : undefined;
+            : undefined;
     const attachedQaSessionError = compiledQaPreset?.checks.attached
         ? params.sessionMode === "fresh"
             ? "qa.attached cannot be used with sessionMode=fresh; attach or launch a session first, then run qa.attached with the current session."
             : undefined
         : undefined;
     const validationError = semanticActionResult.error
-        ?? jobResult.error
         ?? qaResult.error
         ?? sourceLookupResult.error
         ?? networkSourceLookupResult.error
         ?? electronResult.error
-        ?? scriptResult.error
         ?? inputModeError
         ?? generatedStdinError
         ?? outputPathError
         ?? timeoutMsError
-        ?? scriptSessionModeError
         ?? attachedQaSessionError
-        ?? (compiledElectron || compiledScript ? undefined : validateToolArgs(toolArgs) ?? getBatchPreflightValidationError(toolArgs, toolStdin));
+        ?? (compiledElectron ? undefined : validateToolArgs(toolArgs) ?? getBatchPreflightValidationError(toolArgs, toolStdin));
     const redactedCompiledJob = redactCompiledJob(compiledJob);
     const redactedCompiledSemanticAction = compiledSemanticAction
         ? { ...compiledSemanticAction, args: redactInvocationArgs(compiledSemanticAction.args) }
         : undefined;
     const attemptedKind = compiledElectron
         ? "electron"
-        : compiledScript
-            ? "script"
-            : compiledNetworkSourceLookup
-                ? "networkSourceLookup"
-                : compiledSourceLookup
-                    ? "sourceLookup"
-                    : compiledQaPreset
-                        ? "qa"
-                        : jobResult.compiled
-                            ? "job"
-                            : compiledSemanticAction
-                                ? "semanticAction"
-                                : hasExplicitArgs
-                                    ? "args"
-                                    : undefined;
+        : compiledNetworkSourceLookup
+            ? "networkSourceLookup"
+            : compiledSourceLookup
+                ? "sourceLookup"
+                : compiledQaPreset
+                    ? "qa"
+                    : compiledSemanticAction
+                        ? "semanticAction"
+                        : hasExplicitArgs
+                            ? "args"
+                            : undefined;
     const redactedCompiledElectron = redactCompiledElectron(compiledElectron);
     const redactedCompiledNetworkSourceLookup = redactCompiledNetworkSourceLookup(compiledNetworkSourceLookup);
     const redactedCompiledQaPreset = compiledQaPreset && redactedCompiledJob ? { ...redactedCompiledJob, checks: compiledQaPreset.checks } : undefined;
@@ -195,9 +175,6 @@ export function resolveAgentBrowserInput(options) {
     if (compiledElectron && redactedCompiledElectron) {
         return { ...resolvedBase, compiledElectron, kind: "electron", redactedCompiledElectron, status: "valid" };
     }
-    if (compiledScript) {
-        return { ...resolvedBase, compiledScript, kind: "script", status: "valid" };
-    }
     if (compiledNetworkSourceLookup && redactedCompiledNetworkSourceLookup) {
         return {
             ...resolvedBase,
@@ -230,16 +207,6 @@ export function resolveAgentBrowserInput(options) {
             status: "valid",
         };
     }
-    if (jobResult.compiled && redactedCompiledJob) {
-        return {
-            ...resolvedBase,
-            compiledGeneratedBatch: jobResult.compiled,
-            compiledJob: jobResult.compiled,
-            kind: "job",
-            redactedCompiledJob,
-            status: "valid",
-        };
-    }
     if (compiledSemanticAction && redactedCompiledSemanticAction) {
         return { ...resolvedBase, compiledSemanticAction, kind: "semanticAction", redactedCompiledSemanticAction, status: "valid" };
     }

package/dist/extensions/agent-browser/lib/orchestration/native-session-defaults.js

@@ -75,7 +75,7 @@ async function readNativeIdentity(path, cwd, signal, rootFallback, browserComman
 export async function withNativeSessionDefaults(input, cwd, signal, run, root) {
     if (input.kind === "electron" && input.compiledElectron.action === "launch")
         return withAgentBrowserProcessEnvironment({ AGENT_BROWSER_SESSION: undefined }, () => run(input));
-    if (input.kind === "script" || input.kind === "electron" || isPlainTextInspectionArgs(input.toolArgs))
+    if (input.kind === "electron" || isPlainTextInspectionArgs(input.toolArgs))
         return run(input);
     const env = getAgentBrowserProcessEnvironment();
     const configArg = scanUpstreamGlobalFlagOccurrences(input.toolArgs, "--config")[0];
@@ -127,8 +127,8 @@ export async function withNativeSessionDefaults(input, cwd, signal, run, root) {
         ...(configPath !== undefined ? { AGENT_BROWSER_CONFIG: resolve(cwd, configPath) } : {}),
         ...(session !== undefined ? { AGENT_BROWSER_SESSION: session } : {}),
         ...(namespace !== undefined ? { AGENT_BROWSER_NAMESPACE: namespace } : {}),
-    }, () => run({ ...input, toolArgs: args, chromeStartupArgs, persistentChromeArgs, configuredChromeLaunch }, !rootDefault ? undefined : async (browserRun) => {
-        const daemon = await inspectManagedSessionDaemon({ cwd, signal, sessionName: rootDefault,
+    }, () => run({ ...input, toolArgs: args, chromeStartupArgs, persistentChromeArgs, configuredChromeLaunch }, !rootDefault ? undefined : async (browserRun, launchSignal = signal) => {
+        const daemon = await inspectManagedSessionDaemon({ cwd, signal: launchSignal, sessionName: rootDefault,
             namespace: scanUpstreamGlobalFlagOccurrences(args, "--namespace").at(-1)?.value ?? namespace, timeoutMs: 5_000 });
         if (daemon.status === "missing-binary")
             return {

package/dist/extensions/agent-browser/lib/orchestration/operation-paths.js

@@ -0,0 +1,76 @@
+import { isAbsolute, resolve } from "node:path";
+import { projectUpstreamGlobalFlags, scanUpstreamGlobalFlagOccurrences } from "../argv-grammar.js";
+import { getDiffFilePathIndices, getExplicitArtifactDestinationIndex } from "./browser-run/artifact-paths.js";
+import { parseBatchCommandArgument, parseUserBatchStdin } from "./batch-stdin.js";
+/** Bind caller file operands without moving the native process/config or interpreting literal form data. */
+export function resolveOperationPaths(args, stdin, cwd, batchStep = false) {
+    let resolved = args;
+    const setPath = (index, prefix = "") => {
+        const path = index === undefined ? undefined : args[index]?.slice(prefix.length);
+        if (index === undefined || !path || isAbsolute(path))
+            return;
+        if (resolved === args)
+            resolved = [...args];
+        resolved[index] = prefix + resolve(cwd, path);
+    };
+    const { tokens, indices } = batchStep ? { tokens: args, indices: args.map((_, index) => index) } : projectUpstreamGlobalFlags(args);
+    if (!batchStep) {
+        for (const flag of ["--config", "--ca-cert", "--executable-path", "--extension", "--init-script", "--download-path", "--screenshot-dir", "--action-policy", "--state"]) {
+            for (const occurrence of scanUpstreamGlobalFlagOccurrences(args, flag)) {
+                if (flag === "--state" && tokens[0] === "wait" && occurrence.index > indices[0])
+                    continue;
+                setPath(occurrence.index + 1);
+            }
+        }
+        for (const occurrence of scanUpstreamGlobalFlagOccurrences(args, "--profile")) {
+            // Native Chrome profile names are identities, not workspace file paths.
+            if (occurrence.value && /[\\/]/.test(occurrence.value) && !occurrence.value.startsWith("~"))
+                setPath(occurrence.index + 1);
+        }
+    }
+    if (tokens[0] === "batch") {
+        if (tokens.slice(1).some(token => token !== "--bail")) {
+            for (let index = 1; index < tokens.length; index++) {
+                if (tokens[index] === "--bail")
+                    continue;
+                const row = parseBatchCommandArgument(tokens[index]).step;
+                if (!row)
+                    continue;
+                const bound = resolveOperationPaths(row, undefined, cwd, true).args;
+                if (bound === row)
+                    continue;
+                if (resolved === args)
+                    resolved = [...args];
+                resolved[indices[index]] = bound.map(token => `'${token.replaceAll("'", "'\\''")}'`).join(" ");
+            }
+            return { args: resolved, stdin }; // Native ignores stdin when raw rows exist.
+        }
+        const rows = parseUserBatchStdin(stdin).steps;
+        if (!rows?.length)
+            return { args: resolved, stdin };
+        const bound = rows.map(row => resolveOperationPaths(row, undefined, cwd, true).args);
+        return { args: resolved, stdin: bound.some((row, index) => row !== rows[index]) ? JSON.stringify(bound) : stdin };
+    }
+    setPath(indices[getExplicitArtifactDestinationIndex(tokens) ?? -1]);
+    if (tokens[0] === "upload")
+        for (let index = 2; index < tokens.length; index++)
+            setPath(indices[index]);
+    if (tokens[0] === "state" && tokens[1] === "load")
+        setPath(indices[2]);
+    setPath(indices[getDiffFilePathIndices(tokens).baseline ?? -1]);
+    if (tokens[0] === "cookies" && tokens[1] === "set") {
+        const index = tokens.indexOf("--curl");
+        if (index >= 0)
+            setPath(indices[index + 1]);
+    }
+    if (tokens[0] === "webmcp" && tokens[1] === "invoke") {
+        for (let index = 3; index < tokens.length; index++) {
+            if (!["--params", "--frame", "--timeout"].includes(tokens[index]))
+                continue;
+            if (tokens[index] === "--params" && tokens[index + 1]?.startsWith("@"))
+                setPath(indices[index + 1], "@");
+            index++;
+        }
+    }
+    return { args: resolved, stdin };
+}

package/dist/extensions/agent-browser/lib/orchestration/output-file.js

@@ -138,7 +138,7 @@ export async function applyAgentBrowserOutputPath(options) {
         return {
             ...options.result,
             content: appendOutputFileNotice(options.result, `Output file rejected: ${message}`, true),
-            details: { ...details, failureCategory: "validation-error", outputFile, resultCategory: "failure" },
+            details: { ...details, error: message, summary: "Output file rejected.", failureCategory: "validation-error", outputFile, resultCategory: "failure" },
             isError: true,
         };
     }
@@ -163,9 +163,9 @@ export async function applyAgentBrowserOutputPath(options) {
             ? (() => {
                 const rest = { ...options.result.details };
                 delete rest.successCategory;
-                return { ...rest, failureCategory: rest.failureCategory ?? "upstream-error", outputFile, resultCategory: "failure" };
+                return { ...rest, error: message, summary: "Output file failed.", failureCategory: rest.failureCategory ?? "upstream-error", outputFile, resultCategory: "failure" };
             })()
-            : { failureCategory: "upstream-error", outputFile, resultCategory: "failure" };
+            : { error: message, summary: "Output file failed.", failureCategory: "upstream-error", outputFile, resultCategory: "failure" };
         return {
             ...options.result,
             content: appendOutputFileNotice(options.result, `Output file failed: ${requestedPath} (${message}).`, true),