ff-automationv2 1.0.0 → 2.0.1-beta.0

package/src/ai/llmprompts/systemPrompts/actionExtractorPrompt.ts

@@ -1,70 +1,70 @@
-
-type keywordExtractor = {
-  priorAndNextSteps: string[];
-
-};
-export async function keywordExtractor(
-  { priorAndNextSteps }: keywordExtractor
-): Promise<string> {
-
-  const allowedActions: string[] = [
-    "enter",
-    "wait",
-    "verify",
-    "scroll",
-    "navigate",
-    "click",
-    "maximize",
-    "get",
-    "upload",
-    "close",
-    "open",
-    "drag_and_drop",
-    "switch"
-  ];
-
-  const prompt = `
-You are an expert in Web application testing.
-From the step: step, extract ONLY the meaningful keywords so that i can search for the element in the dom.
-
-Rules:
-- Only give response for the current step .
-- understand the step and context from the ${JSON.stringify(priorAndNextSteps)} and keywords should be from step.
-- 3 to 5 keywords maximum.
-- If the step is about entering text or Uploading file, Should NOT include input value from the step into keywords.
-- If the step has words like tag name audio, video, image, svg, checkbox etc, include them in the keywords.
-- If icon is mentioned in step then 'svg' should add in keywords and for Upload action first keyword should be 'file'.
-- First keyword should be from step. Next keywords must be distinct and based on element label meaning only.
-- If keyword is two words, then create an additional keyword by combining them.
-  Example: "Sign In" → "Sign In", "SignIn"
-- Keywords can be string or number.
-- Do NOT include generic UI words or action words.
-- Do NOT include status or technical words.
-- If element label contains multiple words, keep them together as ONE keyword.
-- element_name: extract name of the element from step.
-  Capitalize first letter.
-  If not mentioned, return action as element_name.
-- action rules:
-  click, enter, wait, scroll, navigate, get, maximize, close, open,
-  upload, drag_and_drop, switch
-- action must be one of ${JSON.stringify(allowedActions)}.
-  If not, return "0".
-- For navigate:
-  - keywords must contain ONLY full URL
-  - element_name must be "URL"
-  - navigate allowed ONLY if step intent is purely navigation
-- If verification intent exists, MUST return "verify"
-
-Respond only with JSON using this format:
-{
-  "keywords": [],
-  "elementName": "",
-  "elementType":""
-  "action": ""
-}
-
-No other text.
-`;
-
-  return prompt;
-}
+
+type keywordExtractor = {
+  priorAndNextSteps: string[];
+
+};
+export async function keywordExtractor(
+  { priorAndNextSteps }: keywordExtractor
+): Promise<string> {
+
+  const allowedActions: string[] = [
+    "enter",
+    "wait",
+    "verify",
+    "scroll",
+    "navigate",
+    "click",
+    "maximize",
+    "get",
+    "upload",
+    "close",
+    "open",
+    "drag_and_drop",
+    "switch"
+  ];
+
+  const prompt = `
+You are an expert in Web application testing.
+From the step: step, extract ONLY the meaningful keywords so that i can search for the element in the dom.
+
+Rules:
+- Only give response for the current step .
+- understand the step and context from the ${JSON.stringify(priorAndNextSteps)} and keywords should be from step.
+- 3 to 5 keywords maximum.
+- If the step is about entering text or Uploading file, Should NOT include input value from the step into keywords.
+- If the step has words like tag name audio, video, image, svg, checkbox etc, include them in the keywords.
+- If icon is mentioned in step then 'svg' should add in keywords and for Upload action first keyword should be 'file'.
+- First keyword should be from step. Next keywords must be distinct and based on element label meaning only.
+- If keyword is two words, then create an additional keyword by combining them.
+  Example: "Sign In" → "Sign In", "SignIn"
+- Keywords can be string or number.
+- Do NOT include generic UI words or action words.
+- Do NOT include status or technical words.
+- If element label contains multiple words, keep them together as ONE keyword.
+- element_name: extract name of the element from step.
+  Capitalize first letter.
+  If not mentioned, return action as element_name.
+- action rules:
+  click, enter, wait, scroll, navigate, get, maximize, close, open,
+  upload, drag_and_drop, switch
+- action must be one of ${JSON.stringify(allowedActions)}.
+  If not, return "0".
+- For navigate:
+  - keywords must contain ONLY full URL
+  - element_name must be "URL"
+  - navigate allowed ONLY if step intent is purely navigation
+- If verification intent exists, MUST return "verify"
+
+Respond only with JSON using this format:
+{
+  "keywords": [],
+  "elementName": "",
+  "elementType":""
+  "action": ""
+}
+
+No other text.
+`;
+
+  return prompt;
+}

package/src/ai/llmprompts/systemPrompts/errorDescriptionPrompt.ts

@@ -1,23 +1,23 @@
-export async function buildErrorDescriptionPrompt(): Promise<string> {
-  const description_prompt = `
-
-Analyze the error and execution history to provide diagnostics in JSON format.
-
-Return a JSON object with the following keys:
-1. "error": A concise, single-line, selenium type(error).__name__ (e.g., "InvalidSessionIdException", "InvalidElementStateException", "ElementNotInteractableException", "NoSuchElementException" etc).
-2. "errorDescription": A very simple, non-technical, and user-friendly explanation of why the automation might have failed.
-   - Use plain English.
-   - All the lines should contain FireFlink. example: "FireFlink was unable to find the element..."
-   - Do not mention code, XPaths, or internal technical terms.
-   - The output MUST be a single paragraph of exactly 10 to 15 lines of text.
-   - Focus on potential real-world causes.
-
-Respond only with JSON using this format:
-{
-  "error": ",
-  "errorDescription": "",
-}
-`;
-
-  return description_prompt;
-}
+export async function buildErrorDescriptionPrompt(): Promise<string> {
+  const description_prompt = `
+
+Analyze the error and execution history to provide diagnostics in JSON format.
+
+Return a JSON object with the following keys:
+1. "error": A concise, single-line, selenium type(error).__name__ (e.g., "InvalidSessionIdException", "InvalidElementStateException", "ElementNotInteractableException", "NoSuchElementException" etc).
+2. "errorDescription": A very simple, non-technical, and user-friendly explanation of why the automation might have failed.
+   - Use plain English.
+   - All the lines should contain FireFlink. example: "FireFlink was unable to find the element..."
+   - Do not mention code, XPaths, or internal technical terms.
+   - The output MUST be a single paragraph of exactly 10 to 15 lines of text.
+   - Focus on potential real-world causes.
+
+Respond only with JSON using this format:
+{
+  "error": ",
+  "errorDescription": "",
+}
+`;
+
+  return description_prompt;
+}

package/src/ai/llmprompts/systemPrompts/fireflinkElementIndexExtactors.ts

@@ -1,198 +1,198 @@
-type FFInspectorArgs = {
-  stepAction: string;
-  extractedDomJson: string;
-  priorAndNextSteps: string[];
-  isAlert?: boolean;
-  isDrag?: boolean;
-};
-export async function ffInspectorNumExtractor({
-  stepAction,
-  extractedDomJson,
-  priorAndNextSteps,
-  isAlert = false,
-  isDrag = false,
-}: FFInspectorArgs
-): Promise<
-  string
-> {
-  const clickActions = ["click", "doubleclick", "rightclick"];
-  const alertActions = [
-    "ClickOkOnAlertPopup",
-    "ClickCancelOnAlertPopup",
-    "EnterDataToAlert",
-    "GetTextPresentOnAlertPopup",
-    "VerifyAlertPopupIsDisplayed",
-    "VerifyAlertPopUpMessageContainsString",
-    "WaitTillAlertIsPresent",
-    "VerifyTextPresentOnAlertPopup"
-  ];
-  const elementType = ["button", "link"]
-
-  let prompt;
-
-  // ---------------- ALERT ----------------
-  if (isAlert) {
-    prompt = `
-You are given a Selenium test step that requires interaction with a browser alert popup.
-
-Rules:
-- if the alert requires entering text or verification comparison, extract that text.
-- If no input is needed, set "input_text" to "None".
-- If step verifies text presence → VerifyTextPresentOnAlertPopup.
-- If step verifies text contains → VerifyAlertPopUpMessageContainsString.
-- Do not add explanations or extra text.
-- Return ONE action from this list only: ${JSON.stringify(alertActions)}.
-- Output must be valid JSON.
-Respond only with JSON using this format:
-
-{
-  "action": x,
-  "input_text": "text to enter" | "None"
-}
-
-`;
-  }
-
-  // ---------------- DRAG & DROP ----------------
-  else if (isDrag) {
-    prompt = `
-You are an intelligent assistant that extracts structured UI action data.
-
-Given a structured UI JSON representation with uniquely identified elements
-(ff-inspect values like Fire-Flink-1, Fire-Flink-2... in DOM order),
-locate the most appropriate elements for a drag-and-drop action.
-
-
-
-Rules:
-- Step: for the given step
-- Use the simplified JSON to find elements.
-- Identify source (drag) and destination (drop).
-- If not found, return Fire-Flink-0 for both.
-- Extract element names from step if possible.
-- Respond with JSON only.
-Simplified JSON:
-${extractedDomJson}
-
-Respond only with JSON using this format:
-{
-  "drag": "Fire-Flink-x",
-  "drop": "Fire-Flink-x",
-  "action": "Builtin_DragAndDropToElement",
-  "drag_element_name": "x",
-  "drop_element_name": "x"
-}
-`;
-  }
-
-  // ---------------- WAIT / SCROLL ----------------
-  else if (stepAction === "wait" || stepAction === "scroll") {
-    prompt = `
-You are an intelligent assistant that extracts structured UI action data.
-
-
-
-Rules:
-- Step: for the given step
-- Use context from test steps: ${JSON.stringify(priorAndNextSteps)}
-- Use simplified JSON to find best FF-inspect match.
-- Choose closest semantic + DOM-distance match.
-- Default direction is down.
-- Action:
-  - scroll → extract num_of_scrolls & direction
-  - wait → extract wait time (number only)
-- No nulls; use empty strings if needed.
-- The element type should be from this list ${elementType}
-
-Simplified JSON:
-${extractedDomJson}
-
-Return ONLY valid JSON:
-{
-  "attribute_value": "Fire-Flink-x",
-  "action": "x",
-  "input_text": "seconds in Interger | input text",
-  "keyword": "x",
-  "num_of_scrolls": "0",
-  "direction": "down",
-  "elementType":"x"
-}
-`;
-  }
-
-  // ---------------- DEFAULT (CLICK / ENTER / UPLOAD etc.) ----------------
-  else {
-    prompt = `
-You are a deterministic UI action extraction engine.
-
-OBJECTIVE:
-Identify the SINGLE best matching element from the Simplified JSON
-for the given step and return structured automation data.
-
------------------------------------------
-
-CONTEXT STEPS:
-${JSON.stringify(priorAndNextSteps)}
-
-CLICK ACTION WORDS:
-${JSON.stringify(clickActions)}
-
-ALLOWED ELEMENT TYPES:
-${elementType}
------------------------------------------
-
-CRITICAL MATCHING LOGIC:
-
-1. First understand intent:
-   - click words → click
-   - enter/type/write → enter_text
-   - upload → upload
-   - drag_and_drop → drag_and_drop
-
-2. For ENTER actions:
-   DO NOT pick first input blindly.
-   Score candidates using priority:
-
-   Priority Order:
-   (1) Exact label text match
-   (2) Placeholder match
-   (3) Name attribute semantic match
-   (4) ID semantic match
-   (5) type="email" if step contains "email"
-   (6) DOM sibling label relationship
-
-   Choose highest scoring match.
-
-   If NO strong semantic match → return:
-   attribute_value = "Fire-Flink-0"
-
-3. For CLICK actions:
-   - Prefer exact text match
-   - If multiple, choose closest contextual match
-   - If icon + svg exists, prefer svg element
-
-4. Never hallucinate.
-5. Only use FF values from provided JSON.
-6. No null values. Use empty string "".
-7. Output must be valid JSON only.
-
------------------------------------------
-SIMPLIFIED JSON:
-${extractedDomJson}
------------------------------------------
-
-Respond ONLY with JSON:
-
-{
-  "attribute_value": "Fire-Flink-x",
-  "action": "click | enter | upload | drag_and_drop",
-  "input_text": "",
-  "keyword": "",
-  "num_of_scrolls": "0",
-  "direction": "down",
-  "elementType": ""
-}
-`;
-  }
-  return prompt;
-}
+type FFInspectorArgs = {
+  stepAction: string;
+  extractedDomJson: string;
+  priorAndNextSteps: string[];
+  isAlert?: boolean;
+  isDrag?: boolean;
+};
+export async function ffInspectorNumExtractor({
+  stepAction,
+  extractedDomJson,
+  priorAndNextSteps,
+  isAlert = false,
+  isDrag = false,
+}: FFInspectorArgs
+): Promise<
+  string
+> {
+  const clickActions = ["click", "doubleclick", "rightclick"];
+  const alertActions = [
+    "ClickOkOnAlertPopup",
+    "ClickCancelOnAlertPopup",
+    "EnterDataToAlert",
+    "GetTextPresentOnAlertPopup",
+    "VerifyAlertPopupIsDisplayed",
+    "VerifyAlertPopUpMessageContainsString",
+    "WaitTillAlertIsPresent",
+    "VerifyTextPresentOnAlertPopup"
+  ];
+  const elementType = ["button", "link"]
+
+  let prompt;
+
+  // ---------------- ALERT ----------------
+  if (isAlert) {
+    prompt = `
+You are given a Selenium test step that requires interaction with a browser alert popup.
+
+Rules:
+- if the alert requires entering text or verification comparison, extract that text.
+- If no input is needed, set "input_text" to "None".
+- If step verifies text presence → VerifyTextPresentOnAlertPopup.
+- If step verifies text contains → VerifyAlertPopUpMessageContainsString.
+- Do not add explanations or extra text.
+- Return ONE action from this list only: ${JSON.stringify(alertActions)}.
+- Output must be valid JSON.
+Respond only with JSON using this format:
+
+{
+  "action": x,
+  "input_text": "text to enter" | "None"
+}
+
+`;
+  }
+
+  // ---------------- DRAG & DROP ----------------
+  else if (isDrag) {
+    prompt = `
+You are an intelligent assistant that extracts structured UI action data.
+
+Given a structured UI JSON representation with uniquely identified elements
+(ff-inspect values like Fire-Flink-1, Fire-Flink-2... in DOM order),
+locate the most appropriate elements for a drag-and-drop action.
+
+
+
+Rules:
+- Step: for the given step
+- Use the simplified JSON to find elements.
+- Identify source (drag) and destination (drop).
+- If not found, return Fire-Flink-0 for both.
+- Extract element names from step if possible.
+- Respond with JSON only.
+Simplified JSON:
+${extractedDomJson}
+
+Respond only with JSON using this format:
+{
+  "drag": "Fire-Flink-x",
+  "drop": "Fire-Flink-x",
+  "action": "Builtin_DragAndDropToElement",
+  "drag_element_name": "x",
+  "drop_element_name": "x"
+}
+`;
+  }
+
+  // ---------------- WAIT / SCROLL ----------------
+  else if (stepAction === "wait" || stepAction === "scroll") {
+    prompt = `
+You are an intelligent assistant that extracts structured UI action data.
+
+
+
+Rules:
+- Step: for the given step
+- Use context from test steps: ${JSON.stringify(priorAndNextSteps)}
+- Use simplified JSON to find best FF-inspect match.
+- Choose closest semantic + DOM-distance match.
+- Default direction is down.
+- Action:
+  - scroll → extract num_of_scrolls & direction
+  - wait → extract wait time (number only)
+- No nulls; use empty strings if needed.
+- The element type should be from this list ${elementType}
+
+Simplified JSON:
+${extractedDomJson}
+
+Return ONLY valid JSON:
+{
+  "attribute_value": "Fire-Flink-x",
+  "action": "x",
+  "input_text": "seconds in Interger | input text",
+  "keyword": "x",
+  "num_of_scrolls": "0",
+  "direction": "down",
+  "elementType":"x"
+}
+`;
+  }
+
+  // ---------------- DEFAULT (CLICK / ENTER / UPLOAD etc.) ----------------
+  else {
+    prompt = `
+You are a deterministic UI action extraction engine.
+
+OBJECTIVE:
+Identify the SINGLE best matching element from the Simplified JSON
+for the given step and return structured automation data.
+
+-----------------------------------------
+
+CONTEXT STEPS:
+${JSON.stringify(priorAndNextSteps)}
+
+CLICK ACTION WORDS:
+${JSON.stringify(clickActions)}
+
+ALLOWED ELEMENT TYPES:
+${elementType}
+-----------------------------------------
+
+CRITICAL MATCHING LOGIC:
+
+1. First understand intent:
+   - click words → click
+   - enter/type/write → enter_text
+   - upload → upload
+   - drag_and_drop → drag_and_drop
+
+2. For ENTER actions:
+   DO NOT pick first input blindly.
+   Score candidates using priority:
+
+   Priority Order:
+   (1) Exact label text match
+   (2) Placeholder match
+   (3) Name attribute semantic match
+   (4) ID semantic match
+   (5) type="email" if step contains "email"
+   (6) DOM sibling label relationship
+
+   Choose highest scoring match.
+
+   If NO strong semantic match → return:
+   attribute_value = "Fire-Flink-0"
+
+3. For CLICK actions:
+   - Prefer exact text match
+   - If multiple, choose closest contextual match
+   - If icon + svg exists, prefer svg element
+
+4. Never hallucinate.
+5. Only use FF values from provided JSON.
+6. No null values. Use empty string "".
+7. Output must be valid JSON only.
+
+-----------------------------------------
+SIMPLIFIED JSON:
+${extractedDomJson}
+-----------------------------------------
+
+Respond ONLY with JSON:
+
+{
+  "attribute_value": "Fire-Flink-x",
+  "action": "click | enter | upload | drag_and_drop",
+  "input_text": "",
+  "keyword": "",
+  "num_of_scrolls": "0",
+  "direction": "down",
+  "elementType": ""
+}
+`;
+  }
+  return prompt;
+}

package/src/ai/llmprompts/systemPrompts/userStoryToListPrompt.ts

@@ -1,24 +1,24 @@
-
-export async function buildStepExtractionPrompt(
-): Promise<string> {
-  const prompt = `
-You are a structured automation assistant. Your task is to extract step data from a provided user story for web automation.
-
-Requirements
-Each step must be a single string (not a tuple or any other data type).
-Preserve the original wording of each step.
-You may rewrite for clarity or formatting consistency only.
-Do not add new steps.
-Do not remove any existing steps.
-Do not introduce additional actions.
-Do not include the example list in the response.
-Do not include any extra text, commentary, markdown, or code fences.
-Output Format
-
-Respond strictly in valid JSON using the following structure:
-{
-"manaulSteps": ["step 1", "step 2", "step 3"]
-}
-`;
-  return prompt;
-}
+
+export async function buildStepExtractionPrompt(
+): Promise<string> {
+  const prompt = `
+You are a structured automation assistant. Your task is to extract step data from a provided user story for web automation.
+
+Requirements
+Each step must be a single string (not a tuple or any other data type).
+Preserve the original wording of each step.
+You may rewrite for clarity or formatting consistency only.
+Do not add new steps.
+Do not remove any existing steps.
+Do not introduce additional actions.
+Do not include the example list in the response.
+Do not include any extra text, commentary, markdown, or code fences.
+Output Format
+
+Respond strictly in valid JSON using the following structure:
+{
+"manaulSteps": ["step 1", "step 2", "step 3"]
+}
+`;
+  return prompt;
+}