@exodus/xqa 1.13.0 → 1.14.1

package/dist/xqa.cjs

@@ -52247,7 +52247,6 @@ function createAccessibilitySnapshotTool(udid = "booted") {
     }
   );
 }
-var accessibilitySnapshotTool = createAccessibilitySnapshotTool();
 var LAUNCH_APP_TOOL_NAME = "launch_app";
 var TERMINATE_APP_TOOL_NAME = "terminate_app";
 var LIST_APPS_TOOL_NAME = "list_apps";
@@ -52348,11 +52347,9 @@ function createListAppsTool(udid = "booted") {
     async () => handleListApps(udid)
   );
 }
-var launchAppTool = createLaunchAppTool();
-var terminateAppTool = createTerminateAppTool();
-var listAppsTool = createListAppsTool();
 var DEFAULT_LONG_PRESS_DURATION_MS = 500;
 var MS_PER_SECOND = 1e3;
+var DEFAULT_SWIPE_DURATION_SECONDS = 0.1;
 var ENTER_KEY_CODE = "0x28";
 var TAP_SCHEMA = { x: external_exports.number(), y: external_exports.number() };
 var LONG_PRESS_SCHEMA = { x: external_exports.number(), y: external_exports.number(), duration: external_exports.number().optional() };
@@ -52360,7 +52357,10 @@ var SWIPE_SCHEMA = {
   x_start: external_exports.number(),
   y_start: external_exports.number(),
   x_end: external_exports.number(),
-  y_end: external_exports.number()
+  y_end: external_exports.number(),
+  duration: external_exports.number().positive().optional().describe(
+    "Gesture duration in seconds. Smaller values (e.g. 0.1) produce a flick that scrolls lists and dismisses sheets; larger values (e.g. 0.5+) produce a slow drag suitable for reordering or panning. Defaults to 0.1s (flick)."
+  )
 };
 var TYPE_TEXT_SCHEMA = { text: external_exports.string(), submit: external_exports.boolean().optional() };
 var PRESS_BUTTON_SCHEMA = {
@@ -52443,7 +52443,9 @@ function buildSwipeArguments(resolvedUdid, input) {
     String(input.xStart),
     String(input.yStart),
     String(input.xEnd),
-    String(input.yEnd)
+    String(input.yEnd),
+    "--duration",
+    String(input.duration)
   ];
 }
 async function handleSwipe(input) {
@@ -52510,17 +52512,22 @@ function createLongPressTool(udid = "booted") {
     async ({ x, y: y6, duration: duration3 }) => handleLongPress({ udid, x, y: y6, duration: duration3 ?? DEFAULT_LONG_PRESS_DURATION_MS })
   );
 }
+function toSwipeInput(udid, params) {
+  return {
+    udid,
+    xStart: params.x_start,
+    yStart: params.y_start,
+    xEnd: params.x_end,
+    yEnd: params.y_end,
+    duration: params.duration ?? DEFAULT_SWIPE_DURATION_SECONDS
+  };
+}
 function createSwipeTool(udid = "booted") {
   return _x(
     "swipe",
-    "Swipe on the screen from one point to another.",
+    "Swipe on the screen from one point to another. `duration` (seconds) controls gesture velocity: the default 0.1s produces a flick that scrolls lists, dismisses sheets and triggers paging; pass a larger value (e.g. 0.5+) for a slow drag suitable for reordering or panning. Omitting `duration` uses the flick default.",
     SWIPE_SCHEMA,
-    async ({
-      x_start,
-      y_start,
-      x_end,
-      y_end
-    }) => handleSwipe({ udid, xStart: x_start, yStart: y_start, xEnd: x_end, yEnd: y_end })
+    async (params) => handleSwipe(toSwipeInput(udid, params))
   );
 }
 function createTypeTextTool(udid = "booted") {
@@ -52539,12 +52546,6 @@ function createPressButtonTool(udid = "booted") {
     async ({ button }) => handlePressButton(udid, button)
   );
 }
-var tapTool = createTapTool();
-var doubleTapTool = createDoubleTapTool();
-var longPressTool = createLongPressTool();
-var swipeTool = createSwipeTool();
-var typeTextTool = createTypeTextTool();
-var pressButtonTool = createPressButtonTool();
 var cache2 = /* @__PURE__ */ new Map();
 function parseDeviceInfo(raw) {
   const parsed = JSON.parse(raw);
@@ -52620,7 +52621,6 @@ function createScreenshotTool(udid = "booted") {
     }
   );
 }
-var screenshotTool = createScreenshotTool();
 var MIN_WAIT_SECONDS = 1;
 var MAX_WAIT_SECONDS = 10;
 var MS_PER_SECOND2 = 1e3;
@@ -62198,121 +62198,24 @@ var import_neverthrow33 = __toESM(require_index_cjs(), 1);
 var import_promises16 = require("node:timers/promises");
 
 // ../../agents/explorer/dist/index.js
-var import_node_child_process3 = require("node:child_process");
-var import_node_fs = require("node:fs");
-var import_node_path4 = __toESM(require("node:path"), 1);
-var import_neverthrow15 = __toESM(require_index_cjs(), 1);
 var import_promises9 = require("node:timers/promises");
+var import_neverthrow15 = __toESM(require_index_cjs(), 1);
 var import_neverthrow16 = __toESM(require_index_cjs(), 1);
-var import_neverthrow17 = __toESM(require_index_cjs(), 1);
 var import_promises10 = require("node:fs/promises");
-var import_node_path5 = __toESM(require("node:path"), 1);
-var import_neverthrow18 = __toESM(require_index_cjs(), 1);
-var import_node_child_process4 = require("node:child_process");
+var import_node_path4 = __toESM(require("node:path"), 1);
+var import_neverthrow17 = __toESM(require_index_cjs(), 1);
+var import_node_child_process3 = require("node:child_process");
 var import_promises11 = require("node:fs/promises");
+var import_neverthrow18 = __toESM(require_index_cjs(), 1);
 var import_neverthrow19 = __toESM(require_index_cjs(), 1);
+var import_node_child_process4 = require("node:child_process");
+var import_node_fs = require("node:fs");
+var import_node_path5 = __toESM(require("node:path"), 1);
 var import_neverthrow20 = __toESM(require_index_cjs(), 1);
 var import_neverthrow21 = __toESM(require_index_cjs(), 1);
 var import_promises12 = require("node:fs/promises");
 var import_node_path6 = __toESM(require("node:path"), 1);
 var import_neverthrow22 = __toESM(require_index_cjs(), 1);
-async function runFfmpeg(arguments_) {
-  const { promise: promise2, resolve, reject } = Promise.withResolvers();
-  (0, import_node_child_process3.execFile)("ffmpeg", arguments_, (error48) => {
-    if (error48) {
-      reject(error48);
-    } else {
-      resolve(true);
-    }
-  });
-  await promise2;
-}
-function spawnRecorder(outputPath) {
-  const safeMkdirSync = (0, import_neverthrow15.fromThrowable)(import_node_fs.mkdirSync, (cause) => cause);
-  const safeSpawn2 = (0, import_neverthrow15.fromThrowable)(
-    (command, arguments_) => (0, import_node_child_process3.spawn)(command, arguments_),
-    (cause) => cause
-  );
-  const mkdirResult = safeMkdirSync(import_node_path4.default.dirname(outputPath), { recursive: true });
-  if (mkdirResult.isErr()) {
-    return mkdirResult.error;
-  }
-  const spawnResult = safeSpawn2("xcrun", [
-    "simctl",
-    "io",
-    "booted",
-    "recordVideo",
-    "--force",
-    outputPath
-  ]);
-  if (spawnResult.isErr()) {
-    return spawnResult.error;
-  }
-  return spawnResult.value;
-}
-function earlyExitError(code) {
-  return new RangeError(`simctl exited with code ${String(code)} before recording started`);
-}
-function waitForRecordingStart(proc) {
-  const { stderr } = proc;
-  if (!stderr) {
-    return (0, import_neverthrow15.errAsync)({ type: "SPAWN_FAILED", cause: new RangeError("proc has no stderr pipe") });
-  }
-  const { promise: promise2, resolve, reject } = Promise.withResolvers();
-  proc.once("error", reject);
-  proc.once("exit", (code) => {
-    if (code !== 0) {
-      reject(earlyExitError(code));
-    }
-  });
-  let accumulated = "";
-  stderr.on("data", (chunk) => {
-    accumulated += chunk.toString();
-    if (accumulated.includes("Recording started")) {
-      resolve({ process: proc, startedAt: Date.now() });
-    }
-  });
-  return import_neverthrow15.ResultAsync.fromPromise(
-    promise2,
-    (cause) => ({ type: "SPAWN_FAILED", cause })
-  );
-}
-function startRecording(outputPath) {
-  const procOrError = spawnRecorder(outputPath);
-  if (procOrError instanceof Error) {
-    return (0, import_neverthrow15.errAsync)({ type: "SPAWN_FAILED", cause: procOrError });
-  }
-  return waitForRecordingStart(procOrError);
-}
-function stopRecording(handle) {
-  const { promise: promise2, resolve, reject } = Promise.withResolvers();
-  handle.process.once("exit", () => {
-    resolve(true);
-  });
-  handle.process.once("error", reject);
-  if (handle.process.exitCode == void 0) {
-    handle.process.kill("SIGINT");
-  } else {
-    resolve(true);
-  }
-  return import_neverthrow15.ResultAsync.fromPromise(
-    promise2,
-    (cause) => ({ type: "STOP_FAILED", cause })
-  );
-}
-function speedUpVideo({
-  inputPath,
-  factor,
-  outputPath
-}) {
-  const pts = String(1 / factor);
-  return (0, import_neverthrow15.fromAsyncThrowable)(
-    runFfmpeg,
-    (cause) => ({ type: "PROCESS_FAILED", cause })
-  )(["-i", inputPath, "-filter:v", `setpts=${pts}*PTS`, "-an", "-y", outputPath]).map(
-    () => outputPath
-  );
-}
 var VIEW_UI_TOOL_NAME = "mcp__mobile-ios__view_ui";
 var VIEW_UI_DESCRIPTION = `Capture current screen state: accessibility tree (element labels, positions, attributes) and screenshot in one call. Use when you need to tap an element, assert element presence or labels, check attributes, or track screen identity via <screen_id>.
 
@@ -62329,8 +62232,8 @@ function deriveScreenLabel(tree, stepIndex) {
   return match[1].replaceAll(/([A-Z])/g, "-$1").toLowerCase().replaceAll(/[^a-z\d]+/g, "-").replaceAll(/^-|-$/g, "");
 }
 async function persistScreenshot(params) {
-  const screenshotPath = import_node_path5.default.join(params.screenshotsDirectory, `${params.screenLabel}.png`);
-  const safeWriteFile = (0, import_neverthrow18.fromAsyncThrowable)(
+  const screenshotPath = import_node_path4.default.join(params.screenshotsDirectory, `${params.screenLabel}.png`);
+  const safeWriteFile = (0, import_neverthrow17.fromAsyncThrowable)(
     import_promises10.writeFile,
     (cause) => ({ type: "WRITE_FAILED", cause })
   );
@@ -62395,11 +62298,11 @@ function buildSnapshotRecord(rawElements, { stepIndex, context }) {
 }
 async function handleScreenshotData(data, params) {
   if (params.context.screenshotsDir !== void 0) {
-    const safeWrite2 = (0, import_neverthrow18.fromAsyncThrowable)(
+    const safeWrite2 = (0, import_neverthrow17.fromAsyncThrowable)(
       import_promises10.writeFile,
       (cause) => ({ type: "WRITE_FAILED", cause })
     );
-    const snapshotPath = import_node_path5.default.join(params.context.screenshotsDir, `${params.screenLabel}.txt`);
+    const snapshotPath = import_node_path4.default.join(params.context.screenshotsDir, `${params.screenLabel}.txt`);
     await safeWrite2(snapshotPath, params.formatted).match(
       () => true,
       () => false
@@ -62562,23 +62465,23 @@ function processMessage(message, state) {
   }
   if (message.type === "result") {
     if (message.subtype !== "success" && !state.timedOut.value && !state.aborted.value) {
-      return (0, import_neverthrow17.err)(message.errors.join("; "));
+      return (0, import_neverthrow16.err)(message.errors.join("; "));
     }
-    return (0, import_neverthrow17.ok)(true);
+    return (0, import_neverthrow16.ok)(true);
   }
-  return (0, import_neverthrow17.ok)(false);
+  return (0, import_neverthrow16.ok)(false);
 }
 async function processMessages(queryRunner, state) {
   for await (const message of queryRunner) {
     const result = processMessage(message, state);
     if (result.isErr()) {
-      return (0, import_neverthrow17.err)(result.error);
+      return (0, import_neverthrow16.err)(result.error);
     }
     if (result.value) {
       break;
     }
   }
-  return (0, import_neverthrow17.ok)(null);
+  return (0, import_neverthrow16.ok)(null);
 }
 var MessageQueue = class {
   pending = [];
@@ -62620,7 +62523,7 @@ var MessageQueue = class {
   }
 };
 function spawnDetached(options) {
-  const child = (0, import_node_child_process4.spawn)(options.command, options.args, {
+  const child = (0, import_node_child_process3.spawn)(options.command, options.args, {
     cwd: options.cwd,
     env: options.env,
     stdio: ["pipe", "pipe", "pipe"],
@@ -62688,7 +62591,6 @@ function buildAgentState({
     pendingToolCallNames: /* @__PURE__ */ new Map(),
     snapshots: [],
     stepCounter: { value: 0 },
-    lastEmittedHash: void 0,
     captureInput: outputTools.captureInput,
     sendMessage: buildSendMessageFunction(inputQueue),
     closeQueue: () => {
@@ -62790,7 +62692,7 @@ var INTERRUPT_DRAIN_TIMEOUT_MS = 1e4;
 async function interruptOrTimeout(queryRunner) {
   await Promise.race([queryRunner.interrupt(), (0, import_promises9.setTimeout)(INTERRUPT_DRAIN_TIMEOUT_MS)]);
 }
-var safeInterruptOrTimeout = import_neverthrow16.ResultAsync.fromThrowable(
+var safeInterruptOrTimeout = import_neverthrow15.ResultAsync.fromThrowable(
   interruptOrTimeout,
   (error48) => error48
 );
@@ -62841,18 +62743,18 @@ function startQueryTimers(config3, context) {
 }
 function awaitMessagesAndResolve({ queryRunner, state }, { cleanup: cleanup3, getOutput }) {
   const messagesPromise = processMessages(queryRunner, state);
-  return import_neverthrow16.ResultAsync.fromPromise(messagesPromise, String).andThen((innerResult) => {
+  return import_neverthrow15.ResultAsync.fromPromise(messagesPromise, String).andThen((innerResult) => {
     cleanup3();
     if (innerResult.isErr()) {
-      return (0, import_neverthrow16.err)(innerResult.error);
+      return (0, import_neverthrow15.err)(innerResult.error);
     }
-    return (0, import_neverthrow16.ok)(getOutput());
+    return (0, import_neverthrow15.ok)(getOutput());
   }).orElse((sdkError) => {
     cleanup3();
     if (!state.timedOut.value && !state.aborted.value) {
-      return (0, import_neverthrow16.err)(sdkError);
+      return (0, import_neverthrow15.err)(sdkError);
     }
-    return (0, import_neverthrow16.ok)(getOutput());
+    return (0, import_neverthrow15.ok)(getOutput());
   });
 }
 function executeQuery({
@@ -62866,9 +62768,9 @@ function executeQuery({
     message: { role: "user", content: prompt },
     parent_tool_use_id: null
   });
-  const queryRunnerResult = (0, import_neverthrow16.fromThrowable)(Qs, String)({ prompt: inputQueue, options });
+  const queryRunnerResult = (0, import_neverthrow15.fromThrowable)(Qs, String)({ prompt: inputQueue, options });
   if (queryRunnerResult.isErr()) {
-    return (0, import_neverthrow16.errAsync)(queryRunnerResult.error);
+    return (0, import_neverthrow15.errAsync)(queryRunnerResult.error);
   }
   const queryRunner = queryRunnerResult.value;
   const cleanup3 = startQueryTimers(config3, { state, queryRunner, inputQueue, linkedController });
@@ -62888,6 +62790,103 @@ function runQuery(prompt, config3) {
 function collectAgentOutput(prompt, config3) {
   return runQuery(prompt, config3).mapErr((cause) => ({ type: "QUERY_FAILED", cause }));
 }
+async function runFfmpeg(arguments_) {
+  const { promise: promise2, resolve, reject } = Promise.withResolvers();
+  (0, import_node_child_process4.execFile)("ffmpeg", arguments_, (error48) => {
+    if (error48) {
+      reject(error48);
+    } else {
+      resolve(true);
+    }
+  });
+  await promise2;
+}
+function spawnRecorder(outputPath) {
+  const safeMkdirSync = (0, import_neverthrow20.fromThrowable)(import_node_fs.mkdirSync, (cause) => cause);
+  const safeSpawn2 = (0, import_neverthrow20.fromThrowable)(
+    (command, arguments_) => (0, import_node_child_process4.spawn)(command, arguments_),
+    (cause) => cause
+  );
+  const mkdirResult = safeMkdirSync(import_node_path5.default.dirname(outputPath), { recursive: true });
+  if (mkdirResult.isErr()) {
+    return mkdirResult.error;
+  }
+  const spawnResult = safeSpawn2("xcrun", [
+    "simctl",
+    "io",
+    "booted",
+    "recordVideo",
+    "--force",
+    outputPath
+  ]);
+  if (spawnResult.isErr()) {
+    return spawnResult.error;
+  }
+  return spawnResult.value;
+}
+function earlyExitError(code) {
+  return new RangeError(`simctl exited with code ${String(code)} before recording started`);
+}
+function waitForRecordingStart(proc) {
+  const { stderr } = proc;
+  if (!stderr) {
+    return (0, import_neverthrow20.errAsync)({ type: "SPAWN_FAILED", cause: new RangeError("proc has no stderr pipe") });
+  }
+  const { promise: promise2, resolve, reject } = Promise.withResolvers();
+  proc.once("error", reject);
+  proc.once("exit", (code) => {
+    if (code !== 0) {
+      reject(earlyExitError(code));
+    }
+  });
+  let accumulated = "";
+  stderr.on("data", (chunk) => {
+    accumulated += chunk.toString();
+    if (accumulated.includes("Recording started")) {
+      resolve({ process: proc, startedAt: Date.now() });
+    }
+  });
+  return import_neverthrow20.ResultAsync.fromPromise(
+    promise2,
+    (cause) => ({ type: "SPAWN_FAILED", cause })
+  );
+}
+function startRecording(outputPath) {
+  const procOrError = spawnRecorder(outputPath);
+  if (procOrError instanceof Error) {
+    return (0, import_neverthrow20.errAsync)({ type: "SPAWN_FAILED", cause: procOrError });
+  }
+  return waitForRecordingStart(procOrError);
+}
+function stopRecording(handle) {
+  const { promise: promise2, resolve, reject } = Promise.withResolvers();
+  handle.process.once("exit", () => {
+    resolve(true);
+  });
+  handle.process.once("error", reject);
+  if (handle.process.exitCode == void 0) {
+    handle.process.kill("SIGINT");
+  } else {
+    resolve(true);
+  }
+  return import_neverthrow20.ResultAsync.fromPromise(
+    promise2,
+    (cause) => ({ type: "STOP_FAILED", cause })
+  );
+}
+function speedUpVideo({
+  inputPath,
+  factor,
+  outputPath
+}) {
+  const pts = String(1 / factor);
+  return (0, import_neverthrow20.fromAsyncThrowable)(
+    runFfmpeg,
+    (cause) => ({ type: "PROCESS_FAILED", cause })
+  )(["-i", inputPath, "-filter:v", `setpts=${pts}*PTS`, "-an", "-y", outputPath]).map(
+    () => outputPath
+  );
+}
 var SPEED_2X = 2;
 var SPEED_4X = 4;
 function applySpeedUpVariants({
@@ -62898,7 +62897,7 @@ function applySpeedUpVariants({
   const { findings, snapshots } = result;
   const { videoPath, videoPath2x, videoPath4x, signal } = params;
   if (signal?.aborted) {
-    return (0, import_neverthrow20.okAsync)({ findings, snapshots });
+    return (0, import_neverthrow19.okAsync)({ findings, snapshots });
   }
   return speedUpVideo({ inputPath: videoPath, factor: SPEED_2X, outputPath: videoPath2x }).mapErr(toRecordingError).andThen(
     () => speedUpVideo({ inputPath: videoPath, factor: SPEED_4X, outputPath: videoPath4x }).mapErr(toRecordingError).map(() => ({ findings, snapshots }))
@@ -62912,7 +62911,7 @@ function runWithRecording(handle, collectOutput) {
   return collectOutput().andThen(
     ({ findings, snapshots }) => stopRecording(handle).mapErr(toRecordingError).map(() => ({ findings, snapshots }))
   ).orElse(
-    (error48) => stopRecording(handle).mapErr(toRecordingError).andThen(() => (0, import_neverthrow20.errAsync)(error48)).orElse(() => (0, import_neverthrow20.errAsync)(error48))
+    (error48) => stopRecording(handle).mapErr(toRecordingError).andThen(() => (0, import_neverthrow19.errAsync)(error48)).orElse(() => (0, import_neverthrow19.errAsync)(error48))
   );
 }
 function startAndRun(params) {
@@ -62953,12 +62952,18 @@ At every reasoning step, maintain a mental ledger:
 
 Consult the ledger before every action. Always prefer navigating to a QUEUE screen over a VISITED one.`;
 var SESSION_START_RULE = `Before taking any other action \u2014 including initializing the Working State ledger or emitting findings \u2014 call \`view_ui\` once to observe the starting screen`;
-var BACK_NAV_RULE = `After navigating forward to any new screen: tap back, then confirm you returned to the expected parent in PATH \u2014 use \`screenshot\` if the parent screen is visually unambiguous, \`view_ui\` if you need element labels or coordinates to verify or to act next \u2014 if confirmation fails, emit a \`back-nav-failure\` finding, then navigate forward again to continue`;
-var STUCK_LOOP_RULE = `Stuck loop: emit a \`stuck-loop\` finding when any of these occur: (1) \`view_ui\` returns the same \`<screen_id>\` across 3 or more consecutive \`view_ui\` calls, (2) the same element has been tapped more than twice with no screen change confirmed by \`view_ui\`, (3) PATH shows the same screen at two non-adjacent positions \u2014 before emitting, try one alternative action (scroll, long-press, swipe) to rule out a gesture mismatch \u2014 note: \`screenshot\`-only calls do not update the stuck-loop counter; only \`view_ui\` calls count`;
+var BACK_NAV_RULE = `After navigating forward to any new screen: attempt to return to the expected parent in PATH \u2014 consult App Knowledge first for the correct exit gesture on this screen, then try in order: (1) any visible back/close button, (2) OS back gesture, (3) swipe up, (4) swipe down, (5) swipe left, (6) swipe right \u2014 confirm return via \`screenshot\` if the parent is visually unambiguous, \`view_ui\` otherwise \u2014 only after ALL attempts fail emit a \`back-nav-failure\` finding, then navigate forward again to continue`;
+var STUCK_LOOP_RULE = `Stuck loop: emit a \`stuck-loop\` finding when any of these signals occur: (1) \`view_ui\` returns the same \`<screen_id>\` across 3 or more consecutive \`view_ui\` calls, (2) the same element has been tapped more than twice with no screen change confirmed by \`view_ui\`, (3) PATH shows the same screen at two non-adjacent positions \u2014 i.e. you visited a screen, left it, and returned. Example: \`Portfolio \u2192 Asset \u2192 Swap \u2192 Asset\` (Asset revisited). Indicates circular navigation. NOT tripped by adjacent repeats like \`Portfolio \u2192 Asset \u2192 Asset\` (normal state refresh), (4) a scrollable list's element positions \u2014 x for horizontal, y for vertical \u2014 are identical across 2 or more consecutive swipe+\`view_ui\` cycles (zero-delta scroll stall) \u2014 before emitting for (1)\u2013(3), try one alternative action (scroll, long-press, swipe) to rule out a gesture mismatch \u2014 before emitting for (4), baseline element positions via \`view_ui\`, then vary one spatial parameter on the next attempt: shift swipe start coord by \u226540% of visible container dimension, OR reverse direction to probe opposite bound, OR extend throw to span \u226580% of viewport dimension \u2014 then \`view_ui\` and compare. Example: if tab bar positions \`Tokens(-31) ETH(65) ... Tron(352)\` are unchanged across 2 swipes starting near x=350, shift start to x=10 or x=395 and extend end past the opposite viewport edge. If delta still zero, emit \`stuck-loop\` with axis, before/after positions, and variation attempted \u2014 note: \`screenshot\`-only calls do not update the stuck-loop counter; only \`view_ui\` calls count. Zero-delta scroll stall is not a separate finding type \u2014 report as \`stuck-loop\`.`;
 var LOADING_STATE_RULE = `Transient loading state: when the screen shows spinners, skeleton screens, progress bars, "Loading..." text, or placeholder content NOT described in spec or app context \u2014 use \`screenshot\` to poll for resolution (up to 3 retries); switch to \`view_ui\` only on the final check or when you need element data to act \u2014 if loading persists after 3 retries, proceed with what is visible; if spec or app context explicitly describes a loading screen as a step, do not retry \u2014 call \`view_ui\` and assert normally`;
 var EXPECTED_CONTENT_MISSING_RULE = `Expected content missing: when \`view_ui\` shows no loading indicator yet omits an element named or strongly implied by spec or app context \u2014 and its absence is not semantically consistent with the current screen \u2014 call \`wait_seconds\` with 2\u20135 seconds and retry \`view_ui\` up to 2 times; if element remains absent, emit a \`missing-content\` finding stating what was expected and what was observed`;
 var CLIPPED_ELEMENT_RULE = `Never tap an element tagged \`[clipped-top]\`, \`[clipped-bottom]\`, \`[clipped-left]\`, or \`[clipped-right]\` \u2014 scroll to fully reveal it first, then re-call \`view_ui\` before tapping`;
-var A11Y_FALLBACK_RULE = `Missing a11y element: if you intend to tap or interact with a UI element and that element is absent from the most recent \`view_ui\` a11y tree \u2014 do NOT estimate its coordinates from the screenshot or attempt any pixel-based tap \u2014 instead, emit a \`missing-a11y-element\` finding that states: (1) your intent (what you were trying to do), (2) the approximate visual region where the element appeared (coords/size from the screenshot), (3) nearby labeled elements from the a11y tree that serve as landmarks \u2014 then continue: in freestyle mode keep exploring other reachable screens; in spec mode advance to the next step`;
+var SCROLL_FOLD_RULE = `Scrollable lists: elements outside the visible viewport are absent from the a11y tree by design \u2014 this applies to elements below the fold in vertical lists AND elements clipped off-left or off-right in horizontal lists \u2014 scroll or swipe in the appropriate axis to reveal before asserting presence or absence; never emit a finding solely because list items, rows, or tabs are missing from the tree on a scrollable screen; if swipe attempts yield no position change across 2+ cycles, apply the scroll-stall path in STUCK_LOOP_RULE.`;
+var FINDING_TAXONOMY_SECTION = `## Finding Types
+
+You may emit only these trigger types: \`back-nav-failure\`, \`dead-end\`, \`stuck-modal\`, \`stuck-loop\`, \`missing-a11y-element\`, \`missing-content\`, \`spec-deviation\`, \`destructive-only-exit\`. Do NOT emit \`design-system-violation\`, \`motion-regression\`, \`continuity-regression\`, \`interaction-regression\`, or \`loading-regression\` \u2014 those belong to other agents.`;
+var A11Y_FALLBACK_RULE = `Missing a11y element: if you intend to tap or interact with a UI element and that element is absent from the most recent \`view_ui\` a11y tree \u2014 visible in the screenshot does NOT imply interactable; the a11y tree is authoritative \u2014 do NOT estimate its coordinates from the screenshot, do NOT attempt any pixel-based tap, do NOT retry at different coordinates, do NOT long-press or swipe in the element's visual region as a fallback; a failed pixel tap is never an \`interaction-regression\` \u2014 it is a \`missing-a11y-element\` \u2014 instead, emit a \`missing-a11y-element\` finding that states: (1) your intent (what you were trying to do), (2) the approximate visual region where the element appeared (coords/size from the screenshot), (3) nearby labeled elements from the a11y tree that serve as landmarks \u2014 then continue: in freestyle mode keep exploring other reachable screens; in spec mode advance to the next step`;
+var OUTCOME_LITERAL_RULE = `When verifying a step outcome or assertion, interpret all quantifiers literally and apply them exhaustively: any keyword that imposes a universal or count-bound constraint \u2014 including but not limited to \`only\`, \`all\`, \`every\`, \`each\`, \`both\`, \`no\`, \`none\`, \`neither\`, \`exactly N\`, \`at least N\`, \`fewer than N\`, \`more than N\` \u2014 a single counter-example observed in \`view_ui\` or \`screenshot\` constitutes a failed constraint; this rule applies only when the outcome text contains a universal or count-bound quantifier; positive-existence outcomes (\`X appears\`, \`Y becomes active\`, \`screen is visible\`) are not subject to counter-example scanning and are governed by SPEC_STEP_READING_SECTION; on a scrollable list, scroll through the full list before concluding pass or fail on a universal quantifier \u2014 apply SCROLL_FOLD_RULE to reveal all items, then evaluate across the complete visible set; if the screen shows any loading indicator at the time of observation, apply LOADING_STATE_RULE first and re-verify after resolution before emitting; do NOT soften (\`mostly\`, \`essentially\`, \`largely\`) and do NOT narrow the target class post-hoc to exclude the counter-example; when the counter-evidence is an element absent from the a11y tree, A11Y_FALLBACK_RULE takes precedence and determines the finding type (\`missing-a11y-element\`) \u2014 OUTCOME_LITERAL_RULE governs unwanted-presence violations only; if one item violates the constraint, emit \`spec-deviation\` immediately with: (a) the literal outcome text, (b) the specific element(s) that violate it, (c) the constraint keyword that is broken.`;
+var ANTI_RATIONALIZATION_RULE = `During outcome verification, monitor your own reasoning for reconciliation hypotheses: if you generate any reasoning that re-frames, redefines, or reinterprets the observed counter-example or target class in order to produce agreement with the spec outcome \u2014 regardless of phrasing \u2014 treat that reasoning as a deviation signal, not a resolution; stop, do NOT mark the step complete, and emit \`spec-deviation\` with: (a) the literal outcome text, (b) the specific observation that triggered the hypothesis, (c) the reconciliation reasoning itself verbatim; before marking any quantifier-bearing outcome complete, state explicitly in your reasoning: \`No reconciliation hypothesis generated. Counter-examples found: [list or none].\` If you cannot make that statement honestly, a hypothesis exists \u2014 emit \`spec-deviation\`; when outcome verification is ambiguous and no reconciliation hypothesis is generated, still default to emitting \`spec-deviation\`; silence is not a pass.`;
 var WHAT_TO_TEST_SECTION = `## What to Test
 
 Test navigation elements first, interactions second.
@@ -62989,9 +62994,9 @@ Rules:
 - When in doubt about reversibility, treat the action as destructive and back out.`;
 var DEAD_END_SECTION = `## Dead End and Modal Detection
 
-**Dead end** \u2014 when \`view_ui\` shows no interactive exit affordance, first consult App Knowledge for gesture-based navigation on this screen, then attempt ALL of before emitting a finding: (1) any visible back/close button, (2) swipe from the left edge (back gesture), (3) swipe down (dismiss gesture). If all fail, emit a \`dead-end\` finding describing what was visible and what was attempted.
+**Dead end** \u2014 emitted ONLY when every attempted exit fails to leave the screen. Consult App Knowledge first for the exit gesture on this screen, then attempt ALL before emitting: (1) any visible back/close button, (2) OS back gesture, (3) swipe up, (4) swipe down, (5) swipe left, (6) swipe right. If any succeeds, do NOT emit a finding \u2014 absence of an explicit "Cancel" or "Close" button is not a dead-end when a gesture dismisses the screen. If all fail, emit a \`dead-end\` finding describing what was visible and what was attempted.
 
-**Stuck modal** \u2014 when a modal or bottom sheet blocks the screen, attempt dismissal in order: (1) close/X button if present, (2) tap outside the modal, (3) swipe down, (4) swipe from the left edge. If all fail, emit a \`stuck-modal\` finding listing the modal, the screen it appeared on, and the methods attempted.`;
+**Stuck modal** \u2014 when a modal or bottom sheet blocks the screen, attempt dismissal in order: (1) close/X button if present, (2) tap outside the modal, (3) swipe down, (4) swipe up, (5) swipe left, (6) swipe right. If any succeeds, do NOT emit a finding. If all fail, emit a \`stuck-modal\` finding listing the modal, the screen it appeared on, and the methods attempted.`;
 var SPEC_WHAT_TO_TEST_SECTION = `## What to Test
 
 Test only the elements and interactions described in the spec. Do not interact with elements outside the spec path.
@@ -62999,9 +63004,9 @@ Test only the elements and interactions described in the spec. Do not interact w
 If you observe obvious breakage while navigating to a spec step \u2014 a broken control, unexpected error, missing screen, or crash \u2014 flag it as a passive observation without stopping to investigate it.`;
 var SPEC_DEAD_END_SECTION = `## Dead End and Modal Detection
 
-**Dead end** \u2014 if a spec step leaves the agent on a screen with no path to the next spec step, attempt: (1) any visible back/close button, (2) swipe from the left edge, (3) swipe down. If all fail, emit a \`dead-end\` finding and halt \u2014 do not attempt further exploration to recover.
+**Dead end** \u2014 if a spec step leaves the agent on a screen with no path to the next spec step, consult App Knowledge first for the exit gesture, then attempt ALL: (1) any visible back/close button, (2) OS back gesture, (3) swipe up, (4) swipe down, (5) swipe left, (6) swipe right. If all fail, emit a \`dead-end\` finding and halt \u2014 do not attempt further exploration to recover.
 
-**Stuck modal** \u2014 when a modal or bottom sheet blocks spec step execution, attempt dismissal in order: (1) close/X button if present, (2) tap outside the modal, (3) swipe down, (4) swipe from the left edge. If all fail, emit a \`stuck-modal\` finding listing the modal, the screen it appeared on, and the methods attempted.`;
+**Stuck modal** \u2014 when a modal or bottom sheet blocks spec step execution, attempt dismissal in order: (1) close/X button if present, (2) tap outside the modal, (3) swipe down, (4) swipe up, (5) swipe left, (6) swipe right. If all fail, emit a \`stuck-modal\` finding listing the modal, the screen it appeared on, and the methods attempted.`;
 var SPEC_STEP_READING_SECTION = `## Reading Spec Steps
 
 Each step has this shape:
@@ -63049,7 +63054,10 @@ var SPEC_RULES_SECTION = `## Rules
 - ${LOADING_STATE_RULE}
 - ${EXPECTED_CONTENT_MISSING_RULE}
 - ${CLIPPED_ELEMENT_RULE}
+- ${SCROLL_FOLD_RULE}
 - ${A11Y_FALLBACK_RULE}
+- ${OUTCOME_LITERAL_RULE}
+- ${ANTI_RATIONALIZATION_RULE}
 - Each item in \`**Assertions**\` is a mandatory pass/fail check \u2014 verify using \`view_ui\` when the assertion targets an element attribute, label, or presence in the tree; use \`screenshot\` when the assertion is purely visual; if neither can confirm, emit a \`spec-deviation\` finding based on what is observable
 - Flag crash dialogs, unexpected system errors, or navigation failures that occur as a direct result of executing a spec step; if you observe a visibly broken element in passing while navigating, note it without interacting with it`;
 function buildSpecModeBody({
@@ -63079,6 +63087,8 @@ ${SPEC_STEP_READING_SECTION}
 
 ${SPEC_OPTIONAL_STEPS_SECTION}
 
+${FINDING_TAXONOMY_SECTION}
+
 ## Specs
 
 ${specContent}${environmentSection}
@@ -63112,6 +63122,7 @@ ${TOOL_SELECTION_SECTION}
 - ${LOADING_STATE_RULE}
 - ${EXPECTED_CONTENT_MISSING_RULE}
 - ${CLIPPED_ELEMENT_RULE}
+- ${SCROLL_FOLD_RULE}
 - ${A11Y_FALLBACK_RULE}
 
 ## Exploration Strategy
@@ -63122,7 +63133,9 @@ ${WHAT_TO_TEST_SECTION}
 
 ${DESTRUCTIVE_ACTIONS_SECTION}
 
-${DEAD_END_SECTION}${environmentSection}
+${DEAD_END_SECTION}
+
+${FINDING_TAXONOMY_SECTION}${environmentSection}
 
 ## Output
 
@@ -63380,7 +63393,7 @@ function buildPrompt(safeConfig, specs) {
   });
 }
 function parseSpecs(resolvedSpecs) {
-  return import_neverthrow19.Result.combine(
+  return import_neverthrow18.Result.combine(
     resolvedSpecs.map(
       (spec) => parseTestSpec(spec.name, spec.content).mapErr(
         (cause) => ({ type: "SPEC_PARSE_FAILED", specName: spec.name, cause })
@@ -63439,7 +63452,7 @@ function collectAndFinalize({
 }
 function resolveAndParseSpecs(safeConfig) {
   if (safeConfig.mode === "freestyle") {
-    return (0, import_neverthrow19.okAsync)([]);
+    return (0, import_neverthrow18.okAsync)([]);
   }
   return resolveSpecs(safeConfig).mapErr((cause) => ({ type: "SPEC_RESOLVE_FAILED", cause })).andThen((specs) => parseSpecs(specs));
 }
@@ -63462,11 +63475,11 @@ function runExplorer(config3) {
     date: date5
   });
   if (runPathsResult.isErr()) {
-    return (0, import_neverthrow19.errAsync)({ type: "RUN_PATHS_FAILED", cause: runPathsResult.error });
+    return (0, import_neverthrow18.errAsync)({ type: "RUN_PATHS_FAILED", cause: runPathsResult.error });
   }
   const runPaths = runPathsResult.value;
   if (config3.signal?.aborted) {
-    return (0, import_neverthrow19.okAsync)(toArtifacts({ findings: [], snapshots: [] }, runPaths));
+    return (0, import_neverthrow18.okAsync)(toArtifacts({ findings: [], snapshots: [] }, runPaths));
   }
   const safeConfig = {
     ...config3,
@@ -63474,7 +63487,7 @@ function runExplorer(config3) {
     onEvent: config3.onEvent === void 0 ? void 0 : safeHandler(config3.onEvent)
   };
   safeConfig.onEvent?.({ type: "STAGE_START", agent: "explorer" });
-  return import_neverthrow19.ResultAsync.fromSafePromise(
+  return import_neverthrow18.ResultAsync.fromSafePromise(
     (0, import_promises11.mkdir)(runPaths.screenshotsDir, { recursive: true }).catch(() => null)
   ).andThen(() => runPipeline({ safeConfig, runPaths, start }));
 }
@@ -66969,6 +66982,13 @@ var FsDesignStore = class {
 // ../../packages/pipeline/dist/index.js
 var RETRY_MAX_ATTEMPTS = 3;
 var RETRY_BASE_DELAY_MS = 1e3;
+var CONFIDENCE_THRESHOLD = 0.7;
+function filterByConfidence(findings, threshold) {
+  return {
+    kept: findings.filter((finding) => finding.confidence >= threshold),
+    dropped: findings.filter((finding) => finding.confidence < threshold).map((finding) => ({ finding, reason: "low-confidence" }))
+  };
+}
 function attemptRetry(options) {
   const { factory, config: config3, delayFunction, onRetry, attempt } = options;
   return factory().orElse((error48) => {
@@ -67284,11 +67304,12 @@ function buildExplorerConfig({
   };
 }
 function buildOutput(consolidationResult, options) {
-  const { runId, runPaths, signal } = options;
+  const { runId, runPaths, signal, confidenceThreshold } = options;
+  const filtered = filterByConfidence(consolidationResult.findings, confidenceThreshold);
   const output = {
     runId,
-    findings: consolidationResult.findings,
-    dismissed: consolidationResult.dismissed,
+    findings: filtered.kept,
+    dismissed: [...consolidationResult.dismissed, ...filtered.dropped],
     findingsPath: runPaths.findingsPath,
     aborted: signal?.aborted ? true : void 0
   };
@@ -67342,7 +67363,8 @@ function executePipeline(setup, config3) {
     (consolidationResult) => buildOutput(consolidationResult, {
       runId: setup.runId,
       runPaths: setup.runPaths,
-      signal
+      signal,
+      confidenceThreshold: config3.confidenceThreshold ?? CONFIDENCE_THRESHOLD
     })
   );
 }
@@ -76144,6 +76166,9 @@ var DEFAULT_FREESTYLE_TIMEOUT_SECONDS2 = 300;
 function buildDeviceInstruction(simulatorUdid) {
   return `You MUST use device "${simulatorUdid}" for ALL mobile tool calls. The simulator UDID is already configured \u2014 use it directly.`;
 }
+function composeAppContext(parts) {
+  return parts.filter((part) => part !== void 0 && part.length > 0).join("\n\n");
+}
 function ensureWorkerCwd(simulatorUdid) {
   const workerDirectory = import_node_path21.default.join(import_node_os4.default.tmpdir(), "xqa-workers", simulatorUdid);
   (0, import_node_fs11.mkdirSync)(workerDirectory, { recursive: true });
@@ -76157,7 +76182,7 @@ function resolveFreestyleTimeout(item, config3) {
 }
 function buildFreestylePipelineConfig(input) {
   const { item, context, signal, simulatorUdid, onEvent } = input;
-  const { config: config3, xqaDirectory, runId, date: date5 } = context;
+  const { config: config3, xqaDirectory, runId, date: date5, appContext } = context;
   return {
     outputDir: import_node_path21.default.join(xqaDirectory, "output", item.id),
     runId,
@@ -76170,9 +76195,11 @@ function buildFreestylePipelineConfig(input) {
       mcpServers: createDefaultMcpServers(simulatorUdid),
       allowedTools: ALLOWED_TOOLS,
       timeoutMs: resolveFreestyleTimeout(item, config3) * MS_PER_SECOND4,
-      appContext: item.prompt ? `${buildDeviceInstruction(simulatorUdid)}
-
-${item.prompt}` : buildDeviceInstruction(simulatorUdid),
+      appContext: composeAppContext([
+        buildDeviceInstruction(simulatorUdid),
+        appContext,
+        item.prompt
+      ]),
       buildEnv: config3.QA_BUILD_ENV,
       cwd: ensureWorkerCwd(simulatorUdid)
     }
@@ -76183,7 +76210,7 @@ function executeFreestyleItem(input) {
 }
 function buildSpecPipelineConfig(input) {
   const { item, context, signal, simulatorUdid, onEvent } = input;
-  const { config: config3, xqaDirectory, runId } = context;
+  const { config: config3, xqaDirectory, runId, appContext } = context;
   return {
     outputDir: import_node_path21.default.join(xqaDirectory, "output", item.id),
     runId,
@@ -76195,7 +76222,7 @@ function buildSpecPipelineConfig(input) {
       specFiles: [item.specPath],
       mcpServers: createDefaultMcpServers(simulatorUdid),
       allowedTools: ALLOWED_TOOLS,
-      appContext: buildDeviceInstruction(simulatorUdid),
+      appContext: composeAppContext([buildDeviceInstruction(simulatorUdid), appContext]),
       buildEnv: config3.QA_BUILD_ENV,
       cwd: ensureWorkerCwd(simulatorUdid)
     }
@@ -76271,17 +76298,30 @@ function deriveSuiteIdFromMode(mode) {
   }
   return deriveSuiteId({ mode: "spec", globs: mode.globs });
 }
+async function loadAppContext(xqaDirectory) {
+  const result = await readAppContext(xqaDirectory);
+  if (result.isErr()) {
+    const cause = result.error.cause instanceof Error ? result.error.cause.message : JSON.stringify(result.error.cause);
+    process.stderr.write(`Failed to read app context: ${result.error.type}
+${cause}
+`);
+    return void 0;
+  }
+  return result.value;
+}
 async function buildSuiteRunContext(input) {
   const suiteId = deriveSuiteIdFromMode(input.mode);
   const date5 = (/* @__PURE__ */ new Date()).toISOString().slice(0, ISO_DATE_LENGTH3);
   const outputDirectory = import_node_path22.default.join(input.xqaDirectory, "output");
   const existingDirectories = await listRunDirectories({ outputDirectory, suiteId, date: date5 });
   const runId = computeNextRunId(existingDirectories);
+  const appContext = await loadAppContext(input.xqaDirectory);
   const context = {
     config: input.config,
     xqaDirectory: input.xqaDirectory,
     runId,
-    date: date5
+    date: date5,
+    appContext
   };
   return { suiteId, date: date5, outputDirectory, runId, context };
 }
@@ -76475,7 +76515,7 @@ function resolveXqaDirectory() {
   return result.value;
 }
 var program2 = new Command();
-program2.name("xqa").description("AI-powered QA agent CLI").version(`${"1.13.0"}${false ? ` (dev build +${"5e72ba0"})` : ""}`);
+program2.name("xqa").description("AI-powered QA agent CLI").version(`${"1.14.1"}${false ? ` (dev build +${"18d343b"})` : ""}`);
 program2.command("init").description("Initialize a new xqa project in the current directory").action(() => {
   runInitCommand();
 });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@exodus/xqa",
-  "version": "1.13.0",
+  "version": "1.14.1",
   "type": "module",
   "engines": {
     "node": ">=22"
@@ -26,11 +26,11 @@
     "typescript": "^5.8.3",
     "vitest": "^3.2.1",
     "zod": "^3.0.0",
-    "@qa-agents/display": "0.0.0",
     "@qa-agents/explorer": "0.0.0",
-    "@qa-agents/eslint-config": "0.0.0",
     "@qa-agents/mobile-ios": "0.0.0",
     "@qa-agents/pipeline": "0.0.0",
+    "@qa-agents/display": "0.0.0",
+    "@qa-agents/eslint-config": "0.0.0",
     "@qa-agents/shared": "0.0.0",
     "@qa-agents/typescript-config": "0.0.0"
   },