@empiricalrun/test-gen 0.31.13 → 0.31.16

package/CHANGELOG.md

@@ -1,5 +1,23 @@
 # @empiricalrun/test-gen
 
+## 0.31.16
+
+### Patch Changes
+
+- ad4af33: fix: update test agent halucinating in larger context
+
+## 0.31.15
+
+### Patch Changes
+
+- 71898fe: fix: handle json string responses of o1
+
+## 0.31.14
+
+### Patch Changes
+
+- 89226da: fix: xpath leading to root element
+
 ## 0.31.13
 
 ### Patch Changes

package/dist/actions/utils/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/actions/utils/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAElC,OAAO,CAAC,MAAM,CAAC;IACb,UAAU,MAAM;QACd,UAAU,EAAE,GAAG,CAAC;QAChB,MAAM,EAAE,GAAG,CAAC;KACb;CACF;AAED,wBAAsB,oCAAoC,CACxD,WAAW,EAAE,MAAM,EACnB,KAAK,EAAE,MAAM,EACb,IAAI,EAAE,IAAI,gBA8CX;AAED,wBAAgB,cAAc,WAG7B"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/actions/utils/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAElC,OAAO,CAAC,MAAM,CAAC;IACb,UAAU,MAAM;QACd,UAAU,EAAE,GAAG,CAAC;QAChB,MAAM,EAAE,GAAG,CAAC;KACb;CACF;AAED,wBAAsB,oCAAoC,CACxD,WAAW,EAAE,MAAM,EACnB,KAAK,EAAE,MAAM,EACb,IAAI,EAAE,IAAI,gBAmEX;AAED,wBAAgB,cAAc,WAG7B"}

package/dist/actions/utils/index.js

@@ -8,17 +8,38 @@ async function getPlaywrightLocatorUsingCssSelector(cssSelector, xpath, page) {
         .replaceAll(":has-text", ":contains")
         .replaceAll(":text", ":contains");
     return await page.evaluate((locator) => {
-        // @ts-ignore
-        function getElementByXpath(path) {
-            return document.evaluate(path, document, null, XPathResult.FIRST_ORDERED_NODE_TYPE, null).singleNodeValue;
-        }
         let selectedElem;
-        try {
-            const xpath = locator.xpath?.replaceAll("text()", "normalize-space()");
-            selectedElem = getElementByXpath(xpath);
-        }
-        catch (e) {
-            // ignore error
+        if (locator.xpath) {
+            try {
+                const xpath = locator.xpath.replaceAll("text()", "normalize-space()");
+                const result = document.evaluate(xpath, document, null, XPathResult.ORDERED_NODE_ITERATOR_TYPE, null);
+                let node = result.iterateNext();
+                const elements = [];
+                // Store all matched elements
+                while (node) {
+                    elements.push(node);
+                    node = result.iterateNext();
+                }
+                // Find the least deep nesting (i.e., with the fewest child elements)
+                let minChildCount = Infinity;
+                const leastNestedElements = [];
+                elements.forEach((el) => {
+                    //@ts-ignore
+                    const childCount = el.children.length; // Count direct child elements
+                    if (childCount < minChildCount) {
+                        minChildCount = childCount;
+                        leastNestedElements.length = 0;
+                        leastNestedElements.push(el);
+                    }
+                    else if (childCount === minChildCount) {
+                        leastNestedElements.push(el);
+                    }
+                });
+                selectedElem = leastNestedElements[0];
+            }
+            catch (e) {
+                // ignore error
+            }
         }
         if (!selectedElem) {
             const elements = window.jQuery(locator.cssForJq);
@@ -32,7 +53,7 @@ async function getPlaywrightLocatorUsingCssSelector(cssSelector, xpath, page) {
             selectedElem = elements[elIdx];
         }
         else {
-            console.log("using xpath for locator");
+            console.log("using xpath locator");
         }
         if (!selectedElem) {
             throw Error(`Unable to find element, css: ${locator.css}`);

package/dist/agent/browsing/o1-completion.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"o1-completion.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/o1-completion.ts"],"names":[],"mappings":"AAAA,OAAO,EAAO,WAAW,EAAE,MAAM,mBAAmB,CAAC;AACrD,OAAO,EACL,qBAAqB,EACrB,0BAA0B,EAC1B,kBAAkB,EACnB,MAAM,4BAA4B,CAAC;AAKpC,wBAAsB,eAAe,CAAC,EACpC,QAAQ,EACR,KAAK,EACL,KAAK,GACN,EAAE;IACD,QAAQ,EAAE,0BAA0B,EAAE,CAAC;IACvC,KAAK,EAAE,kBAAkB,EAAE,CAAC;IAC5B,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,GAAG,OAAO,CAAC,qBAAqB,GAAG,SAAS,CAAC,CA0D7C"}
+{"version":3,"file":"o1-completion.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/o1-completion.ts"],"names":[],"mappings":"AAAA,OAAO,EAAO,WAAW,EAAE,MAAM,mBAAmB,CAAC;AACrD,OAAO,EACL,qBAAqB,EACrB,0BAA0B,EAC1B,kBAAkB,EACnB,MAAM,4BAA4B,CAAC;AAKpC,wBAAsB,eAAe,CAAC,EACpC,QAAQ,EACR,KAAK,EACL,KAAK,GACN,EAAE;IACD,QAAQ,EAAE,0BAA0B,EAAE,CAAC;IACvC,KAAK,EAAE,kBAAkB,EAAE,CAAC;IAC5B,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,GAAG,OAAO,CAAC,qBAAqB,GAAG,SAAS,CAAC,CA6D7C"}

package/dist/agent/browsing/o1-completion.js

@@ -50,7 +50,9 @@ async function getO1Completion({ messages, tools, trace, }) {
             type: "function",
             function: {
                 name: toolRespJSON.function.name,
-                arguments: JSON.stringify(toolRespJSON.function.parameters, null, 2),
+                arguments: typeof parameters === "string"
+                    ? parameters
+                    : JSON.stringify(parameters, null, 2),
             },
         };
         completion = {

package/dist/agent/codegen/update-flow.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"update-flow.d.ts","sourceRoot":"","sources":["../../../src/agent/codegen/update-flow.ts"],"names":[],"mappings":"AAAA,OAAO,EAKL,WAAW,EACZ,MAAM,mBAAmB,CAAC;AAqB3B,OAAO,EAAE,QAAQ,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAI7D,KAAK,eAAe,GAAG,QAAQ,GAAG;IAChC,YAAY,EAAE,MAAM,EAAE,CAAC;CACxB,CAAC;AAqIF,wBAAsB,UAAU,CAC9B,QAAQ,EAAE,QAAQ,EAClB,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,oBAAoB,GAAG,SAAS,EACzC,OAAO,GAAE,OAAc,EACvB,QAAQ,GAAE,OAAc,GACvB,OAAO,CAAC,eAAe,EAAE,CAAC,CAuF5B;AAED,wBAAsB,qBAAqB,CAAC,EAC1C,QAAQ,EACR,IAAI,EACJ,OAAO,EACP,KAAK,EACL,aAAoB,GACrB,EAAE;IACD,QAAQ,EAAE,QAAQ,CAAC;IACnB,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,CAAC,EAAE,oBAAoB,CAAC;IAC/B,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,aAAa,CAAC,EAAE,OAAO,CAAC;CACzB,GAAG,OAAO,CAAC,eAAe,EAAE,CAAC,CA6E7B"}
+{"version":3,"file":"update-flow.d.ts","sourceRoot":"","sources":["../../../src/agent/codegen/update-flow.ts"],"names":[],"mappings":"AAAA,OAAO,EAKL,WAAW,EACZ,MAAM,mBAAmB,CAAC;AAqB3B,OAAO,EAAE,QAAQ,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAI7D,KAAK,eAAe,GAAG,QAAQ,GAAG;IAChC,YAAY,EAAE,MAAM,EAAE,CAAC;CACxB,CAAC;AAqIF,wBAAsB,UAAU,CAC9B,QAAQ,EAAE,QAAQ,EAClB,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,oBAAoB,GAAG,SAAS,EACzC,OAAO,GAAE,OAAc,EACvB,QAAQ,GAAE,OAAc,GACvB,OAAO,CAAC,eAAe,EAAE,CAAC,CA2F5B;AAED,wBAAsB,qBAAqB,CAAC,EAC1C,QAAQ,EACR,IAAI,EACJ,OAAO,EACP,KAAK,EACL,aAAoB,GACrB,EAAE;IACD,QAAQ,EAAE,QAAQ,CAAC;IACnB,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,CAAC,EAAE,oBAAoB,CAAC;IAC/B,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,aAAa,CAAC,EAAE,OAAO,CAAC;CACzB,GAAG,OAAO,CAAC,eAAe,EAAE,CAAC,CA6E7B"}

package/dist/agent/codegen/update-flow.js

@@ -141,13 +141,19 @@ async function updateTest(testCase, file, options, logging = true, validate = tr
             return text;
         }, "") + testCase.name
         : testCase.name;
+    const { testBlock: currentScenarioCodeBlock } = (0, web_1.getTypescriptTestBlock)({
+        scenarioName: testCase?.name,
+        content: testFileContent,
+        suites: testCase?.suites || [],
+    });
     const instruction = await (0, llm_1.getPrompt)(promptName, {
         testFiles: codePrompt,
         pageFiles: pomPrompt,
         scenarioName,
         scenarioSteps: testCase.steps.join("\n"),
         scenarioFile: file,
-    }, 14);
+        currentScenarioCodeBlock,
+    });
     promptSpan.end({ output: { instruction } });
     const llm = new llm_1.LLM({
         trace,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@empiricalrun/test-gen",
-  "version": "0.31.13",
+  "version": "0.31.16",
   "publishConfig": {
     "registry": "https://registry.npmjs.org/",
     "access": "public"
@@ -44,9 +44,9 @@
     "ts-morph": "^23.0.0",
     "tsx": "^4.16.2",
     "typescript": "^5.3.3",
-    "@empiricalrun/r2-uploader": "^0.3.3",
+    "@empiricalrun/llm": "^0.9.3",
     "@empiricalrun/reporter": "^0.20.2",
-    "@empiricalrun/llm": "^0.9.3"
+    "@empiricalrun/r2-uploader": "^0.3.3"
   },
   "devDependencies": {
     "@types/detect-port": "^1.3.5",