@empiricalrun/test-gen 0.78.0 → 0.78.1

package/dist/tools/test-run-fetcher/index.js

@@ -1,13 +1,13 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.fetchTestRunReportTool = void 0;
+exports.fetchTestRunDetailsTool = void 0;
 exports.extractPathAfterSourceRepo = extractPathAfterSourceRepo;
 const zod_1 = require("zod");
-const urls_1 = require("../utils/urls");
 const TestRunSchema = zod_1.z.object({
-    testRunUrl: zod_1.z
-        .string()
-        .describe("The full URL of the test run (e.g. https://dash.empirical.run/sortment-tests/test-runs/20269 or with query params like ?status=failed)"),
+    testRunId: zod_1.z
+        .number()
+        .describe("Numeric identifier (e.g. 20269) for a test run"),
+    projectSlug: zod_1.z.string().describe("Slug for the project of the test run"),
 });
 function extractPathAfterSourceRepo(fullPath) {
     const parts = fullPath.split("source-repo");
@@ -18,10 +18,24 @@ function extractPathAfterSourceRepo(fullPath) {
         return parts[1].replace(/^\/|\/$/g, "");
     }
 }
-exports.fetchTestRunReportTool = {
+exports.fetchTestRunDetailsTool = {
     schema: {
         name: "fetchTestRunDetails",
-        description: "Fetch details about a test run using its URL",
+        description: `Fetch details about a test run using its numeric identifier.
+This tool returns the status of the test run, along with information about the environment
+(e.g. "staging") or build that was used for the test run.
+
+## How to use this tool?
+
+When you are given a URL like https://dash.empirical.run/<project-slug>/test-runs/20269?query=params
+- "testRunId": Extract the numeric id from the URL - in this case #20269
+- "projectSlug": Extract the project-slug from the URL 
+
+## Related tools
+
+- To learn more about a particular test case that failed in the test, you can use the fetch diagnosis tool
+  with the diagnosis id (which is passed to the test run URL as the ?detail= query parameter)
+`,
         parameters: TestRunSchema,
     },
     needsBrowser: false,
@@ -30,9 +44,11 @@ exports.fetchTestRunReportTool = {
         if (!apiClient) {
             throw new Error("Dashboard API client is not available.");
         }
-        const { testRunUrl } = input;
+        const { testRunId: runId, projectSlug } = input;
         let data = null;
-        const { repoName, runId } = (0, urls_1.extractProjectFromUrl)(testRunUrl);
+        const repoName = projectSlug.includes("-tests")
+            ? projectSlug
+            : `${projectSlug}-tests`;
         try {
             data = await apiClient.request(`/api/test-runs/${runId}?repo_name=${repoName}`, { method: "GET" });
         }
@@ -42,79 +58,64 @@ exports.fetchTestRunReportTool = {
                 result: `Failed to fetch test run details: ${error instanceof Error ? error.message : String(error)}`,
             };
         }
-        // To efficiently use input_tokens, we
-        //   1. Truncate stack trace to last 300 characters
-        //   2. Remove request/response headers from network metadata
-        //   3. Focus on failed tests only
         const testRun = data.data?.test_run;
-        if (testRun?.flattenedSummaryDetails) {
-            testRun.flattenedSummaryDetails = testRun.flattenedSummaryDetails.map((detail) => {
-                // Truncate stack trace to last 300 characters
-                if (detail.failed_run_metadata?.stack) {
-                    const stack = detail.failed_run_metadata.stack;
-                    // TODO: change to last 10 lines
-                    detail.failed_run_metadata.stack =
-                        stack.length > 300 ? `...${stack.slice(-300)}` : stack;
-                }
-                // Remove headers from network metadata
-                if (detail.network_metadata?.failed_calls) {
-                    detail.network_metadata.failed_calls =
-                        detail.network_metadata.failed_calls.map((call) => ({
-                            ...call,
-                            request: { ...call.request, headers: [] },
-                            response: { ...call.response, headers: [] },
-                        }));
-                }
-                if (detail.network_metadata?.failed_calls_within_time_range) {
-                    detail.network_metadata.failed_calls_within_time_range =
-                        detail.network_metadata.failed_calls_within_time_range.map((call) => ({
-                            ...call,
-                            request: { ...call.request, headers: [] },
-                            response: { ...call.response, headers: [] },
-                        }));
-                }
-                return detail;
-            });
+        if (!testRun) {
+            return {
+                isError: true,
+                result: `Test run not found.`,
+            };
         }
-        const failedTests = testRun?.flattenedSummaryDetails.filter((detail) => detail.status === "failed");
-        const failedTestInMarkdown = failedTests
-            ?.map((detail) => {
+        const buildInfoIfAvailable = testRun.testRun.build_url
+            ? `
+    ## Build info
+    - Build URL: ${testRun.testRun.build_url}
+    - Git branch: ${testRun.testRun.build_branch}
+    - Git commit: ${testRun.testRun.commit} 
+    `
+            : ``;
+        const failedTestsInMarkdown = (testRun.flattenedSummaryDetails || [])
+            .filter((detail) => detail.status === "failed")
+            .map((detail) => {
             const { nesting } = detail.failed_run_metadata;
             return {
                 testName: nesting.slice(1).join(" > "),
                 fileName: extractPathAfterSourceRepo(detail.failed_run_metadata.location.file),
                 project: detail.test_project,
-                detail,
+                diagnosisId: detail.slug,
             };
         })
-            .map((res) => {
+            .map((res, idx) => {
             return `
-### Test name: ${res.testName}
-File: ${res.fileName}
-Project: ${res.project}
+    ### Failed test ${idx + 1}
 
-Failure data below
+    Test case name: ${res.testName}
+    File: ${res.fileName}
+    Project: ${res.project}
+    Diagnosis ID: ${res.diagnosisId}
 
-\`\`\`json
-${JSON.stringify(res.detail, null, 2)}
-\`\`\`
-`;
+    `;
         })
-            .join("\n\n");
+            .join("\n");
         const markdownResponse = `
-# Test run details
+    Test run #${testRun.testRun.id}
+
+    ## Info
+    - Run ID: ${runId}
+    - Status: ${testRun.testRun.state}
+    - Started at: ${testRun.testRun.run_started_at}
+    - Ended at: ${testRun.testRun.run_ended_at}
+    - Duration: ${testRun.testRun.duration} seconds
+
+    ## Environment
+    - Name: ${testRun?.testRun.environment_name}
+    - Slug: ${testRun?.testRun.environment_slug}
+
+    ${buildInfoIfAvailable}
 
-## Run info
-- Run ID: ${runId}
-- Repository: ${repoName}
-- Started at: ${testRun?.testRun.run_started_at}
-- Ended at: ${testRun?.testRun.run_ended_at}
-- Duration: ${testRun?.testRun.duration} seconds
-- Environment name: ${testRun?.testRun.environment_name}
+    ## Failed tests
 
-## Failed tests
-${failedTestInMarkdown}
-`;
+    ${failedTestsInMarkdown.length > 0 ? failedTestsInMarkdown : "None"}
+    `;
         return {
             result: markdownResponse,
             isError: false,

package/dist/tools/trace-dot-zip/index.d.ts

@@ -1,3 +1,3 @@
-import { Tool } from "@empiricalrun/shared-types";
+import { Tool } from "@empiricalrun/shared-types/chat-agent";
 export declare const traceDotZipTool: Tool;
 //# sourceMappingURL=index.d.ts.map

package/dist/tools/trace-dot-zip/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/trace-dot-zip/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAc,MAAM,4BAA4B,CAAC;AAa9D,eAAO,MAAM,eAAe,EAAE,IAwD7B,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/trace-dot-zip/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAc,MAAM,uCAAuC,CAAC;AAazE,eAAO,MAAM,eAAe,EAAE,IAwD7B,CAAC"}

package/dist/tools/trace-dot-zip/utils/network-trace.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"network-trace.d.ts","sourceRoot":"","sources":["../../../../src/tools/trace-dot-zip/utils/network-trace.ts"],"names":[],"mappings":"AAUA,OAAO,EACL,sBAAsB,EACtB,sBAAsB,EACtB,sBAAsB,EACvB,MAAM,UAAU,CAAC;AAUlB,KAAK,KAAK,GAAG;IACX,GAAG,EAAE,MAAM,CAAC;IACZ,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE;QACR,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,QAAQ,EACJ;YACE,IAAI,EAAE,MAAM,CAAC;YACb,IAAI,EAAE,MAAM,CAAC;YACb,MAAM,EAAE,MAAM,CAAC;SAChB,GACD,SAAS,CAAC;KACf,CAAC;CACH,CAAC;AAEF,eAAO,MAAM,oBAAoB,GAAU,8BAIxC,KAAK,KAAG,OAAO,CAAC;IACjB,cAAc,EAAE,sBAAsB,EAAE,CAAC;IACzC,0BAA0B,EAAE,sBAAsB,EAAE,CAAC;CACtD,CA+KA,CAAC;AAEF,eAAO,MAAM,8BAA8B,GAAU,UAElD,KAAK,KAAG,OAAO,CAAC;IACjB,mBAAmB,EAAE,sBAAsB,EAAE,CAAC;IAC9C,mBAAmB,EAAE,sBAAsB,EAAE,CAAC;CAC/C,CAyEA,CAAC;AAsBF,eAAO,MAAM,0CAA0C,GACrD,UAAU,sBAAsB,EAAE,KACjC,OAAO,CAAC,sBAAsB,CAAC,EAoBjC,CAAC;AAmCF,eAAO,MAAM,iBAAiB,GAC5B,UAAU,sBAAsB,EAAE,EAClC,QAAQ,MAAM,KACb,OAAO,CAAC,sBAAsB,EAAE,CA+BlC,CAAC;AAEF,eAAO,MAAM,wBAAwB,GACnC,0CAA0C,sBAAsB,EAAE,KACjE,sBAAsB,EA2BxB,CAAC"}
+{"version":3,"file":"network-trace.d.ts","sourceRoot":"","sources":["../../../../src/tools/trace-dot-zip/utils/network-trace.ts"],"names":[],"mappings":"AAUA,OAAO,EACL,sBAAsB,EACtB,sBAAsB,EACtB,sBAAsB,EACvB,MAAM,UAAU,CAAC;AAUlB,KAAK,KAAK,GAAG;IACX,GAAG,EAAE,MAAM,CAAC;IACZ,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE;QACR,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,QAAQ,EACJ;YACE,IAAI,EAAE,MAAM,CAAC;YACb,IAAI,EAAE,MAAM,CAAC;YACb,MAAM,EAAE,MAAM,CAAC;SAChB,GACD,SAAS,CAAC;KACf,CAAC;CACH,CAAC;AAEF,eAAO,MAAM,oBAAoB,GAAU,8BAIxC,KAAK,KAAG,OAAO,CAAC;IACjB,cAAc,EAAE,sBAAsB,EAAE,CAAC;IACzC,0BAA0B,EAAE,sBAAsB,EAAE,CAAC;CACtD,CA+KA,CAAC;AAEF,eAAO,MAAM,8BAA8B,GAAU,UAElD,KAAK,KAAG,OAAO,CAAC;IACjB,mBAAmB,EAAE,sBAAsB,EAAE,CAAC;IAC9C,mBAAmB,EAAE,sBAAsB,EAAE,CAAC;CAC/C,CAkEA,CAAC;AAsBF,eAAO,MAAM,0CAA0C,GACrD,UAAU,sBAAsB,EAAE,KACjC,OAAO,CAAC,sBAAsB,CAAC,EAoBjC,CAAC;AAmCF,eAAO,MAAM,iBAAiB,GAC5B,UAAU,sBAAsB,EAAE,EAClC,QAAQ,MAAM,KACb,OAAO,CAAC,sBAAsB,EAAE,CA+BlC,CAAC;AAEF,eAAO,MAAM,wBAAwB,GACnC,0CAA0C,sBAAsB,EAAE,KACjE,sBAAsB,EA2BxB,CAAC"}

package/dist/tools/trace-dot-zip/utils/network-trace.js

@@ -147,10 +147,10 @@ const generateNetworkTraceFromZipUrl = async ({ url, }) => {
                         const hasValidResponseStatus = !!parsedNetworkTrace?.snapshot?.response?.status;
                         const responseStatus = Number(parsedNetworkTrace?.snapshot?.response?.status);
                         allNetworkCalls.push(parsedNetworkTrace);
-                        // only failures & redirects for detailed analysis
+                        // only failures for detailed analysis
                         if (hasValidResponseStatus &&
-                            (responseStatus === 308 ||
-                                (400 <= responseStatus && responseStatus < 600))) {
+                            400 <= responseStatus &&
+                            responseStatus < 600) {
                             relevantNetworkCalls.push(parsedNetworkTrace);
                         }
                     });
@@ -165,13 +165,8 @@ const generateNetworkTraceFromZipUrl = async ({ url, }) => {
         const sequenceCalls = allNetworkCalls.filter((c) => {
             const hasValidResponseStatus = !!c.snapshot.response.status;
             const status = Number(c.snapshot.response.status);
-            return (hasValidResponseStatus &&
-                ((400 <= status && status < 600) ||
-                    status === 307 ||
-                    status === 308 ||
-                    status === 301 ||
-                    status === 302));
-            // excluding: 304 (cached), 2xx success
+            return hasValidResponseStatus && 400 <= status && status < 600;
+            // excluding: 304 (cached), 2xx success, 3xx redirects
         });
         const networkCallSequence = (0, exports.buildNetworkCallSequence)(sequenceCalls);
         const formattedFailedCalls = failedCalls.map(formatNetworkFailure);

package/dist/tools/triage-summary/index.d.ts

@@ -1,3 +1,3 @@
-import { Tool } from "@empiricalrun/shared-types";
+import { Tool } from "@empiricalrun/shared-types/chat-agent";
 export declare const sendTriageSummaryTool: Tool;
 //# sourceMappingURL=index.d.ts.map

package/dist/tools/triage-summary/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/triage-summary/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAc,MAAM,4BAA4B,CAAC;AAmB9D,eAAO,MAAM,qBAAqB,EAAE,IA6CnC,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/triage-summary/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAEL,IAAI,EAEL,MAAM,uCAAuC,CAAC;AAmB/C,eAAO,MAAM,qBAAqB,EAAE,IA0DnC,CAAC"}

package/dist/tools/triage-summary/index.js

@@ -19,7 +19,7 @@ exports.sendTriageSummaryTool = {
     },
     needsBrowser: false,
     isInlineTool: true,
-    execute: async ({ input, apiClient, chatSession }) => {
+    execute: async ({ input, apiClient, chatSession, logger, }) => {
         try {
             const triagedIssuesList = input.issues;
             if (!triagedIssuesList) {
@@ -41,7 +41,10 @@ exports.sendTriageSummaryTool = {
             };
         }
         catch (error) {
-            console.error("Error sending triage summary to slack", error);
+            logger.error("Error sending triage summary to slack", {
+                tool: "sendTriageSummaryTool",
+                error,
+            });
             return {
                 isError: true,
                 result: error instanceof Error ? error.message : String(error),

package/dist/tools/triage-summary/utils.d.ts

@@ -1,4 +1,4 @@
-import { IDashboardAPIClient } from "@empiricalrun/shared-types";
+import { IDashboardAPIClient } from "@empiricalrun/shared-types/api/base";
 import { TriagedIssue } from "./types";
 export declare const triggerSlackMessageWithTriageSummary: (issues: TriagedIssue[], apiClient: IDashboardAPIClient, chatSessionId: number) => Promise<string>;
 //# sourceMappingURL=utils.d.ts.map

package/dist/tools/triage-summary/utils.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../../src/tools/triage-summary/utils.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,mBAAmB,EAAE,MAAM,4BAA4B,CAAC;AAEjE,OAAO,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AAEvC,eAAO,MAAM,oCAAoC,GAC/C,QAAQ,YAAY,EAAE,EACtB,WAAW,mBAAmB,EAC9B,eAAe,MAAM,KACpB,OAAO,CAAC,MAAM,CAahB,CAAC"}
+{"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../../src/tools/triage-summary/utils.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,mBAAmB,EAAE,MAAM,qCAAqC,CAAC;AAE1E,OAAO,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AAEvC,eAAO,MAAM,oCAAoC,GAC/C,QAAQ,YAAY,EAAE,EACtB,WAAW,mBAAmB,EAC9B,eAAe,MAAM,KACpB,OAAO,CAAC,MAAM,CAahB,CAAC"}

package/dist/tools/upgrade-packages/index.d.ts

@@ -1,3 +1,3 @@
-import type { Tool } from "@empiricalrun/shared-types";
+import type { Tool } from "@empiricalrun/shared-types/chat-agent";
 export declare const upgradePackagesTool: Tool;
 //# sourceMappingURL=index.d.ts.map

package/dist/tools/upgrade-packages/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/upgrade-packages/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,4BAA4B,CAAC;AAoBvD,eAAO,MAAM,mBAAmB,EAAE,IA8HjC,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/upgrade-packages/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,uCAAuC,CAAC;AAmBlE,eAAO,MAAM,mBAAmB,EAAE,IAqIjC,CAAC"}

package/dist/tools/upgrade-packages/index.js

@@ -14,13 +14,13 @@ const pr_description_1 = require("../executor/utils/pr-description");
 const utils_2 = require("./utils");
 exports.upgradePackagesTool = {
     ...upgrade_packages_1.upgradePackagesTool,
-    execute: async ({ input, repoPath, apiClient, chatSession }) => {
+    execute: async ({ input, repoPath, apiClient, chatSession, logger }) => {
         const { packages: packagesToUpdate } = input;
         try {
-            const repoName = path_1.default.basename(repoPath);
             const branchName = await (0, git_1.getCurrentBranchName)(repoPath);
             const packageJsonPath = path_1.default.join(repoPath, "package.json");
             const packageJson = JSON.parse(fs_1.default.readFileSync(packageJsonPath, "utf-8"));
+            const repoName = packageJson.name;
             const changes = [];
             for (let { name: pkgName, version } of packagesToUpdate) {
                 const isDevDep = !!packageJson.devDependencies?.[pkgName];
@@ -30,11 +30,12 @@ exports.upgradePackagesTool = {
                         pkgName,
                         version,
                         isDevDep,
+                        logger: logger,
                     });
                     changes.push(updatedPackage);
                 }
                 catch (err) {
-                    console.error(`Failed to upgrade ${pkgName}:`, err);
+                    logger.error("Failed to upgrade package", { pkgName, error: err });
                     return {
                         result: `Failed to upgrade ${pkgName}: ${err.message}`,
                         isError: true,
@@ -83,9 +84,14 @@ exports.upgradePackagesTool = {
                 if (!pr.number) {
                     throw new Error("Failed to create PR");
                 }
-                shouldMerge = await (0, utils_2.shouldMergePR)({ repoName, prNumber, apiClient });
+                shouldMerge = await (0, utils_2.shouldMergePR)({
+                    repoName,
+                    prNumber,
+                    apiClient,
+                    logger: logger,
+                });
                 if (shouldMerge) {
-                    console.log("All changes are patch updates, proceeding with merge");
+                    logger.info("All changes are patch updates, proceeding with merge");
                     merged = await (0, git_1.mergePullRequest)({
                         repoName,
                         prNumber,
@@ -93,11 +99,11 @@ exports.upgradePackagesTool = {
                     });
                 }
                 else {
-                    console.log(`PR #${prNumber} created but not merged - contains non-patch updates`);
+                    logger.info("PR created but not merged - contains non-patch updates", { prNumber });
                 }
             }
             catch (error) {
-                console.error("Failed to handle PR operations:", error);
+                logger.error("Failed to handle PR operations", { error });
                 const prLink = prNumber
                     ? ` PR: https://github.com/empirical-run/${repoName}/pull/${prNumber}`
                     : "";
@@ -118,7 +124,7 @@ exports.upgradePackagesTool = {
             };
         }
         catch (error) {
-            console.error("Failed to upgrade packages:", error);
+            logger.error("Failed to upgrade packages", { error });
             return {
                 result: `Failed to upgrade packages: ${error.message}`,
                 isError: true,

package/dist/tools/upgrade-packages/utils.d.ts

@@ -1,22 +1,26 @@
-import { IDashboardAPIClient } from "@empiricalrun/shared-types";
+import { IDashboardAPIClient } from "@empiricalrun/shared-types/api/base";
+import { Logger } from "@empiricalrun/shared-types/chat-agent";
 export declare function isPatchUpdate(currentVersion: string | undefined, newVersion: string | undefined): boolean;
-export declare function getLatestVersion(packageName: string): Promise<any>;
-export declare function parsePackageJsonDiff({ diff, }: {
+export declare function getLatestVersion(packageName: string, logger: Logger): Promise<any>;
+export declare function parsePackageJsonDiff({ diff, logger, }: {
     diff: string;
+    logger: Logger;
 }): Record<string, {
     old?: string;
     new?: string;
 }>;
-export declare function shouldMergePR({ repoName, prNumber, apiClient, }: {
+export declare function shouldMergePR({ repoName, prNumber, apiClient, logger, }: {
     repoName: string;
     prNumber: number;
     apiClient: IDashboardAPIClient;
+    logger: Logger;
 }): Promise<boolean>;
-export declare function upgradeAndStagePackage({ repoPath, pkgName, version, isDevDep, }: {
+export declare function upgradeAndStagePackage({ repoPath, pkgName, version, isDevDep, logger, }: {
     repoPath: string;
     pkgName: string;
     version?: string;
     isDevDep: boolean;
+    logger: Logger;
 }): Promise<{
     name: string;
     version: any;

package/dist/tools/upgrade-packages/utils.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../../src/tools/upgrade-packages/utils.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,mBAAmB,EAAE,MAAM,4BAA4B,CAAC;AAGjE,wBAAgB,aAAa,CAC3B,cAAc,EAAE,MAAM,GAAG,SAAS,EAClC,UAAU,EAAE,MAAM,GAAG,SAAS,GAC7B,OAAO,CAWT;AAED,wBAAsB,gBAAgB,CAAC,WAAW,EAAE,MAAM,gBAkBzD;AA4BD,wBAAgB,oBAAoB,CAAC,EACnC,IAAI,GACL,EAAE;IACD,IAAI,EAAE,MAAM,CAAC;CACd,GAAG,MAAM,CAAC,MAAM,EAAE;IAAE,GAAG,CAAC,EAAE,MAAM,CAAC;IAAC,GAAG,CAAC,EAAE,MAAM,CAAA;CAAE,CAAC,CAuBjD;AAED,wBAAsB,aAAa,CAAC,EAClC,QAAQ,EACR,QAAQ,EACR,SAAS,GACV,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,mBAAmB,CAAC;CAChC,oBA8BA;AAED,wBAAsB,sBAAsB,CAAC,EAC3C,QAAQ,EACR,OAAO,EACP,OAAO,EACP,QAAQ,GACT,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,OAAO,CAAC;CACnB;;;GAcA"}
+{"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../../src/tools/upgrade-packages/utils.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,mBAAmB,EAAE,MAAM,qCAAqC,CAAC;AAC1E,OAAO,EAAE,MAAM,EAAE,MAAM,uCAAuC,CAAC;AAG/D,wBAAgB,aAAa,CAC3B,cAAc,EAAE,MAAM,GAAG,SAAS,EAClC,UAAU,EAAE,MAAM,GAAG,SAAS,GAC7B,OAAO,CAWT;AAED,wBAAsB,gBAAgB,CAAC,WAAW,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,gBA0BzE;AAkCD,wBAAgB,oBAAoB,CAAC,EACnC,IAAI,EACJ,MAAM,GACP,EAAE;IACD,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,MAAM,CAAC;CAChB,GAAG,MAAM,CAAC,MAAM,EAAE;IAAE,GAAG,CAAC,EAAE,MAAM,CAAC;IAAC,GAAG,CAAC,EAAE,MAAM,CAAA;CAAE,CAAC,CA0BjD;AAED,wBAAsB,aAAa,CAAC,EAClC,QAAQ,EACR,QAAQ,EACR,SAAS,EACT,MAAM,GACP,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,mBAAmB,CAAC;IAC/B,MAAM,EAAE,MAAM,CAAC;CAChB,oBAkDA;AAED,wBAAsB,sBAAsB,CAAC,EAC3C,QAAQ,EACR,OAAO,EACP,OAAO,EACP,QAAQ,EACR,MAAM,GACP,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,OAAO,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;CAChB;;;GAqBA"}

package/dist/tools/upgrade-packages/utils.js

@@ -16,7 +16,7 @@ function isPatchUpdate(currentVersion, newVersion) {
     const next = newVersion.replace(/^[\^~]/, "").split(".");
     return (current[0] === next[0] && current[1] === next[1] && current[2] !== next[2]);
 }
-async function getLatestVersion(packageName) {
+async function getLatestVersion(packageName, logger) {
     const url = `https://registry.npmjs.org/${packageName}/latest`;
     try {
         const response = await fetch(url);
@@ -27,15 +27,23 @@ async function getLatestVersion(packageName) {
             throw new Error(errorMsg);
         }
         const packageInfo = await response.json();
-        console.log(`Latest version of ${packageName}: ${packageInfo.version}`);
+        logger.info(`Latest version of ${packageName}: ${packageInfo.version}`, {
+            tool: "upgradePackages",
+            packageName,
+            version: packageInfo.version,
+        });
         return packageInfo.version;
     }
     catch (error) {
-        console.error(`Error fetching package info: ${error.message}`);
+        logger.error(`Error fetching package info: ${error.message}`, {
+            tool: "upgradePackages",
+            packageName,
+            error,
+        });
         throw error;
     }
 }
-async function getGitDiffForPackageJson({ repoName, prNumber, apiClient, }) {
+async function getGitDiffForPackageJson({ repoName, prNumber, apiClient, logger, }) {
     const files = await apiClient.callGitHubProxy({
         method: "GET",
         url: `/repos/empirical-run/${repoName}/pulls/${prNumber}/files`,
@@ -45,12 +53,16 @@ async function getGitDiffForPackageJson({ repoName, prNumber, apiClient, }) {
     }
     const packageJsonChanges = files.find((file) => file.filename === "package.json");
     if (!packageJsonChanges) {
-        console.log("No package.json changes found in PR");
+        logger.info("No package.json changes found in PR", {
+            tool: "upgradePackages",
+            repoName,
+            prNumber,
+        });
         return null;
     }
     return packageJsonChanges.patch;
 }
-function parsePackageJsonDiff({ diff, }) {
+function parsePackageJsonDiff({ diff, logger, }) {
     const patchLines = diff.split("\n");
     const changes = {};
     for (let i = 0; i < patchLines.length; i++) {
@@ -60,7 +72,10 @@ function parsePackageJsonDiff({ diff, }) {
         if (match) {
             const [, pkg, version] = match;
             if (!pkg) {
-                console.warn("No package name found in line:", line);
+                logger.warn("No package name found in line:", {
+                    tool: "upgradePackages",
+                    line,
+                });
                 continue;
             }
             const changeType = line.startsWith("+") ? "new" : "old";
@@ -72,38 +87,63 @@ function parsePackageJsonDiff({ diff, }) {
     }
     return changes;
 }
-async function shouldMergePR({ repoName, prNumber, apiClient, }) {
+async function shouldMergePR({ repoName, prNumber, apiClient, logger, }) {
     const gitDiff = await getGitDiffForPackageJson({
         repoName,
         prNumber,
         apiClient,
+        logger,
     });
     if (!gitDiff) {
-        console.log("No package.json changes found, skipping merge");
+        logger.info("No package.json changes found, skipping merge", {
+            tool: "upgradePackages",
+            repoName,
+            prNumber,
+        });
         return false;
     }
-    const changes = parsePackageJsonDiff({ diff: gitDiff });
+    const changes = parsePackageJsonDiff({ diff: gitDiff, logger });
     if (Object.keys(changes).length === 0) {
-        console.log("No package.json changes found, skipping merge");
+        logger.info("No package.json changes found, skipping merge", {
+            tool: "upgradePackages",
+            repoName,
+            prNumber,
+        });
         return false;
     }
     let allPatchUpdates = true;
     for (const [pkg, versions] of Object.entries(changes)) {
         const { old: oldVersion, new: newVersion } = versions;
         if (!isPatchUpdate(oldVersion, newVersion)) {
-            console.log(`${pkg}: ${oldVersion} -> ${newVersion} is not a patch update`);
+            logger.info(`${pkg}: ${oldVersion} -> ${newVersion} is not a patch update`, {
+                tool: "upgradePackages",
+                package: pkg,
+                oldVersion,
+                newVersion,
+            });
             allPatchUpdates = false;
         }
         else {
-            console.log(`${pkg}: ${oldVersion} -> ${newVersion} is a patch update`);
+            logger.info(`${pkg}: ${oldVersion} -> ${newVersion} is a patch update`, {
+                tool: "upgradePackages",
+                package: pkg,
+                oldVersion,
+                newVersion,
+            });
         }
     }
     return allPatchUpdates;
 }
-async function upgradeAndStagePackage({ repoPath, pkgName, version, isDevDep, }) {
-    const pkgVersion = version ?? (await getLatestVersion(pkgName));
+async function upgradeAndStagePackage({ repoPath, pkgName, version, isDevDep, logger, }) {
+    const pkgVersion = version ?? (await getLatestVersion(pkgName, logger));
     (0, child_process_1.execSync)(`npm i ${pkgName}@${pkgVersion} ${isDevDep ? "--save-dev" : "--save"}`, { cwd: repoPath });
-    console.log(`Updated package: ${pkgName} version to ${pkgVersion} in ${repoPath}`);
+    logger.info(`Updated package: ${pkgName} version to ${pkgVersion} in ${repoPath}`, {
+        tool: "upgradePackages",
+        pkgName,
+        pkgVersion,
+        repoPath,
+        isDevDep,
+    });
     (0, child_process_1.execSync)("git add package.json package-lock.json", { cwd: repoPath });
     return {
         name: pkgName,

package/dist/tools/utils/queue.d.ts

@@ -1,4 +1,4 @@
-import { PendingToolCall } from "@empiricalrun/shared-types";
+import { PendingToolCall } from "@empiricalrun/shared-types/chat-agent";
 export declare function appendSuffixToQueueUrl(queueUrl: string, suffix: string): string;
 export declare function replaceRegionInUrl(url: string, newRegion: string): string;
 export declare function getQueueUrl(toolCalls: PendingToolCall[], needsBrowser: boolean): string;

package/dist/tools/utils/queue.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"queue.d.ts","sourceRoot":"","sources":["../../../src/tools/utils/queue.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,4BAA4B,CAAC;AAE7D,wBAAgB,sBAAsB,CACpC,QAAQ,EAAE,MAAM,EAChB,MAAM,EAAE,MAAM,GACb,MAAM,CAKR;AAED,wBAAgB,kBAAkB,CAAC,GAAG,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,GAAG,MAAM,CAOzE;AAkBD,wBAAgB,WAAW,CACzB,SAAS,EAAE,eAAe,EAAE,EAC5B,YAAY,EAAE,OAAO,GACpB,MAAM,CAeR"}
+{"version":3,"file":"queue.d.ts","sourceRoot":"","sources":["../../../src/tools/utils/queue.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,uCAAuC,CAAC;AAExE,wBAAgB,sBAAsB,CACpC,QAAQ,EAAE,MAAM,EAChB,MAAM,EAAE,MAAM,GACb,MAAM,CAKR;AAED,wBAAgB,kBAAkB,CAAC,GAAG,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,GAAG,MAAM,CAOzE;AAkBD,wBAAgB,WAAW,CACzB,SAAS,EAAE,eAAe,EAAE,EAC5B,YAAY,EAAE,OAAO,GACpB,MAAM,CAeR"}

package/dist/tools/utils/validate-schema.d.ts

@@ -0,0 +1,3 @@
+import { PendingToolCall, ToolResult } from "@empiricalrun/shared-types/chat-agent";
+export declare function validateToolInput<T>(toolCall: PendingToolCall, schema: T): ToolResult | null;
+//# sourceMappingURL=validate-schema.d.ts.map

package/dist/tools/utils/validate-schema.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"validate-schema.d.ts","sourceRoot":"","sources":["../../../src/tools/utils/validate-schema.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,eAAe,EACf,UAAU,EACX,MAAM,uCAAuC,CAAC;AAE/C,wBAAgB,iBAAiB,CAAC,CAAC,EACjC,QAAQ,EAAE,eAAe,EACzB,MAAM,EAAE,CAAC,GACR,UAAU,GAAG,IAAI,CAwBnB"}

package/dist/tools/utils/validate-schema.js

@@ -0,0 +1,22 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.validateToolInput = validateToolInput;
+function validateToolInput(toolCall, schema) {
+    const { name: toolName, input } = toolCall;
+    const strictSchema = schema._def?.typeName === "ZodObject"
+        ? schema.strict()
+        : schema;
+    const validation = strictSchema.safeParse(input);
+    if (!validation.success) {
+        const errorMessage = [
+            `Schema validation failed for tool "${toolName}".`,
+            `\nValidation errors: ${JSON.stringify(validation.error.issues, null, 2)}`,
+        ].join("\n");
+        return {
+            isError: true,
+            result: errorMessage,
+            artifacts: null,
+        };
+    }
+    return null;
+}

package/dist/utils/SQSClient.d.ts

@@ -1,4 +1,4 @@
-import { ServicePayload } from "@empiricalrun/shared-types";
+import { ServicePayload } from "@empiricalrun/shared-types/tool-execute-service";
 export declare class SQSClient {
     private region;
     private accessKeyId;

package/dist/utils/SQSClient.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"SQSClient.d.ts","sourceRoot":"","sources":["../../src/utils/SQSClient.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAE5D,qBAAa,SAAS;IACpB,OAAO,CAAC,MAAM,CAAS;IACvB,OAAO,CAAC,WAAW,CAAS;IAC5B,OAAO,CAAC,eAAe,CAAS;IAChC,OAAO,CAAC,YAAY,CAAC,CAAS;gBAG5B,MAAM,EAAE,MAAM,EACd,WAAW,EAAE,MAAM,EACnB,eAAe,EAAE,MAAM,EACvB,YAAY,CAAC,EAAE,MAAM;YAQT,MAAM;YAON,UAAU;YAkBV,eAAe;YAgBf,IAAI;IASZ,WAAW,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,cAAc,GAAG,OAAO,CAAC,IAAI,CAAC;CAyG5E"}
+{"version":3,"file":"SQSClient.d.ts","sourceRoot":"","sources":["../../src/utils/SQSClient.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,iDAAiD,CAAC;AAEjF,qBAAa,SAAS;IACpB,OAAO,CAAC,MAAM,CAAS;IACvB,OAAO,CAAC,WAAW,CAAS;IAC5B,OAAO,CAAC,eAAe,CAAS;IAChC,OAAO,CAAC,YAAY,CAAC,CAAS;gBAG5B,MAAM,EAAE,MAAM,EACd,WAAW,EAAE,MAAM,EACnB,eAAe,EAAE,MAAM,EACvB,YAAY,CAAC,EAAE,MAAM;YAQT,MAAM;YAON,UAAU;YAkBV,eAAe;YAgBf,IAAI;IASZ,WAAW,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,cAAc,GAAG,OAAO,CAAC,IAAI,CAAC;CAyG5E"}

package/dist/utils/dedup/dedup-image.js

@@ -16,8 +16,8 @@ async function deduplicateImages({ base64Images, threshold, logPrefix = "dedup-i
             previousImage = currentImage.image;
             continue;
         }
-        const similarity = await (0, find_threshold_1.findSimilarityPercentage)(previousImage, currentImage.image);
-        if (similarity < threshold) {
+        const diffFraction = await (0, find_threshold_1.findDiffFraction)(previousImage, currentImage.image);
+        if (diffFraction > threshold) {
             uniqueImages.push(currentImage);
             previousImage = currentImage.image;
         }

package/dist/utils/dedup/find-threshold.d.ts

@@ -1,2 +1,2 @@
-export declare function findSimilarityPercentage(base64Image1: string, base64Image2: string): Promise<number>;
+export declare function findDiffFraction(base64Image1: string, base64Image2: string): Promise<number>;
 //# sourceMappingURL=find-threshold.d.ts.map

package/dist/utils/dedup/find-threshold.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"find-threshold.d.ts","sourceRoot":"","sources":["../../../src/utils/dedup/find-threshold.ts"],"names":[],"mappings":"AAsCA,wBAAsB,wBAAwB,CAC5C,YAAY,EAAE,MAAM,EACpB,YAAY,EAAE,MAAM,GACnB,OAAO,CAAC,MAAM,CAAC,CASjB"}
+{"version":3,"file":"find-threshold.d.ts","sourceRoot":"","sources":["../../../src/utils/dedup/find-threshold.ts"],"names":[],"mappings":"AAsCA,wBAAsB,gBAAgB,CACpC,YAAY,EAAE,MAAM,EACpB,YAAY,EAAE,MAAM,GACnB,OAAO,CAAC,MAAM,CAAC,CASjB"}

package/dist/utils/dedup/find-threshold.js

@@ -3,7 +3,7 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
     return (mod && mod.__esModule) ? mod : { "default": mod };
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.findSimilarityPercentage = findSimilarityPercentage;
+exports.findDiffFraction = findDiffFraction;
 const pixelmatch_1 = __importDefault(require("pixelmatch"));
 const sharp_1 = __importDefault(require("sharp"));
 async function compareImageBuffers(buffer1, buffer2, pixelmatchThreshold) {
@@ -29,7 +29,7 @@ async function compareImageBuffers(buffer1, buffer2, pixelmatchThreshold) {
     const diffFraction = diffPixels / totalPixels;
     return { diffPixels, totalPixels, diffFraction };
 }
-async function findSimilarityPercentage(base64Image1, base64Image2) {
+async function findDiffFraction(base64Image1, base64Image2) {
     try {
         const buffer1 = Buffer.from(base64Image1, "base64");
         const buffer2 = Buffer.from(base64Image2, "base64");