@validate.qa/runner 1.0.33 → 1.0.35

package/dist/cli.js

@@ -1349,11 +1349,13 @@ function escapeEnvValue(value) {
   const escaped = value.replace(/\\/g, "\\\\").replace(/"/g, '\\"').replace(/\n/g, "\\n").replace(/\r/g, "\\r");
   return `"${escaped}"`;
 }
-var API_PREFIX, AUDIO_CHUNK_DURATION_MS, ACTION_BATCH_INTERVAL_MS, ACTION_BATCH_MAX_SIZE, MOBILE_SCREENSHOT_INTERVAL_MS, MOBILE_ACTION_BATCH_MAX_SIZE, MOBILE_BRIDGE_PORT, MOBILE_SNAPSHOT_MAX_SIZE_BYTES, MOBILE_APP_ID_MAX_LENGTH, MOBILE_APP_ID_PATTERN, MOBILE_FOCUS_SETTLE_MS, MOBILE_DEFAULT_SWIPE_DISTANCE, MOBILE_MAX_SWIPE_DISTANCE, SELECTOR_PRIORITIES, MOBILE_STEP_TYPES, BLOCKED_ENV_VARS, SPAWNED_ENV_ALLOWLIST;
+var API_PREFIX, SUPPORT_EMAIL, PLATFORM_ADMIN_EMAIL, AUDIO_CHUNK_DURATION_MS, ACTION_BATCH_INTERVAL_MS, ACTION_BATCH_MAX_SIZE, MOBILE_SCREENSHOT_INTERVAL_MS, MOBILE_ACTION_BATCH_MAX_SIZE, MOBILE_BRIDGE_PORT, MOBILE_SNAPSHOT_MAX_SIZE_BYTES, MOBILE_APP_ID_MAX_LENGTH, MOBILE_APP_ID_PATTERN, MOBILE_FOCUS_SETTLE_MS, MOBILE_DEFAULT_SWIPE_DISTANCE, MOBILE_MAX_SWIPE_DISTANCE, SELECTOR_PRIORITIES, MOBILE_STEP_TYPES, BLOCKED_ENV_VARS, SPAWNED_ENV_ALLOWLIST;
 var init_constants = __esm({
   "../shared/dist/constants.js"() {
     "use strict";
     API_PREFIX = "/api";
+    SUPPORT_EMAIL = "support@validate.qa";
+    PLATFORM_ADMIN_EMAIL = "admin@validate.qa";
     AUDIO_CHUNK_DURATION_MS = 3e4;
     ACTION_BATCH_INTERVAL_MS = 2e3;
     ACTION_BATCH_MAX_SIZE = 10;
@@ -2557,6 +2559,8 @@ var init_audit_pricing = __esm({
       // Current defaults
       "grok-build-0.1": { input: 1, output: 2 },
       "claude-opus-4-7": { input: 5, output: 25 },
+      // Introductory pricing through 2026-08-31 ($2/$10); standard is $3/$15.
+      "claude-sonnet-5": { input: 2, output: 10 },
       "claude-sonnet-4-6": { input: 3, output: 15 },
       "claude-haiku-4-5-20251001": { input: 1, output: 5 },
       "gpt-4o": { input: 2.5, output: 10 },
@@ -3593,7 +3597,7 @@ var init_agent_env = __esm({
       "CommonProgramFiles"
     ]);
     ALLOW_PREFIX = [
-      // AI provider / CLI config (grok-build, claude-code, claude-code-minimax)
+      // AI provider / CLI config (grok-build, claude-code)
       "GROK_",
       "XAI_",
       "ANTHROPIC_",
@@ -3900,8 +3904,10 @@ __export(dist_exports, {
   MOBILE_STEP_TYPES: () => MOBILE_STEP_TYPES,
   PLAN_DEFINITIONS: () => PLAN_DEFINITIONS,
   PLAN_ORDER: () => PLAN_ORDER,
+  PLATFORM_ADMIN_EMAIL: () => PLATFORM_ADMIN_EMAIL,
   SELECTOR_PRIORITIES: () => SELECTOR_PRIORITIES,
   SHARED_PRICING_PLAN_MAP: () => SHARED_PRICING_PLAN_MAP,
+  SUPPORT_EMAIL: () => SUPPORT_EMAIL,
   TRANSIENT_CONTROL_ROLES: () => TRANSIENT_CONTROL_ROLES,
   TRIAL_DURATION_DAYS: () => TRIAL_DURATION_DAYS,
   UI_VARIANTS: () => UI_VARIANTS,
@@ -65163,7 +65169,7 @@ Return ONLY one valid JSON object matching the schema below. Do NOT include <thi
     Example - BAD happy path: \`["Navigate to /feature", "Click Create", "Verify dialog opens"]\` \u2192 only tests the trigger.
     Example - GOOD happy path (generic, async): \`["Navigate to landing and log in", "Navigate to the feature page", "Note the current count of items on the list", "Click the primary action button", "Fill required inputs and submit", "Wait for completion signal (up to 60s)", "Verify the item count increased by 1", "Open the newly created item", "Verify its content is non-empty and contains the input I provided"]\`
     Example - GOOD happy path (sync CRUD): \`["Navigate to landing and log in", "Navigate to the list page", "Click New", "Fill required fields", "Click Save", "Verify the new record appears in the list", "Click the record and verify its detail view shows the values I entered"]\`
-12. **Auth happy paths require concrete session proof.** A login/auth HAPPY_PATH is incomplete if it only proves "user reaches dashboard" or "authenticated page is visible." It MUST verify at least one durable signed-in signal that was observed in the evidence: user/account menu opens, Logout/Sign out is available, the observed display name/email/avatar/account item is present, authenticated navigation remains available after a reload, or another app-specific session control is usable. Prefer concrete observed labels from the evidence (for example "Logout", "Demo User", "Account", "Profile") over generic phrases like "personalized content." If the explorer only observed a redirect and no signed-in signal, keep the route outcome but mark the session-proof outcome reliability "low" and state what concrete signal was missing.
+12. **Auth happy paths require concrete session proof.** A login/auth HAPPY_PATH is incomplete if it only proves "user reaches dashboard" or "authenticated page is visible." It MUST verify at least one durable signed-in signal that was observed in the evidence: user/account menu opens, Logout/Sign out is available, the observed display name/email/avatar/account item is present, authenticated navigation remains available after a reload, or another app-specific session control is usable. Prefer concrete observed labels from the evidence (for example "Logout", "Sign out", "Account", "Profile", or the user's own displayed name/email from evidence) over generic phrases like "personalized content." If the explorer only observed a redirect and no signed-in signal, keep the route outcome but mark the session-proof outcome reliability "low" and state what concrete signal was missing.
 13. **Independent execution contract.** Every scenario MUST pass when run alone, first, last, in random order, and in parallel against an environment with no relevant feature data except explicitly guaranteed immutable fixtures. A scenario that depends on a sibling scenario's side effects (e.g. "delete the conversation that another test created") is invalid. If two flows are inherently coupled, bundle them as a single USER_JOURNEY scenario.
 14. **Scenario-owned state.** If a scenario reads, searches, opens, updates, messages, cancels, or deletes an entity, that scenario MUST create or seed the entity in its own setup steps. "Use an existing record" / "the conversation in the demo account" / "the existing user named X" are forbidden phrasings \u2014 they assume mutable pre-existing app data. Carve-out: baseline observations (current count, list length, "the dashboard heading is visible") are allowed as prerequisite reads ONLY when the durable assertion targets a self-created entity. Phrase setup as: "Click New, fill required fields with run-scoped values, submit and verify the new record appears, THEN proceed to <feature under test>."
 15. **Run-scoped test data.** Generated entities MUST use unique, run-scoped identifiers \u2014 phrase as "test message <Date.now()>" or "QA-<unique id>" in steps. Subsequent assertions MUST target that unique identifier, not "the first item", "the latest message", "the last row", or any visible display name observed during exploration. Hardcoding any specific display name, username, email, conversation title, or other identifier observed in the explorer's snapshot is forbidden \u2014 that data is incidental, not a stable contract. The same identifiers will not exist (or will look different) when the test runs in a fresh environment or against a different account.
@@ -66123,7 +66129,7 @@ Return ONLY one valid JSON object matching the schema below. Do NOT include <thi
 9. **No duplicates** - check existing test names before adding
 10. **All tests start from the app launch entry** - every generated test starts by launching the target app. The runner clears app data when the device/reset policy safely supports it; otherwise it performs a terminate + activate relaunch, which can preserve app session data. Therefore scenarios MUST navigate through the app's own UI from whatever launch state is valid, and MUST NOT rely on mutable data from a prior scenario disappearing. Steps must reflect this: say "Launch the app and tap Sign In" not "deep-link to the login screen". Do NOT navigate by deep-link; the app's launch + tab/drawer navigation is the only entry path. \`targetPages\` lists the screen-ids involved in the flow.
 11. **targetPages must be copied CHARACTER-FOR-CHARACTER from the screen-ids in the evidence** (they look like \`com.example.app.MainActivity#ab12cd34\`). NEVER invent, prettify, or abbreviate them \u2014 no "Landing", "AUTH", "Home Screen", or any name that is not literally in the provided screens. A scenario whose targetPages contains even ONE id not present in the evidence is REJECTED WHOLESALE by the validator, so a single made-up route silently destroys the entire scenario. If you are unsure which screen-id a step touches, use the screen-id of the screen where the flow STARTS.
-12. **Happy paths must verify the DURABLE OUTCOME, not just the trigger.** A HAPPY_PATH is NOT complete until it verifies the feature's real output. "Button tapped" / "Sheet opened" / "Spinner appeared" is NOT a happy path - that is a UI interaction test. Apply this to any feature where an action produces a result the user then sees or uses: CRUD creates, photo/file uploads, searches that return results, imports, exports, payments, messages sent, AI/ML outputs, generated documents, sync runs, analysis runs, conversion, data processing. The scenario MUST include:
+12. **Happy paths must verify the DURABLE OUTCOME, not just the trigger.** A HAPPY_PATH is NOT complete until it verifies the feature's real output. "Button tapped" / "Sheet opened" / "Spinner appeared" / "Settings screen opened" is NOT a happy path - that is a UI interaction test. Form-mutation happy paths (change password, create/edit/save a record, send a message, checkout, etc.) MUST include typing into the relevant fields AND committing the form (submit/save/send or keyboard Enter/Done) \u2014 navigating to the form screen alone is insufficient. Apply this to any feature where an action produces a result the user then sees or uses: CRUD creates, photo/file uploads, searches that return results, imports, exports, payments, messages sent, AI/ML outputs, generated documents, sync runs, analysis runs, conversion, data processing. The scenario MUST include:
     - The trigger step (Tap Submit / Generate / Send / Upload / Run / Analyze / Save / Pay / etc.)
     - A wait step for the operation to complete. Phrase it by SIGNAL, not wall-clock ("Wait for the result to appear - up to 60 seconds"). Prefer signals the app actually uses: spinner disappears, status badge flips to a success state, a new row appears, an item count increments, a success toast/snackbar shows, the screen transitions to a result screen.
     - A verification step on the DURABLE RESULT - the thing that outlives the transient UI: a new list row, a persisted record, a received message, a committed state change, visible output text. Do NOT assert on the spinner disappearing alone; assert on what replaced it.
@@ -66794,6 +66800,9 @@ var require_mobile_generator = __commonJS({
   "../runner-core/dist/services/mobile/mobile-generator.js"(exports2) {
     "use strict";
     Object.defineProperty(exports2, "__esModule", { value: true });
+    exports2.collapseConsecutiveBacks = collapseConsecutiveBacks;
+    exports2.scenarioIntentMismatchReason = scenarioIntentMismatchReason;
+    exports2.dedupeConsecutiveIdenticalActions = dedupeConsecutiveIdenticalActions;
     exports2.collapseRetryCycles = collapseRetryCycles;
     exports2.runMobileGeneratePhase = runMobileGeneratePhase;
     var crypto_1 = require("crypto");
@@ -67064,7 +67073,7 @@ var require_mobile_generator = __commonJS({
 ## Your job
 1. mobile_snapshot the current screen.
 2. Execute the scenario's steps using the real UI: tap buttons, fill fields (mobile_type), submit, navigate.
-3. COMPLETE THE ACTION the scenario is about \u2014 do not stop half-way. If the scenario submits a form (sign in, sign up, search, create), you MUST tap the submit/primary button; filling fields is NOT the behavior under test. If it navigates, you MUST land on the destination. A "happy path" that fills a login form but never taps "Sign In" is WRONG.
+3. COMPLETE THE ACTION the scenario is about \u2014 do not stop half-way. If the scenario submits a form (sign in, sign up, search, create, change password, edit profile), you MUST commit it \u2014 tap the submit/primary button OR press the keyboard's Enter/Done action on the last field (mobile_type with a trailing newline on the focused field when no button is visible). Filling fields alone is NOT the behavior under test. Navigating to a form screen without changing its fields is also NOT enough for mutation scenarios (change/create/update/edit). If it navigates, you MUST land on the destination. A "happy path" that fills a form but never commits \u2014 or only opens the form without typing \u2014 is WRONG.
 4. When you reach the target screen / complete the flow, PROVE the outcome with an assert tool:
    - mobile_assert_visible({ref}) \u2014 the target screen's stable element is on screen.
    - mobile_assert_text({ref, expected}) \u2014 an element contains expected text (only durable when the element has a stable accessibility-id / resource-id).
@@ -67079,6 +67088,8 @@ var require_mobile_generator = __commonJS({
 - Stay inside this app.${targetLine} Do not open or use app stores, browsers, email apps, settings, or any other app. If an action leaves the target app, the runner will relaunch the target app and reject that action.
 - NEVER tap the phone's home screen / launcher (app icons, "Predicted app" chips, search bars, widgets). Those are NOT part of the app under test \u2014 tapping them is junk that poisons the test. If the app ever backgrounds or you find yourself on the launcher, use mobile_relaunch to cold-start the app; do NOT tap a launcher icon or app-prediction to get back in.
 - Do NOT re-do a sequence you already completed. If you filled a form then navigated away, do NOT come back and fill it again \u2014 the scenario needs ONE clean pass. Filling the login form, going back, and re-filling it is exactly the kind of wandering that produces a bloated, broken test.
+- Do NOT thrash the system Back button. One intentional back/navigation is fine; a chain of backs while lost is not a test \u2014 re-snapshot, re-orient, and take the productive path once.
+- Prefer product identity locators (accessibility-id, resource-id, visible label). Never assert on bare hierarchy roots or unlabeled full-screen containers \u2014 those prove nothing.
 - Do ONLY the actions this scenario needs. Do NOT re-open a menu you already opened, re-select an option that is already selected, toggle a setting back and forth, or repeat a step that already succeeded \u2014 EVERY action you take is baked into the generated test verbatim, so each extra action becomes a permanent junk step.
 - Each action must make NEW progress toward the goal. If an action leaves the screen unchanged, do NOT repeat it \u2014 pick a different element or move on.
 - Finish promptly once the outcome is proven: the instant the expected outcome is visible, assert it and call finish_generate. Do not keep exploring after the goal is proven.
@@ -67097,7 +67108,7 @@ This app has stored login credentials under these keys:
 ${keyList}
 
 When a scenario needs a signed-in session, complete the FULL login sequence \u2014 filling the fields alone does NOT log you in:
-1. Reach the email/password form. 2. Fill each credential field with \`mobile_fill_credential\` (pass the field ref and the KEY, e.g. \`{ ref: "e5", credentialKey: "TEST_USER_EMAIL" }\`) \u2014 the runner types the real value on the device and the generated test stores a \`{{KEY}}\` placeholder that replay resolves; you never see the secret. 3. TAP THE SUBMIT button (Sign In / Log In / Continue). 4. Re-snapshot and VERIFY an authenticated screen appeared. If login fails twice, do NOT keep retrying or wander: call finish_generate with verified:false and say the login could not be completed (the credentials may be invalid for this app).
+1. Reach the email/password form. 2. Fill each credential field with \`mobile_fill_credential\` (pass the field ref and the KEY, e.g. \`{ ref: "e5", credentialKey: "TEST_USER_EMAIL" }\`) \u2014 the runner types the real value on the device and the generated test stores a \`{{KEY}}\` placeholder that replay resolves; you never see the secret. 3. COMMIT the form \u2014 tap the submit/primary button (Sign In / Log In / Continue) OR, if no button is visible, focus the last field and \`mobile_type\` a trailing newline (\`\\n\`) to trigger the keyboard's Enter/Done action. 4. Re-snapshot and VERIFY an authenticated screen appeared. If login fails twice, do NOT keep retrying or wander: call finish_generate with verified:false and say the login could not be completed (the credentials may be invalid for this app).
 
 SECURITY RULES:
 - NEVER call \`mobile_type\` with a real email, password, or token value \u2014 the runner will REJECT the call and you lose an iteration. Use \`mobile_fill_credential\` for every credential field.
@@ -67447,6 +67458,7 @@ ${statePacket}` },
                 if (action.type === "ASSERT_VISIBLE" || action.type === "ASSERT_TEXT")
                   assertCount++;
               },
+              recordedActions: actions,
               log: log2
             });
             pushResult(dispatchResult);
@@ -67462,7 +67474,7 @@ ${statePacket}` },
       return { actions, assertCount, finish, lastSnapshot: getLatest(), lastScreenName };
     }
     async function dispatchGenerateTool(args) {
-      const { toolName, toolArgs, driver, platform: platform3, latest, absorbObservation, resolveRef, credentials, screenName, recordAction, log: log2 } = args;
+      const { toolName, toolArgs, driver, platform: platform3, latest, absorbObservation, resolveRef, credentials, screenName, recordAction, recordedActions, log: log2 } = args;
       const unknownRef = (ref) => ({
         ok: false,
         error: `Unknown element ref "${String(ref)}". Call mobile_snapshot first; refs are only valid for the latest snapshot.`
@@ -67527,14 +67539,23 @@ ${statePacket}` },
           const screenChanged = changedScreen(before, absorbed);
           const identityTap = isIdentityBearingTapTarget(element);
           if (shouldRecordAction(result, absorbed) && (screenChanged || identityTap)) {
-            if (!screenChanged && identityTap) {
-              log2(`  tap left content signature unchanged but target has identity \u2014 recording commit control "${labelForElement(element)}".`);
-            }
-            recordAction({
+            const tapAction = {
               type: "TAP",
               selector: (0, mobile_source_parser_js_2.preferredLocator)(element),
-              metadata: metadataForElement(element, beforeScreenName, platform3)
-            });
+              metadata: metadataForElement(element, beforeScreenName, platform3),
+              innerText: null,
+              url: "",
+              value: null
+            };
+            const lastRecorded = recordedActions[recordedActions.length - 1];
+            if (lastRecorded && actionCycleKey(lastRecorded) === actionCycleKey(tapAction)) {
+              log2(`  tap left content signature unchanged but duplicates the prior step \u2014 not recording again.`);
+            } else if (!screenChanged && identityTap) {
+              log2(`  tap left content signature unchanged but target has identity \u2014 recording commit control "${labelForElement(element)}".`);
+              recordAction(tapAction);
+            } else {
+              recordAction(tapAction);
+            }
           } else if (shouldRecordAction(result, absorbed)) {
             log2("  tap changed nothing on screen \u2014 not recording it as a step (dead action).");
           }
@@ -67878,6 +67899,105 @@ ${statePacket}` },
         ...scenario.expectedOutcomes.flatMap((outcome) => [outcome.description, outcome.hint])
       ].filter((value) => typeof value === "string" && value.trim().length > 0).join(" ").toLowerCase();
     }
+    var INSPECTION_OUTCOME_TYPES = /* @__PURE__ */ new Set([
+      "element-visible",
+      "text-visible",
+      "element-hidden",
+      "screen-visible"
+    ]);
+    function isInspectionOnlyScenario(scenario, actions) {
+      if (actions.length === 0 || scenario.expectedOutcomes.length === 0)
+        return false;
+      if (!scenario.expectedOutcomes.every((o) => INSPECTION_OUTCOME_TYPES.has(o.type))) {
+        return false;
+      }
+      if (actions.some((a) => a.type === "TYPE" || a.type === "CLEAR")) {
+        return false;
+      }
+      const text = scenarioText(scenario);
+      const visibilityOnlyGoal = (/\b(entry\s+points?|providers?|options?|provider\s+buttons?)\b/.test(text) || /\b(show|display)\s+(all|the)?\s*(provider|option|button|link|entry)/.test(text) || /\b(all\s+(three\s+)?providers?)\b/.test(text)) && !/\b(change|update|create|save|send|register|edit|with\s+valid|successful\s+(sign|log)|complete\s+(the\s+)?(login|sign[- ]?in)|submit\s+(the\s+)?(form|credentials)|enter\s+(an?\s+)?(email|password))\b/.test(text);
+      if (visibilityOnlyGoal) {
+        const assertCount2 = actions.filter((a) => a.type === "ASSERT_VISIBLE" || a.type === "ASSERT_TEXT").length;
+        return assertCount2 > 0;
+      }
+      const completionIntent = scenario.type === "HAPPY_PATH" || /\b(happy\s*path|successful|with\s+valid|complete\s+(the\s+)?(login|sign[- ]?in|registration|checkout|purchase))\b/.test(text);
+      const formIntent = /\b(sign\s*in|log\s*in|login|authenticate|register|sign\s*up|checkout|create|save|submit|send|change|update|edit)\b/.test(text);
+      if (completionIntent && formIntent)
+        return false;
+      const assertCount = actions.filter((a) => a.type === "ASSERT_VISIBLE" || a.type === "ASSERT_TEXT").length;
+      return assertCount > 0;
+    }
+    function isNavigationFocusedScenario(scenario) {
+      const text = scenarioText(scenario);
+      const navigates = /\b(back|return|go\s+home|navigate\s+back|leave|exit|cancel\s+out|dismiss|previous\s+screen)\b/.test(text);
+      const completesFlow = /\b(happy\s*path|successful|complete\s+(the\s+)?(login|sign[- ]?in|registration|checkout)|with\s+valid)\b/.test(text);
+      return navigates && !completesFlow;
+    }
+    function isFormValidationScenario(scenario) {
+      if (scenario.variant === "validation")
+        return true;
+      if (scenario.type !== "EDGE_CASE")
+        return false;
+      const text = scenarioText(scenario);
+      return /\b(empty|required|missing|blank|invalid|reject|error|validation|wrong|incorrect|mismatch)\b/.test(text);
+    }
+    function isFormMutationHappyPath(scenario) {
+      if (isNavigationFocusedScenario(scenario))
+        return false;
+      if (isFormValidationScenario(scenario))
+        return false;
+      const text = scenarioText(scenario);
+      const isHappy = scenario.type === "HAPPY_PATH" || /\b(happy\s*path|successful|success)\b/.test(text);
+      if (!isHappy)
+        return false;
+      if (/\b(sign\s*in|log\s*in|login|authenticate|sign\s*out|log\s*out)\b/.test(text) && !/\b(change|update|reset)\s+(the\s+)?password\b/.test(text)) {
+        return false;
+      }
+      if (/\b(show|display|visible|options?|providers?|links?)\b/.test(text) && !/\b(change|update|create|save|send|register|edit|post|write|compose|delete|add|set)\b/.test(text)) {
+        return false;
+      }
+      const mutates = /\b(change|update|create|save|send|register|sign\s*up|checkout|edit|post|write|compose|set|add|delete|remove)\b/.test(text);
+      const hasObject = /\b(password|email|profile|message|post|item|record|form|settings|account|name|title|description|comment|bio|phone|address|payment|card|user|note|task|entry)\b/.test(text);
+      return mutates && hasObject;
+    }
+    function isNavigationChromeTap(action) {
+      if (action.type !== "TAP")
+        return false;
+      const textForAction = actionText(action);
+      return /\b(go\s+back|back\s+to|←|return\s+to|cancel|close|dismiss|not\s+now|skip|maybe\s+later)\b/.test(textForAction);
+    }
+    function isAuthMethodChooserTap(action) {
+      if (action.type !== "TAP")
+        return false;
+      const textForAction = actionText(action);
+      if (/\b(submit|confirm|sign\s*in|log\s*in|login)\b/.test(textForAction) && !/\b(with|using|via)\b/.test(textForAction)) {
+        return false;
+      }
+      return /\b(continue\s+with|sign\s+in\s+with|log\s+in\s+with|use\s+(email|phone)|choose|pick|select)\b/.test(textForAction);
+    }
+    function isCredentialSubmitTap(action) {
+      if (action.type !== "TAP")
+        return false;
+      if (isNavigationChromeTap(action) || isAuthMethodChooserTap(action))
+        return false;
+      const textForAction = actionText(action);
+      return /\b(sign\s*in|log\s*in|login|continue|submit|done|next|confirm)\b/.test(textForAction);
+    }
+    function isKeyboardFormCommit(action) {
+      return action.type === "TYPE" && (action.value ?? "").endsWith("\n");
+    }
+    function isFormCommitAction(action) {
+      if (isKeyboardFormCommit(action))
+        return true;
+      if (isCredentialSubmitTap(action))
+        return true;
+      if (action.type !== "TAP")
+        return false;
+      if (isNavigationChromeTap(action) || isAuthMethodChooserTap(action))
+        return false;
+      const textForAction = actionText(action);
+      return /\b(save|create|send|confirm|next|apply|post|checkout|pay|finish|update|register|sign\s*up|done|continue|submit)\b/i.test(textForAction);
+    }
     function actionText(action) {
       const bundle = action.metadata.locatorBundle;
       return [
@@ -67894,9 +68014,80 @@ ${statePacket}` },
         bundle?.xpath
       ].filter((value) => typeof value === "string" && value.trim().length > 0).join(" ").toLowerCase();
     }
+    function isStructuralOnlyAssert(action) {
+      if (action.type !== "ASSERT_VISIBLE" && action.type !== "ASSERT_TEXT")
+        return false;
+      const sel = (action.selector ?? "").trim();
+      if (/^\/hierarchy(\[\d+\])?(\/android\.widget\.FrameLayout\[\d+\])*$/i.test(sel))
+        return true;
+      if (isStructuralAssertionIdentity(sel) || isStructuralAssertionIdentity(action.metadata.targetLabel)) {
+        const bundle = action.metadata.locatorBundle;
+        const human = Boolean(nonEmpty2(action.innerText) || nonEmpty2(action.metadata.targetLabel) || nonEmpty2(bundle?.text) || nonEmpty2(bundle?.contentDesc));
+        if (!human)
+          return true;
+      }
+      if (/^\/hierarchy/i.test(sel)) {
+        const bundle = action.metadata.locatorBundle;
+        const hasIdentity = Boolean(nonEmpty2(action.metadata.targetLabel) || nonEmpty2(action.innerText) || nonEmpty2(bundle?.text) || nonEmpty2(bundle?.contentDesc) || nonEmpty2(bundle?.accessibilityId) || nonEmpty2(bundle?.resourceId) && !isStructuralAssertionIdentity(bundle?.resourceId));
+        if (!hasIdentity)
+          return true;
+      }
+      return false;
+    }
+    function collapseConsecutiveBacks(actions) {
+      if (actions.length <= 1)
+        return actions;
+      const out = [actions[0]];
+      for (let i = 1; i < actions.length; i++) {
+        if (actions[i].type === "BACK" && out[out.length - 1].type === "BACK")
+          continue;
+        out.push(actions[i]);
+      }
+      return out;
+    }
     function scenarioIntentMismatchReason(scenario, actions) {
       const text = scenarioText(scenario);
-      const isCredentialSignIn = /\b(sign\s*in|log\s*in|login|authenticate|auth)\b/.test(text) && /\b(email|e-mail|password|credential|valid credentials?)\b/.test(text) && !/\b(reset|forgot|recover|logout|log\s*out|sign\s*out)\b/.test(text);
+      if (isInspectionOnlyScenario(scenario, actions)) {
+        return null;
+      }
+      if (isNavigationFocusedScenario(scenario)) {
+        return null;
+      }
+      if (isFormValidationScenario(scenario)) {
+        const typedFields = actions.some((a) => a.type === "TYPE");
+        const hasCommit = actions.some((a) => isFormCommitAction(a));
+        const hasValidationAssert = actions.some((a) => {
+          if (a.type !== "ASSERT_VISIBLE" && a.type !== "ASSERT_TEXT")
+            return false;
+          return /\b(error|required|invalid|empty|missing|incorrect|wrong|try again|mismatch|does not match|must match)\b/.test(actionText(a));
+        });
+        const emptyFieldIntent = /\b(empty|required|missing|blank)\b/.test(text);
+        const invalidValueIntent = /\b(wrong|incorrect|invalid|mismatch|reject)\b/.test(text);
+        if (emptyFieldIntent && !invalidValueIntent) {
+          if (!hasCommit && !hasValidationAssert) {
+            return "validation scenario did not record a form commit or validation outcome";
+          }
+          return null;
+        }
+        if (typedFields && !hasCommit) {
+          return "validation scenario filled fields but never committed the form";
+        }
+        if (!hasCommit && !hasValidationAssert) {
+          return "validation scenario did not record a form commit or validation outcome";
+        }
+        return null;
+      }
+      if (isFormMutationHappyPath(scenario)) {
+        const hasType = actions.some((a) => a.type === "TYPE");
+        const hasCommit = actions.some((a) => isFormCommitAction(a));
+        if (!hasType) {
+          return "form-mutation happy path did not record any field entry";
+        }
+        if (!hasCommit) {
+          return "form-mutation happy path did not record a form commit";
+        }
+      }
+      const isCredentialSignIn = /\b(sign\s*in|log\s*in|login|authenticate|auth)\b/.test(text) && /\b(email|e-mail|password|credential|valid credentials?)\b/.test(text) && !/\b(reset|forgot|recover|logout|log\s*out|sign\s*out|change\s+password|update\s+password)\b/.test(text);
       const typedEmailIndexes = [];
       const typedPasswordIndexes = [];
       actions.forEach((action, index) => {
@@ -67913,19 +68104,19 @@ ${statePacket}` },
       const lastCredentialIndex = Math.max(...typedEmailIndexes, ...typedPasswordIndexes);
       const hasAnyCredentialEntry = Number.isFinite(lastCredentialIndex);
       const submitAfterCredentials = actions.some((action, index) => {
-        if (action.type !== "TAP" || index <= lastCredentialIndex)
+        if (index < lastCredentialIndex)
           return false;
-        const textForAction = actionText(action);
-        if (!/\b(sign\s*in|log\s*in|login|continue|submit)\b/.test(textForAction))
+        if (!isFormCommitAction(action))
           return false;
-        return !/\b(with\s+(email|e-mail|google|facebook|apple)|email\s+form|showemailformbtn)\b/i.test(textForAction);
+        if (action.type === "TYPE" && index === lastCredentialIndex)
+          return true;
+        return index > lastCredentialIndex;
       });
       const isAuthLikeScenario = /\b(sign(?:ed)?\s*in|log(?:ged)?\s*in|login|authenticate|auth|credential|email|e-mail|password)\b/.test(text);
       if (isAuthLikeScenario && hasAnyCredentialEntry && !submitAfterCredentials) {
-        return "credential fields were filled but no sign-in submit tap was recorded after them";
+        return "credential fields were filled but no form commit was recorded after them";
       }
       const uiOnlyCredentialScenario = /\b(mask|masked|visibility|show password|hide password|toggle|keyboard|keypad|max(?:imum)?\s*length|field focus|autofill)\b/.test(text);
-      const emptyFieldValidationScenario = /\b(empty|required|missing|blank)\b/.test(text);
       const sessionLifecycleScenario = /\b(logout|log\s*out|sign\s*out|session persistence|remain(?:s)? signed in|remember me)\b/.test(text);
       const loginTapWithoutCredentialPath = actions.some((action) => {
         if (action.type !== "TAP")
@@ -67933,9 +68124,11 @@ ${statePacket}` },
         const textForAction = actionText(action);
         if (/\b(logout|log\s*out|sign\s*out)\b/.test(textForAction))
           return false;
+        if (isAuthMethodChooserTap(action) || isNavigationChromeTap(action))
+          return false;
         return /\b(sign\s*in|log\s*in|login)\b/.test(textForAction);
       });
-      if (loginTapWithoutCredentialPath && !hasAnyCredentialEntry && !uiOnlyCredentialScenario && !emptyFieldValidationScenario && (sessionLifecycleScenario || /\b(valid credentials?|happy path|successful sign\s*in|successful login)\b/.test(text))) {
+      if (loginTapWithoutCredentialPath && !hasAnyCredentialEntry && !uiOnlyCredentialScenario && (sessionLifecycleScenario || /\b(valid credentials?|happy path|successful sign\s*in|successful login)\b/.test(text))) {
         return "auth/session scenario tapped a login or sign-in control without recording a credential path";
       }
       if (isCredentialSignIn) {
@@ -67947,7 +68140,7 @@ ${statePacket}` },
           const missing = [
             typedEmail ? null : "email/username entry",
             typedPassword ? null : "password entry",
-            tappedSubmit ? null : "sign-in submit tap"
+            tappedSubmit ? null : "form commit (submit tap or keyboard Enter)"
           ].filter((value) => value !== null);
           return `credential sign-in scenario did not record ${missing.join(", ")}`;
         }
@@ -67968,11 +68161,12 @@ ${statePacket}` },
       const typeScreen = actions[lastTypeIndex]?.metadata?.screenName?.trim() ?? "";
       const assertScreen = actions[lastAssertIndex]?.metadata?.screenName?.trim() ?? "";
       const screensDiffer = typeScreen !== "" && assertScreen !== "" && typeScreen !== assertScreen;
-      const commitAfterType = actions.some((action, index) => {
-        if (action.type !== "TAP" || index <= lastTypeIndex || index >= lastAssertIndex)
+      const lastTypeAction = actions[lastTypeIndex];
+      const keyboardCommitOnLastType = lastTypeAction != null && isKeyboardFormCommit(lastTypeAction);
+      const commitAfterType = keyboardCommitOnLastType || actions.some((action, index) => {
+        if (index <= lastTypeIndex || index >= lastAssertIndex)
           return false;
-        const textForAction = actionText(action);
-        return /\b(sign\s*in|log\s*in|login|continue|submit|save|create|send|confirm|done|next|apply|post|checkout|pay|finish|update|register|sign\s*up)\b/i.test(textForAction);
+        return isFormCommitAction(action);
       });
       if (screensDiffer && !commitAfterType) {
         return "form fields were filled then a success assert was recorded without a commit/submit tap between them";
@@ -67983,6 +68177,17 @@ ${statePacket}` },
       return `${action.type}|${action.selector}|${action.value ?? ""}`;
     }
     var MAX_RETRY_CYCLE_LENGTH = 8;
+    function dedupeConsecutiveIdenticalActions(actions) {
+      if (actions.length <= 1)
+        return actions;
+      const out = [actions[0]];
+      for (let i = 1; i < actions.length; i++) {
+        if (actionCycleKey(actions[i]) === actionCycleKey(out[out.length - 1]))
+          continue;
+        out.push(actions[i]);
+      }
+      return out;
+    }
     function collapseRetryCycles(actions) {
       const out = [...actions];
       let changed = true;
@@ -68023,15 +68228,30 @@ ${statePacket}` },
     }
     function assembleTest(args) {
       const { scenario, recording, ctx, platform: platform3, appDataCleared, log: log2 } = args;
-      let actions = collapseRetryCycles(recording.actions);
+      let actions = dedupeConsecutiveIdenticalActions(recording.actions);
       if (actions.length < recording.actions.length) {
-        log2(`  collapsed ${recording.actions.length - actions.length} retry-cycle step(s) out of the recorded trace (${recording.actions.length} \u2192 ${actions.length}).`);
+        log2(`  deduped ${recording.actions.length - actions.length} consecutive identical step(s) (${recording.actions.length} \u2192 ${actions.length}).`);
+      }
+      const beforeBackCollapse = actions.length;
+      actions = collapseConsecutiveBacks(actions);
+      if (actions.length < beforeBackCollapse) {
+        log2(`  collapsed ${beforeBackCollapse - actions.length} consecutive BACK step(s) (${beforeBackCollapse} \u2192 ${actions.length}).`);
+      }
+      const beforeCycleCollapse = actions.length;
+      actions = collapseRetryCycles(actions);
+      if (actions.length < beforeCycleCollapse) {
+        log2(`  collapsed ${beforeCycleCollapse - actions.length} retry-cycle step(s) out of the recorded trace (${beforeCycleCollapse} \u2192 ${actions.length}).`);
       }
       const beforeContainerDrop = actions.length;
       actions = actions.filter((a) => !isBareContainerTap(a));
       if (actions.length < beforeContainerDrop) {
         log2(`  dropped ${beforeContainerDrop - actions.length} bare-container tap step(s) (no identity \u2014 un-replayable ambient taps).`);
       }
+      const beforeWeakAssertDrop = actions.length;
+      actions = actions.filter((a) => !isStructuralOnlyAssert(a));
+      if (actions.length < beforeWeakAssertDrop) {
+        log2(`  dropped ${beforeWeakAssertDrop - actions.length} structural-only assert step(s) (no product identity).`);
+      }
       const intentMismatch = scenarioIntentMismatchReason(scenario, actions);
       if (intentMismatch) {
         log2(`  skipped scenario "${scenario.name}": ${intentMismatch}.`);
@@ -68843,10 +69063,22 @@ For a fix, return the COMPLETE steps array (not a diff).`;
       if (!/\b(tap|click|press)\b/i.test(step.action))
         return false;
       const text = mobileStepSearchText(step);
-      if (/\b(logout|log\s*out|sign\s*out|reset|forgot|google|apple|facebook|github|sso|showemailform|with\s+email|email\s+form)\b/.test(text)) {
+      if (/\b(logout|log\s*out|sign\s*out|reset|forgot|go\s+back|back\s+to|cancel|close|dismiss)\b/.test(text)) {
         return false;
       }
-      return /\b(sign\s*in|signin|log\s*in|login|continue|submit)\b/.test(text);
+      if (/\b(continue\s+with|sign\s+in\s+with|log\s+in\s+with|use\s+(email|phone)|choose|pick|select)\b/.test(text) && !/\b(submit|confirm)\b/.test(text)) {
+        return false;
+      }
+      return /\b(sign\s*in|signin|log\s*in|login|continue|submit|done|next|confirm)\b/.test(text);
+    }
+    function isKeyboardFormCommitStep(step) {
+      if (!/\b(type|fill|input|enter)\b/i.test(step.action))
+        return false;
+      const value = typeof step.value === "string" ? step.value : "";
+      return value.endsWith("\n");
+    }
+    function isFormCommitStep(step) {
+      return isCredentialSubmitTap(step) || isKeyboardFormCommitStep(step);
     }
     function detectInvalidAuthValidationSetup(steps, error2) {
       if (!/\b(element not found|no such element|not visible|timeout|assert|could not be located|unexpected screen|screen drift)\b/i.test(error2 ?? "")) {
@@ -68871,20 +69103,23 @@ For a fix, return the COMPLETE steps array (not a diff).`;
           if ((step.order ?? -1) >= lastCredentialEditOrder)
             return false;
           const text = mobileStepSearchText(step);
-          return /\b(showemailformbtn|with\s+email|email\s+form|sign in with email|loginform|login form|email|e-mail|username|user\s*name)\b/.test(text);
+          return /\b(sign\s+in\s+with\s+email|log\s+in\s+with\s+email|email\s+and\s+password|login\s+form|credential\s+form|email|e-mail|username|user\s*name)\b/.test(text);
         });
         if (hasCredentialFormEvidence)
           return null;
         return "credential-control test edits a credential field without recording a path to the email/password form; stale auth-control test is invalid until generation records a complete form path";
       }
-      const hasSubmitAfterCredentialEdit = steps.some((step) => (step.order ?? -1) > lastCredentialEditOrder && isCredentialSubmitTap(step));
+      const hasSubmitAfterCredentialEdit = steps.some((step) => (step.order ?? -1) > lastCredentialEditOrder && isFormCommitStep(step));
       if (hasSubmitAfterCredentialEdit)
         return null;
+      const hasKeyboardCommitOnCredential = steps.some((step) => (step.order ?? -1) === lastCredentialEditOrder && isKeyboardFormCommitStep(step));
+      if (hasKeyboardCommitOnCredential)
+        return null;
       const hasValidationAssertion = steps.some((step) => {
         if (!/\b(assert|wait)\b/i.test(step.action))
           return false;
         const text = mobileStepSearchText(step);
-        return /\b(invalid|required|error|empty|missing|incorrect|wrong|try again|validation|must enter|enter (an )?(email|password)|email is required|password is required)\b/.test(text);
+        return /\b(invalid|required|error|empty|missing|incorrect|wrong|try again|validation|must enter|enter (an )?(email|password)|email is required|password is required|mismatch|does not match)\b/.test(text);
       });
       if (hasValidationAssertion)
         return null;
@@ -298740,7 +298975,8 @@ var require_bootstrap_grok_auth = __commonJS({
       const home = process.env.HOME ?? "/home/appuser";
       const authPath = node_path_1.default.join(home, ".grok", "auth.json");
       const b64 = process.env.GROK_AUTH_JSON_B64;
-      if (node_fs_1.default.existsSync(authPath)) {
+      const force = process.env.GROK_AUTH_FORCE_RESEED === "true" || process.env.GROK_AUTH_FORCE_RESEED === "1";
+      if (!force && node_fs_1.default.existsSync(authPath)) {
         return;
       }
       if (!b64) {
@@ -298751,7 +298987,7 @@ var require_bootstrap_grok_auth = __commonJS({
         node_fs_1.default.mkdirSync(dir, { recursive: true, mode: 448 });
         node_fs_1.default.writeFileSync(authPath, Buffer.from(b64, "base64"));
         node_fs_1.default.chmodSync(authPath, 384);
-        console.log(`[grok-auth][runner] Seeded ${authPath} from GROK_AUTH_JSON_B64`);
+        console.log(force ? `[grok-auth][runner] Force-reseeded ${authPath} from GROK_AUTH_JSON_B64` : `[grok-auth][runner] Seeded ${authPath} from GROK_AUTH_JSON_B64`);
       } catch (err) {
         const message = err instanceof Error ? err.message : String(err);
         console.warn(`[grok-auth][runner] Failed to seed auth.json: ${message}`);
@@ -299396,6 +299632,37 @@ var require_proven_actions_adapter = __commonJS({
   }
 });
 
+// ../runner-core/dist/services/agentic-cli-constants.js
+var require_agentic_cli_constants = __commonJS({
+  "../runner-core/dist/services/agentic-cli-constants.js"(exports2) {
+    "use strict";
+    Object.defineProperty(exports2, "__esModule", { value: true });
+    exports2.GROK_CLI_DEFAULT_REASONING_EFFORT = exports2.GROK_CLI_DEFAULT_MODEL = exports2.CLAUDE_CODE_DEFAULT_EFFORT = exports2.CLAUDE_CODE_DEFAULT_MODEL = exports2.MODEL_CLAUDE_CODE = exports2.MODEL_GROK_BUILD = void 0;
+    exports2.isGrokBuildModel = isGrokBuildModel;
+    exports2.isClaudeCodeModel = isClaudeCodeModel;
+    exports2.getAgenticEngine = getAgenticEngine;
+    exports2.MODEL_GROK_BUILD = "grok-build";
+    function isGrokBuildModel(id) {
+      return id === exports2.MODEL_GROK_BUILD;
+    }
+    exports2.MODEL_CLAUDE_CODE = "claude-code";
+    function isClaudeCodeModel(id) {
+      return id === exports2.MODEL_CLAUDE_CODE;
+    }
+    function getAgenticEngine(modelId) {
+      if (isGrokBuildModel(modelId))
+        return "grok-build";
+      if (isClaudeCodeModel(modelId))
+        return "claude-code";
+      return "native";
+    }
+    exports2.CLAUDE_CODE_DEFAULT_MODEL = "claude-sonnet-5";
+    exports2.CLAUDE_CODE_DEFAULT_EFFORT = "max";
+    exports2.GROK_CLI_DEFAULT_MODEL = "grok-4.5";
+    exports2.GROK_CLI_DEFAULT_REASONING_EFFORT = "high";
+  }
+});
+
 // ../runner-core/dist/services/grok-cli.js
 var require_grok_cli = __commonJS({
   "../runner-core/dist/services/grok-cli.js"(exports2) {
@@ -299406,6 +299673,7 @@ var require_grok_cli = __commonJS({
     exports2.runGrok = runGrok;
     var node_child_process_1 = require("child_process");
     var shared_1 = (init_dist(), __toCommonJS(dist_exports));
+    var agentic_cli_constants_js_1 = require_agentic_cli_constants();
     function extractAssistantJson(stdout) {
       const tryParse = (s) => {
         const trimmed = s.trim();
@@ -299500,7 +299768,8 @@ var require_grok_cli = __commonJS({
     function runGrok(options) {
       const {
         prompt,
-        model = "grok-build",
+        model = agentic_cli_constants_js_1.GROK_CLI_DEFAULT_MODEL,
+        reasoningEffort = agentic_cli_constants_js_1.GROK_CLI_DEFAULT_REASONING_EFFORT,
         maxTurns,
         // undefined = let grok decide when to quit
         timeoutMs = 6e5,
@@ -299529,6 +299798,9 @@ var require_grok_cli = __commonJS({
           "--always-approve",
           "--no-alt-screen"
         ];
+        if (reasoningEffort) {
+          args.push("--reasoning-effort", reasoningEffort);
+        }
         if (typeof maxTurns === "number" && maxTurns > 0) {
           args.push("--max-turns", String(maxTurns));
         }
@@ -299674,7 +299946,7 @@ var require_grok_per_test_healer = __commonJS({
     var HEAL_MCP_CAPABILITIES = ["core", "core-navigation", "core-input", "core-tabs", "testing"];
     var grok_cli_js_1 = require_grok_cli();
     var DEFAULT_MAX_ATTEMPTS = 3;
-    var DEFAULT_TIMEOUT_MS = 30 * 60 * 1e3;
+    var DEFAULT_TIMEOUT_MS2 = 30 * 60 * 1e3;
     var IDLE_WATCHDOG_MS = 10 * 60 * 1e3;
     function getMaxAttempts() {
       const raw = process.env.GROK_PER_TEST_HEAL_MAX_ATTEMPTS;
@@ -299684,7 +299956,7 @@ var require_grok_per_test_healer = __commonJS({
     function getAttemptTimeoutMs() {
       const raw = process.env.GROK_PER_TEST_HEAL_TIMEOUT_MS;
       const parsed = raw ? parseInt(raw, 10) : NaN;
-      return Number.isFinite(parsed) && parsed > 0 ? parsed : DEFAULT_TIMEOUT_MS;
+      return Number.isFinite(parsed) && parsed > 0 ? parsed : DEFAULT_TIMEOUT_MS2;
     }
     async function executeGrokPerTestHeal2(playwrightCode, steps, baseUrl, options = {}) {
       const externalLogSink = options.onLog;
@@ -300097,7 +300369,7 @@ Output the JSON envelope exactly ONCE as your final message, then exit.`;
         "-p",
         'Read BRIEF.md in this directory. Heal TEST_UNDER_HEAL.spec.ts as instructed. Your FINAL assistant message MUST be ONLY a JSON object \u2014 no prose around it \u2014 with this shape: {"verdict":"fix|app_bug|config_issue|prereq_unmet|state_stale|test_invalid","playwrightCode":"<full file if verdict=fix>","changeDescription":"<one sentence>","reason":"<terse, when not fix>","confidence":"high|medium|low"}',
         "-m",
-        "grok-build",
+        "grok-4.5",
         // streaming-json: per-event NDJSON so the UI's GrokSessionLog drawer can
         // show progress instead of a frozen "starting…" until the final envelope.
         "--output-format",
@@ -301317,7 +301589,7 @@ When done with all ${input.tests.length} tests in this chunk, exit.`;
         "-p",
         "Read BRIEF.md in this directory. Heal every test under tests/*.spec.ts as instructed and emit a verdict file per test under verdicts/<testCaseId>.json (write to .tmp then rename \u2014 the runner is polling). When done, exit.",
         "-m",
-        "grok-build",
+        "grok-4.5",
         // streaming-json so per-token output reaches the UI for live progress.
         "--output-format",
         "streaming-json",
@@ -308806,47 +309078,6 @@ var require_sitemap_builder = __commonJS({
   }
 });
 
-// ../runner-core/dist/services/agentic-cli-constants.js
-var require_agentic_cli_constants = __commonJS({
-  "../runner-core/dist/services/agentic-cli-constants.js"(exports2) {
-    "use strict";
-    Object.defineProperty(exports2, "__esModule", { value: true });
-    exports2.CLAUDE_CODE_MINIMAX_FALLBACK_CHAIN = exports2.CLAUDE_CODE_MINIMAX_DEFAULT_MODEL = exports2.CLAUDE_CODE_DEFAULT_MODEL = exports2.MODEL_CLAUDE_CODE_MINIMAX = exports2.MODEL_CLAUDE_CODE = exports2.MODEL_GROK_BUILD = void 0;
-    exports2.isGrokBuildModel = isGrokBuildModel;
-    exports2.isClaudeCodeModel = isClaudeCodeModel;
-    exports2.isClaudeCodeMinimaxModel = isClaudeCodeMinimaxModel;
-    exports2.getAgenticEngine = getAgenticEngine;
-    exports2.MODEL_GROK_BUILD = "grok-build";
-    function isGrokBuildModel(id) {
-      return id === exports2.MODEL_GROK_BUILD;
-    }
-    exports2.MODEL_CLAUDE_CODE = "claude-code";
-    function isClaudeCodeModel(id) {
-      return id === exports2.MODEL_CLAUDE_CODE;
-    }
-    exports2.MODEL_CLAUDE_CODE_MINIMAX = "claude-code-minimax";
-    function isClaudeCodeMinimaxModel(id) {
-      return id === exports2.MODEL_CLAUDE_CODE_MINIMAX;
-    }
-    function getAgenticEngine(modelId) {
-      if (isGrokBuildModel(modelId))
-        return "grok-build";
-      if (isClaudeCodeMinimaxModel(modelId))
-        return "claude-code-minimax";
-      if (isClaudeCodeModel(modelId))
-        return "claude-code";
-      return "native";
-    }
-    exports2.CLAUDE_CODE_DEFAULT_MODEL = "claude-sonnet-4-6";
-    exports2.CLAUDE_CODE_MINIMAX_DEFAULT_MODEL = "MiniMax-M3";
-    exports2.CLAUDE_CODE_MINIMAX_FALLBACK_CHAIN = [
-      "MiniMax-M3",
-      "MiniMax-M2.7-highspeed",
-      "MiniMax-M2.7"
-    ];
-  }
-});
-
 // ../runner-core/dist/services/auth-capture-run.js
 var require_auth_capture_run = __commonJS({
   "../runner-core/dist/services/auth-capture-run.js"(exports2) {
@@ -309650,7 +309881,7 @@ var require_grok_survey_runner = __commonJS({
     var scrub_credentials_js_1 = require_scrub_credentials();
     var agentic_survey_prompts_js_1 = require_agentic_survey_prompts();
     var MCP_CAPABILITIES = ["core", "core-navigation", "core-input", "core-tabs", "testing"];
-    var DEFAULT_TIMEOUT_MS = 15 * 60 * 1e3;
+    var DEFAULT_TIMEOUT_MS2 = 15 * 60 * 1e3;
     var RETRY_TIMEOUT_MS = 5 * 60 * 1e3;
     function extractCredentials(vars) {
       if (!vars)
@@ -310086,7 +310317,7 @@ var require_grok_survey_runner = __commonJS({
     }
     async function runGrokSurveyOnRunner(ctx, options = {}) {
       const maxTurns = options.maxTurns;
-      const timeoutMs = options.timeoutMs ?? DEFAULT_TIMEOUT_MS;
+      const timeoutMs = options.timeoutMs ?? DEFAULT_TIMEOUT_MS2;
       const logs2 = [];
       const log2 = (line) => {
         logs2.push(line);
@@ -310220,7 +310451,7 @@ ${(0, inbox_tools_js_1.buildInboxAgentsGuidance)(inboxReg.email, inboxReg.passwo
       try {
         const result = await (0, grok_cli_js_1.runGrok)({
           prompt: kickoffPrompt,
-          model: "grok-build",
+          model: "grok-4.5",
           maxTurns,
           timeoutMs,
           homeDir: grokHomeDir,
@@ -310267,7 +310498,7 @@ ${(0, inbox_tools_js_1.buildInboxAgentsGuidance)(inboxReg.email, inboxReg.passwo
         try {
           const retryResult = await (0, grok_cli_js_1.runGrok)({
             prompt: fixPrompt,
-            model: "grok-build",
+            model: "grok-4.5",
             // Retry: also let grok decide turns; the shorter retry budget lives
             // in the wall-clock timeout below.
             timeoutMs: RETRY_TIMEOUT_MS,
@@ -310400,20 +310631,19 @@ var require_canvas_detector = __commonJS({
         return { kind: "override", canvasMode: false };
       if (opts.engine === "claude-code")
         return { kind: "canvas-mode", canvasMode: true };
-      if (opts.engine === "claude-code-minimax" || opts.engine === "native")
+      if (opts.engine === "native")
         return { kind: "bail" };
       return { kind: "proceed", canvasMode: false };
     }
     function modelSupportsVision(modelId) {
       if (!modelId)
         return false;
-      if (modelId === "claude-code-minimax")
-        return false;
       if (modelId === "claude-code")
         return true;
       if (modelId === "grok-build")
         return false;
       const visionCapable = /* @__PURE__ */ new Set([
+        "claude-sonnet-5",
         "claude-sonnet-4-6",
         "claude-opus-4-7",
         "claude-haiku-4-5",
@@ -311590,7 +311820,7 @@ var require_grok_discovery_runner = __commonJS({
       "core-tabs",
       "testing"
     ];
-    var DEFAULT_TIMEOUT_MS = 20 * 60 * 1e3;
+    var DEFAULT_TIMEOUT_MS2 = 20 * 60 * 1e3;
     var RETRY_TIMEOUT_MS = 6 * 60 * 1e3;
     var DEFAULT_TARGET = { min: 6, max: 15 };
     var WATCHER_DEBOUNCE_MS = 500;
@@ -311754,7 +311984,7 @@ var require_grok_discovery_runner = __commonJS({
       const manifestFeatureRoutes = Array.isArray(manifestShape?.feature?.pages) ? manifestShape.feature.pages.filter((r) => typeof r === "string" && r.length > 0) : [];
       const featureRoutes = manifestFeatureRoutes.length > 0 ? manifestFeatureRoutes : ctx.featureContext?.pageIntel?.map((p) => p.route) ?? ctx.focusAreas ?? [];
       const targetTestCount = options.targetTestCount ?? DEFAULT_TARGET;
-      const timeoutMs = options.timeoutMs ?? DEFAULT_TIMEOUT_MS;
+      const timeoutMs = options.timeoutMs ?? DEFAULT_TIMEOUT_MS2;
       const logs2 = [];
       const log2 = (line) => {
         logs2.push(line);
@@ -312359,7 +312589,7 @@ ${(0, inbox_tools_js_1.buildInboxAgentsGuidance)(inboxReg.email, inboxReg.passwo
         try {
           const result = await (0, grok_cli_js_1.runGrok)({
             prompt: kickoffPrompt,
-            model: "grok-build",
+            model: "grok-4.5",
             timeoutMs,
             homeDir: grokHomeDir,
             cwd: runDir,
@@ -312439,7 +312669,7 @@ ${(0, inbox_tools_js_1.buildInboxAgentsGuidance)(inboxReg.email, inboxReg.passwo
           try {
             const retryResult = await (0, grok_cli_js_1.runGrok)({
               prompt: fixPrompt,
-              model: "grok-build",
+              model: "grok-4.5",
               timeoutMs: RETRY_TIMEOUT_MS,
               homeDir: grokHomeDir,
               cwd: runDir,
@@ -312671,7 +312901,7 @@ ${(0, inbox_tools_js_1.buildInboxAgentsGuidance)(inboxReg.email, inboxReg.passwo
       const manifestFeatureRoutes = Array.isArray(manifestShape?.feature?.pages) ? manifestShape.feature.pages.filter((r) => typeof r === "string" && r.length > 0) : [];
       const featureRoutes = manifestFeatureRoutes.length > 0 ? manifestFeatureRoutes : ctx.featureContext?.pageIntel?.map((p) => p.route) ?? ctx.focusAreas ?? [];
       const targetTestCount = options.targetTestCount ?? DEFAULT_TARGET;
-      const callerTimeoutMs = options.timeoutMs ?? DEFAULT_TIMEOUT_MS;
+      const callerTimeoutMs = options.timeoutMs ?? DEFAULT_TIMEOUT_MS2;
       const call1TimeoutMs = Math.max(callerTimeoutMs, GROK_V2_CALL1_TIMEOUT_MS);
       const logs2 = [];
       const log2 = (line) => {
@@ -313188,24 +313418,12 @@ var require_claude_cli = __commonJS({
       return s.includes("authentication_failed") || s.includes("oauth_org_not_allowed") || s.includes("invalid_grant") || s.includes("401 unauthorized") || s.includes("invalid api key") || s.includes("invalid bearer token") || /\boauth token\b.*\b(expired|invalid)\b/.test(s) || /\b(token|credentials?)\b.*\bexpired\b/.test(s);
     }
     function runClaudeOnce(options) {
-      const { prompt, model = "claude-sonnet-4-6", provider = "anthropic", maxTurns, timeoutMs = 6e5, cwd, mcpConfigPath, addDirs, appendSystemPrompt, disallowedTools = ["Task"], onChunk, idleKillMs = 5 * 60 * 1e3, abortRef } = options;
+      const { prompt, model = agentic_cli_constants_js_1.CLAUDE_CODE_DEFAULT_MODEL, effort = agentic_cli_constants_js_1.CLAUDE_CODE_DEFAULT_EFFORT, provider = "anthropic", maxTurns, timeoutMs = 6e5, cwd, mcpConfigPath, addDirs, appendSystemPrompt, disallowedTools = ["Task"], onChunk, idleKillMs = 5 * 60 * 1e3, abortRef } = options;
       return new Promise((resolve, reject) => {
-        let oauthToken;
-        let minimaxBaseUrl;
-        let minimaxApiKey;
-        if (provider === "minimax") {
-          minimaxBaseUrl = process.env.MINIMAX_ANTHROPIC_BASE_URL;
-          minimaxApiKey = process.env.MINIMAX_API_KEY;
-          if (!minimaxBaseUrl || !minimaxApiKey) {
-            reject(new ClaudeSpawnError("Claude Code (MiniMax) engine selected but MINIMAX_ANTHROPIC_BASE_URL and/or MINIMAX_API_KEY are not set. Set both as Railway env vars on the runner service.", -1, "", "", true));
-            return;
-          }
-        } else {
-          oauthToken = process.env.CLAUDE_CODE_OAUTH_TOKEN;
-          if (!oauthToken) {
-            reject(new ClaudeSpawnError("Claude Code engine selected but CLAUDE_CODE_OAUTH_TOKEN is not set. Set it as a Railway env var on the runner service.", -1, "", "", true));
-            return;
-          }
+        const oauthToken = process.env.CLAUDE_CODE_OAUTH_TOKEN;
+        if (!oauthToken) {
+          reject(new ClaudeSpawnError("Claude Code engine selected but CLAUDE_CODE_OAUTH_TOKEN is not set. Set it as a Railway env var on the runner service.", -1, "", "", true));
+          return;
         }
         const args = [
           "-p",
@@ -313227,8 +313445,8 @@ var require_claude_cli = __commonJS({
           // Stateless container — don't write session transcripts to disk.
           "--no-session-persistence"
         ];
-        if (provider === "minimax") {
-          args.push("--bare");
+        if (effort && effort.trim().length > 0) {
+          args.push("--effort", effort.trim());
         }
         if (typeof maxTurns === "number" && maxTurns > 0) {
           args.push("--max-turns", String(maxTurns));
@@ -313253,21 +313471,9 @@ var require_claude_cli = __commonJS({
             MCP_TIMEOUT: process.env.MCP_TIMEOUT ?? "120000"
           }
         });
-        if (provider === "minimax") {
-          spawnEnv.ANTHROPIC_BASE_URL = minimaxBaseUrl;
-          spawnEnv.ANTHROPIC_API_KEY = minimaxApiKey;
-          spawnEnv.ANTHROPIC_AUTH_TOKEN = minimaxApiKey;
-          delete spawnEnv.CLAUDE_CODE_OAUTH_TOKEN;
-          delete spawnEnv.ANTHROPIC_MODEL;
-          delete spawnEnv.ANTHROPIC_DEFAULT_SONNET_MODEL;
-          delete spawnEnv.ANTHROPIC_DEFAULT_OPUS_MODEL;
-          delete spawnEnv.ANTHROPIC_DEFAULT_HAIKU_MODEL;
-          delete spawnEnv.ANTHROPIC_SMALL_FAST_MODEL;
-        } else {
-          spawnEnv.CLAUDE_CODE_OAUTH_TOKEN = oauthToken;
-          delete spawnEnv.ANTHROPIC_API_KEY;
-          delete spawnEnv.ANTHROPIC_AUTH_TOKEN;
-        }
+        spawnEnv.CLAUDE_CODE_OAUTH_TOKEN = oauthToken;
+        delete spawnEnv.ANTHROPIC_API_KEY;
+        delete spawnEnv.ANTHROPIC_AUTH_TOKEN;
         const isWindows = process.platform === "win32";
         const proc = (0, node_child_process_1.spawn)("claude", args, {
           env: spawnEnv,
@@ -313379,7 +313585,7 @@ var require_claude_cli = __commonJS({
           }
           if (exitCode !== 0) {
             const authFailure = looksLikeAuthFailure(stdout + "\n" + stderr);
-            const msg = authFailure ? provider === "minimax" ? "claude --bare session failed to authenticate against MiniMax \u2014 MINIMAX_API_KEY is missing, expired, or rejected by MINIMAX_ANTHROPIC_BASE_URL. Verify both env vars on the runner service." : "claude session failed to authenticate \u2014 CLAUDE_CODE_OAUTH_TOKEN is missing, expired, or invalid. Regenerate it with `claude setup-token` and update the Railway env var on the runner service." : `claude exited with code ${exitCode}${signal ? ` (signal=${signal})` : ""}`;
+            const msg = authFailure ? "claude session failed to authenticate \u2014 CLAUDE_CODE_OAUTH_TOKEN is missing, expired, or invalid. Regenerate it with `claude setup-token` and update the Railway env var on the runner service." : `claude exited with code ${exitCode}${signal ? ` (signal=${signal})` : ""}`;
             reject(new ClaudeSpawnError(msg, exitCode, stderr.trim(), stdout.trim(), authFailure));
             return;
           }
@@ -313392,39 +313598,8 @@ var require_claude_cli = __commonJS({
         });
       });
     }
-    function minimaxClaudeFallbackModels(requested) {
-      const chain = agentic_cli_constants_js_1.CLAUDE_CODE_MINIMAX_FALLBACK_CHAIN;
-      const idx = chain.indexOf(requested);
-      if (idx === -1)
-        return [requested];
-      return chain.slice(idx);
-    }
     async function runClaude(options) {
-      if (options.provider !== "minimax")
-        return runClaudeOnce(options);
-      const requested = options.model ?? agentic_cli_constants_js_1.CLAUDE_CODE_MINIMAX_DEFAULT_MODEL;
-      const chain = minimaxClaudeFallbackModels(requested);
-      if (chain.length <= 1)
-        return runClaudeOnce({ ...options, model: requested });
-      let lastError;
-      for (let i = 0; i < chain.length; i++) {
-        const model = chain[i];
-        const isLast = i === chain.length - 1;
-        try {
-          const result = await runClaudeOnce({ ...options, model });
-          if (i > 0)
-            console.warn(`[claude-cli] MiniMax CLI failover succeeded on ${model} (requested ${requested})`);
-          return result;
-        } catch (err) {
-          lastError = err;
-          const spawnErr = err instanceof ClaudeSpawnError ? err : null;
-          if (isLast || spawnErr?.isAuthError || spawnErr?.isTimeout)
-            throw err;
-          const reason = err instanceof Error ? err.message : String(err);
-          console.warn(`[claude-cli] MiniMax CLI ${model} failed (${reason}) \u2014 failing over to ${chain[i + 1]}`);
-        }
-      }
-      throw lastError;
+      return runClaudeOnce(options);
     }
   }
 });
@@ -313809,6 +313984,8 @@ var require_claude_pricing = __commonJS({
     Object.defineProperty(exports2, "__esModule", { value: true });
     exports2.computeClaudeCostUsd = computeClaudeCostUsd;
     var CLAUDE_PRICE_PER_M = {
+      // Introductory pricing through 2026-08-31 ($2/$10); standard is $3/$15.
+      "claude-sonnet-5": { input: 2, output: 10 },
       "claude-sonnet-4-6": { input: 3, output: 15 },
       "claude-haiku-4-5-20251001": { input: 1, output: 5 },
       "claude-opus-4-7": { input: 5, output: 25 }
@@ -313848,7 +314025,7 @@ var require_claude_survey_runner = __commonJS({
     var agentic_survey_prompts_js_1 = require_agentic_survey_prompts();
     var MCP_CAPABILITIES = ["core", "core-navigation", "core-input", "core-tabs", "testing"];
     var CANVAS_MODE_EXTRA_CAPABILITIES = ["vision"];
-    var DEFAULT_TIMEOUT_MS = 15 * 60 * 1e3;
+    var DEFAULT_TIMEOUT_MS2 = 15 * 60 * 1e3;
     var RETRY_TIMEOUT_MS = 5 * 60 * 1e3;
     async function logResultFile(path, log2) {
       try {
@@ -313860,10 +314037,10 @@ ${preview}`);
         log2("[diagnostic] result file does not exist \u2014 Claude never wrote it");
       }
     }
-    async function runClaudeSurveyOnRunner(ctx, options = {}, provider = "anthropic") {
+    async function runClaudeSurveyOnRunner(ctx, options = {}) {
       const maxTurns = options.maxTurns;
-      const timeoutMs = options.timeoutMs ?? DEFAULT_TIMEOUT_MS;
-      const claudeDefaultModel = provider === "minimax" ? agentic_cli_constants_js_1.CLAUDE_CODE_MINIMAX_DEFAULT_MODEL : agentic_cli_constants_js_1.CLAUDE_CODE_DEFAULT_MODEL;
+      const timeoutMs = options.timeoutMs ?? DEFAULT_TIMEOUT_MS2;
+      const claudeDefaultModel = agentic_cli_constants_js_1.CLAUDE_CODE_DEFAULT_MODEL;
       const logs2 = [];
       const log2 = (line) => {
         logs2.push(line);
@@ -314007,7 +314184,7 @@ ${(0, inbox_tools_js_1.buildInboxAgentsGuidance)(inboxReg.email, inboxReg.passwo
         console.log(`[claude-survey] ${line}`);
       };
       const onAudit = options.onAudit;
-      const claudeProviderLabel = provider === "minimax" ? "minimax" : "anthropic";
+      const claudeProviderLabel = "anthropic";
       const claudeModelLabel = claudeDefaultModel;
       const surveyExploreCallType = "surveyExplore";
       const sessionStartedAt = Date.now();
@@ -314043,7 +314220,7 @@ ${(0, inbox_tools_js_1.buildInboxAgentsGuidance)(inboxReg.email, inboxReg.passwo
             groupId: auditGroupId,
             callType: surveyExploreCallType,
             model: claudeModelLabel,
-            provider: claudeProviderLabel,
+            provider: "anthropic",
             durationMs: turn.durationMs ?? 0,
             promptTokens,
             completionTokens,
@@ -314086,7 +314263,6 @@ ${(0, inbox_tools_js_1.buildInboxAgentsGuidance)(inboxReg.email, inboxReg.passwo
         const result = await (0, claude_cli_js_1.runClaude)({
           prompt: kickoffPrompt,
           model: claudeDefaultModel,
-          provider,
           maxTurns,
           timeoutMs,
           cwd: runDir,
@@ -314146,7 +314322,6 @@ ${(0, inbox_tools_js_1.buildInboxAgentsGuidance)(inboxReg.email, inboxReg.passwo
           const retryResult = await (0, claude_cli_js_1.runClaude)({
             prompt: fixPrompt,
             model: claudeDefaultModel,
-            provider,
             timeoutMs: RETRY_TIMEOUT_MS,
             cwd: runDir,
             mcpConfigPath,
@@ -314202,7 +314377,7 @@ ${(0, inbox_tools_js_1.buildInboxAgentsGuidance)(inboxReg.email, inboxReg.passwo
             groupId: auditGroupId,
             callType: surveyExploreCallType,
             model: claudeModelLabel,
-            provider: claudeProviderLabel,
+            provider: "anthropic",
             durationMs: sessionDurationMs,
             promptTokens: sessionInputTokens,
             completionTokens: sessionOutputTokens,
@@ -314578,11 +314753,11 @@ var require_claude_discovery_adapter = __commonJS({
     var PER_TEST_IDLE_KILL_MS = 10 * 60 * 1e3;
     var PER_TEST_MAX_TURNS = 120;
     function createClaudeDiscoveryAdapter(args) {
-      const { provider, claudeDefaultModel } = args;
+      const { claudeDefaultModel } = args;
       return {
-        engine: provider === "minimax" ? "claude-code-minimax" : "claude-code",
+        engine: "claude-code",
         auditModel: claudeDefaultModel,
-        auditProvider: provider,
+        auditProvider: "anthropic",
         computeSessionCostUsd(usage) {
           return (0, claude_pricing_js_1.computeClaudeCostUsd)(claudeDefaultModel, {
             inputTokens: usage.inputTokens,
@@ -314623,7 +314798,7 @@ var require_claude_discovery_adapter = __commonJS({
               prompt,
               appendSystemPrompt,
               model: claudeDefaultModel,
-              provider,
+              provider: "anthropic",
               maxTurns: PER_TEST_MAX_TURNS,
               // A per-test session needs only browser MCP + the file tools — never a
               // shell. Disallow Bash to shrink the off-task surface (and Task, the
@@ -314695,17 +314870,17 @@ var require_claude_discovery_runner = __commonJS({
     ];
     var CALL1_TIMEOUT_MS = 38 * 60 * 1e3;
     var DISCOVERY_IDLE_KILL_MS = 15 * 60 * 1e3;
-    var DEFAULT_TIMEOUT_MS = 20 * 60 * 1e3;
+    var DEFAULT_TIMEOUT_MS2 = 20 * 60 * 1e3;
     var DEFAULT_TARGET = { min: 6, max: 15 };
-    async function runClaudeDiscoveryOnRunner(ctx, callbackOptions, options = {}, provider = "anthropic") {
-      const claudeDefaultModel = provider === "minimax" ? agentic_cli_constants_js_1.CLAUDE_CODE_MINIMAX_DEFAULT_MODEL : agentic_cli_constants_js_1.CLAUDE_CODE_DEFAULT_MODEL;
-      const claudeProviderLabel = provider === "minimax" ? "minimax" : "anthropic";
+    async function runClaudeDiscoveryOnRunner(ctx, callbackOptions, options = {}) {
+      const claudeDefaultModel = agentic_cli_constants_js_1.CLAUDE_CODE_DEFAULT_MODEL;
+      const claudeProviderLabel = "anthropic";
       const featureName = ctx.featureName ?? "Untitled Feature";
       const manifestShape = (0, grok_discovery_runner_js_1.narrowManifest)(ctx.featureContextManifest);
       const manifestFeatureRoutes = Array.isArray(manifestShape?.feature?.pages) ? manifestShape.feature.pages.filter((r) => typeof r === "string" && r.length > 0) : [];
       const featureRoutes = manifestFeatureRoutes.length > 0 ? manifestFeatureRoutes : ctx.featureContext?.pageIntel?.map((p) => p.route) ?? ctx.focusAreas ?? [];
       const targetTestCount = options.targetTestCount ?? DEFAULT_TARGET;
-      const callerTimeoutMs = options.timeoutMs ?? DEFAULT_TIMEOUT_MS;
+      const callerTimeoutMs = options.timeoutMs ?? DEFAULT_TIMEOUT_MS2;
       const call1TimeoutMs = Math.max(callerTimeoutMs, CALL1_TIMEOUT_MS);
       const logs2 = [];
       const log2 = (line) => {
@@ -314952,7 +315127,7 @@ ${(0, inbox_tools_js_1.buildInboxAgentsGuidance)(inboxReg.email, inboxReg.passwo
               groupId: ctx.sessionId ?? featureName,
               callType: featureExploreCallType,
               model: claudeModelLabel,
-              provider: claudeProviderLabel,
+              provider: "anthropic",
               durationMs: turn.durationMs ?? 0,
               promptTokens,
               completionTokens,
@@ -315063,7 +315238,6 @@ ${(0, inbox_tools_js_1.buildInboxAgentsGuidance)(inboxReg.email, inboxReg.passwo
             // they actually get put there.
             appendSystemPrompt: canvasMode ? (0, agentic_discovery_prompts_js_1.buildAuthoringRules)() + (0, canvas_detector_js_1.buildCanvasModeAuthoringAddendum)() : (0, agentic_discovery_prompts_js_1.buildAuthoringRules)(),
             model: claudeDefaultModel,
-            provider,
             timeoutMs: call1TimeoutMs,
             idleKillMs: DISCOVERY_IDLE_KILL_MS,
             cwd: runDir,
@@ -315173,7 +315347,7 @@ ${(0, inbox_tools_js_1.buildInboxAgentsGuidance)(inboxReg.email, inboxReg.passwo
               groupId: ctx.sessionId ?? featureName,
               callType: featureExploreCallType,
               model: claudeModelLabel,
-              provider: claudeProviderLabel,
+              provider: "anthropic",
               durationMs: sessionDurationMs,
               promptTokens: sessionInputTokens,
               completionTokens: sessionOutputTokens,
@@ -315265,7 +315439,7 @@ ${(0, inbox_tools_js_1.buildInboxAgentsGuidance)(inboxReg.email, inboxReg.passwo
             },
             onAudit,
             auditGroupId: ctx.sessionId ?? featureName,
-            adapter: (0, claude_discovery_adapter_js_1.createClaudeDiscoveryAdapter)({ provider, claudeDefaultModel })
+            adapter: (0, claude_discovery_adapter_js_1.createClaudeDiscoveryAdapter)({ claudeDefaultModel })
           });
           log2(`Verify + Call 2: complete \u2014 ${capturedTests.length} test(s) saved, ${localUnverified.length} unverified`);
         }
@@ -316191,7 +316365,7 @@ var require_discover_agent = __commonJS({
       } else if (cached === "standard") {
         return { renderType: "standard", canvasMode: false, probeRan: false };
       } else {
-        const bailEligible = engine === "native" || engine === "claude-code-minimax";
+        const bailEligible = engine === "native";
         const cannotConfirm = (reason) => {
           const note = bailEligible ? ` \u26A0 engine=${engine} would have bailed on canvas \u2014 if discovery loops, set Settings \u2192 App renderer manually.` : "";
           log2(`[canvas] preflight could not run: ${reason} \u2014 proceeding as standard.${note}`);
@@ -316510,7 +316684,7 @@ Respond with ONLY valid JSON:
             // wiring is in place when it does.
             onAudit: onAICall
           };
-          const cliResult = surveyEngine === "claude-code-minimax" ? await (0, claude_survey_runner_js_1.runClaudeSurveyOnRunner)(ctx, surveyOpts, "minimax") : surveyEngine === "claude-code" ? await (0, claude_survey_runner_js_1.runClaudeSurveyOnRunner)(ctx, surveyOpts, "anthropic") : await (0, grok_survey_runner_js_1.runGrokSurveyOnRunner)(ctx, surveyOpts);
+          const cliResult = surveyEngine === "claude-code" ? await (0, claude_survey_runner_js_1.runClaudeSurveyOnRunner)(ctx, surveyOpts) : await (0, grok_survey_runner_js_1.runGrokSurveyOnRunner)(ctx, surveyOpts);
           logs2.push(`${surveyEngine} survey runner finished cleanly.`);
           onLog?.("done", logs2.join("\n"));
           return {
@@ -317032,7 +317206,7 @@ Fatal error: ${msg}`);
             // through the existing DiscoveryAuditCallback (→ aiAuditBatch).
             onAudit: onAICall
           };
-          const cliResult = deepEngine === "claude-code-minimax" ? await (0, claude_discovery_runner_js_1.runClaudeDiscoveryOnRunner)(ctx, deepCallbacks, deepOpts, "minimax") : deepEngine === "claude-code" ? await (0, claude_discovery_runner_js_1.runClaudeDiscoveryOnRunner)(ctx, deepCallbacks, deepOpts, "anthropic") : await (0, grok_discovery_runner_js_1.runGrokDiscoveryOnRunner)(ctx, deepCallbacks, deepOpts);
+          const cliResult = deepEngine === "claude-code" ? await (0, claude_discovery_runner_js_1.runClaudeDiscoveryOnRunner)(ctx, deepCallbacks, deepOpts) : await (0, grok_discovery_runner_js_1.runGrokDiscoveryOnRunner)(ctx, deepCallbacks, deepOpts);
           logs2.push(`${deepEngine} deep discovery finished cleanly. ${cliResult.testsEmitted} tests emitted, ${cliResult.unverified.length} unverified.`);
           onLog?.("done", logs2.join("\n"));
           return {
@@ -327912,9 +328086,8 @@ var require_framework_import_runner = __commonJS({
       const imp = ctx.frameworkImport;
       if (!imp)
         throw new Error("runFrameworkImportOnRunner: ctx.frameworkImport is required");
-      const engine = (0, agentic_cli_constants_js_1.getAgenticEngine)(ctx.exploreModel);
-      const provider = engine === "claude-code-minimax" ? "minimax" : "anthropic";
-      const model = provider === "minimax" ? agentic_cli_constants_js_1.CLAUDE_CODE_MINIMAX_DEFAULT_MODEL : agentic_cli_constants_js_1.CLAUDE_CODE_DEFAULT_MODEL;
+      const provider = "anthropic";
+      const model = agentic_cli_constants_js_1.CLAUDE_CODE_DEFAULT_MODEL;
       log2("=== Framework Import Runner (Claude CLI, no browser) ===");
       log2(`Source: ${imp.source}${imp.source === "github" ? ` (${imp.repoOwner}/${imp.repoName}@${imp.branch ?? "default"})` : ""}`);
       const runDir = await (0, promises_1.mkdtemp)((0, node_path_1.join)((0, node_os_1.tmpdir)(), "framework-import-"));
@@ -329044,8 +329217,8 @@ var require_repo_analysis_runner = __commonJS({
       const maxEndpointsPerFeature = intEnv("REPO_ANALYSIS_MAX_ENDPOINTS_PER_FEATURE", 80, 1, 400);
       const maxFacts = intEnv("REPO_ANALYSIS_MAX_FACTS", 40, 1, 120);
       const engine = (0, agentic_cli_constants_js_1.getAgenticEngine)(ctx.exploreModel);
-      const provider = engine === "claude-code-minimax" ? "minimax" : "anthropic";
-      const model = provider === "minimax" ? agentic_cli_constants_js_1.CLAUDE_CODE_MINIMAX_DEFAULT_MODEL : agentic_cli_constants_js_1.CLAUDE_CODE_DEFAULT_MODEL;
+      const provider = "anthropic";
+      const model = agentic_cli_constants_js_1.CLAUDE_CODE_DEFAULT_MODEL;
       log2("=== Repo Analysis Runner (Claude CLI, no browser) ===");
       log2(`Source: ${imp.source}${imp.source === "github" ? ` (${imp.repoOwner}/${imp.repoName}@${imp.branch ?? "default"})` : ""}`);
       const runDir = await (0, promises_1.mkdtemp)((0, node_path_1.join)((0, node_os_1.tmpdir)(), "repo-analysis-"));
@@ -329528,7 +329701,7 @@ var require_process_inventory = __commonJS({
     var node_util_1 = require("util");
     var execFileAsync3 = (0, node_util_1.promisify)(node_child_process_1.execFile);
     var DEFAULT_LIMIT = 60;
-    var DEFAULT_TIMEOUT_MS = 800;
+    var DEFAULT_TIMEOUT_MS2 = 800;
     var MAX_COMMAND_LENGTH = 320;
     function roundOne(value) {
       return Math.round(value * 10) / 10;
@@ -329677,7 +329850,7 @@ var require_process_inventory = __commonJS({
     }
     async function getBrowserProcessInventory2(options = {}) {
       const limit = Math.max(1, Math.min(200, options.limit ?? DEFAULT_LIMIT));
-      const timeoutMs = Math.max(100, Math.min(5e3, options.timeoutMs ?? DEFAULT_TIMEOUT_MS));
+      const timeoutMs = Math.max(100, Math.min(5e3, options.timeoutMs ?? DEFAULT_TIMEOUT_MS2));
       try {
         const rawProcesses = process.platform === "win32" ? await collectWindowsProcesses(timeoutMs) : await collectUnixProcesses(timeoutMs);
         const processes = [...rawProcesses].sort((a, b) => (b.cpuPct ?? 0) - (a.cpuPct ?? 0) || (b.rssMB ?? 0) - (a.rssMB ?? 0)).slice(0, limit);
@@ -330644,7 +330817,7 @@ var require_package4 = __commonJS({
   "package.json"(exports2, module2) {
     module2.exports = {
       name: "@validate.qa/runner",
-      version: "1.0.33",
+      version: "1.0.35",
       description: "validate local test runner - execute Playwright tests on your machine, connected to validate.qa cloud.",
       bin: {
         "validate-runner": "dist/cli.js"
@@ -330652,6 +330825,7 @@ var require_package4 = __commonJS({
       scripts: {
         build: "tsup",
         dev: "tsx src/cli.ts",
+        test: "node --import tsx --test src/version-check.test.ts",
         "test:mobile": "node --import tsx --test src/runner.mobile-ai-env.test.ts src/services/mobile/mobile-network-capture.smoke.test.ts src/services/mobile/mobile-host-proxy.test.ts",
         prepublishOnly: "npm run build"
       },
@@ -330780,6 +330954,136 @@ function shouldUseAuthState(tags, playwrightCode) {
   return !tags.includes(INLINE_LOGIN_TAG) && !hasLegacyInlineLogin;
 }
 
+// src/version-check.ts
+var RUNNER_NPM_PACKAGE = "@validate.qa/runner";
+var NPM_LATEST_URL = `https://registry.npmjs.org/${encodeURIComponent(RUNNER_NPM_PACKAGE)}/latest`;
+var DEFAULT_TIMEOUT_MS = 8e3;
+function parseSemver(raw) {
+  const cleaned = raw.trim().replace(/^v/i, "");
+  const match = cleaned.match(/^(\d+)\.(\d+)\.(\d+)(?:-([0-9A-Za-z.-]+))?(?:\+[0-9A-Za-z.-]+)?$/);
+  if (!match) return null;
+  return {
+    major: Number(match[1]),
+    minor: Number(match[2]),
+    patch: Number(match[3]),
+    prerelease: match[4] ?? null
+  };
+}
+function compareSemver(a, b) {
+  const pa = parseSemver(a);
+  const pb = parseSemver(b);
+  if (!pa || !pb) return null;
+  if (pa.major !== pb.major) return pa.major < pb.major ? -1 : 1;
+  if (pa.minor !== pb.minor) return pa.minor < pb.minor ? -1 : 1;
+  if (pa.patch !== pb.patch) return pa.patch < pb.patch ? -1 : 1;
+  if (pa.prerelease === pb.prerelease) return 0;
+  if (pa.prerelease === null) return 1;
+  if (pb.prerelease === null) return -1;
+  return pa.prerelease < pb.prerelease ? -1 : pa.prerelease > pb.prerelease ? 1 : 0;
+}
+function isVersionOutdated(current, latest) {
+  const cmp = compareSemver(current, latest);
+  return cmp === -1;
+}
+function shouldSkipVersionCheck() {
+  if (process.env.RUNNER_CLOUD === "true") {
+    return "cloud runner (RUNNER_CLOUD=true)";
+  }
+  if (process.env.VALIDATE_RUNNER_ALLOW_OUTDATED === "1" || process.env.VALIDATE_RUNNER_ALLOW_OUTDATED === "true") {
+    return "VALIDATE_RUNNER_ALLOW_OUTDATED is set";
+  }
+  return null;
+}
+async function fetchLatestRunnerVersion(timeoutMs = DEFAULT_TIMEOUT_MS, fetchImpl = fetch) {
+  const controller = new AbortController();
+  const timer = setTimeout(() => controller.abort(), timeoutMs);
+  try {
+    const res = await fetchImpl(NPM_LATEST_URL, {
+      headers: {
+        Accept: "application/json",
+        // Identify ourselves; helps npm operators + caching intermediaries.
+        "User-Agent": `${RUNNER_NPM_PACKAGE}-version-check`
+      },
+      signal: controller.signal
+    });
+    if (!res.ok) return null;
+    const body = await res.json();
+    return typeof body.version === "string" && body.version.length > 0 ? body.version : null;
+  } catch {
+    return null;
+  } finally {
+    clearTimeout(timer);
+  }
+}
+async function checkRunnerVersion(currentVersion, opts = {}) {
+  const skip = shouldSkipVersionCheck();
+  if (skip) {
+    return { outdated: false, currentVersion, latestVersion: null, skippedReason: skip };
+  }
+  const latestVersion = await fetchLatestRunnerVersion(opts.timeoutMs, opts.fetchImpl);
+  if (!latestVersion) {
+    return {
+      outdated: false,
+      currentVersion,
+      latestVersion: null,
+      skippedReason: "could not reach npm registry (or response was invalid)"
+    };
+  }
+  if (!parseSemver(currentVersion) || !parseSemver(latestVersion)) {
+    return {
+      outdated: false,
+      currentVersion,
+      latestVersion,
+      skippedReason: "unparseable version string"
+    };
+  }
+  return {
+    outdated: isVersionOutdated(currentVersion, latestVersion),
+    currentVersion,
+    latestVersion
+  };
+}
+function formatUpgradeInstructions(latestVersion) {
+  const lines = [
+    "",
+    "\u2554\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2557",
+    "\u2551  VALIDATE.QA RUNNER UPDATE REQUIRED                              \u2551",
+    "\u2560\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2563",
+    "\u2551  A newer runner is published on npm. Each runner release changes \u2551",
+    "\u2551  the workflow, so this version can no longer run.                \u2551",
+    `\u2551  Latest:  ${latestVersion.padEnd(54)}\u2551`,
+    "\u2551                                                                  \u2551",
+    "\u2551  Upgrade with:                                                   \u2551",
+    `\u2551    npm install -g ${RUNNER_NPM_PACKAGE}@latest`.padEnd(69) + "\u2551",
+    "\u2551                                                                  \u2551",
+    "\u2551  Then restart:                                                   \u2551",
+    "\u2551    validate-runner start                                         \u2551",
+    "\u2551    # or: npx @validate.qa/runner@latest start                    \u2551",
+    "\u255A\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u255D",
+    ""
+  ];
+  return lines.join("\n");
+}
+async function enforceRunnerUpToDate(currentVersion, opts = {}) {
+  const log2 = opts.log ?? ((msg) => console.error(msg));
+  const result = await checkRunnerVersion(currentVersion, opts);
+  if (result.skippedReason && !result.outdated) {
+    if (result.skippedReason.startsWith("could not reach") || result.skippedReason.startsWith("unparseable")) {
+      log2(`\u26A0  Runner version check skipped: ${result.skippedReason}`);
+    }
+    return result;
+  }
+  if (result.outdated && result.latestVersion) {
+    log2(formatUpgradeInstructions(result.latestVersion));
+    log2(`  Installed: ${result.currentVersion}`);
+    log2(`  Required:  ${result.latestVersion} (or newer)`);
+    log2("");
+    const exit = opts.exitFn ?? ((code) => process.exit(code));
+    exit(1);
+  }
+  return result;
+}
+
 // src/services/appium-executor.ts
 init_dist();
 
@@ -335813,13 +336117,31 @@ function buildHeaders(authToken, projectId) {
   return {
     "Content-Type": "application/json",
     Authorization: `Bearer ${authToken}`,
-    "X-Project-Id": projectId
+    "X-Project-Id": projectId,
+    // Lets the server force-upgrade outdated local installs on /pending too
+    // (not only heartbeat), so they cannot claim work after a new npm publish.
+    "X-Runner-Version": RUNNER_VERSION,
+    "X-Runner-Name": RUNNER_NAME
   };
 }
+var RunnerUpgradeRequiredError = class extends Error {
+  latestVersion;
+  constructor(message, latestVersion) {
+    super(message);
+    this.name = "RunnerUpgradeRequiredError";
+    this.latestVersion = latestVersion;
+  }
+};
 async function fetchPendingRuns(serverUrl, headers) {
   const res = await fetch(`${serverUrl}/api/runner/pending`, { headers, signal: AbortSignal.timeout(3e4) });
   if (!res.ok) {
     if (res.status === 401) throw new Error("Runner token is invalid or expired. Run: validate.qa-runner login");
+    if (res.status === 426) {
+      const body = await res.json().catch(() => null);
+      const latest = typeof body?.latestRunnerVersion === "string" ? body.latestRunnerVersion : null;
+      const msg = typeof body?.error === "string" ? body.error : "A newer @validate.qa/runner is required. Run: npm install -g @validate.qa/runner@latest";
+      throw new RunnerUpgradeRequiredError(msg, latest);
+    }
     if (res.status === 429) {
       const retryAfter = res.headers.get("Retry-After");
       const waitSecs = retryAfter ? parseInt(retryAfter, 10) : 60;
@@ -338332,10 +338654,34 @@ async function startRunner(opts) {
   }
   console.log("");
   log.ok(`Runner started - waiting for test runs...${enableMobile ? " (web + mobile)" : ""}`);
+  log.info(`Version:  ${RUNNER_VERSION}`);
   log.dim("Press Ctrl+C to stop");
   console.log("");
   let runsCompleted = 0;
   let activeRuns = 0;
+  const VERSION_RECHECK_MS = Math.max(
+    6e4,
+    envInt("VALIDATE_RUNNER_VERSION_CHECK_MS", 15 * 6e4)
+  );
+  const exitForUpgrade = (latestVersion, installed) => {
+    console.error(formatUpgradeInstructions(latestVersion));
+    console.error(`  Installed: ${installed}`);
+    console.error(`  Required:  ${latestVersion} (or newer)`);
+    console.error("");
+    process.exit(1);
+  };
+  const recheckRunnerVersion = async () => {
+    if (activeRuns > 0) return;
+    const result = await checkRunnerVersion(RUNNER_VERSION);
+    if (result.outdated && result.latestVersion) {
+      log.warn("A newer runner was published on npm \u2014 shutting down so you can upgrade.");
+      exitForUpgrade(result.latestVersion, RUNNER_VERSION);
+    }
+  };
+  const versionRecheckTimer = setInterval(() => {
+    void recheckRunnerVersion();
+  }, VERSION_RECHECK_MS);
+  versionRecheckTimer.unref?.();
   const MAX_CONCURRENT = Math.max(1, envInt("MAX_CONCURRENT", DEFAULT_MAX_CONCURRENT));
   const MAX_PROJECT_CONCURRENT = Math.max(1, envInt("MAX_PROJECT_CONCURRENT", DEFAULT_MAX_PROJECT_CONCURRENT));
   const TYPE_LIMITS = {
@@ -338561,6 +338907,22 @@ async function startRunner(opts) {
         if (body && Array.isArray(body.commands) && body.commands.length > 0) {
           void handleControlCommands(body.commands);
         }
+        if (body && body.forceUpgrade === true && typeof body.latestRunnerVersion === "string" && isVersionOutdated(RUNNER_VERSION, body.latestRunnerVersion) && activeRuns === 0) {
+          log.warn(
+            typeof body.upgradeMessage === "string" && body.upgradeMessage ? body.upgradeMessage : "Server requires a newer runner version \u2014 shutting down."
+          );
+          exitForUpgrade(body.latestRunnerVersion, RUNNER_VERSION);
+        }
+      } else if (response.status === 426) {
+        const body = await response.json().catch(() => null);
+        const latest = typeof body?.latestRunnerVersion === "string" ? body.latestRunnerVersion : null;
+        if (latest) {
+          exitForUpgrade(latest, RUNNER_VERSION);
+        }
+        log.warn(
+          typeof body?.error === "string" ? body.error : "Server rejected this runner version (HTTP 426). Upgrade @validate.qa/runner."
+        );
+        process.exit(1);
       }
     } catch {
     }
@@ -338801,6 +339163,10 @@ async function startRunner(opts) {
         consecutivePollErrors = 0;
       }
     } catch (err) {
+      if (err instanceof RunnerUpgradeRequiredError) {
+        log.warn(err.message);
+        exitForUpgrade(err.latestVersion ?? "latest", RUNNER_VERSION);
+      }
       const message = err instanceof Error ? err.message : String(err);
       consecutivePollErrors++;
       const backoffMs = Math.min(pollIntervalMs * Math.pow(2, consecutivePollErrors - 1), MAX_BACKOFF_MS);
@@ -338820,7 +339186,11 @@ async function startRunner(opts) {
 var pkg = require_package4();
 var program = new import_commander.Command();
 program.name("validate-runner").description("validate.qa local test runner - execute Playwright and Appium tests on your machine").version(pkg.version);
+async function requireLatestRunner() {
+  await enforceRunnerUpToDate(pkg.version);
+}
 program.command("login").description("Save runner credentials for future use").requiredOption("--token <token>", "Runner token (urt_... from validate.qa dashboard)").requiredOption("--server <url>", "validate.qa server URL").option("--project <id>", "Project ID (from dashboard URL)").action(async (opts) => {
+  await requireLatestRunner();
   const serverUrl = opts.server.replace(/\/+$/, "");
   if (!opts.token.startsWith("urt_") && !opts.token.startsWith("rt_") && !opts.token.startsWith("eyJ")) {
     console.error("\u26A0  Token should start with urt_ (from the dashboard Runner Token)");
@@ -338852,6 +339222,7 @@ program.command("login").description("Save runner credentials for future use").r
   console.log("Run: validate-runner start");
 });
 program.command("start").description("Start the local test runner").option("--token <token>", "Runner token (overrides saved credentials)").option("--server <url>", "Server URL (overrides saved credentials)").option("--project <id>", "Project ID (overrides saved credentials)").option("--mode <mode>", "Runner mode: playwright-native (default)", "playwright-native").option("--poll-interval <ms>", "Poll interval in milliseconds", "2000").option("--mobile", "Enable mobile testing (Appium + mobile bridge server)").option("--tunnel", "Enable ngrok tunnel for mobile bridge (allows HTTPS dashboard to connect)").action(async (opts) => {
+  await requireLatestRunner();
   const saved = loadCredentials();
   const authToken = opts.token || process.env.AUTH_TOKEN || saved?.token;
   const serverUrl = (opts.server || process.env.SERVER_URL || saved?.serverUrl)?.replace(/\/+$/, "");
@@ -338904,6 +339275,7 @@ program.command("logout").description("Clear saved runner credentials").action((
   console.log("\u2713  Credentials cleared");
 });
 program.command("status").description("Check connection to validate.qa server").action(async () => {
+  await requireLatestRunner();
   const saved = loadCredentials();
   if (!saved) {
     console.log("No saved credentials. Run: validate-runner login");