@empiricalrun/test-gen 0.31.19 → 0.31.21

package/dist/agent/browsing/index.js

@@ -1,229 +1,141 @@
 "use strict";
-var __importDefault = (this && this.__importDefault) || function (mod) {
-    return (mod && mod.__esModule) ? mod : { "default": mod };
-};
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.browsingAgentUsingMasterAgent = void 0;
-const llm_1 = require("@empiricalrun/llm");
-const crypto_1 = __importDefault(require("crypto"));
-const actions_1 = require("../../actions");
-const logger_1 = require("../../bin/logger");
+exports.executeTaskUsingBrowsingAgent = void 0;
 const constants_1 = require("../../constants");
 const reporter_1 = require("../../reporter");
 const session_1 = require("../../session");
 const html_1 = require("../../utils/html");
-const run_1 = require("../master/run");
 const verification_1 = require("../verification");
 const o1_completion_1 = require("./o1-completion");
 const utils_1 = require("./utils");
-async function browsingAgentUsingMasterAgent(task, page, options) {
-    const logger = new logger_1.CustomLogger({ useReporter: false });
-    const testgenUpdatesReporter = new reporter_1.TestGenUpdatesReporter();
-    // add timeout for the page to settle in
-    await page.waitForTimeout(3000);
-    const trace = llm_1.langfuseInstance.trace({
-        name: "test-generator",
-        id: crypto_1.default.randomUUID(),
-        version: (0, session_1.getSessionDetails)().version,
-        metadata: {
-            generationId: (0, session_1.getSessionDetails)().generationId,
-            sessionId: (0, session_1.getSessionDetails)().sessionId,
-        },
-        tags: [
-            options.metadata?.projectName,
-            options.metadata?.environment,
-        ].filter((s) => !!s),
-    });
-    const llm = new llm_1.LLM({
-        trace,
-        provider: options.modelProvider || constants_1.DEFAULT_MODEL_PROVIDER,
-        defaultModel: options.model || constants_1.DEFAULT_MODEL,
-        providerApiKey: constants_1.MODEL_API_KEYS[options.modelProvider || constants_1.DEFAULT_MODEL_PROVIDER],
-        // we will be using google model for larger context window, in such cases 1 million tokens is not enough
-        maxTokens: options.modelProvider === "google" ? 3000000 : 1000000,
-    });
-    const actions = new actions_1.PlaywrightActions(page);
-    const tools = actions.getActionSchemas();
-    await (0, utils_1.injectPwLocatorGenerator)(page);
-    trace.update({ input: { task } });
+async function executeTaskUsingBrowsingAgent({ trace, action, logger, page, options, llm, actions, }) {
+    let isTaskDone = false;
+    const executedActions = [];
     let lastActionExecTrace = "";
-    let isGivenTaskDone = false;
-    const masterAgentActions = [];
-    try {
-        while (!isGivenTaskDone) {
-            const masterAgentSpan = trace.span({ name: "master-agent" });
-            if (masterAgentActions.length > 0) {
-                const verificationAgentResp = await (0, verification_1.verificationAgent)({
-                    llm,
-                    trace: masterAgentSpan,
-                    task,
-                    conversation: [
-                        "Successfully executed actions",
-                        ...masterAgentActions,
-                    ],
+    const tools = actions.getBrowsingActionSchemas();
+    const testgenUpdatesReporter = new reporter_1.TestGenUpdatesReporter();
+    while (!isTaskDone) {
+        const browsingAgentSpan = trace.span({
+            name: `browsing-agent`,
+        });
+        const sessionState = await (0, session_1.getSessionState)();
+        if (sessionState === "request_complete") {
+            break;
+        }
+        const pageContentSpan = browsingAgentSpan.span({
+            name: "page-content",
+        });
+        const pageContent = await page.content();
+        pageContentSpan.end({ output: { pageContent } });
+        const sanitizationSpan = browsingAgentSpan.span({
+            name: "page-sanitization",
+        });
+        const pageSnapshot = (0, html_1.sanitizeHtml)(pageContent, options.htmlSanitize);
+        sanitizationSpan.end({ output: { pageSnapshot } });
+        const promptSpan = browsingAgentSpan.span({ name: "page-prompt" });
+        // extract all successful actions
+        const successfulActions = executedActions
+            .filter((a) => !a.isError)
+            .map((a) => a.action);
+        if (successfulActions.length > 0) {
+            const verificationAgentResp = await (0, verification_1.verificationAgent)({
+                llm,
+                trace: browsingAgentSpan,
+                task: action,
+                conversation: ["Successfully executed actions", ...successfulActions],
+            });
+            isTaskDone = verificationAgentResp.isDone;
+            logger.log(`isTaskDone: ${isTaskDone}`);
+            logger.log(`reason: ${verificationAgentResp.reason}`);
+            if (isTaskDone) {
+                browsingAgentSpan.event({ name: "task-done" });
+                browsingAgentSpan.end({
+                    output: { taskDone: true, reason: verificationAgentResp.reason },
                 });
-                isGivenTaskDone = verificationAgentResp.isDone;
-                if (isGivenTaskDone) {
-                    await testgenUpdatesReporter.sendMessage(`${verificationAgentResp.reason} Marking the task as done.`);
-                    break;
-                }
+                break;
             }
+        }
+        const messages = await (0, utils_1.getPromptForNextAction)({
+            pageSnapshot,
+            previousActions: successfulActions,
+            task: action,
+            lastActionErrors: lastActionExecTrace ? [lastActionExecTrace] : [],
+            promptType: "browsing-agent-as-tool",
+        });
+        promptSpan.end({ output: { messages } });
+        let completion;
+        completion = await (0, o1_completion_1.getO1Completion)({
+            //@ts-ignore
+            messages,
+            tools,
+            trace: browsingAgentSpan,
+        });
+        // If O1 completion fails due to any reason, resort to old flow
+        if (!completion) {
+            completion = await llm.createChatCompletion({
+                messages,
+                tools,
+                trace: browsingAgentSpan,
+                model: options.model || constants_1.DEFAULT_MODEL,
+                modelParameters: {
+                    ...constants_1.DEFAULT_MODEL_PARAMETERS,
+                    ...options.modelParameters,
+                    tool_choice: "required",
+                },
+            });
+        }
+        const toolCalls = completion?.tool_calls || [];
+        // LLM might respond with empty tool_calls and we can go into endless loop
+        // if we donot record this action and mark it as error
+        if (!toolCalls.length) {
+            executedActions.push({
+                isError: true,
+                action: "",
+            });
+        }
+        const toolCallsSpan = browsingAgentSpan.span({ name: "tool-calls" });
+        for (const i in toolCalls) {
+            const toolCall = toolCalls[i];
             const sessionState = await (0, session_1.getSessionState)();
             if (sessionState === "request_complete") {
-                await testgenUpdatesReporter.sendMessage("Aborting task, marking the task as done.");
-                break;
-            }
-            const { action, reason } = await (0, run_1.masterAgent)(task, page, masterAgentActions, masterAgentSpan, llm, options);
-            logger.log(`Next action: ${action} \n reason: ${reason}`);
-            if (!action) {
                 break;
             }
-            if (isGivenTaskDone) {
-                break;
-            }
-            let isTaskDone = false;
-            const executedActions = [];
-            while (!isTaskDone) {
-                const browsingAgentSpan = masterAgentSpan.span({
-                    name: `browsing-agent`,
-                });
-                const sessionState = await (0, session_1.getSessionState)();
-                if (sessionState === "request_complete") {
-                    break;
-                }
-                const pageContentSpan = browsingAgentSpan.span({
-                    name: "page-content",
-                });
-                const pageContent = await page.content();
-                pageContentSpan.end({ output: { pageContent } });
-                const sanitizationSpan = browsingAgentSpan.span({
-                    name: "page-sanitization",
-                });
-                const pageSnapshot = (0, html_1.sanitizeHtml)(pageContent, options.htmlSanitize);
-                sanitizationSpan.end({ output: { pageSnapshot } });
-                const promptSpan = browsingAgentSpan.span({ name: "page-prompt" });
-                // extract all successful actions
-                const successfulActions = executedActions
-                    .filter((a) => !a.isError)
-                    .map((a) => a.action);
-                if (successfulActions.length > 0) {
-                    const verificationAgentResp = await (0, verification_1.verificationAgent)({
-                        llm,
-                        trace: browsingAgentSpan,
-                        task: action,
-                        conversation: [
-                            "Successfully executed actions",
-                            ...successfulActions,
-                        ],
-                    });
-                    isTaskDone = verificationAgentResp.isDone;
-                    logger.log(`isTaskDone: ${isTaskDone}`);
-                    logger.log(`reason: ${verificationAgentResp.reason}`);
-                    if (isTaskDone) {
-                        browsingAgentSpan.event({ name: "task-done" });
-                        browsingAgentSpan.end({
-                            output: { taskDone: true, reason: verificationAgentResp.reason },
-                        });
-                        break;
-                    }
-                }
-                const messages = await (0, utils_1.getPromptForNextAction)({
-                    pageSnapshot,
-                    previousActions: successfulActions,
-                    task: action,
-                    lastActionErrors: lastActionExecTrace ? [lastActionExecTrace] : [],
-                    promptType: "browsing-agent-as-tool",
+            try {
+                await actions.executeAction(toolCall.function.name, JSON.parse(toolCall.function.arguments), toolCallsSpan);
+                executedActions.push({
+                    isError: false,
+                    action: JSON.stringify(toolCall),
                 });
-                promptSpan.end({ output: { messages } });
-                let completion;
-                completion = await (0, o1_completion_1.getO1Completion)({
-                    //@ts-ignore
-                    messages,
-                    tools,
-                    trace: browsingAgentSpan,
+                lastActionExecTrace = "";
+            }
+            catch (e) {
+                // TODO: implement feedback loop to llm
+                executedActions.push({
+                    isError: true,
+                    action: JSON.stringify(toolCall.function.arguments)?.reason,
                 });
-                // If O1 completion fails due to any reason, resort to old flow
-                if (!completion) {
-                    completion = await llm.createChatCompletion({
-                        messages,
-                        tools,
-                        trace: browsingAgentSpan,
-                        model: options.model || constants_1.DEFAULT_MODEL,
-                        modelParameters: {
-                            ...constants_1.DEFAULT_MODEL_PARAMETERS,
-                            ...options.modelParameters,
-                            tool_choice: "required",
-                        },
-                    });
-                }
-                const toolCalls = completion?.tool_calls || [];
-                // LLM might respond with empty tool_calls and we can go into endless loop
-                // if we donot record this action and mark it as error
-                if (!toolCalls.length) {
-                    executedActions.push({
-                        isError: true,
-                        action: "",
-                    });
-                }
-                const toolCallsSpan = browsingAgentSpan.span({ name: "tool-calls" });
-                for (const i in toolCalls) {
-                    const toolCall = toolCalls[i];
-                    const sessionState = await (0, session_1.getSessionState)();
-                    if (sessionState === "request_complete") {
-                        break;
-                    }
-                    try {
-                        await actions.executeAction(toolCall.function.name, JSON.parse(toolCall.function.arguments));
-                        executedActions.push({
-                            isError: false,
-                            action: JSON.stringify(toolCall),
-                        });
-                        lastActionExecTrace = "";
-                    }
-                    catch (e) {
-                        // TODO: implement feedback loop to llm
-                        executedActions.push({
-                            isError: true,
-                            action: JSON.stringify(toolCall.function.arguments)
-                                ?.reason,
-                        });
-                        lastActionExecTrace = e.message;
-                        void testgenUpdatesReporter.sendMessage(e.message);
-                        logger.error(lastActionExecTrace, e);
-                    }
-                }
-                toolCallsSpan.end({ output: { toolCalls } });
-                // mark task as done if llm is stuck in loop
-                if (executedActions.length >= 3) {
-                    const lastThreeActions = executedActions.slice(-3);
-                    const lastThreeActionsFailed = lastThreeActions.every((a) => a.isError);
-                    // get last 3 lines of code
-                    const lastThreeLinesOfCode = actions.getLastCodeLines(3);
-                    const areLastActionsRepeatitive = lastThreeLinesOfCode.length === 3 &&
-                        lastThreeLinesOfCode.every((a) => a === lastThreeLinesOfCode[0]);
-                    if (lastThreeActionsFailed || areLastActionsRepeatitive) {
-                        // TODO: this should be sent to dashboard
-                        logger.error("Agent is not able to figure out next action, marking task as done");
-                        await testgenUpdatesReporter.sendMessage("Agent is not able to figure out next action, marking task as done");
-                        isGivenTaskDone = true;
-                        break;
-                    }
-                }
+                lastActionExecTrace = e.message;
+                void testgenUpdatesReporter.sendMessage(e.message);
+                logger.error(lastActionExecTrace, e);
+            }
+        }
+        toolCallsSpan.end({ output: { toolCalls } });
+        // mark task as done if llm is stuck in loop
+        if (executedActions.length >= 3) {
+            const lastThreeActions = executedActions.slice(-3);
+            const lastThreeActionsFailed = lastThreeActions.every((a) => a.isError);
+            // get last 3 lines of code
+            const lastThreeLinesOfCode = actions.getLastCodeLines(3);
+            const areLastActionsRepeatitive = lastThreeLinesOfCode.length === 3 &&
+                lastThreeLinesOfCode.every((a) => a === lastThreeLinesOfCode[0]);
+            if (lastThreeActionsFailed || areLastActionsRepeatitive) {
+                // TODO: this should be sent to dashboard
+                const error = "Agent is not able to figure out next browser action, ending retries";
+                logger.error(error);
+                await testgenUpdatesReporter.sendMessage(error);
+                throw Error(error);
             }
-            masterAgentSpan.end({ output: { action, reason } });
-            masterAgentActions.push(action);
         }
     }
-    catch (e) {
-        console.error("Failed to generate code for the given task. Please retry again.", e);
-        await testgenUpdatesReporter.sendMessage(`Failed to generate code for the given task. Please retry again.`);
-    }
-    await page.close();
-    const code = actions.generateCode();
-    trace.update({ input: { task }, output: { code } });
-    logger.success("Successfully generated code for the given task");
-    await testgenUpdatesReporter.sendMessage(`Successfully generated code for the given task. \n View [trace](${trace.getTraceUrl()})`);
-    logger.log(`Trace: ${trace.getTraceUrl()}`);
-    return code;
 }
-exports.browsingAgentUsingMasterAgent = browsingAgentUsingMasterAgent;
+exports.executeTaskUsingBrowsingAgent = executeTaskUsingBrowsingAgent;

package/dist/agent/codegen/create-test-block.js

@@ -16,7 +16,7 @@ async function createEmptyTestCaseBlock({ testCase, file, options, trace, }) {
     trace =
         trace ||
             llm_1.langfuseInstance.trace({
-                name: "update-test",
+                name: "create-empty-test-block",
                 id: crypto.randomUUID(),
                 release: session.version,
                 tags: [

package/dist/agent/codegen/skills-retriever.d.ts

@@ -0,0 +1,13 @@
+import { TraceClient } from "@empiricalrun/llm";
+import { TestCase, TestGenConfigOptions } from "../../types";
+export declare function getAppropriateSkills({ testCase, options, trace, }: {
+    testCase: TestCase;
+    options?: TestGenConfigOptions;
+    trace?: TraceClient;
+}): Promise<{
+    testStep: string;
+    filePath: string;
+    usageExample: string;
+    reason: string;
+}[]>;
+//# sourceMappingURL=skills-retriever.d.ts.map

package/dist/agent/codegen/skills-retriever.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"skills-retriever.d.ts","sourceRoot":"","sources":["../../../src/agent/codegen/skills-retriever.ts"],"names":[],"mappings":"AAAA,OAAO,EAAkB,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAYhE,OAAO,EAAE,QAAQ,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAG7D,wBAAsB,oBAAoB,CAAC,EACzC,QAAQ,EACR,OAAO,EACP,KAAK,GACN,EAAE;IACD,QAAQ,EAAE,QAAQ,CAAC;IACnB,OAAO,CAAC,EAAE,oBAAoB,CAAC;IAC/B,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB;;;;;KAmDA"}

package/dist/agent/codegen/skills-retriever.js

@@ -0,0 +1,61 @@
+"use strict";
+var __importDefault = (this && this.__importDefault) || function (mod) {
+    return (mod && mod.__esModule) ? mod : { "default": mod };
+};
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.getAppropriateSkills = void 0;
+const llm_1 = require("@empiricalrun/llm");
+const fs_1 = __importDefault(require("fs"));
+const logger_1 = require("../../bin/logger");
+const context_1 = require("../../bin/utils/context");
+const fs_2 = require("../../bin/utils/fs");
+const constants_1 = require("../../constants");
+const utils_1 = require("./utils");
+async function getAppropriateSkills({ testCase, options, trace, }) {
+    const logger = new logger_1.CustomLogger({ useReporter: false });
+    logger.log("getting skill set for the repository");
+    const filter = await (0, context_1.createGitIgnoreFileFilter)();
+    const pomFiles = await (0, fs_2.generatePromptFromDirectory)("./pages", filter);
+    const fetchSkillsSpan = trace?.span({
+        name: "fetch-pom-skills",
+        input: {
+            testCase,
+        },
+    });
+    const promptSpan = fetchSkillsSpan?.span({
+        name: "fetch-pom-skills-prompt",
+    });
+    const prompt = await (0, llm_1.getPrompt)("fetch-skills-prompt", {
+        pageFiles: pomFiles,
+        scenarioName: testCase.name,
+        scenario: testCase.steps.join("\n"),
+    });
+    promptSpan?.end({ output: { prompt } });
+    const llm = new llm_1.LLM({
+        trace: fetchSkillsSpan,
+        provider: options?.modelProvider || constants_1.DEFAULT_MODEL_PROVIDER,
+        defaultModel: options?.model || constants_1.DEFAULT_MODEL,
+        providerApiKey: constants_1.MODEL_API_KEYS[options?.modelProvider || constants_1.DEFAULT_MODEL_PROVIDER],
+    });
+    const firstShotMessage = await llm.createChatCompletion({
+        messages: prompt,
+        traceName: "fetch-pom-skills-llm",
+        modelParameters: {
+            ...constants_1.DEFAULT_MODEL_PARAMETERS,
+            ...options?.modelParameters,
+        },
+    });
+    let response = firstShotMessage?.content || "";
+    const skills = (0, utils_1.extractTestStepsSuggestions)(response);
+    const validateSkillsSpan = fetchSkillsSpan?.span({
+        name: "validate-skills",
+        input: {
+            skills,
+        },
+    });
+    const validatedSkills = skills.filter((skill) => fs_1.default.existsSync(skill.filePath));
+    validateSkillsSpan?.end({ output: { validatedSkills } });
+    fetchSkillsSpan?.end({ output: { validatedSkills } });
+    return validatedSkills;
+}
+exports.getAppropriateSkills = getAppropriateSkills;

package/dist/agent/codegen/use-skill.d.ts

@@ -0,0 +1,9 @@
+import { TraceClient } from "@empiricalrun/llm";
+export declare function generateSkillUsageCode({ task, sampleUsageMethod, scopeVariablesMapStr, pageVariableName, trace, }: {
+    task: string;
+    sampleUsageMethod: string;
+    scopeVariablesMapStr: string;
+    pageVariableName: string;
+    trace?: TraceClient;
+}): Promise<string>;
+//# sourceMappingURL=use-skill.d.ts.map

package/dist/agent/codegen/use-skill.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"use-skill.d.ts","sourceRoot":"","sources":["../../../src/agent/codegen/use-skill.ts"],"names":[],"mappings":"AAAA,OAAO,EAAkB,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAShE,wBAAsB,sBAAsB,CAAC,EAC3C,IAAI,EACJ,iBAAiB,EACjB,oBAAoB,EACpB,gBAAgB,EAChB,KAAK,GACN,EAAE;IACD,IAAI,EAAE,MAAM,CAAC;IACb,iBAAiB,EAAE,MAAM,CAAC;IAC1B,oBAAoB,EAAE,MAAM,CAAC;IAC7B,gBAAgB,EAAE,MAAM,CAAC;IACzB,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,mBA0CA"}

package/dist/agent/codegen/use-skill.js

@@ -0,0 +1,49 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.generateSkillUsageCode = void 0;
+const llm_1 = require("@empiricalrun/llm");
+const constants_1 = require("../../constants");
+async function generateSkillUsageCode({ task, sampleUsageMethod, scopeVariablesMapStr, pageVariableName, trace, }) {
+    const skillUsageSpan = trace?.span({
+        name: "skill-usage",
+        input: {
+            task,
+            sampleUsageMethod,
+            scopeVariablesMapStr,
+            pageVariableName,
+        },
+    });
+    const promptSpan = skillUsageSpan?.span({
+        name: "apply-skills-prompt",
+    });
+    const prompt = await (0, llm_1.getPrompt)("apply-skills-prompt", {
+        task,
+        sampleUsageMethod,
+        scopeVariablesMapStr,
+        pageVariableName,
+    });
+    promptSpan?.end({ output: prompt });
+    const llm = new llm_1.LLM({
+        trace: skillUsageSpan,
+        provider: constants_1.DEFAULT_MODEL_PROVIDER,
+        defaultModel: constants_1.DEFAULT_MODEL,
+        providerApiKey: constants_1.MODEL_API_KEYS[constants_1.DEFAULT_MODEL_PROVIDER],
+    });
+    const firstShotMessage = await llm.createChatCompletion({
+        trace: skillUsageSpan,
+        messages: prompt,
+        traceName: "generate-skill-usage-code",
+        modelParameters: {
+            ...constants_1.DEFAULT_MODEL_PARAMETERS,
+        },
+    });
+    let response = firstShotMessage?.content || "";
+    skillUsageSpan?.end({
+        output: {
+            code: response,
+        },
+    });
+    console.log(`generated usage code`, response);
+    return response;
+}
+exports.generateSkillUsageCode = generateSkillUsageCode;

package/dist/agent/codegen/utils.d.ts

@@ -16,4 +16,13 @@ export declare function extractTestUpdates(input: string): {
     newCode: string | undefined;
     reason: string | undefined;
 }[];
+/**
+ *
+ */
+export declare function extractTestStepsSuggestions(input: string): {
+    testStep: string;
+    filePath: string;
+    usageExample: string;
+    reason: string;
+}[];
 //# sourceMappingURL=utils.d.ts.map

package/dist/agent/codegen/utils.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../../src/agent/codegen/utils.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AACH,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,MAAM,GAAG;IACjD,QAAQ,EAAE,MAAM,GAAG,SAAS,CAAC;IAC7B,OAAO,EAAE,MAAM,GAAG,SAAS,CAAC;IAC5B,OAAO,EAAE,MAAM,GAAG,SAAS,CAAC;IAC5B,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC;CAC5B,EAAE,CAiBF"}
+{"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../../src/agent/codegen/utils.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AACH,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,MAAM,GAAG;IACjD,QAAQ,EAAE,MAAM,GAAG,SAAS,CAAC;IAC7B,OAAO,EAAE,MAAM,GAAG,SAAS,CAAC;IAC5B,OAAO,EAAE,MAAM,GAAG,SAAS,CAAC;IAC5B,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC;CAC5B,EAAE,CAiBF;AAED;;GAEG;AAEH,wBAAgB,2BAA2B,CAAC,KAAK,EAAE,MAAM,GAAG;IAC1D,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,MAAM,EAAE,MAAM,CAAC;CAChB,EAAE,CAeF"}

package/dist/agent/codegen/utils.js

@@ -1,6 +1,6 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.extractTestUpdates = void 0;
+exports.extractTestStepsSuggestions = exports.extractTestUpdates = void 0;
 /**
  *
  * method to extract file path and code updates for the LLM response of update flow
@@ -30,3 +30,22 @@ function extractTestUpdates(input) {
     return result;
 }
 exports.extractTestUpdates = extractTestUpdates;
+/**
+ *
+ */
+function extractTestStepsSuggestions(input) {
+    const result = [];
+    const regex = /<subtask>(.*?)<\/subtask>[\s\S]*?<file_import_path>([\s\S]*?)<\/file_import_path>[\s\S]*?<usage_example>([\s\S]*?)<\/usage_example>[\s\S]*?<reason>([\s\S]*?)<\/reason>/g;
+    let match;
+    while ((match = regex.exec(input)) !== null) {
+        const [, testStep, filePath, usageExample, reason] = match;
+        result.push({
+            testStep: testStep?.trim() || "",
+            filePath: filePath?.trim() || "",
+            usageExample: usageExample?.trim() || "",
+            reason: reason?.trim() || "",
+        });
+    }
+    return result.filter((r) => !!r.filePath && !!r.usageExample);
+}
+exports.extractTestStepsSuggestions = extractTestStepsSuggestions;

package/dist/agent/master/run.d.ts

@@ -1,16 +1,27 @@
 import { LLM, TraceClient } from "@empiricalrun/llm";
 import { Page } from "playwright";
-import { TestGenConfigOptions } from "../../types";
-type BrowsingAgentOptions = Partial<TestGenConfigOptions>;
-export declare function getNextAction({ task, executedActions, page, trace, llm, options, pageScreenshot, }: {
+import { PlaywrightActions } from "../../actions";
+import { TestCase } from "../../types";
+import { BrowsingAgentOptions } from "../browsing";
+export declare function getNextAction({ task, executedActions, failedActions, page, trace, llm, options, pageScreenshot, actions, disableSkills, }: {
     task: string;
     executedActions: string[];
+    failedActions: any[];
     page: Page;
     trace?: TraceClient;
     llm: LLM;
     options: BrowsingAgentOptions;
     pageScreenshot: string;
+    actions: PlaywrightActions;
+    disableSkills: boolean;
 }): Promise<import("openai/resources/index.mjs").ChatCompletionMessageToolCall | undefined>;
-export declare function masterAgent(task: string, page: Page, executedActions: string[], trace: TraceClient, llm: LLM, options: BrowsingAgentOptions): Promise<any>;
-export {};
+export declare function createTestUsingMasterAgent({ task, page, testCase, options, }: {
+    task: string;
+    page: Page;
+    testCase: TestCase;
+    options: BrowsingAgentOptions;
+}): Promise<{
+    code: string;
+    importPaths: string[];
+}>;
 //# sourceMappingURL=run.d.ts.map

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

@@ -1 +1 @@
-{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/master/run.ts"],"names":[],"mappings":"AAAA,OAAO,EAAa,GAAG,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAGhE,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAQlC,OAAO,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAEnD,KAAK,oBAAoB,GAAG,OAAO,CAAC,oBAAoB,CAAC,CAAC;AAE1D,wBAAsB,aAAa,CAAC,EAClC,IAAI,EACJ,eAAe,EACf,IAAI,EACJ,KAAK,EACL,GAAG,EACH,OAAO,EACP,cAAc,GACf,EAAE;IACD,IAAI,EAAE,MAAM,CAAC;IACb,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,IAAI,EAAE,IAAI,CAAC;IACX,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,GAAG,EAAE,GAAG,CAAC;IACT,OAAO,EAAE,oBAAoB,CAAC;IAC9B,cAAc,EAAE,MAAM,CAAC;CACxB,2FA+CA;AAED,wBAAsB,WAAW,CAC/B,IAAI,EAAE,MAAM,EACZ,IAAI,EAAE,IAAI,EACV,eAAe,EAAE,MAAM,EAAE,EACzB,KAAK,EAAE,WAAW,EAClB,GAAG,EAAE,GAAG,EACR,OAAO,EAAE,oBAAoB,gBAsC9B"}
+{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/master/run.ts"],"names":[],"mappings":"AAAA,OAAO,EAGL,GAAG,EACH,WAAW,EACZ,MAAM,mBAAmB,CAAC;AAG3B,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAElC,OAAO,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAYlD,OAAO,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AACvC,OAAO,EACL,oBAAoB,EAErB,MAAM,aAAa,CAAC;AAOrB,wBAAsB,aAAa,CAAC,EAClC,IAAI,EACJ,eAAe,EACf,aAAa,EACb,IAAI,EACJ,KAAK,EACL,GAAG,EACH,OAAO,EACP,cAAc,EACd,OAAO,EACP,aAAa,GACd,EAAE;IACD,IAAI,EAAE,MAAM,CAAC;IACb,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,aAAa,EAAE,GAAG,EAAE,CAAC;IACrB,IAAI,EAAE,IAAI,CAAC;IACX,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,GAAG,EAAE,GAAG,CAAC;IACT,OAAO,EAAE,oBAAoB,CAAC;IAC9B,cAAc,EAAE,MAAM,CAAC;IACvB,OAAO,EAAE,iBAAiB,CAAC;IAC3B,aAAa,EAAE,OAAO,CAAC;CACxB,2FAwDA;AAED,wBAAsB,0BAA0B,CAAC,EAC/C,IAAI,EACJ,IAAI,EACJ,QAAQ,EACR,OAAO,GACR,EAAE;IACD,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,IAAI,CAAC;IACX,QAAQ,EAAE,QAAQ,CAAC;IACnB,OAAO,EAAE,oBAAoB,CAAC;CAC/B;;;GAmLA"}