@wa008/ui-audit-mcp 2.2.1 → 2.3.0

package/README.md

@@ -54,16 +54,15 @@ npm install
 | `launch_app` | Launch app by Bundle ID |
 | `take_screenshot` | Capture screen & track evaluation step |
 | `tap` / `swipe` | Interact with UI (ratio coordinates), screenshot, and track step |
-| `get_evaluation_criteria` | Get dimensions and scoring rubrics (Prompts) |
-| `submit_dimension_score` | Submit score (1-10) for a single dimension on a tracked step |
+| `evaluate` | Unified evaluation: get dimension prompt (initial) or submit score + get next (submit mode) |
 | `get_audit_status` | View missing evaluations dashboard or final markdown report |
 
 ## Typical Workflow
 1. `launch_app("com.example.app")`
 2. `take_screenshot("MyTestCase", 1, "Verify Initial Screen")` or `tap(0.5, 0.5, "MyTestCase", 2, "Click Login")`
-3. `get_evaluation_criteria("overlap")` → Read the prompt for overlap
-4. Agent analyzes UI → `submit_dimension_score("MyTestCase", 1, "overlap", 9, "No overlap")`
-5. Repeat 3-4 until all dimensions are evaluated for the step
+3. `evaluate("MyTestCase", 1)` → Returns the first dimension prompt + `evaluationToken`
+4. Agent analyzes UI → `evaluate("MyTestCase", 1, token, 9, "No overlap")` → Records score, returns next dimension prompt + new token
+5. Repeat step 4 until all 5 dimensions are evaluated
 6. `get_audit_status(["MyTestCase"])` → Output the full markdown report
 
 ## Data

package/dist/src/evaluation/checklist.js

@@ -3,21 +3,21 @@ export const DIMENSIONS = [
     {
         id: "overlap",
         name: "Element Overlap & Safe Areas",
-        description: "You are a strict QA inspector. Your job is to find overlap problems, not to praise the UI.\n" +
-            "Assume there ARE problems until you have carefully verified every region.\n\n" +
-            "CHECK EACH REGION METHODICALLY — report what you see in each:\n" +
-            "1. TOP-RIGHT & TOP-LEFT CORNERS (CRITICAL): Look at the black pill-shaped camera cutout (Dynamic Island) or the notch at the top center. Draw an imaginary horizontal line across the VERY BOTTOM EDGE of that cutout. Are there ANY interactive elements (like a '+' button, settings icon, profile picture) that are positioned at the same height or ABOVE this line on the left or right sides? If YES, this is a CRITICAL UI BUG. Score this 0-2 immediately.\n" +
-            "2. STATUS BAR ZONE: Does ANY app content extend behind the system time (top left), battery, or signal indicators (top right)? Look extremely closely. Even partial intrusion = failure.\n" +
-            "3. BOTTOM SAFE AREA: Does any content overlap with the horizontal home indicator bar at the very bottom?\n" +
-            "4. INTER-ELEMENT: Do any two app elements (buttons, text, images, cards) visually overlap each other?\n" +
-            "5. MODAL/POPUP: If present, does it unintentionally cover critical content?\n\n" +
-            "SCORING ATTITUDE: Visual AI models often miss top-corner overlaps. Be explicitly paranoid about buttons placed too high up near the battery or time icons. If you are unsure whether something overlaps or is too close, score LOW.\n" +
-            "NOTE: Do NOT evaluate spacing aesthetics (layout), colors (style), or text meaning (info_clarity).",
-        scoringGuide: "0-2: Any element overlaps the status bar zone (e.g. above the bottom of the Dynamic Island/notch), touches the battery/time, overlaps the home indicator, or covers another element.\n" +
-            "3-4: An element is dangerously close to a safe area boundary (e.g. just barely below the Dynamic Island with no breathing room).\n" +
-            "5-6: All elements are within safe areas but margins are uncomfortably tight.\n" +
-            "7-8: All elements are clearly within safe areas with comfortable margins. No inter-element overlap detected.\n" +
-            "9-10: Perfect — generous safe area margins on all sides, zero overlap risk. ONLY give 9-10 if you are ABSOLUTELY CERTAIN the top corners are completely clear of buttons.",
+        description: "You are a user-centric QA inspector evaluating visual interference and interactive crowding.\n" +
+            "Evaluate the screen from the perspective of a real human user trying to read and tap efficiently.\n" +
+            "Assume there ARE overlap issues until you have visually scanned the entire screen.\n\n" +
+            "CHECK FOR VISUAL INTERFERENCE & COGNITIVE NOISE:\n" +
+            "1. SYSTEM EDGE INTERFERENCE (CRITICAL): Scan the very top, bottom, and corners. Do any app elements (e.g., buttons, avatars, titles) visually mix with system indicators (time, battery, notch, home indicator)? If an app button sits awkwardly close to system icons, it creates cognitive confusion and accidental click risks. This is a CRITICAL BUG.\n" +
+            "2. INTERACTIVE CROWDING: Are touch targets (buttons, links) so close to each other or to the screen edges that a user with a large thumb might accidentally trigger the wrong action or a system gesture?\n" +
+            "3. TEXT & CONTENT CLASH: Do any app elements (text, images, floating buttons) visually collide with or bleed into each other, making them harder to read or recognize?\n" +
+            "4. MODAL/POPUP OBSTRUCTION: If a popup or sheet is present, does it awkwardly obscure essential context that the user still needs to see?\n\n" +
+            "SCORING ATTITUDE: Focus on the HUMAN visual experience. If an element 'feels' like it's stepping on another element's toes or invading system space, it is an overlap failure. Punish visual crowding heavily. If unsure, score LOW.\n" +
+            "NOTE: Do NOT evaluate logical spacing consistency (layout), colors (style), or text meaning (info_clarity).",
+        scoringGuide: "0-2: Severe visual interference. Elements clearly collide (e.g. app button invading system status bar area), causing immediate cognitive confusion or severe accidental click risks.\n" +
+            "3-4: Elements don't perfectly collide, but are awkwardly close to each other or to system edges, creating noticeable visual noise or crowding.\n" +
+            "5-6: All elements are separate, but touch targets might be a bit too close for comfort. Risk of fat-finger errors.\n" +
+            "7-8: Clean separation. No visual interference. App UI respects system boundaries and provides clear touch targets.\n" +
+            "9-10: Perfect logical grouping, generous touch targets, and completely unambiguous visual separation over the entire screen.",
     },
     {
         id: "layout",

package/dist/src/evaluation/checklist.js.map

@@ -1 +1 @@
-{"version":3,"file":"checklist.js","sourceRoot":"","sources":["../../../src/evaluation/checklist.ts"],"names":[],"mappings":"AAEA,+DAA+D;AAC/D,MAAM,CAAC,MAAM,UAAU,GAAoB;IACvC;QACI,EAAE,EAAE,SAAS;QACb,IAAI,EAAE,8BAA8B;QACpC,WAAW,EACP,8FAA8F;YAC9F,+EAA+E;YAC/E,iEAAiE;YACjE,scAAsc;YACtc,2LAA2L;YAC3L,4GAA4G;YAC5G,yGAAyG;YACzG,iFAAiF;YACjF,uOAAuO;YACvO,oGAAoG;QACxG,YAAY,EACR,wLAAwL;YACxL,oIAAoI;YACpI,gFAAgF;YAChF,gHAAgH;YAChH,2KAA2K;KAClL;IACD;QACI,EAAE,EAAE,QAAQ;QACZ,IAAI,EAAE,0BAA0B;QAChC,WAAW,EACP,2FAA2F;YAC3F,8DAA8D;YAC9D,sBAAsB;YACtB,2HAA2H;YAC3H,qJAAqJ;YACrJ,uKAAuK;YACvK,mJAAmJ;YACnJ,yIAAyI;YACzI,qFAAqF;YACrF,kHAAkH;QACtH,YAAY,EACR,0HAA0H;YAC1H,uGAAuG;YACvG,qJAAqJ;YACrJ,6GAA6G;YAC7G,qHAAqH;KAC5H;IACD;QACI,EAAE,EAAE,cAAc;QAClB,IAAI,EAAE,mCAAmC;QACzC,WAAW,EACP,kGAAkG;YAClG,iFAAiF;YACjF,sBAAsB;YACtB,sJAAsJ;YACtJ,2IAA2I;YAC3I,uIAAuI;YACvI,4GAA4G;YAC5G,uGAAuG;YACvG,kFAAkF;YAClF,4EAA4E;YAC5E,iDAAiD;YACjD,sFAAsF;QAC1F,YAAY,EACR,6FAA6F;YAC7F,2FAA2F;YAC3F,kHAAkH;YAClH,+GAA+G;YAC/G,iIAAiI;KACxI;IACD;QACI,EAAE,EAAE,OAAO;QACX,IAAI,EAAE,4BAA4B;QAClC,WAAW,EACP,qGAAqG;YACrG,yEAAyE;YACzE,sBAAsB;YACtB,mHAAmH;YACnH,uHAAuH;YACvH,sIAAsI;YACtI,kHAAkH;YAClH,oJAAoJ;YACpJ,uHAAuH;YACvH,mGAAmG;QACvG,YAAY,EACR,8FAA8F;YAC9F,kHAAkH;YAClH,0FAA0F;YAC1F,qFAAqF;YACrF,sFAAsF;KAC7F;IACD;QACI,EAAE,EAAE,eAAe;QACnB,IAAI,EAAE,4BAA4B;QAClC,WAAW,EACP,+EAA+E;YAC/E,wGAAwG;YACxG,sBAAsB;YACtB,gIAAgI;YAChI,2HAA2H;YAC3H,wHAAwH;YACxH,qHAAqH;YACrH,oHAAoH;YACpH,oEAAoE;QACxE,YAAY,EACR,6FAA6F;YAC7F,6GAA6G;YAC7G,0GAA0G;YAC1G,8FAA8F;YAC9F,sHAAsH;KAC7H;CACJ,CAAC;AAEF,MAAM,CAAC,MAAM,aAAa,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;AACvD,MAAM,CAAC,MAAM,aAAa,GAAG,CAAC,CAAC"}
+{"version":3,"file":"checklist.js","sourceRoot":"","sources":["../../../src/evaluation/checklist.ts"],"names":[],"mappings":"AAEA,+DAA+D;AAC/D,MAAM,CAAC,MAAM,UAAU,GAAoB;IACvC;QACI,EAAE,EAAE,SAAS;QACb,IAAI,EAAE,8BAA8B;QACpC,WAAW,EACP,gGAAgG;YAChG,qGAAqG;YACrG,wFAAwF;YACxF,oDAAoD;YACpD,8VAA8V;YAC9V,6MAA6M;YAC7M,0KAA0K;YAC1K,+IAA+I;YAC/I,wOAAwO;YACxO,6GAA6G;QACjH,YAAY,EACR,wLAAwL;YACxL,kJAAkJ;YAClJ,sHAAsH;YACtH,sHAAsH;YACtH,8HAA8H;KACrI;IACD;QACI,EAAE,EAAE,QAAQ;QACZ,IAAI,EAAE,0BAA0B;QAChC,WAAW,EACP,2FAA2F;YAC3F,8DAA8D;YAC9D,sBAAsB;YACtB,2HAA2H;YAC3H,qJAAqJ;YACrJ,uKAAuK;YACvK,mJAAmJ;YACnJ,yIAAyI;YACzI,qFAAqF;YACrF,kHAAkH;QACtH,YAAY,EACR,0HAA0H;YAC1H,uGAAuG;YACvG,qJAAqJ;YACrJ,6GAA6G;YAC7G,qHAAqH;KAC5H;IACD;QACI,EAAE,EAAE,cAAc;QAClB,IAAI,EAAE,mCAAmC;QACzC,WAAW,EACP,kGAAkG;YAClG,iFAAiF;YACjF,sBAAsB;YACtB,sJAAsJ;YACtJ,2IAA2I;YAC3I,uIAAuI;YACvI,4GAA4G;YAC5G,uGAAuG;YACvG,kFAAkF;YAClF,4EAA4E;YAC5E,iDAAiD;YACjD,sFAAsF;QAC1F,YAAY,EACR,6FAA6F;YAC7F,2FAA2F;YAC3F,kHAAkH;YAClH,+GAA+G;YAC/G,iIAAiI;KACxI;IACD;QACI,EAAE,EAAE,OAAO;QACX,IAAI,EAAE,4BAA4B;QAClC,WAAW,EACP,qGAAqG;YACrG,yEAAyE;YACzE,sBAAsB;YACtB,mHAAmH;YACnH,uHAAuH;YACvH,sIAAsI;YACtI,kHAAkH;YAClH,oJAAoJ;YACpJ,uHAAuH;YACvH,mGAAmG;QACvG,YAAY,EACR,8FAA8F;YAC9F,kHAAkH;YAClH,0FAA0F;YAC1F,qFAAqF;YACrF,sFAAsF;KAC7F;IACD;QACI,EAAE,EAAE,eAAe;QACnB,IAAI,EAAE,4BAA4B;QAClC,WAAW,EACP,+EAA+E;YAC/E,wGAAwG;YACxG,sBAAsB;YACtB,gIAAgI;YAChI,2HAA2H;YAC3H,wHAAwH;YACxH,qHAAqH;YACrH,oHAAoH;YACpH,oEAAoE;QACxE,YAAY,EACR,6FAA6F;YAC7F,6GAA6G;YAC7G,0GAA0G;YAC1G,8FAA8F;YAC9F,sHAAsH;KAC7H;CACJ,CAAC;AAEF,MAAM,CAAC,MAAM,aAAa,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;AACvD,MAAM,CAAC,MAAM,aAAa,GAAG,CAAC,CAAC"}

package/dist/src/evaluation/state-machine.d.ts

@@ -0,0 +1,43 @@
+export declare function getPendingTaskState(caseName: string, stepIndex: number): {
+    completed: boolean;
+    message: string;
+    dimensionId?: undefined;
+    dimensionName?: undefined;
+    description?: undefined;
+    scoringGuide?: undefined;
+    token?: undefined;
+    stepIndex?: undefined;
+} | {
+    completed: boolean;
+    dimensionId: string;
+    dimensionName: string;
+    description: string;
+    scoringGuide: string;
+    token: string;
+    stepIndex: number;
+    message?: undefined;
+};
+export declare function validateAndAdvanceState(caseName: string, stepIndex: number, token: string, score: number, reason: string): {
+    completedDimId: string;
+    score: number;
+    nextState: {
+        completed: boolean;
+        message: string;
+        dimensionId?: undefined;
+        dimensionName?: undefined;
+        description?: undefined;
+        scoringGuide?: undefined;
+        token?: undefined;
+        stepIndex?: undefined;
+    } | {
+        completed: boolean;
+        dimensionId: string;
+        dimensionName: string;
+        description: string;
+        scoringGuide: string;
+        token: string;
+        stepIndex: number;
+        message?: undefined;
+    };
+};
+//# sourceMappingURL=state-machine.d.ts.map

package/dist/src/evaluation/state-machine.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"state-machine.d.ts","sourceRoot":"","sources":["../../../src/evaluation/state-machine.ts"],"names":[],"mappings":"AAQA,wBAAgB,mBAAmB,CAAC,QAAQ,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM;;;;;;;;;;;;;;;;;;EAwCtE;AAED,wBAAgB,uBAAuB,CAAC,QAAQ,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM;;;;;;;;;;;;;;;;;;;;;;EA4CxH"}

package/dist/src/evaluation/state-machine.js

@@ -0,0 +1,79 @@
+import crypto from "crypto";
+import { readLog, writeLog } from "../logger/audit-log.js";
+import { DIMENSIONS, REQUIRED_DIMS } from "./checklist.js";
+function generateToken() {
+    return "tok_" + crypto.randomBytes(4).toString("hex");
+}
+export function getPendingTaskState(caseName, stepIndex) {
+    const log = readLog(caseName);
+    if (!log) {
+        throw new Error(`Test case not found: ${caseName}`);
+    }
+    const step = log.steps[stepIndex];
+    if (!step) {
+        throw new Error(`Step ${stepIndex} not found in case ${caseName}`);
+    }
+    if (step.currentDimIndex >= REQUIRED_DIMS.length) {
+        return {
+            completed: true,
+            message: `All ${REQUIRED_DIMS.length} dimensions for Step ${stepIndex} are fully evaluated! You may now proceed to interact with the device further (e.g. tap, swipe) or call get_audit_status.`
+        };
+    }
+    // Generate a token if there isn't one
+    if (!step.evaluationToken) {
+        step.evaluationToken = generateToken();
+        writeLog(log);
+    }
+    const currentDimId = REQUIRED_DIMS[step.currentDimIndex];
+    const criteria = DIMENSIONS.find(d => d.id === currentDimId);
+    if (!criteria) {
+        throw new Error(`Dimension '${currentDimId}' not found.`);
+    }
+    return {
+        completed: false,
+        dimensionId: currentDimId,
+        dimensionName: criteria.name,
+        description: criteria.description,
+        scoringGuide: criteria.scoringGuide,
+        token: step.evaluationToken,
+        stepIndex: stepIndex
+    };
+}
+export function validateAndAdvanceState(caseName, stepIndex, token, score, reason) {
+    const log = readLog(caseName);
+    if (!log) {
+        throw new Error(`Test case not found: ${caseName}`);
+    }
+    const step = log.steps[stepIndex];
+    if (!step) {
+        throw new Error(`Step ${stepIndex} not found in case ${caseName}`);
+    }
+    if (step.currentDimIndex >= REQUIRED_DIMS.length) {
+        throw new Error(`Step ${stepIndex} is already fully evaluated.`);
+    }
+    if (!step.evaluationToken || step.evaluationToken !== token) {
+        throw new Error("Invalid token. Do not guess tokens or run multiple submissions at once.");
+    }
+    const currentDimId = REQUIRED_DIMS[step.currentDimIndex];
+    // Record the score
+    step.evaluations[currentDimId] = {
+        score,
+        reason,
+    };
+    // Advance the state
+    step.currentDimIndex += 1;
+    // Clear the old token and generate a new one if not completed
+    if (step.currentDimIndex < REQUIRED_DIMS.length) {
+        step.evaluationToken = generateToken();
+    }
+    else {
+        step.evaluationToken = undefined;
+    }
+    writeLog(log);
+    return {
+        completedDimId: currentDimId,
+        score,
+        nextState: getPendingTaskState(caseName, stepIndex)
+    };
+}
+//# sourceMappingURL=state-machine.js.map

package/dist/src/evaluation/state-machine.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"state-machine.js","sourceRoot":"","sources":["../../../src/evaluation/state-machine.ts"],"names":[],"mappings":"AAAA,OAAO,MAAM,MAAM,QAAQ,CAAC;AAC5B,OAAO,EAAE,OAAO,EAAE,QAAQ,EAAE,MAAM,wBAAwB,CAAC;AAC3D,OAAO,EAAE,UAAU,EAAE,aAAa,EAAE,MAAM,gBAAgB,CAAC;AAE3D,SAAS,aAAa;IAClB,OAAO,MAAM,GAAG,MAAM,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;AAC1D,CAAC;AAED,MAAM,UAAU,mBAAmB,CAAC,QAAgB,EAAE,SAAiB;IACnE,MAAM,GAAG,GAAG,OAAO,CAAC,QAAQ,CAAC,CAAC;IAC9B,IAAI,CAAC,GAAG,EAAE,CAAC;QACP,MAAM,IAAI,KAAK,CAAC,wBAAwB,QAAQ,EAAE,CAAC,CAAC;IACxD,CAAC;IAED,MAAM,IAAI,GAAG,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;IAClC,IAAI,CAAC,IAAI,EAAE,CAAC;QACR,MAAM,IAAI,KAAK,CAAC,QAAQ,SAAS,sBAAsB,QAAQ,EAAE,CAAC,CAAC;IACvE,CAAC;IAED,IAAI,IAAI,CAAC,eAAe,IAAI,aAAa,CAAC,MAAM,EAAE,CAAC;QAC/C,OAAO;YACH,SAAS,EAAE,IAAI;YACf,OAAO,EAAE,OAAO,aAAa,CAAC,MAAM,wBAAwB,SAAS,2HAA2H;SACnM,CAAC;IACN,CAAC;IAED,sCAAsC;IACtC,IAAI,CAAC,IAAI,CAAC,eAAe,EAAE,CAAC;QACxB,IAAI,CAAC,eAAe,GAAG,aAAa,EAAE,CAAC;QACvC,QAAQ,CAAC,GAAG,CAAC,CAAC;IAClB,CAAC;IAED,MAAM,YAAY,GAAG,aAAa,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;IACzD,MAAM,QAAQ,GAAG,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,YAAY,CAAC,CAAC;IAE7D,IAAI,CAAC,QAAQ,EAAE,CAAC;QACZ,MAAM,IAAI,KAAK,CAAC,cAAc,YAAY,cAAc,CAAC,CAAC;IAC9D,CAAC;IAED,OAAO;QACH,SAAS,EAAE,KAAK;QAChB,WAAW,EAAE,YAAY;QACzB,aAAa,EAAE,QAAQ,CAAC,IAAI;QAC5B,WAAW,EAAE,QAAQ,CAAC,WAAW;QACjC,YAAY,EAAE,QAAQ,CAAC,YAAY;QACnC,KAAK,EAAE,IAAI,CAAC,eAAe;QAC3B,SAAS,EAAE,SAAS;KACvB,CAAC;AACN,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,QAAgB,EAAE,SAAiB,EAAE,KAAa,EAAE,KAAa,EAAE,MAAc;IACrH,MAAM,GAAG,GAAG,OAAO,CAAC,QAAQ,CAAC,CAAC;IAC9B,IAAI,CAAC,GAAG,EAAE,CAAC;QACP,MAAM,IAAI,KAAK,CAAC,wBAAwB,QAAQ,EAAE,CAAC,CAAC;IACxD,CAAC;IAED,MAAM,IAAI,GAAG,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;IAClC,IAAI,CAAC,IAAI,EAAE,CAAC;QACR,MAAM,IAAI,KAAK,CAAC,QAAQ,SAAS,sBAAsB,QAAQ,EAAE,CAAC,CAAC;IACvE,CAAC;IAED,IAAI,IAAI,CAAC,eAAe,IAAI,aAAa,CAAC,MAAM,EAAE,CAAC;QAC/C,MAAM,IAAI,KAAK,CAAC,QAAQ,SAAS,8BAA8B,CAAC,CAAC;IACrE,CAAC;IAED,IAAI,CAAC,IAAI,CAAC,eAAe,IAAI,IAAI,CAAC,eAAe,KAAK,KAAK,EAAE,CAAC;QAC1D,MAAM,IAAI,KAAK,CAAC,yEAAyE,CAAC,CAAC;IAC/F,CAAC;IAED,MAAM,YAAY,GAAG,aAAa,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;IAEzD,mBAAmB;IACnB,IAAI,CAAC,WAAW,CAAC,YAAY,CAAC,GAAG;QAC7B,KAAK;QACL,MAAM;KACT,CAAC;IAEF,oBAAoB;IACpB,IAAI,CAAC,eAAe,IAAI,CAAC,CAAC;IAE1B,8DAA8D;IAC9D,IAAI,IAAI,CAAC,eAAe,GAAG,aAAa,CAAC,MAAM,EAAE,CAAC;QAC9C,IAAI,CAAC,eAAe,GAAG,aAAa,EAAE,CAAC;IAC3C,CAAC;SAAM,CAAC;QACJ,IAAI,CAAC,eAAe,GAAG,SAAS,CAAC;IACrC,CAAC;IAED,QAAQ,CAAC,GAAG,CAAC,CAAC;IAEd,OAAO;QACH,cAAc,EAAE,YAAY;QAC5B,KAAK;QACL,SAAS,EAAE,mBAAmB,CAAC,QAAQ,EAAE,SAAS,CAAC;KACtD,CAAC;AACN,CAAC"}

package/dist/src/index.d.ts

@@ -7,12 +7,12 @@
  * and provides a rule-based, trackable evaluation workflow.
  *
  * Evaluation workflow for each step:
- *   1. Perform an action (tap / swipe / take_screenshot) — this auto-takes a screenshot and registers a step.
- *   2. For each required dimension (overlap, layout, info_clarity, style, action_result):
- *      a. Call get_evaluation_criteria(dimension) to read the scoring rubric.
- *      b. Visually inspect the screenshot.
- *      c. Call submit_dimension_score to record your score.
- *   3. Call get_audit_status to confirm all dimensions are complete before proceeding to the next action.
+ *   1. Perform an action (tap / swipe / take_screenshot) — auto-takes a screenshot and registers a step.
+ *   2. Call evaluate(caseName, stepIndex) — returns the first dimension prompt + token.
+ *   3. Visually inspect the screenshot based on the prompt.
+ *   4. Call evaluate(caseName, stepIndex, token, score, reason) — records the score, returns the next dimension prompt + new token.
+ *   5. Repeat step 3-4 until all 5 dimensions are evaluated.
+ *   6. Call get_audit_status to get the final markdown report.
  */
 export {};
 //# sourceMappingURL=index.d.ts.map

package/dist/src/index.js

@@ -7,12 +7,12 @@
  * and provides a rule-based, trackable evaluation workflow.
  *
  * Evaluation workflow for each step:
- *   1. Perform an action (tap / swipe / take_screenshot) — this auto-takes a screenshot and registers a step.
- *   2. For each required dimension (overlap, layout, info_clarity, style, action_result):
- *      a. Call get_evaluation_criteria(dimension) to read the scoring rubric.
- *      b. Visually inspect the screenshot.
- *      c. Call submit_dimension_score to record your score.
- *   3. Call get_audit_status to confirm all dimensions are complete before proceeding to the next action.
+ *   1. Perform an action (tap / swipe / take_screenshot) — auto-takes a screenshot and registers a step.
+ *   2. Call evaluate(caseName, stepIndex) — returns the first dimension prompt + token.
+ *   3. Visually inspect the screenshot based on the prompt.
+ *   4. Call evaluate(caseName, stepIndex, token, score, reason) — records the score, returns the next dimension prompt + new token.
+ *   5. Repeat step 3-4 until all 5 dimensions are evaluated.
+ *   6. Call get_audit_status to get the final markdown report.
  */
 import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
 import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
@@ -20,12 +20,11 @@ import { launchAppSchema, launchApp } from "./tools/launch-app.js";
 import { takeScreenshotSchema, takeScreenshot } from "./tools/take-screenshot.js";
 import { tapSchema, tap } from "./tools/tap.js";
 import { swipeSchema, swipe } from "./tools/swipe.js";
-import { getEvaluationCriteriaSchema, getEvaluationCriteria } from "./tools/get-evaluation-criteria.js";
-import { submitDimensionScoreSchema, submitDimensionScore } from "./tools/submit-dimension-score.js";
+import { evaluateSchema, evaluate } from "./tools/evaluate.js";
 import { getAuditStatusSchema, getAuditStatus } from "./tools/get-audit-status.js";
 const server = new McpServer({
     name: "ui-audit-mcp",
-    version: "2.0.0",
+    version: "2.3.0",
 });
 // ─── Device operation tools ────────────────────────────────
 server.tool("launch_app", "Launch an iOS app on the simulator by its bundle ID.", launchAppSchema.shape, async (args) => launchApp(args));
@@ -36,12 +35,10 @@ server.tool("tap", "Tap at a position on screen (ratio 0-1). Automatically captu
 server.tool("swipe", "Swipe on screen (ratio 0-1). Automatically captures a screenshot and registers a tracked step. " +
     "After this call, evaluate the screenshot across all required dimensions before moving to the next action.", swipeSchema.shape, async (args) => swipe(args));
 // ─── Evaluation tools ──────────────────────────────────────
-server.tool("get_evaluation_criteria", "Get UI evaluation dimensions and scoring rubrics. " +
-    "Call without arguments to list all dimension IDs. Pass a dimension ID (e.g. 'overlap') to get its detailed prompt and scoring guide. " +
-    "Required dimensions for every step: overlap, layout, info_clarity, style, action_result.", getEvaluationCriteriaSchema.shape, async (args) => getEvaluationCriteria(args));
-server.tool("submit_dimension_score", "Submit a score (0-10) and reason for one UI dimension on a specific step. " +
-    "Every step requires scores for all 5 dimensions: overlap, layout, info_clarity, style, action_result. " +
-    "After submitting, continue with the remaining dimensions, then call get_audit_status to confirm completion.", submitDimensionScoreSchema.shape, async (args) => submitDimensionScore(args));
+server.tool("evaluate", "Unified evaluation tool. Two modes:\n" +
+    "1) INITIAL: Call with only caseName + stepIndex (omit score/reason/token) → returns the first pending dimension's prompt, scoring guide, and evaluationToken.\n" +
+    "2) SUBMIT: Call with caseName + stepIndex + evaluationToken + score + reason → records the score, advances to the next dimension, and returns its prompt + new token (or completion message).\n" +
+    "Required dimensions per step: overlap, layout, info_clarity, style, action_result.", evaluateSchema.shape, async (args) => evaluate(args));
 server.tool("get_audit_status", "Get the current dashboard or test case report. " +
     "Use this to see which dimensions are missing (Pending) or to get the final markdown report if all steps are fully evaluated.", getAuditStatusSchema.shape, async (args) => getAuditStatus(args));
 import { preflight } from "./device/adapter.js";

package/dist/src/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/index.ts"],"names":[],"mappings":";AAEA;;;;;;;;;;;;;;GAcG;AAEH,OAAO,EAAE,SAAS,EAAE,MAAM,yCAAyC,CAAC;AACpE,OAAO,EAAE,oBAAoB,EAAE,MAAM,2CAA2C,CAAC;AAEjF,OAAO,EAAE,eAAe,EAAE,SAAS,EAAE,MAAM,uBAAuB,CAAC;AACnE,OAAO,EAAE,oBAAoB,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAClF,OAAO,EAAE,SAAS,EAAE,GAAG,EAAE,MAAM,gBAAgB,CAAC;AAChD,OAAO,EAAE,WAAW,EAAE,KAAK,EAAE,MAAM,kBAAkB,CAAC;AACtD,OAAO,EAAE,2BAA2B,EAAE,qBAAqB,EAAE,MAAM,oCAAoC,CAAC;AACxG,OAAO,EAAE,0BAA0B,EAAE,oBAAoB,EAAE,MAAM,mCAAmC,CAAC;AACrG,OAAO,EAAE,oBAAoB,EAAE,cAAc,EAAE,MAAM,6BAA6B,CAAC;AAEnF,MAAM,MAAM,GAAG,IAAI,SAAS,CAAC;IACzB,IAAI,EAAE,cAAc;IACpB,OAAO,EAAE,OAAO;CACnB,CAAC,CAAC;AAEH,8DAA8D;AAE9D,MAAM,CAAC,IAAI,CACP,YAAY,EACZ,sDAAsD,EACtD,eAAe,CAAC,KAAK,EACrB,KAAK,EAAE,IAAI,EAAE,EAAE,CAAC,SAAS,CAAC,IAAI,CAAC,CAClC,CAAC;AAEF,MAAM,CAAC,IAAI,CACP,iBAAiB,EACjB,qFAAqF;IACrF,2GAA2G,EAC3G,oBAAoB,CAAC,KAAK,EAC1B,KAAK,EAAE,IAAI,EAAE,EAAE,CAAC,cAAc,CAAC,IAAI,CAAC,CACvC,CAAC;AAEF,MAAM,CAAC,IAAI,CACP,KAAK,EACL,6GAA6G;IAC7G,2GAA2G,EAC3G,SAAS,CAAC,KAAK,EACf,KAAK,EAAE,IAAI,EAAE,EAAE,CAAC,GAAG,CAAC,IAAI,CAAC,CAC5B,CAAC;AAEF,MAAM,CAAC,IAAI,CACP,OAAO,EACP,iGAAiG;IACjG,2GAA2G,EAC3G,WAAW,CAAC,KAAK,EACjB,KAAK,EAAE,IAAI,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAC9B,CAAC;AAEF,8DAA8D;AAE9D,MAAM,CAAC,IAAI,CACP,yBAAyB,EACzB,oDAAoD;IACpD,uIAAuI;IACvI,0FAA0F,EAC1F,2BAA2B,CAAC,KAAK,EACjC,KAAK,EAAE,IAAI,EAAE,EAAE,CAAC,qBAAqB,CAAC,IAAI,CAAC,CAC9C,CAAC;AAEF,MAAM,CAAC,IAAI,CACP,wBAAwB,EACxB,4EAA4E;IAC5E,wGAAwG;IACxG,6GAA6G,EAC7G,0BAA0B,CAAC,KAAK,EAChC,KAAK,EAAE,IAAI,EAAE,EAAE,CAAC,oBAAoB,CAAC,IAAI,CAAC,CAC7C,CAAC;AAEF,MAAM,CAAC,IAAI,CACP,kBAAkB,EAClB,iDAAiD;IACjD,8HAA8H,EAC9H,oBAAoB,CAAC,KAAK,EAC1B,KAAK,EAAE,IAAI,EAAE,EAAE,CAAC,cAAc,CAAC,IAAI,CAAC,CACvC,CAAC;AAEF,OAAO,EAAE,SAAS,EAAE,MAAM,qBAAqB,CAAC;AAEhD,8DAA8D;AAE9D,KAAK,UAAU,IAAI;IACf,0DAA0D;IAC1D,MAAM,KAAK,GAAG,MAAM,SAAS,EAAE,CAAC;IAChC,IAAI,CAAC,KAAK,CAAC,EAAE,EAAE,CAAC;QACZ,OAAO,CAAC,KAAK,CAAC,0CAA0C,CAAC,CAAC;QAC1D,KAAK,MAAM,CAAC,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC;YAC5B,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;QAC9B,CAAC;QACD,OAAO,CAAC,KAAK,CAAC,sDAAsD,CAAC,CAAC;IAC1E,CAAC;IAED,MAAM,SAAS,GAAG,IAAI,oBAAoB,EAAE,CAAC;IAC7C,MAAM,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;IAChC,OAAO,CAAC,KAAK,CAAC,mDAAmD,CAAC,CAAC;AACvE,CAAC;AAED,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,GAAG,EAAE,EAAE;IACjB,OAAO,CAAC,KAAK,CAAC,6BAA6B,EAAE,GAAG,CAAC,CAAC;IAClD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AACpB,CAAC,CAAC,CAAC"}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/index.ts"],"names":[],"mappings":";AAEA;;;;;;;;;;;;;;GAcG;AAEH,OAAO,EAAE,SAAS,EAAE,MAAM,yCAAyC,CAAC;AACpE,OAAO,EAAE,oBAAoB,EAAE,MAAM,2CAA2C,CAAC;AAEjF,OAAO,EAAE,eAAe,EAAE,SAAS,EAAE,MAAM,uBAAuB,CAAC;AACnE,OAAO,EAAE,oBAAoB,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAClF,OAAO,EAAE,SAAS,EAAE,GAAG,EAAE,MAAM,gBAAgB,CAAC;AAChD,OAAO,EAAE,WAAW,EAAE,KAAK,EAAE,MAAM,kBAAkB,CAAC;AACtD,OAAO,EAAE,cAAc,EAAE,QAAQ,EAAE,MAAM,qBAAqB,CAAC;AAC/D,OAAO,EAAE,oBAAoB,EAAE,cAAc,EAAE,MAAM,6BAA6B,CAAC;AAEnF,MAAM,MAAM,GAAG,IAAI,SAAS,CAAC;IACzB,IAAI,EAAE,cAAc;IACpB,OAAO,EAAE,OAAO;CACnB,CAAC,CAAC;AAEH,8DAA8D;AAE9D,MAAM,CAAC,IAAI,CACP,YAAY,EACZ,sDAAsD,EACtD,eAAe,CAAC,KAAK,EACrB,KAAK,EAAE,IAAI,EAAE,EAAE,CAAC,SAAS,CAAC,IAAI,CAAC,CAClC,CAAC;AAEF,MAAM,CAAC,IAAI,CACP,iBAAiB,EACjB,qFAAqF;IACrF,2GAA2G,EAC3G,oBAAoB,CAAC,KAAK,EAC1B,KAAK,EAAE,IAAI,EAAE,EAAE,CAAC,cAAc,CAAC,IAAI,CAAC,CACvC,CAAC;AAEF,MAAM,CAAC,IAAI,CACP,KAAK,EACL,6GAA6G;IAC7G,2GAA2G,EAC3G,SAAS,CAAC,KAAK,EACf,KAAK,EAAE,IAAI,EAAE,EAAE,CAAC,GAAG,CAAC,IAAI,CAAC,CAC5B,CAAC;AAEF,MAAM,CAAC,IAAI,CACP,OAAO,EACP,iGAAiG;IACjG,2GAA2G,EAC3G,WAAW,CAAC,KAAK,EACjB,KAAK,EAAE,IAAI,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAC9B,CAAC;AAEF,8DAA8D;AAE9D,MAAM,CAAC,IAAI,CACP,UAAU,EACV,uCAAuC;IACvC,iKAAiK;IACjK,iMAAiM;IACjM,oFAAoF,EACpF,cAAc,CAAC,KAAK,EACpB,KAAK,EAAE,IAAI,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,CACjC,CAAC;AAEF,MAAM,CAAC,IAAI,CACP,kBAAkB,EAClB,iDAAiD;IACjD,8HAA8H,EAC9H,oBAAoB,CAAC,KAAK,EAC1B,KAAK,EAAE,IAAI,EAAE,EAAE,CAAC,cAAc,CAAC,IAAI,CAAC,CACvC,CAAC;AAEF,OAAO,EAAE,SAAS,EAAE,MAAM,qBAAqB,CAAC;AAEhD,8DAA8D;AAE9D,KAAK,UAAU,IAAI;IACf,0DAA0D;IAC1D,MAAM,KAAK,GAAG,MAAM,SAAS,EAAE,CAAC;IAChC,IAAI,CAAC,KAAK,CAAC,EAAE,EAAE,CAAC;QACZ,OAAO,CAAC,KAAK,CAAC,0CAA0C,CAAC,CAAC;QAC1D,KAAK,MAAM,CAAC,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC;YAC5B,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;QAC9B,CAAC;QACD,OAAO,CAAC,KAAK,CAAC,sDAAsD,CAAC,CAAC;IAC1E,CAAC;IAED,MAAM,SAAS,GAAG,IAAI,oBAAoB,EAAE,CAAC;IAC7C,MAAM,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;IAChC,OAAO,CAAC,KAAK,CAAC,mDAAmD,CAAC,CAAC;AACvE,CAAC;AAED,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,GAAG,EAAE,EAAE;IACjB,OAAO,CAAC,KAAK,CAAC,6BAA6B,EAAE,GAAG,CAAC,CAAC;IAClD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AACpB,CAAC,CAAC,CAAC"}

package/dist/src/logger/audit-log.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"audit-log.d.ts","sourceRoot":"","sources":["../../../src/logger/audit-log.ts"],"names":[],"mappings":"AAGA,OAAO,EAAE,QAAQ,EAA8B,MAAM,aAAa,CAAC;AAmBnE,wBAAgB,OAAO,CAAC,QAAQ,EAAE,MAAM,GAAG,QAAQ,GAAG,IAAI,CAMzD;AAED,wBAAgB,QAAQ,CAAC,GAAG,EAAE,QAAQ,GAAG,IAAI,CAG5C;AAED,wBAAgB,WAAW,IAAI,QAAQ,EAAE,CAexC;AAED,wBAAgB,UAAU,CACtB,QAAQ,EAAE,MAAM,EAChB,SAAS,EAAE,MAAM,EACjB,WAAW,EAAE,MAAM,EACnB,UAAU,EAAE,KAAK,GAAG,OAAO,GAAG,YAAY,EAC1C,cAAc,EAAE,MAAM,EACtB,WAAW,CAAC,EAAE;IAAE,CAAC,EAAE,MAAM,CAAC;IAAC,CAAC,EAAE,MAAM,CAAA;CAAE,EACtC,eAAe,CAAC,EAAE,MAAM,GACzB,IAAI,CAuBN;AAED,wBAAgB,WAAW,CACvB,QAAQ,EAAE,MAAM,EAChB,SAAS,EAAE,MAAM,EACjB,SAAS,EAAE,MAAM,EACjB,KAAK,EAAE,MAAM,EACb,MAAM,EAAE,MAAM,GACf,IAAI,CAiBN"}
+{"version":3,"file":"audit-log.d.ts","sourceRoot":"","sources":["../../../src/logger/audit-log.ts"],"names":[],"mappings":"AAGA,OAAO,EAAE,QAAQ,EAA8B,MAAM,aAAa,CAAC;AAmBnE,wBAAgB,OAAO,CAAC,QAAQ,EAAE,MAAM,GAAG,QAAQ,GAAG,IAAI,CAMzD;AAED,wBAAgB,QAAQ,CAAC,GAAG,EAAE,QAAQ,GAAG,IAAI,CAG5C;AAED,wBAAgB,WAAW,IAAI,QAAQ,EAAE,CAexC;AAED,wBAAgB,UAAU,CACtB,QAAQ,EAAE,MAAM,EAChB,SAAS,EAAE,MAAM,EACjB,WAAW,EAAE,MAAM,EACnB,UAAU,EAAE,KAAK,GAAG,OAAO,GAAG,YAAY,EAC1C,cAAc,EAAE,MAAM,EACtB,WAAW,CAAC,EAAE;IAAE,CAAC,EAAE,MAAM,CAAC;IAAC,CAAC,EAAE,MAAM,CAAA;CAAE,EACtC,eAAe,CAAC,EAAE,MAAM,GACzB,IAAI,CAyBN;AAED,wBAAgB,WAAW,CACvB,QAAQ,EAAE,MAAM,EAChB,SAAS,EAAE,MAAM,EACjB,SAAS,EAAE,MAAM,EACjB,KAAK,EAAE,MAAM,EACb,MAAM,EAAE,MAAM,GACf,IAAI,CAiBN"}

package/dist/src/logger/audit-log.js

@@ -64,6 +64,8 @@ export function recordStep(caseName, stepIndex, description, actionType, screens
         expectedOutcome,
         timestamp: new Date().toISOString(),
         evaluations: existingStep ? existingStep.evaluations : {},
+        currentDimIndex: existingStep && existingStep.currentDimIndex !== undefined ? existingStep.currentDimIndex : 0,
+        evaluationToken: existingStep ? existingStep.evaluationToken : undefined,
     };
     writeLog(log);
 }

package/dist/src/logger/audit-log.js.map

@@ -1 +1 @@
-{"version":3,"file":"audit-log.js","sourceRoot":"","sources":["../../../src/logger/audit-log.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,IAAI,CAAC;AACpB,OAAO,IAAI,MAAM,MAAM,CAAC;AACxB,OAAO,EAAE,MAAM,IAAI,CAAC;AAGpB,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,EAAE,eAAe,CAAC,CAAC;AAC1D,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC;AAC5C,MAAM,cAAc,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,aAAa,CAAC,CAAC;AAE1D,2BAA2B;AAC3B,SAAS,UAAU;IACf,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,QAAQ,CAAC;QAAE,EAAE,CAAC,SAAS,CAAC,QAAQ,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAC1E,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,OAAO,CAAC;QAAE,EAAE,CAAC,SAAS,CAAC,OAAO,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACxE,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,cAAc,CAAC;QAAE,EAAE,CAAC,SAAS,CAAC,cAAc,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;AAC1F,CAAC;AAED,UAAU,EAAE,CAAC;AAEb,SAAS,UAAU,CAAC,QAAgB;IAChC,OAAO,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,GAAG,QAAQ,OAAO,CAAC,CAAC;AAClD,CAAC;AAED,MAAM,UAAU,OAAO,CAAC,QAAgB;IACpC,MAAM,OAAO,GAAG,UAAU,CAAC,QAAQ,CAAC,CAAC;IACrC,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,OAAO,CAAC,EAAE,CAAC;QAC1B,OAAO,IAAI,CAAC;IAChB,CAAC;IACD,OAAO,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,YAAY,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC;AACzD,CAAC;AAED,MAAM,UAAU,QAAQ,CAAC,GAAa;IAClC,MAAM,OAAO,GAAG,UAAU,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;IACzC,EAAE,CAAC,aAAa,CAAC,OAAO,EAAE,IAAI,CAAC,SAAS,CAAC,GAAG,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC;AACrE,CAAC;AAED,MAAM,UAAU,WAAW;IACvB,MAAM,IAAI,GAAe,EAAE,CAAC;IAC5B,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,OAAO,CAAC;QAAE,OAAO,IAAI,CAAC;IACzC,MAAM,KAAK,GAAG,EAAE,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC;IACtC,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACvB,IAAI,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;YACzB,MAAM,OAAO,GAAG,EAAE,CAAC,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,IAAI,CAAC,EAAE,OAAO,CAAC,CAAC;YACnE,IAAI,CAAC;gBACD,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC;YACnC,CAAC;YAAC,OAAO,CAAC,EAAE,CAAC;gBACT,OAAO,CAAC,KAAK,CAAC,6BAA6B,IAAI,EAAE,EAAE,CAAC,CAAC,CAAC;YAC1D,CAAC;QACL,CAAC;IACL,CAAC;IACD,OAAO,IAAI,CAAC;AAChB,CAAC;AAED,MAAM,UAAU,UAAU,CACtB,QAAgB,EAChB,SAAiB,EACjB,WAAmB,EACnB,UAA0C,EAC1C,cAAsB,EACtB,WAAsC,EACtC,eAAwB;IAExB,IAAI,GAAG,GAAG,OAAO,CAAC,QAAQ,CAAC,CAAC;IAC5B,IAAI,CAAC,GAAG,EAAE,CAAC;QACP,GAAG,GAAG;YACF,QAAQ;YACR,KAAK,EAAE,EAAE;SACZ,CAAC;IACN,CAAC;IAED,MAAM,YAAY,GAAG,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;IAE1C,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,GAAG;QACnB,SAAS;QACT,WAAW;QACX,UAAU;QACV,cAAc;QACd,WAAW;QACX,eAAe;QACf,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;QACnC,WAAW,EAAE,YAAY,CAAC,CAAC,CAAC,YAAY,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE;KAC5D,CAAC;IAEF,QAAQ,CAAC,GAAG,CAAC,CAAC;AAClB,CAAC;AAED,MAAM,UAAU,WAAW,CACvB,QAAgB,EAChB,SAAiB,EACjB,SAAiB,EACjB,KAAa,EACb,MAAc;IAEd,MAAM,GAAG,GAAG,OAAO,CAAC,QAAQ,CAAC,CAAC;IAC9B,IAAI,CAAC,GAAG,EAAE,CAAC;QACP,MAAM,IAAI,KAAK,CAAC,wBAAwB,QAAQ,EAAE,CAAC,CAAC;IACxD,CAAC;IAED,MAAM,IAAI,GAAG,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;IAClC,IAAI,CAAC,IAAI,EAAE,CAAC;QACR,MAAM,IAAI,KAAK,CAAC,QAAQ,SAAS,sBAAsB,QAAQ,EAAE,CAAC,CAAC;IACvE,CAAC;IAED,IAAI,CAAC,WAAW,CAAC,SAAS,CAAC,GAAG;QAC1B,KAAK;QACL,MAAM;KACT,CAAC;IAEF,QAAQ,CAAC,GAAG,CAAC,CAAC;AAClB,CAAC"}
+{"version":3,"file":"audit-log.js","sourceRoot":"","sources":["../../../src/logger/audit-log.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,IAAI,CAAC;AACpB,OAAO,IAAI,MAAM,MAAM,CAAC;AACxB,OAAO,EAAE,MAAM,IAAI,CAAC;AAGpB,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,EAAE,eAAe,CAAC,CAAC;AAC1D,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC;AAC5C,MAAM,cAAc,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,aAAa,CAAC,CAAC;AAE1D,2BAA2B;AAC3B,SAAS,UAAU;IACf,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,QAAQ,CAAC;QAAE,EAAE,CAAC,SAAS,CAAC,QAAQ,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAC1E,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,OAAO,CAAC;QAAE,EAAE,CAAC,SAAS,CAAC,OAAO,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACxE,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,cAAc,CAAC;QAAE,EAAE,CAAC,SAAS,CAAC,cAAc,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;AAC1F,CAAC;AAED,UAAU,EAAE,CAAC;AAEb,SAAS,UAAU,CAAC,QAAgB;IAChC,OAAO,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,GAAG,QAAQ,OAAO,CAAC,CAAC;AAClD,CAAC;AAED,MAAM,UAAU,OAAO,CAAC,QAAgB;IACpC,MAAM,OAAO,GAAG,UAAU,CAAC,QAAQ,CAAC,CAAC;IACrC,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,OAAO,CAAC,EAAE,CAAC;QAC1B,OAAO,IAAI,CAAC;IAChB,CAAC;IACD,OAAO,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,YAAY,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC;AACzD,CAAC;AAED,MAAM,UAAU,QAAQ,CAAC,GAAa;IAClC,MAAM,OAAO,GAAG,UAAU,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;IACzC,EAAE,CAAC,aAAa,CAAC,OAAO,EAAE,IAAI,CAAC,SAAS,CAAC,GAAG,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC;AACrE,CAAC;AAED,MAAM,UAAU,WAAW;IACvB,MAAM,IAAI,GAAe,EAAE,CAAC;IAC5B,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,OAAO,CAAC;QAAE,OAAO,IAAI,CAAC;IACzC,MAAM,KAAK,GAAG,EAAE,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC;IACtC,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACvB,IAAI,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;YACzB,MAAM,OAAO,GAAG,EAAE,CAAC,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,IAAI,CAAC,EAAE,OAAO,CAAC,CAAC;YACnE,IAAI,CAAC;gBACD,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC;YACnC,CAAC;YAAC,OAAO,CAAC,EAAE,CAAC;gBACT,OAAO,CAAC,KAAK,CAAC,6BAA6B,IAAI,EAAE,EAAE,CAAC,CAAC,CAAC;YAC1D,CAAC;QACL,CAAC;IACL,CAAC;IACD,OAAO,IAAI,CAAC;AAChB,CAAC;AAED,MAAM,UAAU,UAAU,CACtB,QAAgB,EAChB,SAAiB,EACjB,WAAmB,EACnB,UAA0C,EAC1C,cAAsB,EACtB,WAAsC,EACtC,eAAwB;IAExB,IAAI,GAAG,GAAG,OAAO,CAAC,QAAQ,CAAC,CAAC;IAC5B,IAAI,CAAC,GAAG,EAAE,CAAC;QACP,GAAG,GAAG;YACF,QAAQ;YACR,KAAK,EAAE,EAAE;SACZ,CAAC;IACN,CAAC;IAED,MAAM,YAAY,GAAG,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;IAE1C,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,GAAG;QACnB,SAAS;QACT,WAAW;QACX,UAAU;QACV,cAAc;QACd,WAAW;QACX,eAAe;QACf,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;QACnC,WAAW,EAAE,YAAY,CAAC,CAAC,CAAC,YAAY,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE;QACzD,eAAe,EAAE,YAAY,IAAI,YAAY,CAAC,eAAe,KAAK,SAAS,CAAC,CAAC,CAAC,YAAY,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC;QAC9G,eAAe,EAAE,YAAY,CAAC,CAAC,CAAC,YAAY,CAAC,eAAe,CAAC,CAAC,CAAC,SAAS;KAC3E,CAAC;IAEF,QAAQ,CAAC,GAAG,CAAC,CAAC;AAClB,CAAC;AAED,MAAM,UAAU,WAAW,CACvB,QAAgB,EAChB,SAAiB,EACjB,SAAiB,EACjB,KAAa,EACb,MAAc;IAEd,MAAM,GAAG,GAAG,OAAO,CAAC,QAAQ,CAAC,CAAC;IAC9B,IAAI,CAAC,GAAG,EAAE,CAAC;QACP,MAAM,IAAI,KAAK,CAAC,wBAAwB,QAAQ,EAAE,CAAC,CAAC;IACxD,CAAC;IAED,MAAM,IAAI,GAAG,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;IAClC,IAAI,CAAC,IAAI,EAAE,CAAC;QACR,MAAM,IAAI,KAAK,CAAC,QAAQ,SAAS,sBAAsB,QAAQ,EAAE,CAAC,CAAC;IACvE,CAAC;IAED,IAAI,CAAC,WAAW,CAAC,SAAS,CAAC,GAAG;QAC1B,KAAK;QACL,MAAM;KACT,CAAC;IAEF,QAAQ,CAAC,GAAG,CAAC,CAAC;AAClB,CAAC"}

package/dist/src/tools/evaluate.d.ts

@@ -0,0 +1,41 @@
+/**
+ * Tool: evaluate
+ *
+ * Unified evaluation tool that combines get_pending_task and submit_evaluation.
+ *
+ * Two modes:
+ *   1. Initial mode (score/reason/token omitted):
+ *      Returns the first pending dimension's prompt, scoring guide, and a token.
+ *
+ *   2. Submit-and-advance mode (score + reason + token all provided):
+ *      Validates the token, records the score, advances to the next dimension.
+ *      If more dimensions remain, returns the next prompt + new token.
+ *      If all done, returns a completion message.
+ */
+import { z } from "zod";
+export declare const evaluateSchema: z.ZodObject<{
+    caseName: z.ZodString;
+    stepIndex: z.ZodNumber;
+    evaluationToken: z.ZodOptional<z.ZodString>;
+    score: z.ZodOptional<z.ZodNumber>;
+    reason: z.ZodOptional<z.ZodString>;
+}, "strip", z.ZodTypeAny, {
+    caseName: string;
+    stepIndex: number;
+    evaluationToken?: string | undefined;
+    score?: number | undefined;
+    reason?: string | undefined;
+}, {
+    caseName: string;
+    stepIndex: number;
+    evaluationToken?: string | undefined;
+    score?: number | undefined;
+    reason?: string | undefined;
+}>;
+export declare function evaluate(input: z.infer<typeof evaluateSchema>): Promise<{
+    content: {
+        type: "text";
+        text: string;
+    }[];
+}>;
+//# sourceMappingURL=evaluate.d.ts.map

package/dist/src/tools/evaluate.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluate.d.ts","sourceRoot":"","sources":["../../../src/tools/evaluate.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAEH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAGxB,eAAO,MAAM,cAAc;;;;;;;;;;;;;;;;;;EAazB,CAAC;AAqBH,wBAAsB,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,cAAc,CAAC;;;;;GAmEnE"}

package/dist/src/tools/evaluate.js

@@ -0,0 +1,87 @@
+/**
+ * Tool: evaluate
+ *
+ * Unified evaluation tool that combines get_pending_task and submit_evaluation.
+ *
+ * Two modes:
+ *   1. Initial mode (score/reason/token omitted):
+ *      Returns the first pending dimension's prompt, scoring guide, and a token.
+ *
+ *   2. Submit-and-advance mode (score + reason + token all provided):
+ *      Validates the token, records the score, advances to the next dimension.
+ *      If more dimensions remain, returns the next prompt + new token.
+ *      If all done, returns a completion message.
+ */
+import { z } from "zod";
+import { getPendingTaskState, validateAndAdvanceState } from "../evaluation/state-machine.js";
+export const evaluateSchema = z.object({
+    caseName: z.string().describe("Name of the test case"),
+    stepIndex: z.number().int().positive().describe("Step index to evaluate"),
+    // Optional: omit all three for initial query, provide all three to submit
+    evaluationToken: z.string().optional().describe("The exact token from the previous evaluate call. Omit on the first call to get the initial dimension."),
+    score: z.number().int().min(0).max(10).optional().describe("Score from 0 (worst) to 10 (perfect). Required when submitting an evaluation."),
+    reason: z.string().optional().describe("Analysis and reason for this score. Required when submitting an evaluation."),
+});
+function renderDimensionPrompt(state, prefix) {
+    return `${prefix}\n\n` +
+        `🔴 CURRENT TASK: ${state.dimensionName}\n\n` +
+        `PROMPT: ${state.description}\n\n` +
+        `SCORING GUIDE: \n${state.scoringGuide}\n\n` +
+        `⚠️ ACTION REQUIRED:\n` +
+        `Inspect the screenshot specifically for this dimension. Then call \`evaluate\` with:\n` +
+        `  "evaluationToken": "${state.token}"\n` +
+        `  "score": <your score 0-10>\n` +
+        `  "reason": "<your analysis>"\n\n` +
+        `Do NOT evaluate other dimensions until you finish this one.`;
+}
+export async function evaluate(input) {
+    const hasScore = input.score !== undefined;
+    const hasReason = input.reason !== undefined;
+    const hasToken = input.evaluationToken !== undefined;
+    const submissionFieldCount = [hasScore, hasReason, hasToken].filter(Boolean).length;
+    // Validate: all three must be present or all absent
+    if (submissionFieldCount > 0 && submissionFieldCount < 3) {
+        return {
+            content: [{
+                    type: "text",
+                    text: "❌ Error: `evaluationToken`, `score`, and `reason` must ALL be provided together, or ALL be omitted.\n" +
+                        "- Omit all three to get the first pending dimension.\n" +
+                        "- Provide all three to submit an evaluation and advance."
+                }]
+        };
+    }
+    // ── Mode 1: Initial query (no score/reason/token) ──
+    if (submissionFieldCount === 0) {
+        const state = getPendingTaskState(input.caseName, input.stepIndex);
+        if (state.completed) {
+            return {
+                content: [{
+                        type: "text",
+                        text: state.message || "✅ All dimensions for this step are fully evaluated."
+                    }]
+            };
+        }
+        const markdown = renderDimensionPrompt(state, `📋 Starting evaluation for Step ${input.stepIndex}:`);
+        return {
+            content: [{ type: "text", text: markdown }],
+        };
+    }
+    // ── Mode 2: Submit score and advance ──
+    const result = validateAndAdvanceState(input.caseName, input.stepIndex, input.evaluationToken, input.score, input.reason);
+    const nextState = result.nextState;
+    if (nextState.completed) {
+        return {
+            content: [{
+                    type: "text",
+                    text: `✅ Score ${result.score} recorded for [${result.completedDimId}].\n\n` +
+                        `🎉 All dimensions for Step ${input.stepIndex} are fully evaluated!\n` +
+                        `You may now proceed to interact with the device (tap, swipe) or call get_audit_status.`
+                }]
+        };
+    }
+    const markdown = renderDimensionPrompt(nextState, `✅ Score ${result.score} recorded for [${result.completedDimId}].\n\n➡️ Next dimension:`);
+    return {
+        content: [{ type: "text", text: markdown }],
+    };
+}
+//# sourceMappingURL=evaluate.js.map

package/dist/src/tools/evaluate.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluate.js","sourceRoot":"","sources":["../../../src/tools/evaluate.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAEH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,mBAAmB,EAAE,uBAAuB,EAAE,MAAM,gCAAgC,CAAC;AAE9F,MAAM,CAAC,MAAM,cAAc,GAAG,CAAC,CAAC,MAAM,CAAC;IACnC,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,uBAAuB,CAAC;IACtD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,wBAAwB,CAAC;IACzE,0EAA0E;IAC1E,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC3C,uGAAuG,CAC1G;IACD,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACtD,+EAA+E,CAClF;IACD,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAClC,6EAA6E,CAChF;CACJ,CAAC,CAAC;AAEH,SAAS,qBAAqB,CAAC,KAM9B,EAAE,MAAc;IACb,OAAO,GAAG,MAAM,MAAM;QAClB,oBAAoB,KAAK,CAAC,aAAa,MAAM;QAC7C,WAAW,KAAK,CAAC,WAAW,MAAM;QAClC,oBAAoB,KAAK,CAAC,YAAY,MAAM;QAC5C,uBAAuB;QACvB,wFAAwF;QACxF,yBAAyB,KAAK,CAAC,KAAK,KAAK;QACzC,gCAAgC;QAChC,mCAAmC;QACnC,6DAA6D,CAAC;AACtE,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,QAAQ,CAAC,KAAqC;IAChE,MAAM,QAAQ,GAAG,KAAK,CAAC,KAAK,KAAK,SAAS,CAAC;IAC3C,MAAM,SAAS,GAAG,KAAK,CAAC,MAAM,KAAK,SAAS,CAAC;IAC7C,MAAM,QAAQ,GAAG,KAAK,CAAC,eAAe,KAAK,SAAS,CAAC;IAErD,MAAM,oBAAoB,GAAG,CAAC,QAAQ,EAAE,SAAS,EAAE,QAAQ,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC;IAEpF,oDAAoD;IACpD,IAAI,oBAAoB,GAAG,CAAC,IAAI,oBAAoB,GAAG,CAAC,EAAE,CAAC;QACvD,OAAO;YACH,OAAO,EAAE,CAAC;oBACN,IAAI,EAAE,MAAe;oBACrB,IAAI,EAAE,uGAAuG;wBACzG,wDAAwD;wBACxD,0DAA0D;iBACjE,CAAC;SACL,CAAC;IACN,CAAC;IAED,sDAAsD;IACtD,IAAI,oBAAoB,KAAK,CAAC,EAAE,CAAC;QAC7B,MAAM,KAAK,GAAG,mBAAmB,CAAC,KAAK,CAAC,QAAQ,EAAE,KAAK,CAAC,SAAS,CAAC,CAAC;QAEnE,IAAI,KAAK,CAAC,SAAS,EAAE,CAAC;YAClB,OAAO;gBACH,OAAO,EAAE,CAAC;wBACN,IAAI,EAAE,MAAe;wBACrB,IAAI,EAAE,KAAK,CAAC,OAAO,IAAI,qDAAqD;qBAC/E,CAAC;aACL,CAAC;QACN,CAAC;QAED,MAAM,QAAQ,GAAG,qBAAqB,CAAC,KAAK,EAAE,mCAAmC,KAAK,CAAC,SAAS,GAAG,CAAC,CAAC;QACrG,OAAO;YACH,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAe,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAC;SACvD,CAAC;IACN,CAAC;IAED,yCAAyC;IACzC,MAAM,MAAM,GAAG,uBAAuB,CAClC,KAAK,CAAC,QAAQ,EACd,KAAK,CAAC,SAAS,EACf,KAAK,CAAC,eAAgB,EACtB,KAAK,CAAC,KAAM,EACZ,KAAK,CAAC,MAAO,CAChB,CAAC;IAEF,MAAM,SAAS,GAAG,MAAM,CAAC,SAAS,CAAC;IAEnC,IAAI,SAAS,CAAC,SAAS,EAAE,CAAC;QACtB,OAAO;YACH,OAAO,EAAE,CAAC;oBACN,IAAI,EAAE,MAAe;oBACrB,IAAI,EAAE,WAAW,MAAM,CAAC,KAAK,kBAAkB,MAAM,CAAC,cAAc,QAAQ;wBACxE,8BAA8B,KAAK,CAAC,SAAS,yBAAyB;wBACtE,wFAAwF;iBAC/F,CAAC;SACL,CAAC;IACN,CAAC;IAED,MAAM,QAAQ,GAAG,qBAAqB,CAAC,SAAS,EAC5C,WAAW,MAAM,CAAC,KAAK,kBAAkB,MAAM,CAAC,cAAc,0BAA0B,CAC3F,CAAC;IAEF,OAAO;QACH,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAe,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAC;KACvD,CAAC;AACN,CAAC"}

package/dist/src/tools/get-audit-status.js

@@ -6,38 +6,38 @@ export const getAuditStatusSchema = z.object({
 });
 // Helper for rendering incomplete template
 function renderIncompleteTemplate(log, missingItems) {
-    let md = `# 📊 审计进度报告: ${log.caseName}\n`;
-    md += `**状态**: 🚧 评估未完成\n\n`;
+    let md = `# 📊 Audit Progress Report: ${log.caseName}\n`;
+    md += `**Status**: 🚧 Evaluation Incomplete\n\n`;
     // Missing summary
-    md += `## 🔴 待办评估任务 (Pending Evaluations)\n`;
-    md += `Agent 请注意：您还需要完成以下维度的评估才能生成最终成绩：\n`;
+    md += `## 🔴 Pending Evaluations\n`;
+    md += `Attention Agent: You still need to complete the evaluations for the following dimensions to generate the final report:\n`;
     for (const item of missingItems) {
-        md += `- Step ${item.stepIndex}: 缺失 [${item.missing.join(", ")}]\n`;
+        md += `- Step ${item.stepIndex}: Missing [${item.missing.join(", ")}]\n`;
     }
     md += `\n`;
     // Step details
-    md += `## 📝 已评估记录\n`;
+    md += `## 📝 Evaluated Records\n`;
     md += renderStepDetails(log);
     return md;
 }
 // Helper for rendering complete template
 function renderCompleteTemplate(log, avg, passed, failedItems) {
-    let md = `# 🏆 最终审计报告: ${log.caseName}\n`;
-    md += `**状态**: ✅ 评估已完成\n\n`;
-    md += `## 📈 总成绩单\n`;
-    md += `- **平均分**: ${avg.toFixed(1)} / 10\n`;
-    md += `- **结论**: ${passed ? "✅ 通过 (Pass)" : "❌ 驳回 (Fail)"}\n`;
+    let md = `# 🏆 Final Audit Report: ${log.caseName}\n`;
+    md += `**Status**: ✅ Evaluation Complete\n\n`;
+    md += `## 📈 Overall Scores\n`;
+    md += `- **Average Score**: ${avg.toFixed(1)} / 10\n`;
+    md += `- **Result**: ${passed ? "✅ Pass" : "❌ Fail"}\n`;
     if (failedItems.length > 0) {
-        md += `- **不合格项**:\n`;
+        md += `- **Failed Items**:\n`;
         for (const f of failedItems) {
-            md += `  - Step ${f.stepIndex} [${f.dim}]: ${f.score}分 - ${f.reason}\n`;
+            md += `  - Step ${f.stepIndex} [${f.dim}]: Score ${f.score} - ${f.reason}\n`;
         }
     }
     else {
-        md += `- **不合格项**: 无\n`;
+        md += `- **Failed Items**: None\n`;
     }
     md += `\n`;
-    md += `## 📝 详细步骤\n`;
+    md += `## 📝 Detailed Steps\n`;
     md += renderStepDetails(log);
     return md;
 }
@@ -46,18 +46,18 @@ function renderStepDetails(log) {
     const steps = Object.values(log.steps).sort((a, b) => a.stepIndex - b.stepIndex);
     for (const step of steps) {
         md += `### Step ${step.stepIndex}: ${step.description}\n`;
-        md += `- **操作**: ${step.actionType} ${step.coordinates ? `(x: ${step.coordinates.x}, y: ${step.coordinates.y})` : ""}\n`;
+        md += `- **Action**: ${step.actionType} ${step.coordinates ? `(x: ${step.coordinates.x}, y: ${step.coordinates.y})` : ""}\n`;
         if (step.expectedOutcome) {
-            md += `- **预期结果**: ${step.expectedOutcome}\n`;
+            md += `- **Expected Outcome**: ${step.expectedOutcome}\n`;
         }
         for (const dim of REQUIRED_DIMS) {
             const ev = step.evaluations ? step.evaluations[dim] : undefined;
             if (ev) {
                 const tag = ev.score >= PASSING_SCORE ? "Pass ✅" : "Fail ❌";
-                md += `- **${dim}**: ${ev.score}分 ${tag} - ${ev.reason}\n`;
+                md += `- **${dim}**: Score ${ev.score} ${tag} - ${ev.reason}\n`;
             }
             else {
-                md += `- **${dim}**: 🔴 缺失评估\n`;
+                md += `- **${dim}**: 🔴 Missing Evaluation\n`;
             }
         }
         md += `\n`;

package/dist/src/tools/get-audit-status.js.map

@@ -1 +1 @@
-{"version":3,"file":"get-audit-status.js","sourceRoot":"","sources":["../../../src/tools/get-audit-status.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,wBAAwB,CAAC;AAC9D,OAAO,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,4BAA4B,CAAC;AAG1E,MAAM,CAAC,MAAM,oBAAoB,GAAG,CAAC,CAAC,MAAM,CAAC;IACzC,SAAS,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,gFAAgF,CAAC;CACvI,CAAC,CAAC;AAEH,2CAA2C;AAC3C,SAAS,wBAAwB,CAAC,GAAa,EAAE,YAAwD;IACrG,IAAI,EAAE,GAAG,gBAAgB,GAAG,CAAC,QAAQ,IAAI,CAAC;IAC1C,EAAE,IAAI,sBAAsB,CAAC;IAE7B,kBAAkB;IAClB,EAAE,IAAI,sCAAsC,CAAC;IAC7C,EAAE,IAAI,oCAAoC,CAAC;IAC3C,KAAK,MAAM,IAAI,IAAI,YAAY,EAAE,CAAC;QAC9B,EAAE,IAAI,UAAU,IAAI,CAAC,SAAS,SAAS,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC;IACxE,CAAC;IACD,EAAE,IAAI,IAAI,CAAC;IAEX,eAAe;IACf,EAAE,IAAI,eAAe,CAAC;IACtB,EAAE,IAAI,iBAAiB,CAAC,GAAG,CAAC,CAAC;IAE7B,OAAO,EAAE,CAAC;AACd,CAAC;AAED,yCAAyC;AACzC,SAAS,sBAAsB,CAAC,GAAa,EAAE,GAAW,EAAE,MAAe,EAAE,WAAkB;IAC3F,IAAI,EAAE,GAAG,gBAAgB,GAAG,CAAC,QAAQ,IAAI,CAAC;IAC1C,EAAE,IAAI,qBAAqB,CAAC;IAE5B,EAAE,IAAI,cAAc,CAAC;IACrB,EAAE,IAAI,cAAc,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC;IAC5C,EAAE,IAAI,aAAa,MAAM,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,aAAa,IAAI,CAAC;IAE9D,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACzB,EAAE,IAAI,eAAe,CAAC;QACtB,KAAK,MAAM,CAAC,IAAI,WAAW,EAAE,CAAC;YAC1B,EAAE,IAAI,YAAY,CAAC,CAAC,SAAS,KAAK,CAAC,CAAC,GAAG,MAAM,CAAC,CAAC,KAAK,OAAO,CAAC,CAAC,MAAM,IAAI,CAAC;QAC5E,CAAC;IACL,CAAC;SAAM,CAAC;QACJ,EAAE,IAAI,iBAAiB,CAAC;IAC5B,CAAC;IACD,EAAE,IAAI,IAAI,CAAC;IAEX,EAAE,IAAI,cAAc,CAAC;IACrB,EAAE,IAAI,iBAAiB,CAAC,GAAG,CAAC,CAAC;IAE7B,OAAO,EAAE,CAAC;AACd,CAAC;AAED,SAAS,iBAAiB,CAAC,GAAa;IACpC,IAAI,EAAE,GAAG,EAAE,CAAC;IACZ,MAAM,KAAK,GAAG,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,GAAG,CAAC,CAAC,SAAS,CAAC,CAAC;IACjF,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACvB,EAAE,IAAI,YAAY,IAAI,CAAC,SAAS,KAAK,IAAI,CAAC,WAAW,IAAI,CAAC;QAC1D,EAAE,IAAI,aAAa,IAAI,CAAC,UAAU,IAAI,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC,OAAO,IAAI,CAAC,WAAW,CAAC,CAAC,QAAQ,IAAI,CAAC,WAAW,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,IAAI,CAAC;QACzH,IAAI,IAAI,CAAC,eAAe,EAAE,CAAC;YACvB,EAAE,IAAI,eAAe,IAAI,CAAC,eAAe,IAAI,CAAC;QAClD,CAAC;QAED,KAAK,MAAM,GAAG,IAAI,aAAa,EAAE,CAAC;YAC9B,MAAM,EAAE,GAAG,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;YAChE,IAAI,EAAE,EAAE,CAAC;gBACL,MAAM,GAAG,GAAG,EAAE,CAAC,KAAK,IAAI,aAAa,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC;gBAC5D,EAAE,IAAI,OAAO,GAAG,OAAO,EAAE,CAAC,KAAK,KAAK,GAAG,MAAM,EAAE,CAAC,MAAM,IAAI,CAAC;YAC/D,CAAC;iBAAM,CAAC;gBACJ,EAAE,IAAI,OAAO,GAAG,eAAe,CAAC;YACpC,CAAC;QACL,CAAC;QACD,EAAE,IAAI,IAAI,CAAC;IACf,CAAC;IACD,OAAO,EAAE,CAAC;AACd,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,cAAc,CAAC,KAA2C;IAC5E,IAAI,IAAI,GAAe,EAAE,CAAC;IAC1B,IAAI,KAAK,CAAC,SAAS,IAAI,KAAK,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAChD,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,SAAS,EAAE,CAAC;YACjC,MAAM,GAAG,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;YAC1B,IAAI,GAAG;gBAAE,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAC5B,CAAC;IACL,CAAC;SAAM,CAAC;QACJ,IAAI,GAAG,WAAW,EAAE,CAAC;IACzB,CAAC;IAED,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACpB,OAAO;YACH,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAe,EAAE,IAAI,EAAE,gFAAgF,EAAE,CAAC;SAC/H,CAAC;IACN,CAAC;IAED,IAAI,cAAc,GAAG,EAAE,CAAC;IAExB,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;QACrB,IAAI,WAAW,GAAG,IAAI,CAAC;QACvB,IAAI,UAAU,GAAG,CAAC,CAAC;QACnB,IAAI,UAAU,GAAG,CAAC,CAAC;QACnB,IAAI,WAAW,GAAU,EAAE,CAAC;QAC5B,IAAI,YAAY,GAA+C,EAAE,CAAC;QAElE,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC;YAC1C,MAAM,SAAS,GAAG,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,WAAW,IAAI,EAAE,CAAC,CAAC;YACtD,MAAM,OAAO,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;YAElE,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBACrB,WAAW,GAAG,KAAK,CAAC;gBACpB,YAAY,CAAC,IAAI,CAAC,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,OAAO,EAAE,CAAC,CAAC;YAC9D,CAAC;YAED,KAAK,MAAM,CAAC,GAAG,EAAE,EAAE,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,WAAW,IAAI,EAAE,CAAC,EAAE,CAAC;gBAC7D,UAAU,IAAI,EAAE,CAAC,KAAK,CAAC;gBACvB,UAAU,IAAI,CAAC,CAAC;gBAChB,IAAI,EAAE,CAAC,KAAK,GAAG,aAAa,EAAE,CAAC;oBAC3B,WAAW,CAAC,IAAI,CAAC,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,MAAM,EAAE,CAAC,CAAC;gBAC7F,CAAC;YACL,CAAC;QACL,CAAC;QAED,IAAI,CAAC,WAAW,EAAE,CAAC;YACf,cAAc,IAAI,wBAAwB,CAAC,GAAG,EAAE,YAAY,CAAC,GAAG,SAAS,CAAC;QAC9E,CAAC;aAAM,CAAC;YACJ,MAAM,GAAG,GAAG,UAAU,GAAG,CAAC,CAAC,CAAC,CAAC,UAAU,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC;YACzD,MAAM,MAAM,GAAG,WAAW,CAAC,MAAM,KAAK,CAAC,CAAC;YACxC,cAAc,IAAI,sBAAsB,CAAC,GAAG,EAAE,GAAG,EAAE,MAAM,EAAE,WAAW,CAAC,GAAG,SAAS,CAAC;QACxF,CAAC;IACL,CAAC;IAED,OAAO;QACH,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAe,EAAE,IAAI,EAAE,cAAc,CAAC,IAAI,EAAE,EAAE,CAAC;KACpE,CAAC;AACN,CAAC"}
+{"version":3,"file":"get-audit-status.js","sourceRoot":"","sources":["../../../src/tools/get-audit-status.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,wBAAwB,CAAC;AAC9D,OAAO,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,4BAA4B,CAAC;AAG1E,MAAM,CAAC,MAAM,oBAAoB,GAAG,CAAC,CAAC,MAAM,CAAC;IACzC,SAAS,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,gFAAgF,CAAC;CACvI,CAAC,CAAC;AAEH,2CAA2C;AAC3C,SAAS,wBAAwB,CAAC,GAAa,EAAE,YAAwD;IACrG,IAAI,EAAE,GAAG,+BAA+B,GAAG,CAAC,QAAQ,IAAI,CAAC;IACzD,EAAE,IAAI,0CAA0C,CAAC;IAEjD,kBAAkB;IAClB,EAAE,IAAI,6BAA6B,CAAC;IACpC,EAAE,IAAI,0HAA0H,CAAC;IACjI,KAAK,MAAM,IAAI,IAAI,YAAY,EAAE,CAAC;QAC9B,EAAE,IAAI,UAAU,IAAI,CAAC,SAAS,cAAc,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC;IAC7E,CAAC;IACD,EAAE,IAAI,IAAI,CAAC;IAEX,eAAe;IACf,EAAE,IAAI,2BAA2B,CAAC;IAClC,EAAE,IAAI,iBAAiB,CAAC,GAAG,CAAC,CAAC;IAE7B,OAAO,EAAE,CAAC;AACd,CAAC;AAED,yCAAyC;AACzC,SAAS,sBAAsB,CAAC,GAAa,EAAE,GAAW,EAAE,MAAe,EAAE,WAAkB;IAC3F,IAAI,EAAE,GAAG,4BAA4B,GAAG,CAAC,QAAQ,IAAI,CAAC;IACtD,EAAE,IAAI,uCAAuC,CAAC;IAE9C,EAAE,IAAI,wBAAwB,CAAC;IAC/B,EAAE,IAAI,wBAAwB,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC;IACtD,EAAE,IAAI,iBAAiB,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,IAAI,CAAC;IAExD,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACzB,EAAE,IAAI,uBAAuB,CAAC;QAC9B,KAAK,MAAM,CAAC,IAAI,WAAW,EAAE,CAAC;YAC1B,EAAE,IAAI,YAAY,CAAC,CAAC,SAAS,KAAK,CAAC,CAAC,GAAG,YAAY,CAAC,CAAC,KAAK,MAAM,CAAC,CAAC,MAAM,IAAI,CAAC;QACjF,CAAC;IACL,CAAC;SAAM,CAAC;QACJ,EAAE,IAAI,4BAA4B,CAAC;IACvC,CAAC;IACD,EAAE,IAAI,IAAI,CAAC;IAEX,EAAE,IAAI,wBAAwB,CAAC;IAC/B,EAAE,IAAI,iBAAiB,CAAC,GAAG,CAAC,CAAC;IAE7B,OAAO,EAAE,CAAC;AACd,CAAC;AAED,SAAS,iBAAiB,CAAC,GAAa;IACpC,IAAI,EAAE,GAAG,EAAE,CAAC;IACZ,MAAM,KAAK,GAAG,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,GAAG,CAAC,CAAC,SAAS,CAAC,CAAC;IACjF,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACvB,EAAE,IAAI,YAAY,IAAI,CAAC,SAAS,KAAK,IAAI,CAAC,WAAW,IAAI,CAAC;QAC1D,EAAE,IAAI,iBAAiB,IAAI,CAAC,UAAU,IAAI,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC,OAAO,IAAI,CAAC,WAAW,CAAC,CAAC,QAAQ,IAAI,CAAC,WAAW,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,IAAI,CAAC;QAC7H,IAAI,IAAI,CAAC,eAAe,EAAE,CAAC;YACvB,EAAE,IAAI,2BAA2B,IAAI,CAAC,eAAe,IAAI,CAAC;QAC9D,CAAC;QAED,KAAK,MAAM,GAAG,IAAI,aAAa,EAAE,CAAC;YAC9B,MAAM,EAAE,GAAG,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;YAChE,IAAI,EAAE,EAAE,CAAC;gBACL,MAAM,GAAG,GAAG,EAAE,CAAC,KAAK,IAAI,aAAa,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC;gBAC5D,EAAE,IAAI,OAAO,GAAG,aAAa,EAAE,CAAC,KAAK,IAAI,GAAG,MAAM,EAAE,CAAC,MAAM,IAAI,CAAC;YACpE,CAAC;iBAAM,CAAC;gBACJ,EAAE,IAAI,OAAO,GAAG,6BAA6B,CAAC;YAClD,CAAC;QACL,CAAC;QACD,EAAE,IAAI,IAAI,CAAC;IACf,CAAC;IACD,OAAO,EAAE,CAAC;AACd,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,cAAc,CAAC,KAA2C;IAC5E,IAAI,IAAI,GAAe,EAAE,CAAC;IAC1B,IAAI,KAAK,CAAC,SAAS,IAAI,KAAK,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAChD,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,SAAS,EAAE,CAAC;YACjC,MAAM,GAAG,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;YAC1B,IAAI,GAAG;gBAAE,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAC5B,CAAC;IACL,CAAC;SAAM,CAAC;QACJ,IAAI,GAAG,WAAW,EAAE,CAAC;IACzB,CAAC;IAED,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACpB,OAAO;YACH,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAe,EAAE,IAAI,EAAE,gFAAgF,EAAE,CAAC;SAC/H,CAAC;IACN,CAAC;IAED,IAAI,cAAc,GAAG,EAAE,CAAC;IAExB,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;QACrB,IAAI,WAAW,GAAG,IAAI,CAAC;QACvB,IAAI,UAAU,GAAG,CAAC,CAAC;QACnB,IAAI,UAAU,GAAG,CAAC,CAAC;QACnB,IAAI,WAAW,GAAU,EAAE,CAAC;QAC5B,IAAI,YAAY,GAA+C,EAAE,CAAC;QAElE,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC;YAC1C,MAAM,SAAS,GAAG,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,WAAW,IAAI,EAAE,CAAC,CAAC;YACtD,MAAM,OAAO,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;YAElE,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBACrB,WAAW,GAAG,KAAK,CAAC;gBACpB,YAAY,CAAC,IAAI,CAAC,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,OAAO,EAAE,CAAC,CAAC;YAC9D,CAAC;YAED,KAAK,MAAM,CAAC,GAAG,EAAE,EAAE,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,WAAW,IAAI,EAAE,CAAC,EAAE,CAAC;gBAC7D,UAAU,IAAI,EAAE,CAAC,KAAK,CAAC;gBACvB,UAAU,IAAI,CAAC,CAAC;gBAChB,IAAI,EAAE,CAAC,KAAK,GAAG,aAAa,EAAE,CAAC;oBAC3B,WAAW,CAAC,IAAI,CAAC,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,MAAM,EAAE,CAAC,CAAC;gBAC7F,CAAC;YACL,CAAC;QACL,CAAC;QAED,IAAI,CAAC,WAAW,EAAE,CAAC;YACf,cAAc,IAAI,wBAAwB,CAAC,GAAG,EAAE,YAAY,CAAC,GAAG,SAAS,CAAC;QAC9E,CAAC;aAAM,CAAC;YACJ,MAAM,GAAG,GAAG,UAAU,GAAG,CAAC,CAAC,CAAC,CAAC,UAAU,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC;YACzD,MAAM,MAAM,GAAG,WAAW,CAAC,MAAM,KAAK,CAAC,CAAC;YACxC,cAAc,IAAI,sBAAsB,CAAC,GAAG,EAAE,GAAG,EAAE,MAAM,EAAE,WAAW,CAAC,GAAG,SAAS,CAAC;QACxF,CAAC;IACL,CAAC;IAED,OAAO;QACH,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAe,EAAE,IAAI,EAAE,cAAc,CAAC,IAAI,EAAE,EAAE,CAAC;KACpE,CAAC;AACN,CAAC"}

package/dist/src/tools/swipe.js

@@ -32,7 +32,7 @@ export async function swipe(input) {
             {
                 type: "text",
                 text: JSON.stringify({
-                    message: `Step ${input.stepIndex} recorded. Expected: "${input.expectedOutcome}". Please now evaluate this screenshot across all 5 dimensions (overlap, layout, info_clarity, style, action_result) using submit_dimension_score before proceeding.`,
+                    message: `Step ${input.stepIndex} recorded. Expected: "${input.expectedOutcome}". Please call evaluate to retrieve the first evaluation dimension and its token before proceeding.`,
                     caseName: input.caseName,
                     stepIndex: input.stepIndex,
                     expectedOutcome: input.expectedOutcome,

package/dist/src/tools/swipe.js.map

@@ -1 +1 @@
-{"version":3,"file":"swipe.js","sourceRoot":"","sources":["../../../src/tools/swipe.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,KAAK,MAAM,MAAM,sBAAsB,CAAC;AAC/C,OAAO,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAEpD,MAAM,CAAC,MAAM,WAAW,GAAG,CAAC,CAAC,MAAM,CAAC;IAChC,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,8BAA8B,CAAC;IACzE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,4BAA4B,CAAC;IACvE,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,4BAA4B,CAAC;IACrE,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,0BAA0B,CAAC;IACnE,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,uBAAuB,CAAC;IACtD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,gCAAgC,CAAC;IACjF,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,2CAA2C,CAAC;IAC7E,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,yFAAyF,CAAC;CAClI,CAAC,CAAC;AAEH,MAAM,CAAC,KAAK,UAAU,KAAK,CAAC,KAAkC;IAC1D,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;IACtF,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;QAClB,OAAO;YACH,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAe,EAAE,IAAI,EAAE,oBAAoB,MAAM,CAAC,KAAK,EAAE,EAAE,CAAC;SACjF,CAAC;IACN,CAAC;IAED,iDAAiD;IACjD,MAAM,IAAI,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC,CAAC;IAExD,MAAM,UAAU,GAAG,MAAM,MAAM,CAAC,cAAc,EAAE,CAAC;IACjD,UAAU,CACN,KAAK,CAAC,QAAQ,EACd,KAAK,CAAC,SAAS,EACf,KAAK,CAAC,WAAW,EACjB,OAAO,EACP,UAAU,CAAC,QAAQ,EACnB,EAAE,CAAC,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,KAAK,CAAC,MAAM,EAAE,EACpC,KAAK,CAAC,eAAe,CACxB,CAAC;IAEF,OAAO;QACH,OAAO,EAAE;YACL;gBACI,IAAI,EAAE,OAAgB;gBACtB,IAAI,EAAE,UAAU,CAAC,WAAW;gBAC5B,QAAQ,EAAE,WAAoB;aACjC;YACD;gBACI,IAAI,EAAE,MAAe;gBACrB,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC;oBACjB,OAAO,EAAE,QAAQ,KAAK,CAAC,SAAS,yBAAyB,KAAK,CAAC,eAAe,sKAAsK;oBACpP,QAAQ,EAAE,KAAK,CAAC,QAAQ;oBACxB,SAAS,EAAE,KAAK,CAAC,SAAS;oBAC1B,eAAe,EAAE,KAAK,CAAC,eAAe;oBACtC,WAAW,EAAE,UAAU,CAAC,KAAK;oBAC7B,YAAY,EAAE,UAAU,CAAC,MAAM;iBAClC,CAAC;aACL;SACJ;KACJ,CAAC;AACN,CAAC"}
+{"version":3,"file":"swipe.js","sourceRoot":"","sources":["../../../src/tools/swipe.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,KAAK,MAAM,MAAM,sBAAsB,CAAC;AAC/C,OAAO,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAEpD,MAAM,CAAC,MAAM,WAAW,GAAG,CAAC,CAAC,MAAM,CAAC;IAChC,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,8BAA8B,CAAC;IACzE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,4BAA4B,CAAC;IACvE,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,4BAA4B,CAAC;IACrE,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,0BAA0B,CAAC;IACnE,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,uBAAuB,CAAC;IACtD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,gCAAgC,CAAC;IACjF,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,2CAA2C,CAAC;IAC7E,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,yFAAyF,CAAC;CAClI,CAAC,CAAC;AAEH,MAAM,CAAC,KAAK,UAAU,KAAK,CAAC,KAAkC;IAC1D,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;IACtF,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;QAClB,OAAO;YACH,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAe,EAAE,IAAI,EAAE,oBAAoB,MAAM,CAAC,KAAK,EAAE,EAAE,CAAC;SACjF,CAAC;IACN,CAAC;IAED,iDAAiD;IACjD,MAAM,IAAI,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC,CAAC;IAExD,MAAM,UAAU,GAAG,MAAM,MAAM,CAAC,cAAc,EAAE,CAAC;IACjD,UAAU,CACN,KAAK,CAAC,QAAQ,EACd,KAAK,CAAC,SAAS,EACf,KAAK,CAAC,WAAW,EACjB,OAAO,EACP,UAAU,CAAC,QAAQ,EACnB,EAAE,CAAC,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,KAAK,CAAC,MAAM,EAAE,EACpC,KAAK,CAAC,eAAe,CACxB,CAAC;IAEF,OAAO;QACH,OAAO,EAAE;YACL;gBACI,IAAI,EAAE,OAAgB;gBACtB,IAAI,EAAE,UAAU,CAAC,WAAW;gBAC5B,QAAQ,EAAE,WAAoB;aACjC;YACD;gBACI,IAAI,EAAE,MAAe;gBACrB,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC;oBACjB,OAAO,EAAE,QAAQ,KAAK,CAAC,SAAS,yBAAyB,KAAK,CAAC,eAAe,qGAAqG;oBACnL,QAAQ,EAAE,KAAK,CAAC,QAAQ;oBACxB,SAAS,EAAE,KAAK,CAAC,SAAS;oBAC1B,eAAe,EAAE,KAAK,CAAC,eAAe;oBACtC,WAAW,EAAE,UAAU,CAAC,KAAK;oBAC7B,YAAY,EAAE,UAAU,CAAC,MAAM;iBAClC,CAAC;aACL;SACJ;KACJ,CAAC;AACN,CAAC"}

package/dist/src/tools/take-screenshot.js

@@ -20,7 +20,7 @@ export async function takeScreenshot(input) {
             {
                 type: "text",
                 text: JSON.stringify({
-                    message: `Step ${input.stepIndex} recorded. Please now evaluate this screenshot across all 5 dimensions (overlap, layout, info_clarity, style, action_result) using submit_dimension_score before proceeding.`,
+                    message: `Step ${input.stepIndex} recorded. Please call evaluate to retrieve the first evaluation dimension and its token before proceeding.`,
                     caseName: input.caseName,
                     stepIndex: input.stepIndex,
                     expectedOutcome: input.expectedOutcome ?? "(pure observation)",

package/dist/src/tools/take-screenshot.js.map

@@ -1 +1 @@
-{"version":3,"file":"take-screenshot.js","sourceRoot":"","sources":["../../../src/tools/take-screenshot.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,KAAK,MAAM,MAAM,sBAAsB,CAAC;AAC/C,OAAO,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAEpD,MAAM,CAAC,MAAM,oBAAoB,GAAG,CAAC,CAAC,MAAM,CAAC;IACzC,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,uBAAuB,CAAC;IACtD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,gCAAgC,CAAC;IACjF,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,mCAAmC,CAAC;IACrE,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,8EAA8E,CAAC;CAClI,CAAC,CAAC;AAEH,MAAM,CAAC,KAAK,UAAU,cAAc,CAAC,KAA2C;IAC5E,MAAM,UAAU,GAAG,MAAM,MAAM,CAAC,cAAc,EAAE,CAAC;IAEjD,UAAU,CACN,KAAK,CAAC,QAAQ,EACd,KAAK,CAAC,SAAS,EACf,KAAK,CAAC,WAAW,EACjB,YAAY,EACZ,UAAU,CAAC,QAAQ,EACnB,SAAS,EACT,KAAK,CAAC,eAAe,CACxB,CAAC;IAEF,OAAO;QACH,OAAO,EAAE;YACL;gBACI,IAAI,EAAE,OAAgB;gBACtB,IAAI,EAAE,UAAU,CAAC,WAAW;gBAC5B,QAAQ,EAAE,WAAoB;aACjC;YACD;gBACI,IAAI,EAAE,MAAe;gBACrB,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC;oBACjB,OAAO,EAAE,QAAQ,KAAK,CAAC,SAAS,8KAA8K;oBAC9M,QAAQ,EAAE,KAAK,CAAC,QAAQ;oBACxB,SAAS,EAAE,KAAK,CAAC,SAAS;oBAC1B,eAAe,EAAE,KAAK,CAAC,eAAe,IAAI,oBAAoB;oBAC9D,WAAW,EAAE,UAAU,CAAC,KAAK;oBAC7B,YAAY,EAAE,UAAU,CAAC,MAAM;iBAClC,CAAC;aACL;SACJ;KACJ,CAAC;AACN,CAAC"}
+{"version":3,"file":"take-screenshot.js","sourceRoot":"","sources":["../../../src/tools/take-screenshot.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,KAAK,MAAM,MAAM,sBAAsB,CAAC;AAC/C,OAAO,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAEpD,MAAM,CAAC,MAAM,oBAAoB,GAAG,CAAC,CAAC,MAAM,CAAC;IACzC,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,uBAAuB,CAAC;IACtD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,gCAAgC,CAAC;IACjF,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,mCAAmC,CAAC;IACrE,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,8EAA8E,CAAC;CAClI,CAAC,CAAC;AAEH,MAAM,CAAC,KAAK,UAAU,cAAc,CAAC,KAA2C;IAC5E,MAAM,UAAU,GAAG,MAAM,MAAM,CAAC,cAAc,EAAE,CAAC;IAEjD,UAAU,CACN,KAAK,CAAC,QAAQ,EACd,KAAK,CAAC,SAAS,EACf,KAAK,CAAC,WAAW,EACjB,YAAY,EACZ,UAAU,CAAC,QAAQ,EACnB,SAAS,EACT,KAAK,CAAC,eAAe,CACxB,CAAC;IAEF,OAAO;QACH,OAAO,EAAE;YACL;gBACI,IAAI,EAAE,OAAgB;gBACtB,IAAI,EAAE,UAAU,CAAC,WAAW;gBAC5B,QAAQ,EAAE,WAAoB;aACjC;YACD;gBACI,IAAI,EAAE,MAAe;gBACrB,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC;oBACjB,OAAO,EAAE,QAAQ,KAAK,CAAC,SAAS,6GAA6G;oBAC7I,QAAQ,EAAE,KAAK,CAAC,QAAQ;oBACxB,SAAS,EAAE,KAAK,CAAC,SAAS;oBAC1B,eAAe,EAAE,KAAK,CAAC,eAAe,IAAI,oBAAoB;oBAC9D,WAAW,EAAE,UAAU,CAAC,KAAK;oBAC7B,YAAY,EAAE,UAAU,CAAC,MAAM;iBAClC,CAAC;aACL;SACJ;KACJ,CAAC;AACN,CAAC"}

package/dist/src/tools/tap.js

@@ -30,7 +30,7 @@ export async function tap(input) {
             {
                 type: "text",
                 text: JSON.stringify({
-                    message: `Step ${input.stepIndex} recorded. Expected: "${input.expectedOutcome}". Please now evaluate this screenshot across all 5 dimensions (overlap, layout, info_clarity, style, action_result) using submit_dimension_score before proceeding.`,
+                    message: `Step ${input.stepIndex} recorded. Expected: "${input.expectedOutcome}". Please call evaluate to retrieve the first evaluation dimension and its token before proceeding.`,
                     caseName: input.caseName,
                     stepIndex: input.stepIndex,
                     expectedOutcome: input.expectedOutcome,

package/dist/src/tools/tap.js.map

@@ -1 +1 @@
-{"version":3,"file":"tap.js","sourceRoot":"","sources":["../../../src/tools/tap.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,KAAK,MAAM,MAAM,sBAAsB,CAAC;AAC/C,OAAO,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAEpD,MAAM,CAAC,MAAM,SAAS,GAAG,CAAC,CAAC,MAAM,CAAC;IAC9B,CAAC,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,iDAAiD,CAAC;IACvF,CAAC,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,+CAA+C,CAAC;IACrF,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,uBAAuB,CAAC;IACtD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,gCAAgC,CAAC;IACjF,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,2CAA2C,CAAC;IAC7E,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,4EAA4E,CAAC;CACrH,CAAC,CAAC;AAEH,MAAM,CAAC,KAAK,UAAU,GAAG,CAAC,KAAgC;IACtD,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC;IAClD,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;QAClB,OAAO;YACH,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAe,EAAE,IAAI,EAAE,kBAAkB,MAAM,CAAC,KAAK,EAAE,EAAE,CAAC;SAC/E,CAAC;IACN,CAAC;IAED,iDAAiD;IACjD,MAAM,IAAI,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC,CAAC;IAExD,MAAM,UAAU,GAAG,MAAM,MAAM,CAAC,cAAc,EAAE,CAAC;IACjD,UAAU,CACN,KAAK,CAAC,QAAQ,EACd,KAAK,CAAC,SAAS,EACf,KAAK,CAAC,WAAW,EACjB,KAAK,EACL,UAAU,CAAC,QAAQ,EACnB,EAAE,CAAC,EAAE,KAAK,CAAC,CAAC,EAAE,CAAC,EAAE,KAAK,CAAC,CAAC,EAAE,EAC1B,KAAK,CAAC,eAAe,CACxB,CAAC;IAEF,OAAO;QACH,OAAO,EAAE;YACL;gBACI,IAAI,EAAE,OAAgB;gBACtB,IAAI,EAAE,UAAU,CAAC,WAAW;gBAC5B,QAAQ,EAAE,WAAoB;aACjC;YACD;gBACI,IAAI,EAAE,MAAe;gBACrB,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC;oBACjB,OAAO,EAAE,QAAQ,KAAK,CAAC,SAAS,yBAAyB,KAAK,CAAC,eAAe,sKAAsK;oBACpP,QAAQ,EAAE,KAAK,CAAC,QAAQ;oBACxB,SAAS,EAAE,KAAK,CAAC,SAAS;oBAC1B,eAAe,EAAE,KAAK,CAAC,eAAe;oBACtC,WAAW,EAAE,UAAU,CAAC,KAAK;oBAC7B,YAAY,EAAE,UAAU,CAAC,MAAM;iBAClC,CAAC;aACL;SACJ;KACJ,CAAC;AACN,CAAC"}
+{"version":3,"file":"tap.js","sourceRoot":"","sources":["../../../src/tools/tap.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,KAAK,MAAM,MAAM,sBAAsB,CAAC;AAC/C,OAAO,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAEpD,MAAM,CAAC,MAAM,SAAS,GAAG,CAAC,CAAC,MAAM,CAAC;IAC9B,CAAC,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,iDAAiD,CAAC;IACvF,CAAC,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,+CAA+C,CAAC;IACrF,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,uBAAuB,CAAC;IACtD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,gCAAgC,CAAC;IACjF,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,2CAA2C,CAAC;IAC7E,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,4EAA4E,CAAC;CACrH,CAAC,CAAC;AAEH,MAAM,CAAC,KAAK,UAAU,GAAG,CAAC,KAAgC;IACtD,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC;IAClD,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;QAClB,OAAO;YACH,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAe,EAAE,IAAI,EAAE,kBAAkB,MAAM,CAAC,KAAK,EAAE,EAAE,CAAC;SAC/E,CAAC;IACN,CAAC;IAED,iDAAiD;IACjD,MAAM,IAAI,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC,CAAC;IAExD,MAAM,UAAU,GAAG,MAAM,MAAM,CAAC,cAAc,EAAE,CAAC;IACjD,UAAU,CACN,KAAK,CAAC,QAAQ,EACd,KAAK,CAAC,SAAS,EACf,KAAK,CAAC,WAAW,EACjB,KAAK,EACL,UAAU,CAAC,QAAQ,EACnB,EAAE,CAAC,EAAE,KAAK,CAAC,CAAC,EAAE,CAAC,EAAE,KAAK,CAAC,CAAC,EAAE,EAC1B,KAAK,CAAC,eAAe,CACxB,CAAC;IAEF,OAAO;QACH,OAAO,EAAE;YACL;gBACI,IAAI,EAAE,OAAgB;gBACtB,IAAI,EAAE,UAAU,CAAC,WAAW;gBAC5B,QAAQ,EAAE,WAAoB;aACjC;YACD;gBACI,IAAI,EAAE,MAAe;gBACrB,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC;oBACjB,OAAO,EAAE,QAAQ,KAAK,CAAC,SAAS,yBAAyB,KAAK,CAAC,eAAe,qGAAqG;oBACnL,QAAQ,EAAE,KAAK,CAAC,QAAQ;oBACxB,SAAS,EAAE,KAAK,CAAC,SAAS;oBAC1B,eAAe,EAAE,KAAK,CAAC,eAAe;oBACtC,WAAW,EAAE,UAAU,CAAC,KAAK;oBAC7B,YAAY,EAAE,UAAU,CAAC,MAAM;iBAClC,CAAC;aACL;SACJ;KACJ,CAAC;AACN,CAAC"}

package/dist/src/types.d.ts

@@ -23,6 +23,8 @@ export interface StepRecord {
     screenshotPath: string;
     timestamp: string;
     evaluations: Record<string, DimensionScore>;
+    currentDimIndex: number;
+    evaluationToken?: string;
 }
 export interface AuditLog {
     caseName: string;

package/dist/src/types.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/types.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,MAAM,WAAW,aAAa;IAC1B,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,cAAc;IAC3B,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,UAAU;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,KAAK,GAAG,OAAO,GAAG,YAAY,CAAC;IAC3C,WAAW,CAAC,EAAE;QAAE,CAAC,EAAE,MAAM,CAAC;QAAC,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;IACvC,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,cAAc,EAAE,MAAM,CAAC;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;CAC/C;AAED,MAAM,WAAW,QAAQ;IACrB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,UAAU,CAAC,CAAC;CACrC"}
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/types.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,MAAM,WAAW,aAAa;IAC1B,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,cAAc;IAC3B,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,UAAU;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,KAAK,GAAG,OAAO,GAAG,YAAY,CAAC;IAC3C,WAAW,CAAC,EAAE;QAAE,CAAC,EAAE,MAAM,CAAC;QAAC,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;IACvC,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,cAAc,EAAE,MAAM,CAAC;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;IAC5C,eAAe,EAAE,MAAM,CAAC;IACxB,eAAe,CAAC,EAAE,MAAM,CAAC;CAC5B;AAED,MAAM,WAAW,QAAQ;IACrB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,UAAU,CAAC,CAAC;CACrC"}

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@wa008/ui-audit-mcp",
-    "version": "2.2.1",
+    "version": "2.3.0",
     "description": "MCP server for iOS app UI evaluation and testing, powered by idb + xcrun simctl",
     "type": "module",
     "main": "dist/src/index.js",

package/dist/src/tools/get-evaluation-criteria.d.ts

@@ -1,15 +0,0 @@
-import { z } from "zod";
-export declare const getEvaluationCriteriaSchema: z.ZodObject<{
-    dimension: z.ZodOptional<z.ZodString>;
-}, "strip", z.ZodTypeAny, {
-    dimension?: string | undefined;
-}, {
-    dimension?: string | undefined;
-}>;
-export declare function getEvaluationCriteria(input: z.infer<typeof getEvaluationCriteriaSchema>): Promise<{
-    content: {
-        type: "text";
-        text: string;
-    }[];
-}>;
-//# sourceMappingURL=get-evaluation-criteria.d.ts.map

package/dist/src/tools/get-evaluation-criteria.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"get-evaluation-criteria.d.ts","sourceRoot":"","sources":["../../../src/tools/get-evaluation-criteria.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAGxB,eAAO,MAAM,2BAA2B;;;;;;EAEtC,CAAC;AAEH,wBAAsB,qBAAqB,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,2BAA2B,CAAC;;;;;GAyB7F"}

package/dist/src/tools/get-evaluation-criteria.js

@@ -1,29 +0,0 @@
-import { z } from "zod";
-import { DIMENSIONS } from "../evaluation/checklist.js";
-export const getEvaluationCriteriaSchema = z.object({
-    dimension: z.string().optional().describe("ID of the dimension (e.g., 'overlap'). Omit to list all available dimensions."),
-});
-export async function getEvaluationCriteria(input) {
-    if (!input.dimension) {
-        return {
-            content: [{
-                    type: "text",
-                    text: JSON.stringify({
-                        availableDimensions: DIMENSIONS.map(d => ({ id: d.id, name: d.name })),
-                        tip: "Call get_evaluation_criteria with a specific dimension ID to see its scoring rubric."
-                    }, null, 2),
-                }],
-        };
-    }
-    const criteria = DIMENSIONS.find(d => d.id === input.dimension);
-    if (!criteria) {
-        throw new Error(`Dimension '${input.dimension}' not found. Available dimensions: ${DIMENSIONS.map(d => d.id).join(", ")}`);
-    }
-    const markdown = `# 审计准则：${criteria.name} [${criteria.id}]\n\n` +
-        `## 🧠 核心考察点 (Description)\n${criteria.description}\n\n` +
-        `## 📊 评分细则 (Scoring Guide)\n${criteria.scoringGuide}\n`;
-    return {
-        content: [{ type: "text", text: markdown }],
-    };
-}
-//# sourceMappingURL=get-evaluation-criteria.js.map

package/dist/src/tools/get-evaluation-criteria.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"get-evaluation-criteria.js","sourceRoot":"","sources":["../../../src/tools/get-evaluation-criteria.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,UAAU,EAAE,MAAM,4BAA4B,CAAC;AAExD,MAAM,CAAC,MAAM,2BAA2B,GAAG,CAAC,CAAC,MAAM,CAAC;IAChD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,+EAA+E,CAAC;CAC7H,CAAC,CAAC;AAEH,MAAM,CAAC,KAAK,UAAU,qBAAqB,CAAC,KAAkD;IAC1F,IAAI,CAAC,KAAK,CAAC,SAAS,EAAE,CAAC;QACnB,OAAO;YACH,OAAO,EAAE,CAAC;oBACN,IAAI,EAAE,MAAe;oBACrB,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC;wBACjB,mBAAmB,EAAE,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;wBACtE,GAAG,EAAE,sFAAsF;qBAC9F,EAAE,IAAI,EAAE,CAAC,CAAC;iBACd,CAAC;SACL,CAAC;IACN,CAAC;IAED,MAAM,QAAQ,GAAG,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,KAAK,CAAC,SAAS,CAAC,CAAC;IAChE,IAAI,CAAC,QAAQ,EAAE,CAAC;QACZ,MAAM,IAAI,KAAK,CAAC,cAAc,KAAK,CAAC,SAAS,sCAAsC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAC/H,CAAC;IAED,MAAM,QAAQ,GAAG,UAAU,QAAQ,CAAC,IAAI,KAAK,QAAQ,CAAC,EAAE,OAAO;QAC3D,8BAA8B,QAAQ,CAAC,WAAW,MAAM;QACxD,+BAA+B,QAAQ,CAAC,YAAY,IAAI,CAAC;IAE7D,OAAO;QACH,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAe,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAC;KACvD,CAAC;AACN,CAAC"}

package/dist/src/tools/submit-dimension-score.d.ts

@@ -1,27 +0,0 @@
-import { z } from "zod";
-export declare const submitDimensionScoreSchema: z.ZodObject<{
-    caseName: z.ZodString;
-    stepIndex: z.ZodNumber;
-    dimension: z.ZodString;
-    score: z.ZodNumber;
-    reason: z.ZodString;
-}, "strip", z.ZodTypeAny, {
-    caseName: string;
-    stepIndex: number;
-    dimension: string;
-    score: number;
-    reason: string;
-}, {
-    caseName: string;
-    stepIndex: number;
-    dimension: string;
-    score: number;
-    reason: string;
-}>;
-export declare function submitDimensionScore(input: z.infer<typeof submitDimensionScoreSchema>): Promise<{
-    content: {
-        type: "text";
-        text: string;
-    }[];
-}>;
-//# sourceMappingURL=submit-dimension-score.d.ts.map

package/dist/src/tools/submit-dimension-score.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"submit-dimension-score.d.ts","sourceRoot":"","sources":["../../../src/tools/submit-dimension-score.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAIxB,eAAO,MAAM,0BAA0B;;;;;;;;;;;;;;;;;;EAMrC,CAAC;AAEH,wBAAsB,oBAAoB,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,0BAA0B,CAAC;;;;;GA0B3F"}

package/dist/src/tools/submit-dimension-score.js

@@ -1,35 +0,0 @@
-import { z } from "zod";
-import { recordScore, readLog } from "../logger/audit-log.js";
-import { REQUIRED_DIMS } from "../evaluation/checklist.js";
-export const submitDimensionScoreSchema = z.object({
-    caseName: z.string().describe("Name of the test case"),
-    stepIndex: z.number().int().positive().describe("Step index this evaluation belongs to"),
-    dimension: z.string().describe(`The dimension ID (must be one of: ${REQUIRED_DIMS.join(", ")})`),
-    score: z.number().int().min(0).max(10).describe("Score from 0 (worst) to 10 (perfect)"),
-    reason: z.string().describe("Analysis and reason for this score"),
-});
-export async function submitDimensionScore(input) {
-    if (!REQUIRED_DIMS.includes(input.dimension)) {
-        throw new Error(`Invalid dimension '${input.dimension}'. Must be one of: ${REQUIRED_DIMS.join(", ")}`);
-    }
-    recordScore(input.caseName, input.stepIndex, input.dimension, input.score, input.reason);
-    // After writing, calculate remaining dimensions for this step to provide a helpful hint
-    const log = readLog(input.caseName);
-    const step = log?.steps[input.stepIndex];
-    const evaluated = step ? Object.keys(step.evaluations || {}) : [];
-    const remaining = REQUIRED_DIMS.filter(d => !evaluated.includes(d));
-    let hint;
-    if (remaining.length === 0) {
-        hint = `All 5 dimensions for Step ${input.stepIndex} are complete. You may call get_audit_status to review progress or proceed to the next action.`;
-    }
-    else {
-        hint = `${remaining.length} dimension(s) still needed for Step ${input.stepIndex}: ${remaining.join(", ")}. Please evaluate them before moving to the next action.`;
-    }
-    return {
-        content: [{
-                type: "text",
-                text: `Recorded [${input.dimension}] for ${input.caseName} Step ${input.stepIndex} — Score: ${input.score}. ${hint}`
-            }]
-    };
-}
-//# sourceMappingURL=submit-dimension-score.js.map

package/dist/src/tools/submit-dimension-score.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"submit-dimension-score.js","sourceRoot":"","sources":["../../../src/tools/submit-dimension-score.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,WAAW,EAAE,OAAO,EAAE,MAAM,wBAAwB,CAAC;AAC9D,OAAO,EAAE,aAAa,EAAE,MAAM,4BAA4B,CAAC;AAE3D,MAAM,CAAC,MAAM,0BAA0B,GAAG,CAAC,CAAC,MAAM,CAAC;IAC/C,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,uBAAuB,CAAC;IACtD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,uCAAuC,CAAC;IACxF,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,qCAAqC,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC;IAChG,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,QAAQ,CAAC,sCAAsC,CAAC;IACvF,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,oCAAoC,CAAC;CACpE,CAAC,CAAC;AAEH,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,KAAiD;IACxF,IAAI,CAAC,aAAa,CAAC,QAAQ,CAAC,KAAK,CAAC,SAAS,CAAC,EAAE,CAAC;QAC3C,MAAM,IAAI,KAAK,CAAC,sBAAsB,KAAK,CAAC,SAAS,sBAAsB,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAC3G,CAAC;IAED,WAAW,CAAC,KAAK,CAAC,QAAQ,EAAE,KAAK,CAAC,SAAS,EAAE,KAAK,CAAC,SAAS,EAAE,KAAK,CAAC,KAAK,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;IAEzF,wFAAwF;IACxF,MAAM,GAAG,GAAG,OAAO,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;IACpC,MAAM,IAAI,GAAG,GAAG,EAAE,KAAK,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;IACzC,MAAM,SAAS,GAAG,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,WAAW,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;IAClE,MAAM,SAAS,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;IAEpE,IAAI,IAAY,CAAC;IACjB,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACzB,IAAI,GAAG,6BAA6B,KAAK,CAAC,SAAS,gGAAgG,CAAC;IACxJ,CAAC;SAAM,CAAC;QACJ,IAAI,GAAG,GAAG,SAAS,CAAC,MAAM,uCAAuC,KAAK,CAAC,SAAS,KAAK,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,0DAA0D,CAAC;IACxK,CAAC;IAED,OAAO;QACH,OAAO,EAAE,CAAC;gBACN,IAAI,EAAE,MAAe;gBACrB,IAAI,EAAE,aAAa,KAAK,CAAC,SAAS,SAAS,KAAK,CAAC,QAAQ,SAAS,KAAK,CAAC,SAAS,aAAa,KAAK,CAAC,KAAK,KAAK,IAAI,EAAE;aACvH,CAAC;KACL,CAAC;AACN,CAAC"}