ff-automationv2 2.2.36-beta.0 → 2.2.36-beta.2

package/dist/ai/llmprompts/systemPrompts/inputContextPrompt.js

@@ -0,0 +1,56 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.inputContextPrompt = inputContextPrompt;
+async function inputContextPrompt({ inputStepsContext }) {
+    const prompt = `
+   You are an intelligent assistant that extracts input data for the current step.
+
+Previous steps data:
+${JSON.stringify(inputStepsContext, null, 2)}
+
+Instructions:
+
+1. Analyze the current step and previous steps carefully.
+
+2. Priority rules (strict order):
+   - If a specific variable name is mentioned (e.g., "atpValue", "atpMulValue", "atpdivValue", "finalResult") → match to the step in "Previous steps data" with that exact "variableName".
+   - If referential calculation phrases are mentioned (e.g., "calculated final result", "calculated value", "final result", "calculation result", "result of calculation") → match to the most recent calculation or arithmetic step in "Previous steps data".
+   - If a step index or step reference is mentioned (e.g., "step 2", "2nd step", "3rd step", "3nd step", "step 1.1", "sub-step 1.1") → match to the corresponding stepindex (e.g., 2nd step -> stepindex 2, step 1.1 -> sub-step with stepindex "1.1"). If the matched step contains "subSteps", extract from the subStep matching the field name or sub-step index, and extract ONLY from that subStep.
+   - If a field name is mentioned (e.g., email, password, first name) → extract from the matching field step or subStep.
+   - If words like "above", "previous", "same" → extract from the most relevant previous step or subStep.
+
+3. If the step is normal (no reference):
+   - Extract input directly from the current step.
+   - Do NOT use previous steps.
+
+4. Conflict rule:
+   - If both source and destination fields exist (e.g., "enter first name in password"):
+     - Use the SOURCE field value ("first name").
+     - Ignore the destination field ("password").
+
+5. Extraction rules:
+   - Use ONLY given data (previous or current step).
+   - If input exists in the current step, then extract that input only, should not use input from previous steps.
+   - If step is about extracting two values from previous steps, extract both values and return them as an inputText as "input1input2".
+   - Do NOT create or merge values.
+   -If the step modifies an input value from the current or previous steps, return the updated value as "inputText".
+      Example: If the input is "abc" and the step says to add "@123" or and add @ and one number, return "inputText": "abc@123" || "abc@5".
+   - If step is saying to genarate a input using previous steps input values, genarate the input value and return it as "inputText". if that is is Password genarate a strong password.
+   - Always return a single value.
+
+6. Output rules:
+   - "inputText": Return ONLY the input value. inputText must NEVER be empty → fallback to most relevant previous value.
+   - "variableName": Do NOT generate a new variable name. Extract and return the "variableName" from the matching step or subStep in "Previous steps data" (inputStepsContext). If the matched previous step has a "variableName", return that exact variableName. If no variableName is present or no previous step is referenced, return "".
+
+7. Never generate new input. Always follow priority rules strictly.
+
+Respond ONLY with valid JSON:
+
+{
+  "inputText": "x",
+  "variableName": "variableNameFromPreviousStep"
+}
+
+`;
+    return prompt;
+}

package/dist/ai/llmprompts/systemPrompts/mobileKeywordExtractor.d.ts

@@ -0,0 +1,2 @@
+import { keywordExtractor } from "../llmPromptTypes/systemPromptTypes";
+export declare function keywordExtractorMobile({ priorAndNextSteps }: keywordExtractor): Promise<string>;

package/dist/ai/llmprompts/systemPrompts/mobileKeywordExtractor.js

@@ -0,0 +1,94 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.keywordExtractorMobile = keywordExtractorMobile;
+async function keywordExtractorMobile({ priorAndNextSteps }) {
+    const allowedActions = ["openApp", "tap", "enter", "wait", "verify", "scroll", "get", "closeApp", "combined", "switchToWeb", "getOtp", "enterOtp", "arithmetic"];
+    const prompt = `
+You are an expert in mobile app testing.
+From the step, extract ONLY the meaningful keywords so that i can search for the element in the dom.
+Rules:
+- understand the step and context from the ${priorAndNextSteps}.and only extract keywords from current step and dont extract from prior and next steps.
+- 3 to 5 keywords maximum.
+- If step is contains more than one keyword  like tap on x button of y ,then include both in keywords like this [x,y].and include full keywords rather than insering part of it.
+- If the step is about entering text, Should NOT include entered input values from the step in keywords.
+- First keywords should be from step next Keywords must be distinct and based on the element's label meaning only.
+- **Keywords can be string or number if the step contains a number it can be any number,must include that number in keywords and give it in string format.**
+- If the step is about launching an app or opening app and if it has capbilities add them in keywords like ["key1:value1","key2:value2"...] and if step has app package and app activity without capability names like this 'com.app.android', 'com.app.activity' in keywords add with capability names
+- For the Open App / Launch App action, if the user mentions any Android app capability, extract it into the keywords array as a key-value pair using the exact format "<capability>:<value>".
+- The only allowed Android capabilities are:
+      - "app" - APK/app path
+      - "appPackage" - Android application package name
+      - "appActivity" - Android activity name
+      - "noReset" - whether the app state should be reset
+      - "fullReset" - whether a complete reset should be performed
+      - "autoGrantPermissions" - whether applicable Android runtime permissions should be automatically granted
+- Every capability mentioned by the user MUST be returned as a key-value pair and capability name must be in above capability names, Never return a capability without its value.
+- if step has keyword with type give two keywords with type and without type. ex: tap on leaving from text field. keywords = ["leaving from",leaving from text field].
+- Should NOT include any other unrelated keywords for step.
+- Should NOT include generic UI words (button, field, etc) and action words (tap, click, press, etc).example:tap on x button -> ['x']
+- Should NOT include status/technical words (displayed, enabled, authenticate, visible).
+- If the step is for the switching to web then provide the action as the switchToWeb
+- If an element label contains multiple words (e.g., "Sign In", "Add to Cart"), keep them together as ONE keyword and do not split them and also for keywords you generated, do not split them.
+-** element_name: extract name of the element that mentioned in the step  not from keywords or other steps.(eg:tap on x -> element_name:x). always try to retuen short and meaning full element name from step**
+- action: openApp for opening or launching of app  and activate app is there in step give action as combined, tap for taping or selecting or clicking or pressing, enter for entering input, wait for waiting or sleeping, verify for verifying or checking,scroll for scrolling and swiping, get for getting and fetching element and gtting logs and getting driver or instance details, closeApp for closing the app but not for terminating app step, arithmetic for performing arithmetic operations (multiply, divide, add, subtract, calculate) or storing/assigning calculation results and captured values into variables.
+- If step is press any key give action as tap.if step is clear or clearandenter give action as enter and if step is open chrome browser app por application and it doesnot have app package or bundle id give action as combined but if step has app package or bundle id give action as openApp.
+- action must be one of from this list ${allowedActions}.if not one of them, return action as 'combined'. if step about set or setting or find, install apk or ipa, uninstall apk or ipa,activate app with app package or bundle id, pinch in or pinch out or related turn wifi or airplane mode return action as 'combined'
+- if the step action is about finding or check app is installed or open chrome browser  or open notification bar or terminate app using app package or bumdle id or running app in the background for some seconds then provide action as combined.
+CRITICAL SPLITTING RULES:
+1. If the step contains multiple actions (two or more, in any combination), you MUST split it into multiple objects in the "steps" array in the same sequence and flow.
+   - Each object must represent EXACTLY ONE action.
+   - Preserve the ORIGINAL ORDER strictly.
+   Example: "Enter 'admin' in Username, '1234' in Password and tap Login" → split into 3 steps:
+   1. Enter 'admin' in Username (action: "enter", elementName: "Username")
+   2. Enter '1234' in Password (action: "enter", elementName: "Password")
+   3. Tap on Login (action: "tap", elementName: "Login")
+   Example: "Enter 'x' in a and tap b" → split into 2 steps:
+   1. Enter 'x' in a (action: "enter", elementName: "a")
+   2. Tap on b (action: "tap", elementName: "b")
+2. If the step contains "select", "choose", "pick" and also contains a connecting reference ("from", "in", "under"):
+   Split into two atomic steps: tap reference element first, then tap target option.
+   Example: "Select 'India' from Country" →
+   1. Tap on Country (action: "tap", elementName: "Country")
+   2. Tap on India (action: "tap", elementName: "India")
+3. If the step performs the SAME action with ONE VALUE into TWO OR MORE DIFFERENT elements:
+   Split into separate steps — one per target element:
+   Example: "Enter '1234' in PIN and Confirm PIN" →
+   1. Enter '1234' in PIN (elementName: "PIN", keywords: ["PIN"])
+   2. Enter '1234' in Confirm PIN (elementName: "Confirm PIN", keywords: ["Confirm PIN"])
+4. Keywords in multiple steps should be extracted based on each individual sub-step only.
+5. If the step performs the SAME action on TWO OR MORE DIFFERENT target elements joined by "and"/"," (with or without a value), split into separate steps — one per target element — each step keeping the same action (and same value, if any) with its own elementName and keywords.
+    Example (with value): "enter x in both a and b" -> 1. Enter x in a (elementName "a", keywords ["a"]), 2. Enter x in b (elementName "b", keywords ["b"]).
+    Example (no value): "get text from a and b" -> 1. Get text from a (elementName "a", keywords ["a"]), 2. Get text from b (elementName "b", keywords ["b"]).
+6. fileType: If the verify action step mentions verifying or checking any file extension or document (e.g. "pdf" or any file extension), extract that file extension in lowercase (without dot, e.g. "pdf") and set it in fileType. For all non-file/UI verification steps, set fileType to null.
+
+Respond ONLY with valid JSON in this EXACT format:
+{
+  "steps": [
+    {
+      "keywords": [key1,key2,key3,key4,key5] if step is opening app then keywords should be like ["key1:value1","key2:value2"...],
+      "elementName": "x",
+      "action": ${allowedActions.join("|")},
+      "isStepValueDependable": boolean (true ONLY if action is verify/enter AND the step contains an actual referential phrase pointing to an earlier step's data — see isStepValueDependable rule below for the full test and examples; false if the value is literal or no referential phrase is present. Never consider future steps.),
+      "liveLog": "Should Generate a short, human-readable description of what the AI is about to do.",
+      "fileType": "pdf" | other lowercase file extension | null,
+      "step": "Clean, complete instruction of the step (or atomic sub-step if split)."
+    }
+  ]
+}
+STEP RULES (VERY IMPORTANT):
+- Each "step" MUST represent EXACTLY ONE action.
+- isStepValueDependable (boolean):
+    • Test is REFERENTIAL vs LITERAL, not sentence completeness — e.g. "enter 2nd step input in name" is a complete sentence but still referential, since the value itself isn't written there.
+    • Set true ONLY when action is 'verify' or 'enter' AND the step contains an actual referential phrase pointing to data from an earlier step (by step number, "previous/prior/earlier/that/same" wording, or naming what was fetched/copied/retrieved there). Confirm the referenced step exists earlier in priorAndNextSteps, never a future step.
+    • Set false when the value is written literally, or no referential phrase is present, or the action isn't verify/enter.
+    • DO NOT infer true from an element/field name alone when that word appears only in the CURRENT step's own target/element — e.g. "get text from Password" is false because "Password" there is just this step's own target, not a value pointing elsewhere.
+    • EXCEPTION — reference-by-element-name: if the VALUE being entered matches (case-insensitively) the elementName or keyword of a 'get' or 'enter' action from an EARLIER step in priorAndNextSteps, treat this as referential (true) even with no explicit pointer wording like "previous"/"same" — this is a common shorthand for "reuse whatever was entered/fetched there". Example: step 2 is "enter santosh@123 in password" (elementName "Password"), step 3 is "enter password in confirm password" → true, because the value "password" matches step 2's elementName, meaning step 3 is reusing step 2's entered value, not typing the literal word "password".
+    • Examples (illustrative, not exhaustive): "enter 2nd step input in name" → true. "enter the OTP received earlier in name" → true. "enter 'John' in name" → false. "get text from Password" → false (no reference present). "enter password in confirm password" (given step 2 entered a value into a "Password" field) → true.
+- Step MUST be rewritten as a CLEAN and COMPLETE instruction containing action, target element, and value (if applicable).
+- DO NOT copy original sentence if it contains multiple actions.
+- ALWAYS normalize mobile action wording (e.g., use "Tap on Y", not "Click on Y").
+- Each step must be SELF-CONTAINED and EXECUTABLE.
+No other text.
+`;
+    return prompt;
+}

package/dist/ai/llmprompts/systemPrompts/multichannelKeywordExtractor.d.ts

@@ -0,0 +1,6 @@
+import { keywordExtractor } from "../llmPromptTypes/systemPromptTypes";
+export type multichannelKeywordExtractor = keywordExtractor & {
+    allSteps: string[];
+    previousPlatform?: string | null;
+};
+export declare function keywordExtractorMultichannel({ priorAndNextSteps, allSteps, previousPlatform, }: multichannelKeywordExtractor): Promise<string>;

package/dist/ai/llmprompts/systemPrompts/multichannelKeywordExtractor.js

@@ -0,0 +1,86 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.keywordExtractorMultichannel = keywordExtractorMultichannel;
+const actionExtractorPrompt_js_1 = require("./actionExtractorPrompt.js");
+const mobileKeywordExtractor_js_1 = require("./mobileKeywordExtractor.js");
+async function keywordExtractorMultichannel({ priorAndNextSteps, allSteps, previousPlatform, }) {
+    const webPrompt = await (0, actionExtractorPrompt_js_1.keywordExtractor)({ priorAndNextSteps });
+    const mobilePrompt = await (0, mobileKeywordExtractor_js_1.keywordExtractorMobile)({ priorAndNextSteps });
+    const allowedActions = [
+        "enter",
+        "refresh",
+        "wait",
+        "verify",
+        "scroll",
+        "navigate",
+        "navigateBack",
+        "click",
+        "maximize",
+        "minimize",
+        "get",
+        "upload",
+        "close",
+        "open",
+        "openApp",
+        "drag_and_drop",
+        "closeApp",
+        "switch",
+        "cleartext",
+        "tap",
+        "combined",
+        "mouseAction",
+        "switchToAndroid",
+        "switchToWeb",
+        "getOtp",
+        "enterOtp",
+        "arithmetic",
+    ];
+    return `
+You are extracting keywords for a MULTICHANNEL test. Platform is not provided by the user.
+You MUST decide the platform for the CURRENT step and return it so downstream execution stays unchanged.
+
+previousPlatform: ${previousPlatform ?? "none"}
+allSteps (full script, use as context only): ${JSON.stringify(allSteps)}
+nearbySteps: ${JSON.stringify(priorAndNextSteps)}
+
+PLATFORM DECISION RULES (mandatory):
+- platform MUST be exactly one of: "web" | "android" | "ios"
+- Decide using the CURRENT step first, then allSteps for context, then previousPlatform.
+- If the current step is in the middle of a flow and does not clearly name a platform, browser, URL, app, package, bundle, or device, you MUST reuse previousPlatform.
+- If previousPlatform is "none" (first step), infer from the current step and allSteps:
+  - web: browser, URL, http/https, navigate, click, open browser, web page, tab, window
+  - android: Android app, APK, appPackage, appActivity, UiAutomator, Android device
+  - ios: iOS app, iPhone, IPA, bundle id, XCUITest, Safari on iOS
+- Do not switch platform on generic UI verbs (tap/click/enter/verify/scroll) unless the step or surrounding allSteps clearly change channel.
+- After you choose platform, follow THAT platform's keyword/action rules below. Do not mix web actions onto mobile or mobile actions onto web.
+  - web actions include: click, open, close, navigate, navigateBack, maximize, minimize, refresh, mouseAction, drag_and_drop, switch, cleartext, upload
+  - android/ios actions include: tap, openApp, closeApp, combined, enterOtp
+  - shared: enter, wait, verify, scroll, get, getOtp, arithmetic, switchToAndroid, switchToWeb
+- action must be one of: ${JSON.stringify(allowedActions)}
+
+===== WEB KEYWORD RULES (use when platform is web) =====
+${webPrompt}
+
+===== ANDROID / iOS KEYWORD RULES (use when platform is android or ios) =====
+${mobilePrompt}
+
+Respond ONLY with valid JSON in this EXACT format (this overrides the JSON formats in the sections above):
+{
+  "steps": [
+    {
+      "platform": "web" | "android" | "ios",
+      "keywords": ["key1", "key2"],
+      "elementName": "element name",
+      "action": ${allowedActions.join("|")},
+      "isStepValueDependable": boolean,
+      "liveLog": "Short human-readable description of what the AI is about to do.",
+      "fileType": "pdf" | other lowercase file extension | null,
+      "step": "Single atomic instruction for this step."
+    }
+  ]
+}
+
+Every object in "steps" MUST include platform. If you split one user step into multiple objects, keep the same platform unless the wording clearly switches channel.
+No other text.
+`;
+}

package/dist/ai/llmprompts/systemPrompts/otpExtractorPrompt.d.ts

@@ -0,0 +1 @@
+export declare function otpExtractorPrompt(): Promise<string>;

package/dist/ai/llmprompts/systemPrompts/otpExtractorPrompt.js

@@ -0,0 +1,22 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.otpExtractorPrompt = otpExtractorPrompt;
+async function otpExtractorPrompt() {
+    return `
+You extract one-time passwords (OTPs) from SMS messages and also the provider of the regex to extract that OTP from the message.
+
+
+Return only valid JSON in exactly this format:
+{
+  "otp": ""
+  "regex":""
+}
+
+Rules:
+- Extract only the OTP/code stated in the supplied SMS body.
+- Do not infer, generate, or guess an OTP.
+- If no single unambiguous OTP is present, return an empty string.
+- Return the regex that can be used to extract that OTP number itself from that message body also don't include any text only give the expression.
+- Return no text, Markdown, or keys other than "otp".
+`;
+}

package/dist/ai/llmprompts/systemPrompts/sliderSwipeVisionPrompt.d.ts

@@ -0,0 +1,2 @@
+import { swipeVisionPrompt } from "../llmPromptTypes/systemPromptTypes.js";
+export declare function swipeVisionPrompt({ priorAndNextSteps }: swipeVisionPrompt): Promise<string>;

package/dist/ai/llmprompts/systemPrompts/sliderSwipeVisionPrompt.js

@@ -0,0 +1,54 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.swipeVisionPrompt = swipeVisionPrompt;
+async function swipeVisionPrompt({ priorAndNextSteps }) {
+    const prompt = `
+You are a mobile automation assistant analyzing a swipe action from a screenshot.
+
+Given:
+- A screenshot with UI elements labeled using Fire-Flink indexes.
+- The user's swipe step.
+- Context: ${JSON.stringify(priorAndNextSteps)}.
+
+Your first task is to identify the TARGET ELEMENT mentioned or implied by the user's step.
+
+Then determine whether that exact target element is visually a SLIDER/DRAG CONTROL in the screenshot.
+
+IMPORTANT:
+- Return NormalSwipe=false ONLY if ALL conditions are true:
+  1. The target from the user's step is present in the screenshot.
+  2. The identified target corresponds to a specific UI control.
+  3. That control is visually designed to be dragged/slid from one position to another.
+  4. A parent/container defining the slider movement area can be identified.
+- If the target is a normal text, button, link, image, icon, label, or any other non-slider element, return NormalSwipe=true.
+- If the target exists but is NOT a slider/drag control, return NormalSwipe=true.
+- If the target cannot be found, return NormalSwipe=true.
+- Never classify an element as a slider only because the step contains words such as "swipe", "slide", "to", or because the element has a suitable size.
+- Do not assume that any horizontally or vertically shaped element is a slider.
+- The slider must have a clear visual indication that a specific control/thumb can be dragged.
+- Do not use surrounding elements as the target unless they are actually the draggable control.
+- For a valid slider, identify both:
+  - targetIndex = draggable thumb/control
+  - parentIndex = slider/container defining the movement boundary
+- Determine the direction from the user's step and UI.
+
+For a valid slider:
+{
+  "NormalSwipe": false,
+  "parentIndex": "Fire-Flink-x",
+  "targetIndex": "Fire-Flink-x",
+  "direction": "left|right|up|down"
+}
+
+For every other case:
+{
+  "NormalSwipe": true,
+  "parentIndex": "Fire-Flink-0",
+  "targetIndex": "Fire-Flink-0",
+  "direction": "left|right|up|down|none"
+}
+
+Return ONLY valid JSON.
+`.trim();
+    return prompt;
+}

package/dist/ai/llmprompts/systemPrompts/userStoryToListPrompt.d.ts

@@ -0,0 +1,2 @@
+export declare function buildStepExtractionPrompt(): Promise<string>;
+export declare function androidbuildStepExtractionPrompt(): Promise<string>;

package/dist/ai/llmprompts/systemPrompts/userStoryToListPrompt.js

@@ -0,0 +1,60 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.buildStepExtractionPrompt = buildStepExtractionPrompt;
+exports.androidbuildStepExtractionPrompt = androidbuildStepExtractionPrompt;
+async function buildStepExtractionPrompt() {
+    const prompt = `
+You are a precise assistant.
+Extract Steps from the following user story for web automation.
+
+Rules:
+1. Each step must be a string — not a tuple.
+2. Keep the wording the same except:
+   - Do not add new steps, only rewrite existing ones.
+   - Do not remove any steps.
+   - Do not remove quotes from the step. if these quotes are present in the step, keep them as it is.
+3. Output only a **valid TypeScript array of strings**. No explanations, markdown, or code fences..
+4. Please paste the “below example array” here, so I can make it dynamic and generic strictly in the same format.
+**Dont include any other text in the array. and also dont include example array in the response.**
+Output Format
+- Return ONLY a valid JSON object; the response must start with '{' and end with '}', with no markdown, code fences, explanations, or extra text.
+Respond strictly in valid JSON using the following structure:
+{
+"manualSteps": ["step 1", "step 2", "step 3"]
+}
+`;
+    return prompt;
+}
+async function androidbuildStepExtractionPrompt() {
+    const prompt = `
+You are a precise assistant.  
+Extract **step data** from the following mobile {user_story_raw}.
+
+Rules:
+Dont take text from one step and use it in another step.
+1. If the step is about luanching an app, like (action, param1, param2) or(action,'param1,param2') then wrap them into a tuple.
+and also give them in the format (launch app, 'com.app.android', 'com.app.activity') like first package name and second activity name. if any separator exists (- / | :), split on it.
+If they mention package:com.app.android and activity:com.app.activity remove variable name and then convert them into (launch app, 'com.app.android, 'com.app.activity')
+2.Dont remove parameters from the step. and never combine or merge steps and dont create any extra steps.
+step may be any formate for enter if input is give seperatly with comma use this
+ example:'enter username in text field','testuser' → ('enter username in text field','testuser').
+if not given seperatly with comma use and mention in step it self than this
+ example:'enter testuser in text field' → ('enter testuser in text field','testuser').
+3. If it's just a single string value, keep it as a string.
+4. Always output a **Always output a valid list of strings.** containing strings or tuples only. 
+ - Do not add explanations, reasoning, or extra text outside the list.
+ - Do not make any assumption if step inputs are missing in user_story_raw
+ - **Dont change the order of steps from user story**
+5. If the step is tuple keep it as tuple like this (step). if not just keep it as a string like this 'step'.
+**strictly follow the output format**
+[
+    ('Launch  Mobile App','com.app.android','com.app.activity'),
+    'Tap on cancel button',
+    'Tap on signin with email',
+    ('enter username in text field','testuser'),
+    ('enter testuser in text field','testuser'),
+    'verify element is displayed'
+]
+`;
+    return prompt;
+}

package/dist/ai/llmprompts/systemPrompts/verifyActionExtractorPrompt.d.ts

@@ -0,0 +1,2 @@
+import { verifyActionArgs } from "../llmPromptTypes/systemPromptTypes";
+export declare function verifyActionExtractorPrompt({ extractedDomJson, priorAndNextSteps, }: verifyActionArgs): Promise<string>;

package/dist/ai/llmprompts/systemPrompts/verifyActionExtractorPrompt.js

@@ -0,0 +1,81 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.verifyActionExtractorPrompt = verifyActionExtractorPrompt;
+async function verifyActionExtractorPrompt({ extractedDomJson, priorAndNextSteps, }) {
+    const nlpList = `VerifyNumberOfElementsByXpath,VerifyElementIsDisabled,VerifyElementIsEnabled,VerifyElementIsSelected,VerifyIfOptionWithIndexIsDeselectedInElement,
+    VerifyIfOptionWithIndexIsSelectedInListBox,VerifyWidthOfElement,VerifyBrowserWindowXLocation,
+    VerifyIfCookieNameIsDeleted,VerifyBrowserWindowWidth,VerifyValueOfElementIsCleared,
+    VerifyTagName,VerifyIfCookieNameIsAdded,VerifyLinkNavigatesToTitle,VerifyXLocationOfElement,
+    VerifyAllOptionsAreDeselected,VerifyBrowserHtmlCodeContainsString,VerifyBrowserWindowYLocation,
+    VerifyNumberOfElementsByTagName,VerifyListBoxOptionsAreSorted,VerifyNavigateURL,
+    VerifyHeightOfWebElement,VerifyValueOfElementContainsString,VerifyBrowserWindowHeight,VerifyIfLinksAreBroken,
+    VerifyIfLinkIsBroken,VerifyValueOfElementIsString,VerifyYLocationOfElement,
+    VerifyAttributeValue,
+    VerifyElementIsDisplayed,VerifyAllOptionsAreSelected,VerifyIfOptionWithValueIsDeselected,
+    VerifyBrowserWindowTitleContainsString,VerifyBrowserWindowTitleIsString,
+    VerifyTitleOfCurrentPage,VerifyPartialTitleOfCurrentPage,VerifyElementIsNotDisplayed,
+    VerifyCssAttributeValue,VerifyPartialAttributeValue,VerifyIfSpecifiedOptionIsSelected,
+    VerifyTitleInAllPage,VerifyIfOptionWithValueIsSelected,VerifyIfVideoIsPlaying,
+    VerifyIfGivenOptionIsDuplicate,VerifyRadioButtonIsSelected,VerifyRadioButtonIsNotSelected,
+    VerifyThatSpecifiedOptionIsNotSelected,
+    VerifyText,VerifyPartialText,VerifyElementNotContainsText,VerifyElementIsClickable,
+    VerifyCheckBoxIsSelected,VerifyCheckBoxIsNotSelected,
+    VerifyUrlContainsExpectedUrl,VerifyUrlIsString,
+    VerifyIfBrowserWindowIsClosed,VerifyAllBrowserWindowsClosed,
+    VerifyIfAllCookiesAreDeleted,VerifyIfCookieObjectIsPresent,
+    VerifyIfXAndYCoordinateOfElement,VerifyIfVideoIsMuted,VerifyIfVideoIsPaused,
+    VerifyIfVideoEnded,VerifyIfVideoPlayingInMute,VerifyIfVideoPlayingInUnmute,
+    VerifyIfAudioIsPlaying,VerifyIfAudioIsPaused,VerifyIfAudioIsMuted,VerifyIfAudioEnded,
+    VerifyIfLinkIsWorking,VerifyIfLinksAreWorking,
+    VerifyIfImageIsBroken,VerifyAllBrokenImages,VerifyIfStatusCodeOfALinkIsStatusCode,
+    VerifyIfStatusCodeOfAllLinksIsStatusCode,VerifyIfOptionWithVisibleTextIsSelected,VerifyBrowserWindowUrlIsString,
+    VerifyIfOptionWithTextIsDeselected,VerifyPerformanceScore,VerifyAccessibilityScore,VerifyBrowserWindowUrlContainsExpectedUrl,
+    VerifyBestPracticesScore,VerifySeoScore,VerifyPwaScore,VerifyHexCodeForGivenXYCoordinates`;
+    const elementType = ['link', 'textfield', 'icon', 'button', 'radioButton', 'text', 'textarea', 'image', 'dropdown', 'checkbox', 'tab', 'action overflow button', 'hamburger icon', 'toggle button', 'suggestion', 'time picker', 'date picker', 'toaster message', 'card', 'tooltip', 'option', 'calender', 'sliders', 'visual testing'];
+    const prompt = `You are an intelligent assistant that extracts structured UI action data.
+Given the step, the simplified DOM: ${extractedDomJson}, and the list of NLP names: ${nlpList} don't provide the other nlp names which are not present in the list in any manner.
+
+
+Select the perfect matching NLP name from the list that best fits the step's intent.
+**The NLP name must match exactly one value from the approved NLP list. Do not generate or suggest any new NLP names under any condition.**
+step may give any way,you need to get most matching and most related nlp name from the list that relates to step.
+for these nlp's VerifyIfSpecifiedOptionIsSelected,VerifyThatSpecifiedOptionIsNotSelected,VerifyIfOptionWithValueIsSelected,VerifyIfOptionWithValueIsDeselected,VerifyIfOptionWithVisibleTextIsSelected,VerifyIfOptionWithTextIsDeselected,use this precedence: if step mentions word "value" then only match with VerifyIfOptionWithValueIsSelected,VerifyIfOptionWithValueIsDeselected. if step mentions word "text" then only match with VerifyIfOptionWithVisibleTextIsSelected,VerifyIfOptionWithTextIsDeselected. Otherwise go with specified option nlp's (VerifyIfSpecifiedOptionIsSelected,VerifyThatSpecifiedOptionIsNotSelected). and only select tags for these nlp's from the given tag list.
+if step is verify option with index is deselected map to VerifyIfOptionWithIndexIsDeselectedInElement and if step is verify option with index is selected then map to VerifyIfOptionWithIndexIsSelectedInListBox.
+example: verify if email element is present
+nlp name should be VerifyElementIsDisplayed
+- Distinguish element-presence from text-presence: if the step says the ELEMENT/component itself is present/displayed/visible (e.g., "verify email element is present"), map to VerifyElementIsDisplayed. If the step says specific TEXT is present/displayed (e.g., "verify login text is present", "verify 'Login' text is displayed"), map to VerifyText instead (or VerifyPartialText if the step says "partial"/"contains"/"includes").
+- Return ONLY a valid JSON object; the response must start with '{' and end with '}', with no markdown, code fences, explanations, or extra text.
+Return **only valid JSON** in the following format:
+{
+  "attributeValue": "Fire-Flink-x",
+  "nlpName": "x",
+  "inputText":"x",
+  "elementType": "x"
+}
+
+Rules:
+- nlpName must be exactly one from the provided list.
+- URL steps (check in order, stop at first match): "title" mentioned → VerifyLinkNavigatesToTitle | contains/includes → VerifyUrlContainsExpectedUrl (if step says "browser"/"window"/"tab" url → VerifyBrowserWindowUrlContainsExpectedUrl instead) | literal word "exactly" or "equals" present → VerifyUrlIsString (if step says "browser"/"window"/"tab" url → VerifyBrowserWindowUrlIsString instead) | everything else, including plain "is"/"navigated to" phrasing → VerifyNavigateURL.
+  example: "verify navigated url is x" → VerifyNavigateURL ("is" here is just a linking verb, not the keyword "exactly"/"equals")
+-while matching nlpName with step (ignore spaces, case differences).If multiple NLPs are similar, choose the MOST SPECIFIC and EXACT match.
+- Use context from the steps : ${priorAndNextSteps}, keyword and json to search for FF-inspecter.
+- **Find the FF-inspecter attribute value of the element in the Simplified JSON whose text or any other attributes best match the step in the simplified DOM.**
+- For some steps you can see two inputs but you need to understand step and send the required inputText. inputText can be string, number, url or xpath.
+        ex: Verify the flights matches the flight, Verify the home has 30 and 50 coordinates if they give x-30,y-50 just send 30,50 and verify hex code '#FFFFFF' for given (34,11) coordinates or they will give x-34,y-11 hexcode-#FFFFFF format than fill inputText in this order "#FFFFFF,34,11"
+        here keyword : "flights" , input : "flight", here input:"30,50"  , here input:"#FFFFFF,34,11" 
+- If step is about checking attribute value then map to attribute nlps and inputText: "attribute_name,expected_value" here attribute_name can be id,text,href etc not fire-flink-x, and expected_value is the value being checked against from the step. inputText should be in this format. if they mention css attribute value then inputText should be "css_property,expected_value" from the step only.
+and also If a step requires comparing two values, provide the input in the format inputText: "value1,value2".
+- if step is about tag name then inputText: "tag name". and if step is about verifying tag name and with count then inputText: "tag name,count". and if any step is about verifying number of elements with xpath then inputText should be "xpath,count"
+- if step is about verifying location then inputText should be the actual coordinate values from the step, formatted as "x,y" (e.g. "30,50") — never the literal word "coordinates".
+- If the step requires verifying text from an element, only select elements capable of containing text. Do not select image or image-related elements (e.g., img, svg, picture, source).
+- If no explicit option is mentioned in the step, map it to the most relevant non-option verification NLP based on the intent of the step.
+- **Extract inputText from the step — this is the specific value/text being verified in the step. If the step does not contain any explicit value being verified, return the element or subject name mentioned in the step as inputText instead.**
+if step is verifying value of element then extract value which is being verified only not any other words in the step. 
+- Only for VerifyIfStatusCodeOfAllLinksIsStatusCode extract all links given in step and give in inputText as '[link1,link2,link3...],statusCode'
+- For performance and accessibility score extraction from step, use the inputs from step: [url,device,score] should be in this order. if any input is missing send remining inputs from step itself.
+- if no matching element found for the step return attributeValue as only 'Fire-Flink-0' dont give Fire-Flink-x or any string .
+- Based on step give most relevant type of element. use this list to choose elementType: ${elementType} and Never change syntax of elementType, follow the syntax of elementType in list.if elementType is not there in list return 'link'.
+- **Respond with valid JSON only. don't return any other text or don't return response in list format**
+`;
+    return prompt;
+}

package/dist/ai/llmprompts/systemPrompts/verifyActionExtractorPromptMob.d.ts

@@ -0,0 +1,2 @@
+import { verifyActionArgs } from "../llmPromptTypes/systemPromptTypes";
+export declare function verifyActionExtractorPromptMob({ extractedDomJson, priorAndNextSteps, }: verifyActionArgs): Promise<string>;

package/dist/ai/llmprompts/systemPrompts/verifyActionExtractorPromptMob.js

@@ -0,0 +1,52 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.verifyActionExtractorPromptMob = verifyActionExtractorPromptMob;
+async function verifyActionExtractorPromptMob({ extractedDomJson, priorAndNextSteps, }) {
+    const nlpList = `  "MOB_VerifyElementIsDisplayed,MOB_VerifyElementIsDisabled,MOB_VerifyElementIsEnabled,MOB_VerifyElementIsSelected,MOB_VerifyText,MOB_VerifyAttributeValue,MOB_VerifyPartialAttributeValue,"
+    "MOB_VerifyElementIsDisplayedInSpecifiedTime,MOB_VerifyElementIsEnabledInSpecifiedTime,"
+    "MOB_VerifyBatteryLevel,MOB_VerifyPartialText,MOB_VerifyElementIsClickable,"
+    "MOB_VerifyIfKeyboardIsHidden,MOB_VerifyIfKeyboardIsShown,MOB_VerifyIfXAndYCoordinateOfElement,"
+    "MOB_VerifyHeightOfElement,MOB_VerifyXLocationOfAndroidElement,MOB_VerifyIfDeviceIsUnlocked,"
+    "MOB_VerifyIfWifiOff,MOB_VerifyElementIsSelectedInSpecifiedTime,MOB_VerifyIfWifiOn,MOB_VerifyIfDeviceIsLocked,"
+    "MOB_VerifyIfAirplaneModeOff,MOB_VerifyIfGpsOff,MOB_VerifyHexCodeForGivenXYCoordinates,"
+    "MOB_VerifyYLocationOfAndroidElement,MOB_VerifyIfGpsOn,MOB_VerifyIfAirplaneModeOn,"
+    "MOB_VerifyWidthOfElement,MOB_VerifyElementIsNotDisplayed,MOB_VerifyElementNotContainsText,"
+    "MOB_VerifyIfAppIsClosed, MOB_CheckIfElementIsDisplayed"`;
+    const elementType = ["link", "textfield", "icon", "button", "radiobutton", "checkbox", "tab", "action overflow button", "hamburger menu", "toggle button", "steppers", "sliders"];
+    const prompt = `You are an intelligent assistant that extracts structured UI action data.
+Given the step: step, the simplified DOM: ${extractedDomJson}, and the list of NLP names: ${nlpList}.
+
+Select the perfect matching NLP name from the list that best fits the step's intent.
+
+Return **only valid JSON** in the following format:
+{
+  "attributeValue": "Fire-Flink-x",
+  "nlpName": "x",
+  "inputText":"x"
+  "elementType": "x",
+  "locatorType":"x"
+}
+Rules:
+- nlpName must be exactly one from the provided list.
+- Use context from the steps : ${priorAndNextSteps}, keyword and json to search for FF-inspecter.
+- **Find the FF-inspecter attribute value of the element in the Simplified JSON whose text or attributes best match the step.**
+- For some steps you can see two inputs but you need to uderstand step and send the requred inputText.
+        ex: Verify the flights matches the flight,Verify the home has 30 and 50 coordinates
+        here keyword : "flights" , input : flight, here input:30,50
+- If step mention time and element name then inputText should be time.
+        example: Verify the home is displayed for 30 seconds 
+        here keyword : "home" , input : 30 only they mention seconds as input in step
+- If step mention verify text then only pick MOB_VerifyText nlp otherwise pick MOB_VerifyElementIsDisplayed.
+- If step mention verify x and y coordinate of element then only pick MOB_VerifyIfXAndYCoordinateOfElement.
+- Extract inputText from the step that is which is being verified in the step if they mention verifying two values in inputs take as it is,if they mention one value take only that as input text, if you can't find any input text in the step, return keyword as inputText.
+- if the step is about checking the element is displayed then pick MOB_CheckIfElementIsDisplayed nlp.
+- For MOB_VerifyHeightOfElement just provide height mentioned in the step as the inputText don't add anything else 
+- Extract keyword from the step, if the step contains any keyword.
+- Use the closest semantic match for the step; return attributeValue as Fire-Flink-0, only if nothing is found.
+- Based on step give most relevant type of element. use this list to choose elementType: ${elementType} and Never change syntax of elementType, follow the syntax of elementType in list.if elementType is not there in list return 'link'.
+- Extract locator_type if mentioned in the step, like(text,resource-id,content-desc) . if not found return 'empty'.
+- Return ONLY a valid JSON object; the response must start with '{' and end with '}', with no markdown, code fences, explanations, or extra text.
+- Respond with valid JSON only.
+`;
+    return prompt;
+}

package/dist/ai/llmprompts/systemPrompts/visionPrompt.d.ts

@@ -0,0 +1,2 @@
+import { visionPrompt } from "../llmPromptTypes/systemPromptTypes";
+export declare function visionPrompt({ priorAndNextSteps }: visionPrompt): Promise<string>;

package/dist/ai/llmprompts/systemPrompts/visionPrompt.js

@@ -0,0 +1,31 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.visionPrompt = visionPrompt;
+async function visionPrompt({ priorAndNextSteps }) {
+    const customPromptText = `
+You are a precise automation assistant.
+Context: ${priorAndNextSteps}
+
+- **Do not provide any additional text or explanation. The output must be strictly in this formate only Fire-Flink-x.**
+- Use the context to understand the step and select the closest semantic match to Step and return only its index, else Fire-Flink-0.
+- Using the provided annotated screenshot and Step, identify the correct target element from the image.
+- First identify the target element that matches the Step using its text, action, role, or visual characteristics.
+- Match the text or action described in Step to the correct target element as accurately as possible.
+- Check the color and the arrow/connector in the screenshot to map the target element to its Fire-Flink index.
+- **The annotation arrow/connector is the primary source of truth for mapping the target element to its Fire-Flink index. First identify the target element, then trace its annotation arrow/connector to the corresponding Fire-Flink number.**
+- **Do not select an index based only on the physical position or proximity of the number to the target element.**
+- **In crowded screenshots, Fire-Flink index numbers may overlap, appear inside other elements, or be close to unrelated elements. Ignore the number's physical position and follow the annotation arrow/connector to determine which index belongs to the target element.**
+- Use color as a secondary confirmation when tracing the annotation arrow/connector.
+- Allow partial and case-insensitive matching. If a clear match exists, return its index.
+- If the Chosen element has more than one index has been provided for it, return the index that smaller in number.
+- **If u can't find the element OR the target element's annotation cannot be reliably mapped to a numbered Fire-Flink index, return Fire-Flink-0. Do not return any other element's index.**
+- Return only the Fire-Flink-x for the target element.
+- if the element that is related to step is there in image but it does not have any bounding box around it, dont give random fire flink index number by asssuming the position, return only "Fire-Flink-0".
+-**return only "Fire-Flink-0" If given image does not have color bounding boxes around elements or if image is not even loaded fully.**
+- Return ONLY a valid JSON object; the response must start with '{' and end with '}', with no markdown, code fences, explanations, or extra text.
+
+Respond only with JSON using this format:
+{"fireflinkIndex": "Fire-Flink-x" | "Fire-Flink-0"}
+`.trim();
+    return customPromptText;
+}

package/dist/ai/llmprompts/systemPrompts/visionPromptMobile.d.ts

@@ -0,0 +1,2 @@
+import { visionPromptMobile } from "../llmPromptTypes/systemPromptTypes";
+export declare function visionPromptMobile({ priorAndNextSteps }: visionPromptMobile): Promise<string>;

package/dist/ai/llmprompts/systemPrompts/visionPromptMobile.js

@@ -0,0 +1,26 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.visionPromptMobile = visionPromptMobile;
+async function visionPromptMobile({ priorAndNextSteps }) {
+    const customPromptText = `
+You are a precise mobile automation assistant. 
+- **Do not provide any additional text or explanation. The output must be strictly in this formate only Fire-Flink-x.**
+- You are given an image containing multiple elements outlined with bounding boxes.Each bounding box has a small black index number label drawn above it (like 1, 2, 3...).
+- Do not treat numbers appearing in the step as index numbers.if any number appears in the step it is not an index number.
+- Use the context from the all steps ${priorAndNextSteps}
+- **Analyze the user's step and action done by that step action determine which UI element in the screenshot corresponds semantically to that action.
+- for action enter check for edit text field and choose index number of that box.
+- **pick the index number that is on the box where element present that best matches semantically or visually to that action of the given step.**
+- Identify which element number (from the image) best matches that step.
+- Return the **index** number as shown in the image (1-based).
+- if there are two elements that match step and one have box and other doesn't have box then prefer the one with box and return that index number.
+- if no correct element found or if none of correct elements has no box,then return 0.
+- **If u can't find the element and also If there is no element with number, return Fire-Flink-0. dont return any other elements index.**
+- if the element that is related to step is there in image but it does not have any bounding box around it, dont give random fire flink index number by asssuming the position, return only "Fire-Flink-0".
+- Return ONLY a valid JSON object; the response must start with '{' and end with '}', with no markdown, code fences, explanations, or extra text.
+Respond only with JSON using this format:
+{fireflinkIndex: Fire-Flink-x | Fire-Flink-0}
+
+`.trim();
+    return customPromptText;
+}