@steipete/oracle 0.15.2 → 0.16.0

package/dist/src/browser/actions/modelSelection.js

@@ -36,9 +36,10 @@ export async function ensureModelSelection(Runtime, desiredModel, logger, strate
     switch (result?.status) {
         case "already-selected":
         case "switched": {
-            const label = result.label?.trim() || (strategy === "current" ? null : desiredModel);
+            const observedLabel = result.label?.trim() || null;
+            const label = observedLabel || (strategy === "current" ? null : desiredModel);
             if (strategy !== "current") {
-                assertResolvedModelSelection(desiredModel, label ?? desiredModel);
+                assertResolvedModelSelection(desiredModel, observedLabel ?? "");
             }
             logger(`Model picker: ${label ?? "current model (label unavailable)"}`);
             return {
@@ -70,6 +71,18 @@ export async function ensureModelSelection(Runtime, desiredModel, logger, strate
 function assertResolvedModelSelection(desiredModel, resolvedLabel) {
     const desired = desiredModel.toLowerCase();
     const resolved = resolvedLabel.toLowerCase();
+    const normalizedDesired = normalizeResolvedModelLabel(desired);
+    const normalizedResolved = normalizeResolvedModelLabel(resolved);
+    const wantsGpt56Sol = /(?:^| )5 6(?: |$)/.test(normalizedDesired) && normalizedDesired.split(" ").includes("sol");
+    if (wantsGpt56Sol) {
+        const resolvedTokens = normalizedResolved.split(" ");
+        if (!/(?:^| )5 6(?: |$)/.test(normalizedResolved) ||
+            !resolvedTokens.includes("sol") ||
+            resolvedTokens.includes("pro")) {
+            throw new Error(`Model picker selected "${resolvedLabel}" while "${desiredModel}" requires GPT-5.6 Sol.`);
+        }
+        return;
+    }
     const wantsGpt55Pro = desired === "pro" ||
         desired === "chatgpt pro" ||
         desired === "gpt-5.5-pro" ||
@@ -152,10 +165,12 @@ function buildModelSelectionExpression(targetModel, strategy) {
       .map((token) => normalizeText(token))
       .filter(Boolean);
     const targetWords = normalizedTarget.split(' ').filter(Boolean);
-    const desiredVersion = normalizedTarget.includes('5 4')
-      ? '5-4'
+    const desiredVersion = normalizedTarget.includes('5 6')
+      ? '5-6'
       : normalizedTarget.includes('5 5')
         ? '5-5'
+        : normalizedTarget.includes('5 4')
+          ? '5-4'
         : normalizedTarget.includes('5 3')
           ? '5-3'
         : normalizedTarget.includes('5 2')
@@ -165,6 +180,7 @@ function buildModelSelectionExpression(targetModel, strategy) {
           : normalizedTarget.includes('5 0')
             ? '5-0'
           : null;
+    const desiredModelVariant = normalizedTarget.includes(' sol') ? 'sol' : null;
     const wantsPro = normalizedTarget.includes(' pro') || normalizedTarget.endsWith(' pro') || normalizedTokens.includes('pro');
     const wantsInstant = normalizedTarget.includes('instant');
     const wantsThinking = normalizedTarget.includes('thinking');
@@ -280,12 +296,14 @@ function buildModelSelectionExpression(targetModel, strategy) {
       label === 'medium' ||
       label === 'high' ||
       label === 'extra high' ||
+      label === 'pro' ||
       label === 'extended' ||
       label === 'standard' ||
       label === 'heavy' ||
       label === 'light';
     const formatModelOptionLabel = (label) => {
       const normalized = normalizeText(label ?? '');
+      if (normalized === '5 6' || normalized === 'gpt 5 6') return 'GPT-5.6';
       if (normalized === '5 5' || normalized === 'gpt 5 5') return 'GPT-5.5';
       if (normalized === '5 4' || normalized === 'gpt 5 4') return 'GPT-5.4';
       if (normalized === '5 3' || normalized === 'gpt 5 3') return 'GPT-5.3';
@@ -319,17 +337,12 @@ function buildModelSelectionExpression(targetModel, strategy) {
       return 1000;
     };
     const versionFromLabel = (label) => {
-      if (label === '5 5' || label === 'gpt 5 5') return '5-5';
-      if (label === '5 4' || label === 'gpt 5 4') return '5-4';
-      if (label === '5 3' || label === 'gpt 5 3') return '5-3';
-      if (label === '5 2' || label === 'gpt 5 2') return '5-2';
-      if (label === '5 1' || label === 'gpt 5 1') return '5-1';
-      if (label === '5 0' || label === 'gpt 5 0') return '5-0';
-      if (label === '4 5' || label === 'gpt 4 5') return '4-5';
-      return null;
+      const match = normalizeText(label).match(/(?:^| )(?:gpt )?(\\d+) (\\d+)(?: |$)/);
+      return match ? match[1] + '-' + match[2] : null;
     };
     const versionFromTestId = (testid) => {
       const normalized = normalizeText(testid);
+      if (normalized.includes('5 6') || normalized.includes('gpt56')) return '5-6';
       if (normalized.includes('5 5') || normalized.includes('gpt55')) return '5-5';
       if (normalized.includes('5 4') || normalized.includes('gpt54')) return '5-4';
       if (normalized.includes('5 3') || normalized.includes('gpt53')) return '5-3';
@@ -355,9 +368,16 @@ function buildModelSelectionExpression(targetModel, strategy) {
       return label;
     };
     const configuredSelectionMatchesTarget = () => {
-      const configuredVersion = versionFromLabel(normalizeText(getConfiguredVersionLabel()));
+      const configuredVersionLabel = normalizeText(getConfiguredVersionLabel());
+      const configuredVersion = versionFromLabel(configuredVersionLabel);
       if (!configuredVersion || configuredVersion !== desiredVersion) return false;
       const configuredVariant = normalizeText(getConfiguredVariantLabel());
+      if (
+        desiredModelVariant &&
+        !configuredVersionLabel.includes(desiredModelVariant) &&
+        !configuredVariant.includes(desiredModelVariant)
+      ) return false;
+      if (desiredModelVariant === 'sol' && labelHasProWord(configuredVariant)) return false;
       if (wantsPro) return labelHasProWord(configuredVariant);
       if (wantsInstant) return configuredVariant.includes('instant');
       if (wantsThinking) {
@@ -372,10 +392,36 @@ function buildModelSelectionExpression(targetModel, strategy) {
         return [variant, version].filter(Boolean).join(' ');
       }
       const composerLabel = getComposerModelLabel();
-      if (composerLabel) return withProPillSignal(composerLabel);
+      const normalizedComposerLabel = normalizeText(composerLabel);
+      if (
+        desiredVersion &&
+        desiredModelVariant &&
+        versionFromLabel(normalizedComposerLabel) === desiredVersion &&
+        normalizedComposerLabel.split(' ').includes(desiredModelVariant)
+      ) {
+        return withProPillSignal(composerLabel);
+      }
       const buttonLabel = getButtonLabel();
       const normalizedButton = normalizeText(buttonLabel);
+      if (
+        desiredVersion &&
+        desiredModelVariant &&
+        versionFromLabel(normalizedButton) === desiredVersion &&
+        normalizedButton.split(' ').includes(desiredModelVariant)
+      ) {
+        return withProPillSignal(buttonLabel);
+      }
       const fallbackLabel = formatModelOptionLabel(fallback);
+      const normalizedFallback = normalizeText(fallbackLabel);
+      if (
+        desiredVersion &&
+        desiredModelVariant &&
+        versionFromLabel(normalizedFallback) === desiredVersion &&
+        normalizedFallback.split(' ').includes(desiredModelVariant)
+      ) {
+        return fallbackLabel;
+      }
+      if (composerLabel) return withProPillSignal(composerLabel);
       if (fallbackLabel && !wantsPro && isIntelligenceEffortLabel(normalizedButton)) {
         return fallbackLabel;
       }
@@ -397,6 +443,7 @@ function buildModelSelectionExpression(targetModel, strategy) {
       if (configuredSelectionMatchesTarget()) return true;
       const normalizedLabel = normalizeText(getButtonLabel());
       if (!normalizedLabel) return false;
+      if (desiredModelVariant === 'sol' && hasProComposerPill()) return false;
       if (wantsThinking && !wantsPro && hasProComposerPill()) return false;
       if (isTargetGpt55VisibleAlias(normalizedLabel)) return true;
       if (
@@ -421,6 +468,7 @@ function buildModelSelectionExpression(targetModel, strategy) {
         return true;
       }
       if (desiredVersion) {
+        if (desiredVersion === '5-6' && !normalizedLabel.includes('5 6')) return false;
         if (desiredVersion === '5-5' && !normalizedLabel.includes('5 5')) return false;
         if (desiredVersion === '5-4' && !normalizedLabel.includes('5 4')) return false;
         if (desiredVersion === '5-3' && !normalizedLabel.includes('5 3')) return false;
@@ -428,6 +476,7 @@ function buildModelSelectionExpression(targetModel, strategy) {
         if (desiredVersion === '5-1' && !normalizedLabel.includes('5 1')) return false;
         if (desiredVersion === '5-0' && !normalizedLabel.includes('5 0')) return false;
       }
+      if (desiredModelVariant && !normalizedLabel.includes(desiredModelVariant)) return false;
       if (wantsPro && labelHasLegacyProVersion(normalizedLabel)) return false;
       if (wantsPro && !labelHasProWord(normalizedLabel)) return false;
       if (wantsInstant && !normalizedLabel.includes('instant')) return false;
@@ -455,6 +504,9 @@ function buildModelSelectionExpression(targetModel, strategy) {
       if (!signal) {
         return COMPOSER_SIGNAL_ALLOW_BLANK;
       }
+      if (desiredModelVariant === 'sol' && hasProComposerPill()) {
+        return false;
+      }
       if (wantsPro && labelHasLegacyProVersion(signal)) {
         return false;
       }
@@ -542,6 +594,9 @@ function buildModelSelectionExpression(targetModel, strategy) {
       let score = 0;
       const normalizedTestId = (testid ?? '').toLowerCase();
       const candidateTextVersion = versionFromLabel(normalizedText);
+      const candidateIsVersionSubmenuTrigger = Boolean(
+        candidateTextVersion && node?.getAttribute?.('aria-haspopup') === 'menu'
+      );
       const candidateIsVersionCombobox =
         node?.getAttribute?.('role') === 'combobox' &&
         node?.getAttribute?.('aria-labelledby') === 'model-selection-label';
@@ -553,6 +608,20 @@ function buildModelSelectionExpression(targetModel, strategy) {
         desiredVersion !== '5-5' &&
         (normalizedTestId === 'model-configure-modal' ||
           (candidateIsVersionCombobox && node?.getAttribute?.('aria-expanded') !== 'true'));
+      const candidateIsConfiguredVersionNavigation =
+        Boolean(getConfigurationDialog()) &&
+        candidateTextVersion === desiredVersion &&
+        (node?.getAttribute?.('role') === 'option' ||
+          node?.getAttribute?.('role') === 'menuitemradio');
+      if (
+        desiredModelVariant &&
+        !normalizedText.includes(desiredModelVariant) &&
+        !candidateIsVersionSubmenuTrigger &&
+        !candidateOpensConfiguration &&
+        !candidateIsConfiguredVersionNavigation
+      ) {
+        return 0;
+      }
       if (candidateOpensConfiguration) {
         return 2000;
       }
@@ -560,6 +629,12 @@ function buildModelSelectionExpression(targetModel, strategy) {
       if (normalizedTestId) {
         if (desiredVersion) {
           // data-testid strings have been observed with both dotted and dashed versions (e.g. gpt-5.2-pro vs gpt-5-2-pro).
+          const has56 =
+            normalizedTestId.includes('5-6') ||
+            normalizedTestId.includes('5.6') ||
+            normalizedTestId.includes('gpt-5-6') ||
+            normalizedTestId.includes('gpt-5.6') ||
+            normalizedTestId.includes('gpt56');
           const has52 =
             normalizedTestId.includes('5-2') ||
             normalizedTestId.includes('5.2') ||
@@ -596,8 +671,10 @@ function buildModelSelectionExpression(targetModel, strategy) {
             normalizedTestId.includes('gpt-5-0') ||
             normalizedTestId.includes('gpt-5.0') ||
             normalizedTestId.includes('gpt50');
-          const candidateVersion = has55
-            ? '5-5'
+          const candidateVersion = has56
+            ? '5-6'
+            : has55
+              ? '5-5'
             : has54
               ? '5-4'
               : has53
@@ -645,10 +722,11 @@ function buildModelSelectionExpression(targetModel, strategy) {
       const candidateClearsProForThinking =
         wantsThinking && !wantsPro && hasActiveProPill && candidateIsNonProThinkingEffort;
       const candidateOpensVersionSubmenu =
-        wantsThinking &&
-        desiredVersion !== '5-5' &&
-        normalizedText === 'gpt 5 5' &&
-        !normalizedTestId.includes('pro');
+        (desiredVersion === '5-6' && candidateIsVersionSubmenuTrigger) ||
+        (wantsThinking &&
+          desiredVersion !== '5-5' &&
+          normalizedText === 'gpt 5 5' &&
+          !normalizedTestId.includes('pro'));
       const candidateSelectsDesiredVersion =
         Boolean(desiredVersion) && candidateTextVersion === desiredVersion;
       if (
@@ -695,6 +773,7 @@ function buildModelSelectionExpression(targetModel, strategy) {
           node?.getAttribute?.('aria-haspopup') === 'menu' ||
           node?.getAttribute?.('data-has-submenu') !== null);
       if (wantsPro && candidateHasThinking) return 0;
+      if (desiredModelVariant === 'sol' && candidateHasPro) return 0;
       if (wantsPro && candidateHasLegacyProVersion && !candidateSelectsDesiredVersion) return 0;
       if (wantsPro && !candidateHasPro && !candidateSelectsDesiredVersion) return 0;
       if (
@@ -850,10 +929,17 @@ function buildModelSelectionExpression(targetModel, strategy) {
       if (getConfigurationDialog() && !configuredSelectionMatchesTarget()) return false;
       const optionVersion = versionFromLabel(normalizedText) ?? versionFromTestId(testid);
       if (desiredVersion && optionVersion !== desiredVersion) return false;
+      if (desiredModelVariant && !normalizedText.includes(desiredModelVariant)) return false;
+      if (
+        desiredModelVariant === 'sol' &&
+        (labelHasProWord(normalizedText) || normalizeText(testid ?? '').includes('pro'))
+      ) return false;
+      if (desiredVersion === '5-6') return !hasProComposerPill();
       const currentButtonLabel = normalizeText(getButtonLabel());
       return !labelHasProWord(currentButtonLabel) && !hasProComposerPill();
     };
     const openedSubmenuKeys = new Set();
+    let clickedTargetOption = false;
     const submenuKey = (normalizedText, testid) =>
       normalizeText(testid ?? '') + '|' + normalizedText;
 
@@ -927,10 +1013,13 @@ function buildModelSelectionExpression(targetModel, strategy) {
     });
     const dispatchHoverSequence = (target) => {
       if (!target || !(target instanceof EventTarget)) return false;
+      const rect = target.getBoundingClientRect?.();
+      const clientX = rect ? rect.x + rect.width / 2 : 0;
+      const clientY = rect ? rect.y + rect.height / 2 : 0;
       const types = ['pointerover', 'pointerenter', 'mouseover', 'mouseenter', 'pointermove', 'mousemove'];
       for (const type of types) {
         try {
-          const common = { bubbles: true, cancelable: true, view: window };
+          const common = { bubbles: true, cancelable: true, view: window, clientX, clientY };
           const event =
             type.startsWith('pointer') && 'PointerEvent' in window
               ? new PointerEvent(type, { ...common, pointerId: 1, pointerType: 'mouse' })
@@ -1002,7 +1091,10 @@ function buildModelSelectionExpression(targetModel, strategy) {
           ) {
             const resolvedLabel = getResolvedLabel(match.label);
             closeMenu();
-            resolve({ status: 'already-selected', label: resolvedLabel });
+            resolve({
+              status: clickedTargetOption ? 'switched' : 'already-selected',
+              label: resolvedLabel,
+            });
             return;
           }
           const previousButtonLabel = normalizeText(getButtonLabel());
@@ -1016,6 +1108,7 @@ function buildModelSelectionExpression(targetModel, strategy) {
             setTimeout(attempt, REOPEN_INTERVAL_MS / 2);
             return;
           }
+          clickedTargetOption = true;
           dispatchClickSequence(match.node);
           // Wait for the selected model signal to settle before reopening the picker.
           waitForTargetSelection(previousButtonLabel, previousComposerSignal).then((selectionSettled) => {
@@ -1052,6 +1145,9 @@ function buildComposerSignalMatchers(targetModel) {
         .replace(/[^a-z0-9]+/g, " ")
         .replace(/\s+/g, " ")
         .trim();
+    if (normalized.includes("5 6")) {
+        return { includesAny: ["5 6 sol"], excludesAny: ["pro"], allowBlank: false };
+    }
     if (normalized.includes("pro")) {
         return { includesAny: ["pro"], excludesAny: ["thinking"], allowBlank: false };
     }
@@ -1086,6 +1182,21 @@ function buildModelMatchersLiteral(targetModel) {
     push(`chatgpt ${dotless}`, labelTokens);
     push(`gpt ${base}`, labelTokens);
     push(`gpt ${dotless}`, labelTokens);
+    // Numeric variations (5.6 <-> 56 <-> gpt-5-6)
+    if (base.includes("5.6") || base.includes("5-6") || base.includes("56")) {
+        push("5.6", labelTokens);
+        push("gpt-5.6", labelTokens);
+        push("gpt5.6", labelTokens);
+        push("gpt-5-6", labelTokens);
+        push("gpt5-6", labelTokens);
+        push("gpt56", labelTokens);
+        push("gpt-5.6 sol", labelTokens);
+        push("chatgpt 5.6", labelTokens);
+        testIdTokens.add("model-switcher-gpt-5-6");
+        testIdTokens.add("gpt-5-6");
+        testIdTokens.add("gpt5-6");
+        testIdTokens.add("gpt56");
+    }
     // Numeric variations (5.5 <-> 55 <-> gpt-5-5)
     if (base.includes("5.5") || base.includes("5-5") || base.includes("55")) {
         push("5.5", labelTokens);

package/dist/src/browser/actions/navigation.js

@@ -124,6 +124,172 @@ export async function navigateToPromptReadyWithFallback(Page, Runtime, options,
         return { usedFallback: true };
     }
 }
+function buildChatModeProbeExpression() {
+    return `(() => {
+    const normalize = (value) => String(value || '').toLowerCase().replace(/\\s+/g, ' ').trim();
+    const isVisible = (node) => {
+      if (!(node instanceof HTMLElement)) return false;
+      const rect = node.getBoundingClientRect();
+      if (rect.width <= 0 || rect.height <= 0) return false;
+      const style = window.getComputedStyle(node);
+      return style.display !== 'none' && style.visibility !== 'hidden' && style.opacity !== '0';
+    };
+    const isSelected = (node) =>
+      node?.getAttribute?.('aria-checked') === 'true' ||
+      node?.getAttribute?.('data-state') === 'on';
+    const conversationIdFromPath = (value) => {
+      const match = String(value || '').match(/\\/c\\/([a-zA-Z0-9-]+)/);
+      return match?.[1] || null;
+    };
+    const isStructuredWorkBadge = (node) =>
+      node instanceof HTMLElement &&
+      node.tagName === 'SPAN' &&
+      normalize(node.textContent) === 'work' &&
+      node.childElementCount === 0 &&
+      !node.hasAttribute('dir') &&
+      node.classList.contains('shrink-0') &&
+      node.parentElement?.matches('span.flex.items-center');
+
+    const pathname = typeof location?.pathname === 'string' ? location.pathname : '';
+    const conversationId = conversationIdFromPath(pathname);
+    if (conversationId) {
+      // Conversation messages can contain same-origin links to the current thread. Only sidebar
+      // history items use ChatGPT's renderer-owned menu-item anchor class.
+      const activeHistoryLinks = Array.from(
+        document.querySelectorAll('a.__menu-item[href*="/c/"]'),
+      ).filter((node) => {
+        try {
+          const candidateUrl = new URL(node.getAttribute('href') || '', location.origin);
+          return candidateUrl.origin === location.origin && conversationIdFromPath(candidateUrl.pathname) === conversationId;
+        } catch {
+          return false;
+        }
+      });
+      if (activeHistoryLinks.length > 0) {
+        const hasWorkBadge = activeHistoryLinks.some((link) =>
+          Array.from(link.querySelectorAll('span')).some(isStructuredWorkBadge),
+        );
+        if (hasWorkBadge) return { status: 'work-conversation' };
+
+        const ariaLabels = activeHistoryLinks
+          .map((link) => normalize(link.getAttribute('aria-label')))
+          .filter(Boolean);
+        if (ariaLabels.length === 0 || ariaLabels.some((aria) => /,\\s*work\\s*$/.test(aria))) {
+          return { status: 'conversation-unresolved' };
+        }
+        return { status: 'chat-conversation' };
+      }
+      return { status: 'conversation-unresolved' };
+    }
+
+    const radios = Array.from(document.querySelectorAll('button[role="radio"]')).filter(isVisible);
+    const chat = radios.find((node) => normalize(node.textContent) === 'chat');
+    const work = radios.find((node) => normalize(node.textContent) === 'work');
+    if (!chat && !work) return { status: 'controls-absent' };
+    if (!chat || !work) return { status: 'mode-unresolved' };
+    if (isSelected(chat)) return { status: 'chat-selected' };
+    if (isSelected(work)) {
+      const rect = chat.getBoundingClientRect();
+      return {
+        status: 'work-selected',
+        chatPoint: { x: rect.left + rect.width / 2, y: rect.top + rect.height / 2 },
+      };
+    }
+    return { status: 'mode-unresolved' };
+  })()`;
+}
+export async function ensureChatMode(Runtime, Input, timeoutMs, logger, options = {}) {
+    const verificationWindowMs = Math.min(Math.max(0, timeoutMs), 10_000);
+    let deadline = Date.now() + verificationWindowMs;
+    const pollMs = Math.max(0, options.pollMs ?? 200);
+    let changedFromWork = false;
+    let clickedChat = false;
+    for (;;) {
+        const outcome = await Runtime.evaluate({
+            expression: buildChatModeProbeExpression(),
+            returnByValue: true,
+        });
+        const probe = outcome.result?.value;
+        if (probe?.status === "chat-selected" || probe?.status === "chat-conversation") {
+            logger(changedFromWork ? "ChatGPT mode: Chat (switched from Work)" : "ChatGPT mode: Chat");
+            return changedFromWork ? "switched" : "chat";
+        }
+        if (probe?.status === "conversation-unresolved") {
+            if (Date.now() < deadline) {
+                await delay(pollMs);
+                continue;
+            }
+            if (changedFromWork)
+                break;
+            if (options.resetWorkConversation) {
+                logger("ChatGPT conversation mode unresolved; opening a new Chat");
+                await options.resetWorkConversation();
+                changedFromWork = true;
+                deadline = Date.now() + verificationWindowMs;
+                await delay(pollMs);
+                continue;
+            }
+            throw new BrowserAutomationError("Oracle could not verify whether the selected ChatGPT conversation uses Chat or Work mode, so it cannot safely resume that conversation.", { stage: "chat-mode-selection", details: { mode: "conversation-unresolved" } });
+        }
+        if (probe?.status === "controls-absent") {
+            if (changedFromWork) {
+                if (Date.now() < deadline) {
+                    await delay(pollMs);
+                    continue;
+                }
+                break;
+            }
+            return "unavailable";
+        }
+        if (probe?.status === "work-conversation") {
+            if (changedFromWork)
+                break;
+            if (options.resetWorkConversation) {
+                logger("ChatGPT mode: Work conversation; opening a new Chat");
+                await options.resetWorkConversation();
+                changedFromWork = true;
+                deadline = Date.now() + verificationWindowMs;
+                await delay(pollMs);
+                continue;
+            }
+            throw new BrowserAutomationError("The selected ChatGPT tab is an existing Work conversation and cannot be resumed as an ordinary Chat conversation. Start a new Chat conversation instead.", { stage: "chat-mode-selection", details: { mode: "work-conversation" } });
+        }
+        if (probe?.status === "work-selected" && !clickedChat && probe.chatPoint) {
+            logger("ChatGPT mode: Work; switching to Chat");
+            const { x, y } = probe.chatPoint;
+            await Input.dispatchMouseEvent({ type: "mouseMoved", x, y });
+            await Input.dispatchMouseEvent({
+                type: "mousePressed",
+                x,
+                y,
+                button: "left",
+                clickCount: 1,
+            });
+            await Input.dispatchMouseEvent({
+                type: "mouseReleased",
+                x,
+                y,
+                button: "left",
+                clickCount: 1,
+            });
+            changedFromWork = true;
+            clickedChat = true;
+            deadline = Date.now() + verificationWindowMs;
+            await delay(pollMs);
+            continue;
+        }
+        if (Date.now() >= deadline)
+            break;
+        await delay(pollMs);
+    }
+    await logDomFailure(Runtime, logger, "chat-mode-selection");
+    throw new BrowserAutomationError(changedFromWork
+        ? "Oracle could not verify Chat mode after leaving Work."
+        : "ChatGPT exposed Chat/Work controls but Oracle could not verify the active mode.", { stage: "chat-mode-selection", details: { changedFromWork, clickedChat } });
+}
+export function buildChatModeProbeExpressionForTest() {
+    return buildChatModeProbeExpression();
+}
 export async function ensureNotBlocked(Runtime, headless, logger) {
     if (await isCloudflareInterstitial(Runtime)) {
         const message = headless
@@ -456,22 +622,77 @@ async function waitForPrompt(Runtime, timeoutMs) {
     }
     return false;
 }
-async function isCloudflareInterstitial(Runtime) {
-    const { result: titleResult } = await Runtime.evaluate({
-        expression: "document.title",
-        returnByValue: true,
-    });
-    const title = typeof titleResult.value === "string" ? titleResult.value : "";
-    const challengeTitle = CLOUDFLARE_TITLE.toLowerCase();
-    if (title.toLowerCase().includes(challengeTitle)) {
-        return true;
+// A single injected verdict. The old check flagged a challenge whenever the
+// '/challenge-platform/' script was present, but Cloudflare bot-management injects that script on
+// NORMAL ChatGPT pages too (including the new GPT-5.6 "Work" UI), so it false-flagged healthy pages
+// as challenges (which then aborted the run). Classification by evidence strength:
+//   strong - the "Just a moment"/"Attention Required" title, a challenge widget, or
+//            verification copy on a short page: a real interstitial, classify immediately.
+//   shell  - the ChatGPT app shell rendered: never a challenge, regardless of injected scripts.
+//   weak   - only the bot-management script on a short, shell-less page. This is INFERENCE, and
+//            right after navigation it also matches a healthy SPA load mid-hydration (readyState
+//            fires before React renders the shell), so weak evidence must PERSIST through a
+//            hydration grace window before it counts (see isCloudflareInterstitial).
+export function buildCloudflareVerdictExpression() {
+    return `(() => {
+    const title = String(document.title || '').toLowerCase();
+    const titleSaysChallenge =
+      title.includes(${JSON.stringify(CLOUDFLARE_TITLE.toLowerCase())}) ||
+      (title.includes('attention required') && title.includes('cloudflare'));
+    const hasAppShell = Boolean(document.querySelector(
+      '#prompt-textarea, [data-testid="prompt-textarea"], [data-testid^="conversation-turn"], [data-testid="profile-button"], main form[data-type], nav a[href*="/c/"]'
+    ));
+    const bodyText = String((document.body && document.body.innerText) || '')
+      .toLowerCase().replace(/\\s+/g, ' ').trim();
+    const isShortPage = bodyText.length < 600;
+    const hasChallengeWidget = Boolean(document.querySelector(
+      '#challenge-form, #challenge-running, #cf-challenge-running, [class*="cf-challenge"], iframe[src*="challenges.cloudflare.com"], iframe[src*="/cdn-cgi/challenge-platform/"]'
+    ));
+    const saysVerifying = /verify(ing)? you are human|checking your browser|needs to review the security of your connection|just a moment/.test(bodyText);
+    const hasChallengeScript = Boolean(document.querySelector(${JSON.stringify(CLOUDFLARE_SCRIPT_SELECTOR)}));
+    return {
+      strong:
+        !hasAppShell &&
+        (titleSaysChallenge || hasChallengeWidget || (isShortPage && saysVerifying)),
+      shell: hasAppShell,
+      // A genuine interstitial is a SHORT page; the content-rich app never is. So the script
+      // only counts on a short page with no app shell.
+      weak: !hasAppShell && hasChallengeScript && isShortPage,
+    };
+  })()`;
+}
+// Boolean composition kept for tests/back-compat: what an instantaneous classifier would say.
+export function buildCloudflareInterstitialExpression() {
+    return `(() => { const v = ${buildCloudflareVerdictExpression()}; return v.strong || v.weak; })()`;
+}
+export const buildCloudflareInterstitialExpressionForTest = buildCloudflareInterstitialExpression;
+export const buildCloudflareVerdictExpressionForTest = buildCloudflareVerdictExpression;
+// Weak (script-only) evidence must persist through a hydration grace window: callers run this
+// right after navigation, when a healthy SPA load can briefly look exactly like the weak case
+// (script present, shell not yet rendered, short body). A real interstitial never grows an app
+// shell, so waiting converts the race into a correct answer at the cost of a few seconds only
+// on genuinely ambiguous pages. Strong evidence still classifies immediately.
+const CLOUDFLARE_HYDRATION_GRACE_MS = 12_000;
+async function isCloudflareInterstitial(Runtime, graceMs = CLOUDFLARE_HYDRATION_GRACE_MS) {
+    const deadline = Date.now() + Math.max(0, graceMs);
+    for (;;) {
+        const { result } = await Runtime.evaluate({
+            expression: buildCloudflareVerdictExpression(),
+            returnByValue: true,
+        });
+        const verdict = (result?.value ?? {});
+        if (verdict.strong)
+            return true;
+        if (verdict.shell)
+            return false;
+        if (!verdict.weak)
+            return false;
+        if (Date.now() >= deadline)
+            return true;
+        await delay(500);
     }
-    const { result } = await Runtime.evaluate({
-        expression: `Boolean(document.querySelector('${CLOUDFLARE_SCRIPT_SELECTOR}'))`,
-        returnByValue: true,
-    });
-    return Boolean(result.value);
 }
+export const isCloudflareInterstitialForTest = isCloudflareInterstitial;
 async function isChatGptAccountSecurityBlock(Runtime) {
     try {
         const outcome = await Runtime.evaluate({