@empiricalrun/test-gen 0.42.9 → 0.42.11

package/CHANGELOG.md

@@ -1,5 +1,22 @@
 # @empiricalrun/test-gen
 
+## 0.42.11
+
+### Patch Changes
+
+- 98a8417: fix: handle JSON parse with try and catch
+
+## 0.42.10
+
+### Patch Changes
+
+- ad2e2c5: fix: annotations for small elements should be outside the container
+- 6471bf9: test: updated timeouts
+- fabdfd6: fix: moved lexical vars check to o3-mini and increased timeout
+- 61bc738: fix: added message for scroll on dashboard
+- Updated dependencies [fabdfd6]
+  - @empiricalrun/llm@0.9.31
+
 ## 0.42.9
 
 ### Patch Changes

package/dist/agent/browsing/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,GAAG,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AACrD,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAElC,OAAO,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAClD,OAAO,EAAE,YAAY,EAAE,MAAM,kBAAkB,CAAC;AAIhD,OAAO,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAMnD,MAAM,MAAM,oBAAoB,GAAG,OAAO,CAAC,oBAAoB,CAAC,GAAG;IACjE,YAAY,CAAC,EAAE;QACb,iBAAiB,CAAC,EAAE,MAAM,EAAE,CAAC;KAC9B,CAAC;CACH,CAAC;AAEF,wBAAsB,6BAA6B,CAAC,EAClD,KAAK,EACL,MAAM,EACN,MAAM,EACN,IAAI,EACJ,OAAO,EACP,GAAG,EACH,OAAO,GACR,EAAE;IACD,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,MAAM,EAAE,YAAY,CAAC;IACrB,IAAI,EAAE,IAAI,CAAC;IACX,OAAO,EAAE,oBAAoB,CAAC;IAC9B,GAAG,EAAE,GAAG,CAAC;IACT,OAAO,EAAE,iBAAiB,CAAC;CAC5B,GAAG,OAAO,CAAC,MAAM,EAAE,GAAG,SAAS,CAAC,CAiJhC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,GAAG,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AACrD,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAElC,OAAO,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAClD,OAAO,EAAE,YAAY,EAAE,MAAM,kBAAkB,CAAC;AAIhD,OAAO,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAOnD,MAAM,MAAM,oBAAoB,GAAG,OAAO,CAAC,oBAAoB,CAAC,GAAG;IACjE,YAAY,CAAC,EAAE;QACb,iBAAiB,CAAC,EAAE,MAAM,EAAE,CAAC;KAC9B,CAAC;CACH,CAAC;AAEF,wBAAsB,6BAA6B,CAAC,EAClD,KAAK,EACL,MAAM,EACN,MAAM,EACN,IAAI,EACJ,OAAO,EACP,GAAG,EACH,OAAO,GACR,EAAE;IACD,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,MAAM,EAAE,YAAY,CAAC;IACrB,IAAI,EAAE,IAAI,CAAC;IACX,OAAO,EAAE,oBAAoB,CAAC;IAC9B,GAAG,EAAE,GAAG,CAAC;IACT,OAAO,EAAE,iBAAiB,CAAC;CAC5B,GAAG,OAAO,CAAC,MAAM,EAAE,GAAG,SAAS,CAAC,CAiJhC"}

package/dist/agent/browsing/index.js

@@ -5,9 +5,10 @@ const constants_1 = require("../../constants");
 const reporter_1 = require("../../reporter");
 const session_1 = require("../../session");
 const html_1 = require("../../utils/html");
+const utils_1 = require("../utils");
 const verification_1 = require("../verification");
 const o1_completion_1 = require("./o1-completion");
-const utils_1 = require("./utils");
+const utils_2 = require("./utils");
 async function executeTaskUsingBrowsingAgent({ trace, action, logger, page, options, llm, actions, }) {
     let isTaskDone = false;
     const executedActions = [];
@@ -61,7 +62,7 @@ async function executeTaskUsingBrowsingAgent({ trace, action, logger, page, opti
                 break;
             }
         }
-        const messages = await (0, utils_1.getPromptForNextAction)({
+        const messages = await (0, utils_2.getPromptForNextAction)({
             pageSnapshot,
             previousActions: successfulActions,
             task: action,
@@ -106,13 +107,13 @@ async function executeTaskUsingBrowsingAgent({ trace, action, logger, page, opti
                 break;
             }
             try {
-                const code = await actions.executeAction(toolCall.function.name, JSON.parse(toolCall.function.arguments), toolCallsSpan);
+                const code = await actions.executeAction(toolCall.function.name, (0, utils_1.parseJson)(toolCall.function.arguments), toolCallsSpan);
                 if (code) {
                     generatedCodeSteps.push(code);
                 }
                 executedActions.push({
                     isError: false,
-                    action: JSON.parse(toolCall.function.arguments)?.reason,
+                    action: (0, utils_1.parseJson)(toolCall.function.arguments)?.reason,
                 });
                 lastActionExecTrace = "";
             }
@@ -120,7 +121,7 @@ async function executeTaskUsingBrowsingAgent({ trace, action, logger, page, opti
                 // TODO: implement feedback loop to llm
                 executedActions.push({
                     isError: true,
-                    action: JSON.parse(toolCall.function.arguments)?.reason,
+                    action: (0, utils_1.parseJson)(toolCall.function.arguments)?.reason,
                 });
                 lastActionExecTrace = e.message;
                 void testgenUpdatesReporter.sendMessage(e.message);

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,CA8D7C"}
+{"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;AAMpC,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,CA8D7C"}

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

@@ -7,6 +7,7 @@ exports.getO1Completion = void 0;
 const llm_1 = require("@empiricalrun/llm");
 const remove_markdown_1 = __importDefault(require("remove-markdown"));
 const constants_1 = require("../../constants");
+const utils_1 = require("../utils");
 async function getO1Completion({ messages, tools, trace, }) {
     let completion;
     try {
@@ -42,7 +43,7 @@ async function getO1Completion({ messages, tools, trace, }) {
         }));
         o1Span?.end({ output: response });
         const toolResponseStr = (0, remove_markdown_1.default)(response.content);
-        const toolRespJSON = JSON.parse(toolResponseStr);
+        const toolRespJSON = (0, utils_1.parseJson)(toolResponseStr);
         const parameters = toolRespJSON.function.parameters || toolRespJSON.function.arguments;
         if (!parameters) {
             throw new Error("No parameters found in tool response");

package/dist/agent/codegen/lexical-scoped-vars.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"lexical-scoped-vars.d.ts","sourceRoot":"","sources":["../../../src/agent/codegen/lexical-scoped-vars.ts"],"names":[],"mappings":"AAAA,OAAO,EAAO,WAAW,EAAE,MAAM,mBAAmB,CAAC;AASrD,OAAO,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAGnD,wBAAsB,oBAAoB,CAAC,EACzC,KAAK,EACL,IAAI,EACJ,cAAc,EACd,OAAO,GACR,EAAE;IACD,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,IAAI,EAAE,MAAM,CAAC;IACb,cAAc,EAAE,MAAM,CAAC;IACvB,OAAO,CAAC,EAAE,oBAAoB,CAAC;CAChC,qBAqDA"}
+{"version":3,"file":"lexical-scoped-vars.d.ts","sourceRoot":"","sources":["../../../src/agent/codegen/lexical-scoped-vars.ts"],"names":[],"mappings":"AAAA,OAAO,EAAO,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAQrD,OAAO,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAGnD,wBAAsB,oBAAoB,CAAC,EACzC,KAAK,EACL,IAAI,EACJ,cAAc,EACd,OAAO,GACR,EAAE;IACD,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,IAAI,EAAE,MAAM,CAAC;IACb,cAAc,EAAE,MAAM,CAAC;IACvB,OAAO,CAAC,EAAE,oBAAoB,CAAC;CAChC,qBAoDA"}

package/dist/agent/codegen/lexical-scoped-vars.js

@@ -16,14 +16,13 @@ async function getLexicalScopedVars({ trace, file, referencePoint, options, }) {
     const llm = new llm_1.LLM({
         trace: fetchLexicalScopedVarsSpan,
         provider: options?.modelProvider || constants_1.DEFAULT_MODEL_PROVIDER,
-        defaultModel: options?.model || constants_1.DEFAULT_MODEL,
+        defaultModel: options?.model || "o3-mini-2025-01-31",
         providerApiKey: constants_1.MODEL_API_KEYS[options?.modelProvider || constants_1.DEFAULT_MODEL_PROVIDER],
     });
     const message = await llm.createChatCompletion({
         messages,
         modelParameters: {
-            ...constants_1.DEFAULT_MODEL_PARAMETERS,
-            ...options?.modelParameters,
+            max_completion_tokens: constants_1.DEFAULT_O1_MODEL_PARAMETERS.max_completion_tokens,
         },
         responseFormat: {
             type: "json_schema",

package/dist/agent/master/element-annotation.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"element-annotation.d.ts","sourceRoot":"","sources":["../../../src/agent/master/element-annotation.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,GAAG,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAGrD,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAOlC,OAAO,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AACnD,OAAO,EAAE,UAAU,EAAE,MAAM,qBAAqB,CAAC;AA0DjD,wBAAsB,oBAAoB,CAAC,EACzC,kBAAkB,EAClB,WAAW,EACX,mBAAmB,EACnB,KAAK,EACL,GAAG,EACH,OAAO,EACP,UAAU,GACX,EAAE;IACD,kBAAkB,EAAE,MAAM,CAAC;IAC3B,WAAW,EAAE,MAAM,CAAC;IACpB,mBAAmB,EAAE,MAAM,CAAC;IAC5B,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,GAAG,CAAC,EAAE,GAAG,CAAC;IACV,OAAO,CAAC,EAAE,oBAAoB,CAAC;IAC/B,UAAU,EAAE,oBAAoB,CAAC;CAClC,GAAG,OAAO,CAAC,MAAM,GAAG,SAAS,CAAC,CA6F9B;AAED,MAAM,MAAM,oBAAoB,GAAG;IACjC,UAAU,EACN,KAAK,GACL,UAAU,CAAC,IAAI,GACf,UAAU,CAAC,WAAW,GACtB,UAAU,CAAC,MAAM,CAAC;IACtB,aAAa,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;CACpC,CAAC;AAEF,wBAAsB,iBAAiB,CAAC,EACtC,IAAI,EACJ,UAAU,EACV,OAAO,GACR,EAAE;IACD,IAAI,EAAE,IAAI,CAAC;IACX,UAAU,EAAE,oBAAoB,CAAC;IACjC,OAAO,EAAE,oBAAoB,CAAC;CAC/B,GAAG,OAAO,CAAC;IACV,cAAc,EAAE;QAAE,SAAS,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,EAAE,CAAC;IACtD,gBAAgB,EAAE,MAAM,CAAC;IACzB,uBAAuB,EAAE,MAAM,CAAC;CACjC,CAAC,CAqDD"}
+{"version":3,"file":"element-annotation.d.ts","sourceRoot":"","sources":["../../../src/agent/master/element-annotation.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,GAAG,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAGrD,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAOlC,OAAO,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAEnD,OAAO,EAAE,UAAU,EAAE,MAAM,qBAAqB,CAAC;AA0DjD,wBAAsB,oBAAoB,CAAC,EACzC,kBAAkB,EAClB,WAAW,EACX,mBAAmB,EACnB,KAAK,EACL,GAAG,EACH,OAAO,EACP,UAAU,GACX,EAAE;IACD,kBAAkB,EAAE,MAAM,CAAC;IAC3B,WAAW,EAAE,MAAM,CAAC;IACpB,mBAAmB,EAAE,MAAM,CAAC;IAC5B,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,GAAG,CAAC,EAAE,GAAG,CAAC;IACV,OAAO,CAAC,EAAE,oBAAoB,CAAC;IAC/B,UAAU,EAAE,oBAAoB,CAAC;CAClC,GAAG,OAAO,CAAC,MAAM,GAAG,SAAS,CAAC,CAgG9B;AAED,MAAM,MAAM,oBAAoB,GAAG;IACjC,UAAU,EACN,KAAK,GACL,UAAU,CAAC,IAAI,GACf,UAAU,CAAC,WAAW,GACtB,UAAU,CAAC,MAAM,CAAC;IACtB,aAAa,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;CACpC,CAAC;AAEF,wBAAsB,iBAAiB,CAAC,EACtC,IAAI,EACJ,UAAU,EACV,OAAO,GACR,EAAE;IACD,IAAI,EAAE,IAAI,CAAC;IACX,UAAU,EAAE,oBAAoB,CAAC;IACjC,OAAO,EAAE,oBAAoB,CAAC;CAC/B,GAAG,OAAO,CAAC;IACV,cAAc,EAAE;QAAE,SAAS,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,EAAE,CAAC;IACtD,gBAAgB,EAAE,MAAM,CAAC;IACzB,uBAAuB,EAAE,MAAM,CAAC;CACjC,CAAC,CAqDD"}

package/dist/agent/master/element-annotation.js

@@ -4,6 +4,7 @@ exports.getAnnotationKeys = exports.getElementAnnotation = void 0;
 const llm_1 = require("@empiricalrun/llm");
 const vision_1 = require("@empiricalrun/llm/vision");
 const constants_1 = require("../../constants");
+const utils_1 = require("../utils");
 const annotationToolAction = {
     name: "element_annotation",
     schema: {
@@ -129,7 +130,7 @@ async function getElementAnnotation({ elementDescription, annotations, annotated
             ...constants_1.DEFAULT_MODEL_PARAMETERS,
             ...options?.modelParameters,
             tool_choice: "required",
-            temperature: 1,
+            temperature: 0.5,
         },
         trace: annotationsSpan,
         traceName: "get-element-from-action",
@@ -138,8 +139,11 @@ async function getElementAnnotation({ elementDescription, annotations, annotated
     });
     const toolCall = completion?.tool_calls?.[0];
     annotationsSpan?.end({ output: toolCall });
-    const args = JSON.parse(toolCall.function.arguments);
-    return args.element_annotation;
+    if (toolCall) {
+        const args = (0, utils_1.parseJson)(toolCall.function.arguments);
+        return args.element_annotation;
+    }
+    return;
 }
 exports.getElementAnnotation = getElementAnnotation;
 async function getAnnotationKeys({ page, preference, options, }) {

package/dist/agent/master/planner.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"planner.d.ts","sourceRoot":"","sources":["../../../src/agent/master/planner.ts"],"names":[],"mappings":"AAAA,OAAO,EAAO,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAGrD,OAAO,EAAE,IAAI,EAAE,MAAM,iBAAiB,CAAC;AAIvC,wBAAsB,4BAA4B,CAAC,EACjD,KAAK,EACL,IAAI,EACJ,YAAY,EACZ,KAAK,EACL,IAAI,EACJ,WAAW,GACZ,EAAE;IACD,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAC5B,IAAI,EAAE,IAAI,CAAC;IACX,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;;;;GAwIA"}
+{"version":3,"file":"planner.d.ts","sourceRoot":"","sources":["../../../src/agent/master/planner.ts"],"names":[],"mappings":"AAAA,OAAO,EAAO,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAGrD,OAAO,EAAE,IAAI,EAAE,MAAM,iBAAiB,CAAC;AAKvC,wBAAsB,4BAA4B,CAAC,EACjD,KAAK,EACL,IAAI,EACJ,YAAY,EACZ,KAAK,EACL,IAAI,EACJ,WAAW,GACZ,EAAE;IACD,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAC5B,IAAI,EAAE,IAAI,CAAC;IACX,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;;;;GAwIA"}

package/dist/agent/master/planner.js

@@ -4,6 +4,7 @@ exports.runtimePlannerWithScreenshot = void 0;
 const llm_1 = require("@empiricalrun/llm");
 const vision_1 = require("@empiricalrun/llm/vision");
 const constants_1 = require("../../constants");
+const utils_1 = require("../utils");
 async function runtimePlannerWithScreenshot({ trace, task, conversation, pages, page, currentPage, }) {
     const buffer = await page.screenshot({
     //This is done to improve element annotation accuracy, anyways it doesn't annotate elements which are out of viewport
@@ -119,7 +120,7 @@ async function runtimePlannerWithScreenshot({ trace, task, conversation, pages,
     });
     const toolCallResp = (response?.tool_calls || [])[0];
     if (toolCallResp) {
-        const toolCall = JSON.parse(toolCallResp.function.arguments);
+        const toolCall = (0, utils_1.parseJson)(toolCallResp.function.arguments);
         const output = {
             pageName: toolCall.pageName,
             isDone: toolCall.isDone,

package/dist/agent/master/run.js

@@ -198,7 +198,7 @@ async function createTestUsingMasterAgent({ task, page, testCase, options, scope
                     });
                     if (annotationKeys.length > 0) {
                         const annotationMapString = annotationKeys
-                            ?.map((a) => `${a.elementID}:${a.text}`)
+                            ?.map((a) => `${a.elementID}: ${a.text}`)
                             .join("\n");
                         // Provides the annotations for all the element present on screen
                         // Also provides the annotation of element on which the action needs to be taken

package/dist/agent/master/scroller.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"scroller.d.ts","sourceRoot":"","sources":["../../../src/agent/master/scroller.ts"],"names":[],"mappings":"AAAA,OAAO,EAAuB,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAGrE,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAElC,OAAO,EAAE,YAAY,EAAE,MAAM,kBAAkB,CAAC;AAahD,MAAM,MAAM,cAAc,GAAG;IAC3B,cAAc,EAAE,MAAM,CAAC;IACvB,eAAe,EAAE,MAAM,CAAC;CACzB,CAAC;AAoZF,wBAAsB,QAAQ,CAAC,EAC7B,kBAAkB,EAClB,IAAI,EACJ,KAAK,EACL,cAAc,EACd,MAAM,GACP,EAAE;IACD,kBAAkB,EAAE,MAAM,CAAC;IAC3B,IAAI,EAAE,IAAI,CAAC;IACX,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,cAAc,CAAC,EAAE,cAAc,CAAC;IAChC,MAAM,CAAC,EAAE,YAAY,CAAC;CACvB,GAAG,OAAO,CAAC,cAAc,EAAE,CAAC,CA6D5B"}
+{"version":3,"file":"scroller.d.ts","sourceRoot":"","sources":["../../../src/agent/master/scroller.ts"],"names":[],"mappings":"AAAA,OAAO,EAAuB,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAGrE,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAElC,OAAO,EAAE,YAAY,EAAE,MAAM,kBAAkB,CAAC;AAchD,MAAM,MAAM,cAAc,GAAG;IAC3B,cAAc,EAAE,MAAM,CAAC;IACvB,eAAe,EAAE,MAAM,CAAC;CACzB,CAAC;AA2ZF,wBAAsB,QAAQ,CAAC,EAC7B,kBAAkB,EAClB,IAAI,EACJ,KAAK,EACL,cAAc,EACd,MAAM,GACP,EAAE;IACD,kBAAkB,EAAE,MAAM,CAAC;IAC3B,IAAI,EAAE,IAAI,CAAC;IACX,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,cAAc,CAAC,EAAE,cAAc,CAAC;IAChC,MAAM,CAAC,EAAE,YAAY,CAAC;CACvB,GAAG,OAAO,CAAC,cAAc,EAAE,CAAC,CA6D5B"}

package/dist/agent/master/scroller.js

@@ -4,6 +4,7 @@ exports.scroller = void 0;
 const llm_1 = require("@empiricalrun/llm");
 const vision_1 = require("@empiricalrun/llm/vision");
 const constants_1 = require("../../constants");
+const reporter_1 = require("../../reporter");
 const utils_1 = require("../utils");
 const action_tool_calls_1 = require("./action-tool-calls");
 const element_annotation_1 = require("./element-annotation");
@@ -43,7 +44,9 @@ async function isScrollable({ scrollDirection, page, divAnnotation, }) {
 // If the div annotation is undefined, we scroll the page
 // else we scroll the element
 async function scroll({ scrollBy, page, direction, divAnnotation, logger, }) {
+    const testGenUpdatesReporter = new reporter_1.TestGenUpdatesReporter();
     if (divAnnotation) {
+        await testGenUpdatesReporter.sendMessage("Scrolling the div since element is not in view");
         logger?.log("Scrolling the div since element is not in view");
         return await page.evaluate(({ scrollBy, direction, divAnnotation }) => {
             if (
@@ -58,6 +61,7 @@ async function scroll({ scrollBy, page, direction, divAnnotation, logger, }) {
             return 0;
         }, { scrollBy, direction, divAnnotation });
     }
+    await testGenUpdatesReporter.sendMessage("Scrolling the page since element is not in view");
     logger?.log("Scrolling the page since element is not in view");
     return await page.evaluate(({ scrollBy, direction }) => {
         let scrollHeight = scrollBy || window.innerHeight;

package/dist/agent/planner/run-time-planner.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"run-time-planner.d.ts","sourceRoot":"","sources":["../../../src/agent/planner/run-time-planner.ts"],"names":[],"mappings":"AAAA,OAAO,EAAO,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAGrD;;;;;;GAMG;AACH,wBAAsB,cAAc,CAAC,EACnC,KAAK,EACL,IAAI,EACJ,YAAY,EACZ,KAAK,EACL,WAAW,GACZ,EAAE;IACD,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAC5B,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;;;;GAiHA"}
+{"version":3,"file":"run-time-planner.d.ts","sourceRoot":"","sources":["../../../src/agent/planner/run-time-planner.ts"],"names":[],"mappings":"AAAA,OAAO,EAAO,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAKrD;;;;;;GAMG;AACH,wBAAsB,cAAc,CAAC,EACnC,KAAK,EACL,IAAI,EACJ,YAAY,EACZ,KAAK,EACL,WAAW,GACZ,EAAE;IACD,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAC5B,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;;;;GAiHA"}

package/dist/agent/planner/run-time-planner.js

@@ -2,6 +2,7 @@
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.runtimePlanner = void 0;
 const llm_1 = require("@empiricalrun/llm");
+const utils_1 = require("../utils");
 /**
  * This agent is used to divide the tasl into individual actions and then
  * compare each action against the actions listed in the conversation.
@@ -101,7 +102,7 @@ ${currentPage}
     });
     const toolCallResp = (response?.tool_calls || [])[0];
     if (toolCallResp) {
-        const toolCall = JSON.parse(toolCallResp.function.arguments);
+        const toolCall = (0, utils_1.parseJson)(toolCallResp.function.arguments);
         const output = {
             pageName: toolCall.pageName,
             isDone: toolCall.isDone,

package/dist/agent/utils.js

@@ -6,7 +6,7 @@ function parseJson(args) {
         return JSON.parse(args);
     }
     catch (e) {
-        console.error("Failed to parse JSON", e);
+        console.error(`Failed to parse JSON with args ${args}`, e);
     }
 }
 exports.parseJson = parseJson;

package/dist/agent/verification/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/verification/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAkB,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAEhE;;GAEG;AACH,wBAAsB,iBAAiB,CAAC,EACtC,KAAK,EACL,IAAI,EACJ,YAAY,GACb,EAAE;IACD,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,IAAI,EAAE,MAAM,CAAC;CACd;;;GA+EA"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/verification/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAkB,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAIhE;;GAEG;AACH,wBAAsB,iBAAiB,CAAC,EACtC,KAAK,EACL,IAAI,EACJ,YAAY,GACb,EAAE;IACD,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,IAAI,EAAE,MAAM,CAAC;CACd;;;GA+EA"}

package/dist/agent/verification/index.js

@@ -2,6 +2,7 @@
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.verificationAgent = void 0;
 const llm_1 = require("@empiricalrun/llm");
+const utils_1 = require("../utils");
 /**
  * This agent is used to verify whether the task is done basis the conversation history
  */
@@ -61,7 +62,7 @@ async function verificationAgent({ trace, task, conversation, }) {
     });
     const toolCallResp = (response?.tool_calls || [])[0];
     if (toolCallResp) {
-        const toolCall = JSON.parse(toolCallResp.function.arguments);
+        const toolCall = (0, utils_1.parseJson)(toolCallResp.function.arguments);
         const output = {
             isDone: toolCall.isDone,
             reason: toolCall.reason,

package/dist/bin/utils/platform/web/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../../src/bin/utils/platform/web/index.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAMhD,OAAO,EAGL,IAAI,EAEJ,UAAU,EAEX,MAAM,UAAU,CAAC;AAGlB,OAAO,EAAE,QAAQ,EAAE,MAAM,mBAAmB,CAAC;AAE7C,eAAO,MAAM,gCAAgC,eAC/B,UAAU,KACrB,MAgBF,CAAC;AAEF;;;;;;GAMG;AACH,wBAAgB,sBAAsB,CAAC,EACrC,YAAY,EACZ,MAAM,EACN,OAAO,GACR,EAAE;IACD,YAAY,EAAE,MAAM,CAAC;IACrB,MAAM,CAAC,EAAE,MAAM,EAAE,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;CACjB,GAAG;IACF,SAAS,EAAE,MAAM,GAAG,SAAS,CAAC;IAC9B,QAAQ,EAAE,IAAI,GAAG,SAAS,CAAC;IAC3B,SAAS,EAAE,MAAM,CAAC;CACnB,CA2CA;AAwBD,wBAAsB,0CAA0C,CAC9D,QAAQ,EAAE,MAAM,oBA+BjB;AAED;;;;;;;;;;;;;;;;;;;;;;GAsBG;AACH,wBAAgB,4BAA4B,CAC1C,IAAI,EAAE,IAAI,GAAG,SAAS,GACrB,IAAI,GAAG,SAAS,CA4BlB;AAED,wBAAgB,iBAAiB,CAAC,SAAS,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,MAAM,CAG5E;AAED,wBAAgB,kBAAkB,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,EAAE,CA8C7D;AAED,wBAAsB,sBAAsB,CAC1C,OAAO,EAAE,MAAM,EACf,QAAQ,EAAE,MAAM,mCAWjB;AAED,wBAAsB,UAAU,CAAC,QAAQ,EAAE,MAAM,iBAWhD;AAED,wBAAsB,UAAU,CAAC,QAAQ,EAAE,MAAM,EAAE,KAAK,CAAC,EAAE,WAAW,iBAgBrE;AAED,wBAAgB,YAAY,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,EAAE,GAAG,EAAE,MAAM,UAE5E;AAED,wBAAsB,cAAc,CAAC,QAAQ,EAAE,MAAM,iBAMpD;AAED,wBAAgB,oBAAoB,CAAC,QAAQ,EAAE,MAAM,UAcpD;AAED,wBAAsB,iCAAiC,CAAC,QAAQ,EAAE,MAAM,+BAoBvE;AA+CD,wBAAgB,4BAA4B,CAC1C,QAAQ,EAAE,MAAM,EAChB,QAAQ,EAAE,MAAM,EAChB,aAAa,EAAE,MAAM,UAoCtB;AAED,eAAO,MAAM,6BAA6B;qBAKvB,MAAM;iBACV,MAAM;YACX,MAAM,EAAE;YA2DjB,CAAC;AAEF,eAAO,MAAM,iCAAiC,cACjC,MAAM,EAAE,gBACL,MAAM,sBAyBrB,CAAC;AAEF,wBAAsB,qBAAqB,CAAC,EAC1C,YAAY,EACZ,QAAQ,EACR,MAAM,GACP,EAAE;IACD,YAAY,EAAE,MAAM,CAAC;IACrB,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,EAAE,CAAC;CAClB,iBAgDA;AAED,wBAAsB,uBAAuB,CAC3C,QAAQ,EAAE,MAAM,EAChB,cAAc,EAAE,MAAM,EAAE,iBAsBzB;AAED,wBAAgB,aAAa,CAAC,EAC5B,QAAQ,EACR,QAAQ,GACT,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,QAAQ,CAAC;CACpB,WAYA;AAED,wBAAgB,mBAAmB,CAAC,EAClC,QAAQ,EACR,MAAM,GACP,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,EAAE,CAAC;CAClB,UAOA;AAED,wBAAgB,+BAA+B,CAAC,UAAU,EAAE,MAAM,GAAG,MAAM,EAAE,CA4B5E;AAED,wBAAgB,aAAa,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAQnD"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../../src/bin/utils/platform/web/index.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAMhD,OAAO,EAGL,IAAI,EAEJ,UAAU,EAEX,MAAM,UAAU,CAAC;AAGlB,OAAO,EAAE,QAAQ,EAAE,MAAM,mBAAmB,CAAC;AAE7C,eAAO,MAAM,gCAAgC,eAC/B,UAAU,KACrB,MAgBF,CAAC;AAEF;;;;;;GAMG;AACH,wBAAgB,sBAAsB,CAAC,EACrC,YAAY,EACZ,MAAM,EACN,OAAO,GACR,EAAE;IACD,YAAY,EAAE,MAAM,CAAC;IACrB,MAAM,CAAC,EAAE,MAAM,EAAE,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;CACjB,GAAG;IACF,SAAS,EAAE,MAAM,GAAG,SAAS,CAAC;IAC9B,QAAQ,EAAE,IAAI,GAAG,SAAS,CAAC;IAC3B,SAAS,EAAE,MAAM,CAAC;CACnB,CA2CA;AAwBD,wBAAsB,0CAA0C,CAC9D,QAAQ,EAAE,MAAM,oBA+BjB;AAED;;;;;;;;;;;;;;;;;;;;;;GAsBG;AACH,wBAAgB,4BAA4B,CAC1C,IAAI,EAAE,IAAI,GAAG,SAAS,GACrB,IAAI,GAAG,SAAS,CA4BlB;AAED,wBAAgB,iBAAiB,CAAC,SAAS,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,MAAM,CAG5E;AAED,wBAAgB,kBAAkB,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,EAAE,CA8C7D;AAED,wBAAsB,sBAAsB,CAC1C,OAAO,EAAE,MAAM,EACf,QAAQ,EAAE,MAAM,mCAWjB;AAED,wBAAsB,UAAU,CAAC,QAAQ,EAAE,MAAM,iBAShD;AAED,wBAAsB,UAAU,CAAC,QAAQ,EAAE,MAAM,EAAE,KAAK,CAAC,EAAE,WAAW,iBAgBrE;AAED,wBAAgB,YAAY,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,EAAE,GAAG,EAAE,MAAM,UAE5E;AAED,wBAAsB,cAAc,CAAC,QAAQ,EAAE,MAAM,iBAMpD;AAED,wBAAgB,oBAAoB,CAAC,QAAQ,EAAE,MAAM,UAcpD;AAED,wBAAsB,iCAAiC,CAAC,QAAQ,EAAE,MAAM,+BAoBvE;AA+CD,wBAAgB,4BAA4B,CAC1C,QAAQ,EAAE,MAAM,EAChB,QAAQ,EAAE,MAAM,EAChB,aAAa,EAAE,MAAM,UAoCtB;AAED,eAAO,MAAM,6BAA6B;qBAKvB,MAAM;iBACV,MAAM;YACX,MAAM,EAAE;YA2DjB,CAAC;AAEF,eAAO,MAAM,iCAAiC,cACjC,MAAM,EAAE,gBACL,MAAM,sBAyBrB,CAAC;AAEF,wBAAsB,qBAAqB,CAAC,EAC1C,YAAY,EACZ,QAAQ,EACR,MAAM,GACP,EAAE;IACD,YAAY,EAAE,MAAM,CAAC;IACrB,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,EAAE,CAAC;CAClB,iBAgDA;AAED,wBAAsB,uBAAuB,CAC3C,QAAQ,EAAE,MAAM,EAChB,cAAc,EAAE,MAAM,EAAE,iBAsBzB;AAED,wBAAgB,aAAa,CAAC,EAC5B,QAAQ,EACR,QAAQ,GACT,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,QAAQ,CAAC;CACpB,WAYA;AAED,wBAAgB,mBAAmB,CAAC,EAClC,QAAQ,EACR,MAAM,GACP,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,EAAE,CAAC;CAClB,UAOA;AAED,wBAAgB,+BAA+B,CAAC,UAAU,EAAE,MAAM,GAAG,MAAM,EAAE,CA4B5E;AAED,wBAAgB,aAAa,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAQnD"}

package/dist/bin/utils/platform/web/index.js

@@ -229,8 +229,6 @@ exports.stripAndPrependImports = stripAndPrependImports;
 async function lintErrors(filePath) {
     const eslint = new eslint_1.ESLint({
         fix: true,
-        // useEslintrc: false,
-        // overrideConfigFile: path.join(process.cwd(), ".eslintrc.js"),
         useEslintrc: true,
     });
     const [result] = await eslint.lintFiles(filePath);

package/dist/browser-injected-scripts/annotate-elements.js

@@ -311,10 +311,7 @@ function annotateElementsWithPreference({
     // Style the marker
     Object.assign(marker.style, {
       position: "absolute",
-      top: `${rect.top + windowToAnnotate.scrollY}px`,
-      left: `${rect.left + windowToAnnotate.scrollX}px`,
-      background:
-        "-webkit-gradient(linear, 0% 0%, 0% 100%, from(rgb(255, 247, 133)), to(rgb(255, 197, 66)))",
+      background: "rgb(255, 197, 66)",
       padding: "1px 3px 0px",
       borderRadius: "3px",
       border: "1px solid rgb(227, 190, 35)",
@@ -331,10 +328,29 @@ function annotateElementsWithPreference({
       fontFamily: "Helvetica, Arial, sans-serif",
       fontWeight: "bold",
       textShadow: "rgba(255, 255, 255, 0.6) 0px 1px 0px",
+      visibility: "hidden", // Setting the visibility to hidden initially, to get the height and width of marker
     });
 
-    // Attach the marker to the specified parent element
     parentElement.appendChild(marker);
+    let markerRect = marker.getBoundingClientRect();
+    let top = rect.top + windowToAnnotate.scrollY;
+    let left = rect.left + windowToAnnotate.scrollX;
+
+    // If the target element is smaller, we annotate outside the container
+    if (markerRect.height > rect.height - markerRect.height) {
+      top = top - markerRect.height;
+    }
+
+    if (markerRect.width > rect.width - markerRect.width) {
+      left = left - markerRect.width / 2;
+    }
+
+    // Applying the position and setting visibility
+    marker.style.top = `${top}px`;
+    marker.style.left = `${left}px`;
+    marker.style.visibility = "visible";
+
+    // Attach the marker to the specified parent element
     parentElements.push(parentElement);
     return marker;
   }
@@ -461,7 +477,7 @@ function annotateElementsWithPreference({
 
       // Use createHintMarker with the specified container
       createHintMarker(el, hint, container, windowToAnnotate);
-      el.style.boxShadow = `inset 0 0 0px 2px red`;
+      el.style.boxShadow = `inset 0 0 0px 1px red`;
 
       // Add element details to the annotations map
       annotationsMap[hint] = {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@empiricalrun/test-gen",
-  "version": "0.42.9",
+  "version": "0.42.11",
   "publishConfig": {
     "registry": "https://registry.npmjs.org/",
     "access": "public"
@@ -72,7 +72,7 @@
     "ts-morph": "^23.0.0",
     "tsx": "^4.16.2",
     "typescript": "^5.3.3",
-    "@empiricalrun/llm": "^0.9.30",
+    "@empiricalrun/llm": "^0.9.31",
     "@empiricalrun/r2-uploader": "^0.3.8",
     "@empiricalrun/reporter": "^0.22.1"
   },