@wa008/ui-audit-mcp 2.0.0 → 2.2.0

package/dist/src/evaluation/checklist.d.ts

@@ -1,5 +1,5 @@
 import { ChecklistItem } from "../types.js";
-/** System-wide UI quality audit dimensions */
+/** System-wide UI quality audit dimensions — strict scoring */
 export declare const DIMENSIONS: ChecklistItem[];
 export declare const REQUIRED_DIMS: string[];
 export declare const PASSING_SCORE = 8;

package/dist/src/evaluation/checklist.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"checklist.d.ts","sourceRoot":"","sources":["../../../src/evaluation/checklist.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAE5C,8CAA8C;AAC9C,eAAO,MAAM,UAAU,EAAE,aAAa,EAyErC,CAAC;AAEF,eAAO,MAAM,aAAa,UAA4B,CAAC;AACvD,eAAO,MAAM,aAAa,IAAI,CAAC"}
+{"version":3,"file":"checklist.d.ts","sourceRoot":"","sources":["../../../src/evaluation/checklist.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAE5C,+DAA+D;AAC/D,eAAO,MAAM,UAAU,EAAE,aAAa,EA+GrC,CAAC;AAEF,eAAO,MAAM,aAAa,UAA4B,CAAC;AACvD,eAAO,MAAM,aAAa,IAAI,CAAC"}

package/dist/src/evaluation/checklist.js

@@ -1,68 +1,104 @@
-/** System-wide UI quality audit dimensions */
+/** System-wide UI quality audit dimensions — strict scoring */
 export const DIMENSIONS = [
     {
         id: "overlap",
         name: "Element Overlap & Safe Areas",
-        description: "Focus ONLY on whether UI elements physically cover or obscure each other. " +
-            "Examine the following in order:\n" +
-            "1. STATUS BAR: Does any app content (header, button, text) intrude into the system status bar area (time, battery, signal icons)? Even 1px overlap is a critical failure.\n" +
-            "2. NOTCH / DYNAMIC ISLAND: Is any content hidden behind the notch or Dynamic Island?\n" +
-            "3. HOME INDICATOR: Does any bottom content overlap with the home indicator bar?\n" +
-            "4. INTER-ELEMENT OVERLAP: Do any buttons, text fields, images, or cards overlap each other within the app content area?\n" +
-            "5. MODAL / POPUP: If a modal or popup is present, does it unintentionally cover critical information behind it?\n" +
-            "NOTE: Do NOT evaluate spacing aesthetics (that is layout), color choices (that is style), or text readability (that is info_clarity).",
-        scoringGuide: "0-3: Critical overlap detected — elements cover system UI or each other, causing functional obstruction.\n" +
-            "4-6: Minor overlap exists but does not block core functionality.\n" +
-            "7-8: No overlap detected, but margins against safe areas are uncomfortably tight (< 4pt).\n" +
-            "9-10: Perfect — all elements are fully within safe areas with comfortable margins, zero overlap.",
+        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 CORNER: Is there ANY element (button, icon, badge) that sits within or touches the status bar zone or Dynamic Island area? " +
+            "Look very carefully at the vertical position of the topmost UI element on the right side. " +
+            "If its top edge is within ~54pt of the screen top on a notch/Dynamic Island device, it is overlapping. This is a CRITICAL failure.\n" +
+            "2. TOP-LEFT CORNER: Same check — any element intruding into the status bar time/signal area?\n" +
+            "3. STATUS BAR ZONE: Does ANY app content (text, header background, button) extend behind the system time, battery, or signal indicators? Even partial intrusion = failure.\n" +
+            "4. BOTTOM SAFE AREA: Does any content overlap with the home indicator bar?\n" +
+            "5. INTER-ELEMENT: Do any two app elements (buttons, text, images, cards) visually overlap each other?\n" +
+            "6. MODAL/POPUP: If present, does it unintentionally cover critical content?\n\n" +
+            "SCORING ATTITUDE: If you are unsure whether something overlaps, score LOW. " +
+            "The purpose of this audit is to catch problems, not to give the benefit of the doubt.\n" +
+            "NOTE: Do NOT evaluate spacing aesthetics (layout), colors (style), or text meaning (info_clarity).",
+        scoringGuide: "0-2: Any element overlaps status bar, Dynamic Island, notch, or home indicator. Or two elements clearly cover each other.\n" +
+            "3-4: An element is dangerously close to a safe area boundary (<8pt margin) — likely overlapping on some devices.\n" +
+            "5-6: All elements are within safe areas but margins are tight (<12pt). Risk of overlap on larger text sizes or different devices.\n" +
+            "7-8: All elements are clearly within safe areas with comfortable margins (>=16pt). 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 after checking every corner.",
     },
     {
         id: "layout",
         name: "Layout & Spatial Balance",
-        description: "Focus ONLY on how the screen space is utilized and whether elements are spatially well-arranged. " +
-            "Examine the following in order:\n" +
-            "1. LETTERBOXING: Does the app appear to run inside a smaller box with large black/empty bars at the top and bottom? This is a critical failure (score 0).\n" +
-            "2. SCREEN FILL: Does the content naturally fill the available screen area without feeling cramped or empty?\n" +
-            "3. WHITESPACE BALANCE: Is the whitespace distributed evenly? Are there large, unexplained empty gaps on one side while the other side is dense?\n" +
-            "4. ALIGNMENT: Are related elements (labels, inputs, buttons) aligned to a consistent vertical or horizontal axis? Look for jagged left edges or uneven column widths.\n" +
-            "5. SPACING CONSISTENCY: Are the gaps between similar elements (e.g., list items, card margins) uniform?\n" +
-            "NOTE: Do NOT evaluate whether elements cover each other (that is overlap), color or font choices (that is style), or text meaning (that is info_clarity).",
-        scoringGuide: "0-3: Letterboxing detected, or layout is severely broken (elements piled up, massive empty areas).\n" +
-            "4-6: Layout is functional but has noticeable imbalance — uneven spacing, misaligned elements, or wasted space.\n" +
-            "7-8: Generally well-structured, but minor spacing or alignment imperfections exist.\n" +
-            "9-10: Perfect — balanced whitespace, consistent spacing, strong alignment, content fills screen naturally.",
+        description: "You are a strict QA inspector for layout quality. Your job is to find spatial problems.\n" +
+            "Assume the layout has issues until you verify otherwise.\n\n" +
+            "CHECK EACH ASPECT:\n" +
+            "1. LETTERBOXING: Does the app appear to run inside a smaller box with black/empty bars? This is score 0, no exceptions.\n" +
+            "2. SCREEN UTILIZATION: Is there a large empty area that serves no purpose? Is content crammed into only half the screen while the rest is wasted?\n" +
+            "3. ALIGNMENT: Pick any two similar elements (e.g., two cards, two labels). Are their left edges aligned? Are the gaps between them equal? Misalignment = deduction.\n" +
+            "4. SPACING: Are gaps between elements consistent? Measure visually: if card A-to-B gap looks different from card B-to-C gap, that is a problem.\n" +
+            "5. RESPONSIVE FIT: Does the layout feel designed for this screen size, or does it look like it was designed for a different device?\n\n" +
+            "SCORING ATTITUDE: Do not be generous. A 'basically fine' layout is a 6, not an 8.\n" +
+            "NOTE: Do NOT evaluate overlap (that is overlap), colors (that is style), or text meaning (that is info_clarity).",
+        scoringGuide: "0-2: Letterboxing detected, or layout is severely broken (elements piled up, massive empty areas, content off-screen).\n" +
+            "3-4: Major layout issues — clearly uneven spacing, significant misalignment, or large wasted areas.\n" +
+            "5-6: Layout is functional but has noticeable imperfections — some uneven spacing or minor alignment issues. This is the 'OK but not great' range.\n" +
+            "7-8: Well-structured layout with only very minor imperfections that require careful inspection to notice.\n" +
+            "9-10: Flawless — pixel-perfect alignment, perfectly consistent spacing, optimal screen utilization. Extremely rare.",
     },
     {
         id: "info_clarity",
         name: "Information Clarity & Readability",
-        description: "Focus ONLY on whether a user can quickly understand what this screen communicates and what action to take. " +
-            "Examine the following in order:\n" +
-            "1. PRIMARY CTA: Is there a clear primary call-to-action? Can the user instantly identify what to do next?\n" +
-            "2. INFORMATION HIERARCHY: Are headings visually dominant over body text? Is the most important information the most prominent?\n" +
-            "3. TEXT TRUNCATION: Is any text cut off with '...' or overflowing its container, losing meaningful content?\n" +
-            "4. LABEL AMBIGUITY: Are button labels, icon meanings, or tab names confusing or misleading? Would a first-time user understand what each element does?\n" +
-            "5. CONTRAST & READABILITY: Is text legible against its background? Are there low-contrast text areas that are hard to read?\n" +
-            "NOTE: Do NOT evaluate spacing or alignment (that is layout), element overlap (that is overlap), or color harmony (that is style).",
-        scoringGuide: "0-3: Key information is impossible to find, CTA is absent or hidden, or labels are severely misleading.\n" +
-            "4-6: Information is present but requires effort to parse — weak hierarchy, some ambiguous labels, or minor truncation.\n" +
-            "7-8: Generally clear with good hierarchy, but 1-2 minor issues (e.g., one slightly ambiguous icon).\n" +
-            "9-10: Perfect — immediate clarity, strong hierarchy, unambiguous labels, fully legible text, obvious CTA.",
+        description: "You are a strict QA inspector for information clarity. Your job is to find confusing elements.\n" +
+            "Imagine you are a FIRST-TIME user of this app who has never seen it before.\n\n" +
+            "CHECK EACH ASPECT:\n" +
+            "1. FIRST IMPRESSION: Within 2 seconds of looking at this screen, can you tell what it does and what the main action is? If not, that is a problem.\n" +
+            "2. BUTTON LABELS: Read every button/link text. Would a first-time user understand what each does? 'Submit' is clearer than a bare icon.\n" +
+            "3. HIERARCHY: Is the most important element (title, CTA) visually the most prominent? Or is secondary info competing for attention?\n" +
+            "4. TEXT TRUNCATION: Is any text cut off with '...' or overflowing? Even one truncated label = deduction.\n" +
+            "5. CONTRAST: Is all text easily readable against its background? Check small/light text especially.\n" +
+            "6. ICON MEANING: Are there any icons without labels that could be ambiguous?\n\n" +
+            "SCORING ATTITUDE: If any single element is confusing, cap the score at 7. " +
+            "Multiple unclear elements = score 5 or below.\n" +
+            "NOTE: Do NOT evaluate spacing (layout), overlap (overlap), or color harmony (style).",
+        scoringGuide: "0-2: Screen purpose is unclear, CTA is missing/hidden, or labels are severely misleading.\n" +
+            "3-4: Core purpose is guessable but requires effort. Multiple ambiguous labels or icons.\n" +
+            "5-6: Generally understandable but has 1-2 confusing elements (ambiguous icon, truncated text, weak hierarchy).\n" +
+            "7-8: Clear and well-organized with only one very minor issue. First-time user would understand immediately.\n" +
+            "9-10: Crystal clear — perfect hierarchy, unambiguous labels, no truncation, excellent contrast. Every element self-explanatory.",
     },
     {
         id: "style",
         name: "Visual Style & Consistency",
-        description: "Focus ONLY on the visual aesthetics and design consistency of this single screen. " +
-            "Examine the following in order:\n" +
-            "1. COLOR PALETTE: Are the colors harmonious? Do primary, secondary, and accent colors feel intentional and cohesive, or do they clash?\n" +
-            "2. TYPOGRAPHY: Are font sizes, weights, and families consistent across the screen? Do headings, body text, and captions follow a clear typographic scale?\n" +
-            "3. COMPONENT CONSISTENCY: Do similar components (buttons, cards, input fields) share the same visual treatment (corner radius, shadow, border style)?\n" +
-            "4. ICONOGRAPHY: Do icons share a consistent style (outline vs filled, stroke width, size)?\n" +
-            "5. OVERALL POLISH: Does the screen feel professionally designed, or does it look like a rough prototype with placeholder styling?\n" +
-            "NOTE: Do NOT evaluate information meaning (that is info_clarity), spatial arrangement (that is layout), or element occlusion (that is overlap).",
-        scoringGuide: "0-3: No coherent visual style — clashing colors, mixed icon styles, inconsistent component design.\n" +
-            "4-6: A basic style exists but with noticeable inconsistencies (e.g., two different button styles, uneven font weights).\n" +
-            "7-8: Mostly consistent and polished, with only minor deviations.\n" +
-            "9-10: Perfect — unified color scheme, consistent typography, cohesive component design, professional polish.",
+        description: "You are a strict QA inspector for visual design consistency. Your job is to find inconsistencies.\n" +
+            "Compare every element against every other element of the same type.\n\n" +
+            "CHECK EACH ASPECT:\n" +
+            "1. COLOR PALETTE: Count the number of distinct colors used. Do they feel intentional and harmonious, or random?\n" +
+            "2. TYPOGRAPHY: Are there more than 2 font sizes that don't follow a clear scale? Mixed font weights without reason?\n" +
+            "3. COMPONENT STYLING: Compare all buttons — same corner radius? Same padding? Compare all cards — same shadow, border, background?\n" +
+            "4. ICONOGRAPHY: Are all icons the same style (outline vs filled, same stroke width)? Mixed styles = deduction.\n" +
+            "5. POLISH: Does it look professionally designed or like a prototype? Placeholder content (e.g., 'Lorem ipsum', stock icons) = major deduction.\n\n" +
+            "SCORING ATTITUDE: Consistency is binary per element — if two buttons look different, they are inconsistent, period.\n" +
+            "NOTE: Do NOT evaluate text meaning (info_clarity), spatial layout (layout), or overlap (overlap).",
+        scoringGuide: "0-2: No design system — clashing colors, random fonts, inconsistent components everywhere.\n" +
+            "3-4: Some design intent visible but multiple clear inconsistencies (different button styles, mixed icon sets).\n" +
+            "5-6: Basic consistency exists but with noticeable deviations. 'Mostly OK' = 6 at best.\n" +
+            "7-8: Strong consistency with only 1 minor deviation requiring careful inspection.\n" +
+            "9-10: Pixel-perfect design system adherence. Every element cohesive. Extremely rare.",
+    },
+    {
+        id: "action_result",
+        name: "Action Result Verification",
+        description: "You are a strict QA inspector verifying whether the action actually worked.\n" +
+            "The step's expectedOutcome describes what SHOULD have happened. Compare the screenshot against it.\n\n" +
+            "CHECK EACH ASPECT:\n" +
+            "1. SCREEN TRANSITION: If the action was a navigation tap, did the correct destination screen appear? Wrong screen = score 0.\n" +
+            "2. STATE CHANGE: If the action should toggle/add/remove something, is that change visually confirmed in the screenshot?\n" +
+            "3. ERROR PRESENCE: Are there any error dialogs, crash screens, or infinite loading spinners? These indicate failure.\n" +
+            "4. NO RESPONSE: Does the screen look identical to before the action? If so, the action likely missed or failed.\n\n" +
+            "SCORING ATTITUDE: If the expectedOutcome is not clearly achieved, score LOW. Partial success is still a problem.\n" +
+            "If this step is a pure observation (no expectedOutcome), score 10.",
+        scoringGuide: "0-2: Action clearly failed — wrong screen, error dialog, crash, or no UI response at all.\n" +
+            "3-4: Action had some effect but the result does not match expectedOutcome (e.g., navigated to wrong tab).\n" +
+            "5-6: Action partially achieved the goal but with unexpected side effects or missing expected elements.\n" +
+            "7-8: Action succeeded — the expected screen/state is present but with minor discrepancies.\n" +
+            "9-10: Action fully succeeded — screenshot exactly matches expectedOutcome. Or no expectedOutcome (pure observation).",
     },
 ];
 export const REQUIRED_DIMS = DIMENSIONS.map(d => d.id);

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

@@ -1 +1 @@
-{"version":3,"file":"checklist.js","sourceRoot":"","sources":["../../../src/evaluation/checklist.ts"],"names":[],"mappings":"AAEA,8CAA8C;AAC9C,MAAM,CAAC,MAAM,UAAU,GAAoB;IACvC;QACI,EAAE,EAAE,SAAS;QACb,IAAI,EAAE,8BAA8B;QACpC,WAAW,EACP,4EAA4E;YAC5E,mCAAmC;YACnC,6KAA6K;YAC7K,wFAAwF;YACxF,mFAAmF;YACnF,2HAA2H;YAC3H,mHAAmH;YACnH,uIAAuI;QAC3I,YAAY,EACR,4GAA4G;YAC5G,oEAAoE;YACpE,6FAA6F;YAC7F,kGAAkG;KACzG;IACD;QACI,EAAE,EAAE,QAAQ;QACZ,IAAI,EAAE,0BAA0B;QAChC,WAAW,EACP,mGAAmG;YACnG,mCAAmC;YACnC,6JAA6J;YAC7J,+GAA+G;YAC/G,mJAAmJ;YACnJ,yKAAyK;YACzK,2GAA2G;YAC3G,2JAA2J;QAC/J,YAAY,EACR,sGAAsG;YACtG,kHAAkH;YAClH,uFAAuF;YACvF,4GAA4G;KACnH;IACD;QACI,EAAE,EAAE,cAAc;QAClB,IAAI,EAAE,mCAAmC;QACzC,WAAW,EACP,6GAA6G;YAC7G,mCAAmC;YACnC,6GAA6G;YAC7G,kIAAkI;YAClI,+GAA+G;YAC/G,0JAA0J;YAC1J,+HAA+H;YAC/H,mIAAmI;QACvI,YAAY,EACR,2GAA2G;YAC3G,0HAA0H;YAC1H,uGAAuG;YACvG,2GAA2G;KAClH;IACD;QACI,EAAE,EAAE,OAAO;QACX,IAAI,EAAE,4BAA4B;QAClC,WAAW,EACP,oFAAoF;YACpF,mCAAmC;YACnC,0IAA0I;YAC1I,6JAA6J;YAC7J,yJAAyJ;YACzJ,8FAA8F;YAC9F,qIAAqI;YACrI,iJAAiJ;QACrJ,YAAY,EACR,sGAAsG;YACtG,2HAA2H;YAC3H,oEAAoE;YACpE,8GAA8G;KACrH;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,8FAA8F;YAC9F,+EAA+E;YAC/E,iEAAiE;YACjE,0IAA0I;YAC1I,4FAA4F;YAC5F,sIAAsI;YACtI,gGAAgG;YAChG,8KAA8K;YAC9K,8EAA8E;YAC9E,yGAAyG;YACzG,iFAAiF;YACjF,6EAA6E;YAC7E,yFAAyF;YACzF,oGAAoG;QACxG,YAAY,EACR,6HAA6H;YAC7H,oHAAoH;YACpH,qIAAqI;YACrI,yHAAyH;YACzH,uJAAuJ;KAC9J;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/index.d.ts

@@ -8,7 +8,7 @@
  *
  * 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):
+ *   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.

package/dist/src/index.js

@@ -8,7 +8,7 @@
  *
  * 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):
+ *   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.
@@ -38,9 +38,9 @@ server.tool("swipe", "Swipe on screen (ratio 0-1). Automatically captures a scre
 // ─── 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.", getEvaluationCriteriaSchema.shape, async (args) => getEvaluationCriteria(args));
+    "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 four dimensions: overlap, layout, info_clarity, style. " +
+    "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("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));

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,2EAA2E,EAC3E,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,4FAA4F;IAC5F,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,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"}

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

@@ -5,6 +5,6 @@ export declare function readAllLogs(): AuditLog[];
 export declare function recordStep(caseName: string, stepIndex: number, description: string, actionType: "tap" | "swipe" | "screenshot", screenshotPath: string, coordinates?: {
     x: number;
     y: number;
-}): void;
+}, expectedOutcome?: string): void;
 export declare function recordScore(caseName: string, stepIndex: number, dimension: string, score: number, reason: string): void;
 //# sourceMappingURL=audit-log.d.ts.map

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,GACvC,IAAI,CAsBN;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,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"}

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

@@ -46,7 +46,7 @@ export function readAllLogs() {
     }
     return logs;
 }
-export function recordStep(caseName, stepIndex, description, actionType, screenshotPath, coordinates) {
+export function recordStep(caseName, stepIndex, description, actionType, screenshotPath, coordinates, expectedOutcome) {
     let log = readLog(caseName);
     if (!log) {
         log = {
@@ -61,6 +61,7 @@ export function recordStep(caseName, stepIndex, description, actionType, screens
         actionType,
         screenshotPath,
         coordinates,
+        expectedOutcome,
         timestamp: new Date().toISOString(),
         evaluations: existingStep ? existingStep.evaluations : {},
     };

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;IAEtC,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,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;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"}

package/dist/src/tools/get-audit-status.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"get-audit-status.d.ts","sourceRoot":"","sources":["../../../src/tools/get-audit-status.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAKxB,eAAO,MAAM,oBAAoB;;;;;;EAE/B,CAAC;AAoEH,wBAAsB,cAAc,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,oBAAoB,CAAC;;;;;GAwD/E"}
+{"version":3,"file":"get-audit-status.d.ts","sourceRoot":"","sources":["../../../src/tools/get-audit-status.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAKxB,eAAO,MAAM,oBAAoB;;;;;;EAE/B,CAAC;AAuEH,wBAAsB,cAAc,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,oBAAoB,CAAC;;;;;GAwD/E"}

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

@@ -47,6 +47,9 @@ function renderStepDetails(log) {
     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`;
+        if (step.expectedOutcome) {
+            md += `- **预期结果**: ${step.expectedOutcome}\n`;
+        }
         for (const dim of REQUIRED_DIMS) {
             const ev = step.evaluations ? step.evaluations[dim] : undefined;
             if (ev) {

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;QAEzH,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,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"}

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

@@ -20,7 +20,7 @@ export async function submitDimensionScore(input) {
     const remaining = REQUIRED_DIMS.filter(d => !evaluated.includes(d));
     let hint;
     if (remaining.length === 0) {
-        hint = `All 4 dimensions for Step ${input.stepIndex} are complete. You may call get_audit_status to review progress or proceed to the next action.`;
+        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.`;

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

@@ -7,10 +7,12 @@ export declare const swipeSchema: z.ZodObject<{
     caseName: z.ZodString;
     stepIndex: z.ZodNumber;
     description: z.ZodString;
+    expectedOutcome: z.ZodString;
 }, "strip", z.ZodTypeAny, {
     caseName: string;
     stepIndex: number;
     description: string;
+    expectedOutcome: string;
     startX: number;
     startY: number;
     endX: number;
@@ -19,6 +21,7 @@ export declare const swipeSchema: z.ZodObject<{
     caseName: string;
     stepIndex: number;
     description: string;
+    expectedOutcome: string;
     startX: number;
     startY: number;
     endX: number;

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

@@ -1 +1 @@
-{"version":3,"file":"swipe.d.ts","sourceRoot":"","sources":["../../../src/tools/swipe.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAIxB,eAAO,MAAM,WAAW;;;;;;;;;;;;;;;;;;;;;;;;EAQtB,CAAC;AAEH,wBAAsB,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,WAAW,CAAC;;;;;;;;;;;;GAwC7D"}
+{"version":3,"file":"swipe.d.ts","sourceRoot":"","sources":["../../../src/tools/swipe.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAIxB,eAAO,MAAM,WAAW;;;;;;;;;;;;;;;;;;;;;;;;;;;EAStB,CAAC;AAEH,wBAAsB,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,WAAW,CAAC;;;;;;;;;;;;GA0C7D"}

package/dist/src/tools/swipe.js

@@ -9,6 +9,7 @@ export const swipeSchema = z.object({
     caseName: z.string().describe("Name of the test case"),
     stepIndex: z.number().int().positive().describe("Current step index (e.g. 1, 2)"),
     description: z.string().describe("Description of the action being performed"),
+    expectedOutcome: z.string().describe("What should happen after this swipe (e.g. 'List should scroll down to show more items')"),
 });
 export async function swipe(input) {
     const result = await device.swipe(input.startX, input.startY, input.endX, input.endY);
@@ -20,7 +21,7 @@ export async function swipe(input) {
     // Wait for UI to settle before taking screenshot
     await new Promise(resolve => setTimeout(resolve, 2000));
     const screenshot = await device.takeScreenshot();
-    recordStep(input.caseName, input.stepIndex, input.description, "swipe", screenshot.filePath, { x: input.startX, y: input.startY });
+    recordStep(input.caseName, input.stepIndex, input.description, "swipe", screenshot.filePath, { x: input.startX, y: input.startY }, input.expectedOutcome);
     return {
         content: [
             {
@@ -31,9 +32,10 @@ export async function swipe(input) {
             {
                 type: "text",
                 text: JSON.stringify({
-                    message: `Step ${input.stepIndex} recorded. Please now evaluate this screenshot across all 4 dimensions (overlap, layout, info_clarity, style) using get_evaluation_criteria and submit_dimension_score before proceeding.`,
+                    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.`,
                     caseName: input.caseName,
                     stepIndex: input.stepIndex,
+                    expectedOutcome: input.expectedOutcome,
                     screenWidth: screenshot.width,
                     screenHeight: screenshot.height,
                 }),

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;CAChF,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,CACvC,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,2LAA2L;oBAC3N,QAAQ,EAAE,KAAK,CAAC,QAAQ;oBACxB,SAAS,EAAE,KAAK,CAAC,SAAS;oBAC1B,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,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"}

package/dist/src/tools/take-screenshot.d.ts

@@ -3,14 +3,17 @@ export declare const takeScreenshotSchema: z.ZodObject<{
     caseName: z.ZodString;
     stepIndex: z.ZodNumber;
     description: z.ZodString;
+    expectedOutcome: z.ZodOptional<z.ZodString>;
 }, "strip", z.ZodTypeAny, {
     caseName: string;
     stepIndex: number;
     description: string;
+    expectedOutcome?: string | undefined;
 }, {
     caseName: string;
     stepIndex: number;
     description: string;
+    expectedOutcome?: string | undefined;
 }>;
 export declare function takeScreenshot(input: z.infer<typeof takeScreenshotSchema>): Promise<{
     content: ({

package/dist/src/tools/take-screenshot.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"take-screenshot.d.ts","sourceRoot":"","sources":["../../../src/tools/take-screenshot.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAIxB,eAAO,MAAM,oBAAoB;;;;;;;;;;;;EAI/B,CAAC;AAEH,wBAAsB,cAAc,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,oBAAoB,CAAC;;;;;;;;;;;;GA8B/E"}
+{"version":3,"file":"take-screenshot.d.ts","sourceRoot":"","sources":["../../../src/tools/take-screenshot.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAIxB,eAAO,MAAM,oBAAoB;;;;;;;;;;;;;;;EAK/B,CAAC;AAEH,wBAAsB,cAAc,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,oBAAoB,CAAC;;;;;;;;;;;;GAiC/E"}

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

@@ -5,10 +5,11 @@ export const takeScreenshotSchema = z.object({
     caseName: z.string().describe("Name of the test case"),
     stepIndex: z.number().int().positive().describe("Current step index (e.g. 1, 2)"),
     description: z.string().describe("Description of the observed state"),
+    expectedOutcome: z.string().optional().describe("Optional: what you expect to see on screen. Omit for pure observation steps."),
 });
 export async function takeScreenshot(input) {
     const screenshot = await device.takeScreenshot();
-    recordStep(input.caseName, input.stepIndex, input.description, "screenshot", screenshot.filePath);
+    recordStep(input.caseName, input.stepIndex, input.description, "screenshot", screenshot.filePath, undefined, input.expectedOutcome);
     return {
         content: [
             {
@@ -19,9 +20,10 @@ export async function takeScreenshot(input) {
             {
                 type: "text",
                 text: JSON.stringify({
-                    message: `Step ${input.stepIndex} recorded. Please now evaluate this screenshot across all 4 dimensions (overlap, layout, info_clarity, style) using get_evaluation_criteria and submit_dimension_score before proceeding.`,
+                    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.`,
                     caseName: input.caseName,
                     stepIndex: input.stepIndex,
+                    expectedOutcome: input.expectedOutcome ?? "(pure observation)",
                     screenWidth: screenshot.width,
                     screenHeight: screenshot.height,
                 }),

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;CACxE,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,CACtB,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,2LAA2L;oBAC3N,QAAQ,EAAE,KAAK,CAAC,QAAQ;oBACxB,SAAS,EAAE,KAAK,CAAC,SAAS;oBAC1B,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,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"}

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

@@ -5,16 +5,19 @@ export declare const tapSchema: z.ZodObject<{
     caseName: z.ZodString;
     stepIndex: z.ZodNumber;
     description: z.ZodString;
+    expectedOutcome: z.ZodString;
 }, "strip", z.ZodTypeAny, {
     caseName: string;
     stepIndex: number;
     description: string;
+    expectedOutcome: string;
     x: number;
     y: number;
 }, {
     caseName: string;
     stepIndex: number;
     description: string;
+    expectedOutcome: string;
     x: number;
     y: number;
 }>;

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

@@ -1 +1 @@
-{"version":3,"file":"tap.d.ts","sourceRoot":"","sources":["../../../src/tools/tap.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAIxB,eAAO,MAAM,SAAS;;;;;;;;;;;;;;;;;;EAMpB,CAAC;AAEH,wBAAsB,GAAG,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,SAAS,CAAC;;;;;;;;;;;;GAwCzD"}
+{"version":3,"file":"tap.d.ts","sourceRoot":"","sources":["../../../src/tools/tap.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAIxB,eAAO,MAAM,SAAS;;;;;;;;;;;;;;;;;;;;;EAOpB,CAAC;AAEH,wBAAsB,GAAG,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,SAAS,CAAC;;;;;;;;;;;;GA0CzD"}

package/dist/src/tools/tap.js

@@ -7,6 +7,7 @@ export const tapSchema = z.object({
     caseName: z.string().describe("Name of the test case"),
     stepIndex: z.number().int().positive().describe("Current step index (e.g. 1, 2)"),
     description: z.string().describe("Description of the action being performed"),
+    expectedOutcome: z.string().describe("What should happen after this tap (e.g. 'Should navigate to login screen')"),
 });
 export async function tap(input) {
     const result = await device.tap(input.x, input.y);
@@ -18,7 +19,7 @@ export async function tap(input) {
     // Wait for UI to settle before taking screenshot
     await new Promise(resolve => setTimeout(resolve, 1500));
     const screenshot = await device.takeScreenshot();
-    recordStep(input.caseName, input.stepIndex, input.description, "tap", screenshot.filePath, { x: input.x, y: input.y });
+    recordStep(input.caseName, input.stepIndex, input.description, "tap", screenshot.filePath, { x: input.x, y: input.y }, input.expectedOutcome);
     return {
         content: [
             {
@@ -29,9 +30,10 @@ export async function tap(input) {
             {
                 type: "text",
                 text: JSON.stringify({
-                    message: `Step ${input.stepIndex} recorded. Please now evaluate this screenshot across all 4 dimensions (overlap, layout, info_clarity, style) using get_evaluation_criteria and submit_dimension_score before proceeding.`,
+                    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.`,
                     caseName: input.caseName,
                     stepIndex: input.stepIndex,
+                    expectedOutcome: input.expectedOutcome,
                     screenWidth: screenshot.width,
                     screenHeight: screenshot.height,
                 }),

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;CAChF,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,CAC7B,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,2LAA2L;oBAC3N,QAAQ,EAAE,KAAK,CAAC,QAAQ;oBACxB,SAAS,EAAE,KAAK,CAAC,SAAS;oBAC1B,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,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"}

package/dist/src/types.d.ts

@@ -19,6 +19,7 @@ export interface StepRecord {
         x: number;
         y: number;
     };
+    expectedOutcome?: string;
     screenshotPath: string;
     timestamp: string;
     evaluations: Record<string, DimensionScore>;

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,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;CAC/C;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": "2.0.0",
+    "version": "2.2.0",
     "description": "MCP server for iOS app UI evaluation and testing, powered by idb + xcrun simctl",
     "type": "module",
     "main": "dist/src/index.js",