@empiricalrun/test-gen 0.64.1 → 0.64.3

package/CHANGELOG.md

@@ -1,5 +1,22 @@
 # @empiricalrun/test-gen
 
+## 0.64.3
+
+### Patch Changes
+
+- d8d624d: fix: pass mapped keys to codegen recordAction for proper code generation
+- addd52e: feat: add JSON value truncation to prevent oversized tool responses
+- f8a53b0: fix: PR description should not get concatenated across versions
+
+## 0.64.2
+
+### Patch Changes
+
+- f05aac2: feat: add environment overrides for projects in tool calls
+- Updated dependencies [f05aac2]
+  - @empiricalrun/test-run@0.10.1
+  - @empiricalrun/llm@0.17.3
+
 ## 0.64.1
 
 ### Patch Changes

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

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/index.ts"],"names":[],"mappings":"AACA,OAAO,EAGL,mBAAmB,EACpB,MAAM,wBAAwB,CAAC;AAyChC,wBAAsB,kBAAkB,CAAC,EACvC,mBAAmB,EACnB,aAAa,EACb,oBAAoB,EACpB,SAAS,GACV,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,mBAAmB,EAAE,OAAO,CAAC;IAC7B,oBAAoB,EAAE,MAAM,GAAG,SAAS,CAAC;IACzC,SAAS,CAAC,EAAE,OAAO,CAAC;CACrB,iBAuHA;AAuBD,wBAAsB,wBAAwB,CAAC,EAC7C,aAAa,EACb,aAAa,GACd,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,aAAa,EAAE,MAAM,CAAC;CACvB,iBA6DA"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/index.ts"],"names":[],"mappings":"AACA,OAAO,EAGL,mBAAmB,EACpB,MAAM,wBAAwB,CAAC;AAyChC,wBAAsB,kBAAkB,CAAC,EACvC,mBAAmB,EACnB,aAAa,EACb,oBAAoB,EACpB,SAAS,GACV,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,mBAAmB,EAAE,OAAO,CAAC;IAC7B,oBAAoB,EAAE,MAAM,GAAG,SAAS,CAAC;IACzC,SAAS,CAAC,EAAE,OAAO,CAAC;CACrB,iBAwHA;AAuBD,wBAAsB,wBAAwB,CAAC,EAC7C,aAAa,EACb,aAAa,GACd,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,aAAa,EAAE,MAAM,CAAC;CACvB,iBA6DA"}

package/dist/agent/chat/index.js

@@ -111,6 +111,7 @@ async function runChatAgentForCLI({ useDiskForChatState, selectedModel, initialP
                 apiKey: process.env.EMPIRICALRUN_API_KEY,
                 trace,
                 featureFlags: [],
+                environmentOverrides: {},
             });
             const fileInfo = await (0, file_tree_1.getFileInfoFromFS)(process.cwd());
             await (0, agent_loop_1.chatAgentLoop)({

package/dist/agent/cua/computer.js

@@ -155,10 +155,10 @@ async function executeModelAction(page, action, codegen) {
             }
             case "keypress": {
                 const { keys } = action;
-                await codegen.recordAction({ type: "keypress", keys });
                 const mappedKeys = keys.map((k) => {
                     return CUA_KEY_TO_PLAYWRIGHT_KEY[k.toLowerCase()] || k;
                 });
+                await codegen.recordAction({ type: "keypress", keys: mappedKeys });
                 const mappedKey = mappedKeys.join("+"); // ["CTRL", "A"] becomes ControlOrMeta+A
                 console.log(`Action: keypress for keys ${keys} -> '${mappedKey}'`);
                 await page.keyboard.press(mappedKey);

package/dist/tool-call-service/index.d.ts

@@ -15,7 +15,8 @@ export declare class ToolCallService {
     apiKey: string | undefined;
     trace?: TraceClient;
     featureFlags: string[];
-    constructor({ chatSessionId, selectedModel, branchName, repoPath, apiKey, trace, featureFlags, }: {
+    environmentOverrides: Record<string, string>;
+    constructor({ chatSessionId, selectedModel, branchName, repoPath, apiKey, trace, featureFlags, environmentOverrides, }: {
         chatSessionId: number | null;
         selectedModel: SupportedChatModels;
         branchName: string;
@@ -23,6 +24,7 @@ export declare class ToolCallService {
         apiKey: string | undefined;
         trace?: TraceClient;
         featureFlags: string[];
+        environmentOverrides?: Record<string, string>;
     });
     sendToQueue(toolCalls: PendingToolCall[]): Promise<void>;
     execute(toolCalls: PendingToolCall[]): Promise<ToolResult[]>;

package/dist/tool-call-service/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/tool-call-service/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,EAEL,eAAe,EACf,mBAAmB,EACnB,UAAU,EACX,MAAM,wBAAwB,CAAC;AAChC,OAAO,EAAiB,IAAI,EAAE,WAAW,EAAE,MAAM,4BAA4B,CAAC;AAiB9E,YAAY,EAAE,mBAAmB,EAAE,CAAC;AAEpC,KAAK,aAAa,GAAG;IACnB,CAAC,GAAG,EAAE,MAAM,GAAG,WAAW,CAAC;CAC5B,CAAC;AAEF,qBAAa,eAAe;IAC1B,KAAK,EAAE,IAAI,EAAE,CAAM;IACnB,aAAa,EAAE,aAAa,CAAM;IAClC,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,aAAa,EAAE,mBAAmB,CAAC;IACnC,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC;IAC3B,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,YAAY,EAAE,MAAM,EAAE,CAAC;gBACX,EACV,aAAa,EACb,aAAa,EACb,UAAU,EACV,QAAQ,EACR,MAAM,EACN,KAAK,EACL,YAAY,GACb,EAAE;QACD,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;QAC7B,aAAa,EAAE,mBAAmB,CAAC;QACnC,UAAU,EAAE,MAAM,CAAC;QACnB,QAAQ,EAAE,MAAM,CAAC;QACjB,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC;QAC3B,KAAK,CAAC,EAAE,WAAW,CAAC;QACpB,YAAY,EAAE,MAAM,EAAE,CAAC;KACxB;IAkCK,WAAW,CAAC,SAAS,EAAE,eAAe,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC;IAexD,OAAO,CAAC,SAAS,EAAE,eAAe,EAAE,GAAG,OAAO,CAAC,UAAU,EAAE,CAAC;CAsEnE"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/tool-call-service/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,EAEL,eAAe,EACf,mBAAmB,EACnB,UAAU,EACX,MAAM,wBAAwB,CAAC;AAChC,OAAO,EAAiB,IAAI,EAAE,WAAW,EAAE,MAAM,4BAA4B,CAAC;AAiB9E,YAAY,EAAE,mBAAmB,EAAE,CAAC;AAEpC,KAAK,aAAa,GAAG;IACnB,CAAC,GAAG,EAAE,MAAM,GAAG,WAAW,CAAC;CAC5B,CAAC;AAEF,qBAAa,eAAe;IAC1B,KAAK,EAAE,IAAI,EAAE,CAAM;IACnB,aAAa,EAAE,aAAa,CAAM;IAClC,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,aAAa,EAAE,mBAAmB,CAAC;IACnC,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC;IAC3B,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,oBAAoB,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;gBACjC,EACV,aAAa,EACb,aAAa,EACb,UAAU,EACV,QAAQ,EACR,MAAM,EACN,KAAK,EACL,YAAY,EACZ,oBAAyB,GAC1B,EAAE;QACD,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;QAC7B,aAAa,EAAE,mBAAmB,CAAC;QACnC,UAAU,EAAE,MAAM,CAAC;QACnB,QAAQ,EAAE,MAAM,CAAC;QACjB,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC;QAC3B,KAAK,CAAC,EAAE,WAAW,CAAC;QACpB,YAAY,EAAE,MAAM,EAAE,CAAC;QACvB,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;KAC/C;IAmCK,WAAW,CAAC,SAAS,EAAE,eAAe,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC;IAkBxD,OAAO,CAAC,SAAS,EAAE,eAAe,EAAE,GAAG,OAAO,CAAC,UAAU,EAAE,CAAC;CAuEnE"}

package/dist/tool-call-service/index.js

@@ -24,7 +24,8 @@ class ToolCallService {
     apiKey;
     trace;
     featureFlags;
-    constructor({ chatSessionId, selectedModel, branchName, repoPath, apiKey, trace, featureFlags, }) {
+    environmentOverrides;
+    constructor({ chatSessionId, selectedModel, branchName, repoPath, apiKey, trace, featureFlags, environmentOverrides = {}, }) {
         this.chatSessionId = chatSessionId;
         this.selectedModel = selectedModel;
         this.branchName = branchName;
@@ -32,6 +33,7 @@ class ToolCallService {
         this.repoPath = repoPath;
         this.apiKey = apiKey;
         this.featureFlags = featureFlags;
+        this.environmentOverrides = environmentOverrides;
         this.tools = [
             grep_1.grepTool,
             test_run_1.runTestTool,
@@ -62,13 +64,16 @@ class ToolCallService {
         if (!requestId) {
             throw new Error("Could not find an id for the tool call.");
         }
-        await (0, utils_1.sendToolRequestToRemoteQueue)({
+        const queueUrl = (0, utils_1.getQueueUrl)(this.featureFlags);
+        if (!queueUrl) {
+            throw new Error(`queueUrl is required for remote execution.`);
+        }
+        await (0, utils_1.sendToolRequestToRemoteQueue)(queueUrl, {
             toolCalls,
             requestId,
             chatSessionId: this.chatSessionId,
             selectedModel: this.selectedModel,
             branchName: this.branchName,
-            featureFlags: this.featureFlags || [],
         });
     }
     async execute(toolCalls) {
@@ -110,6 +115,7 @@ class ToolCallService {
                     apiKey: this.apiKey,
                     trace: this.trace,
                     collectArtifacts: collectArtifactsFn,
+                    environmentOverrides: this.environmentOverrides,
                 });
                 const artifacts = await uploadArtifactsQueue.waitForCompletion();
                 toolResults.push({ ...result, artifacts });

package/dist/tool-call-service/utils.d.ts

@@ -1,11 +1,11 @@
 import { SupportedChatModels } from "@empiricalrun/llm/chat";
 import { PendingToolCall } from "@empiricalrun/shared-types";
-export declare function sendToolRequestToRemoteQueue(payload: {
+export declare function getQueueUrl(featureFlags: string[]): string | undefined;
+export declare function sendToolRequestToRemoteQueue(queueUrl: string, payload: {
     toolCalls: PendingToolCall[];
     requestId: string;
     chatSessionId: number;
     selectedModel: SupportedChatModels;
     branchName: string;
-    featureFlags: string[];
 }): Promise<void>;
 //# sourceMappingURL=utils.d.ts.map

package/dist/tool-call-service/utils.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../src/tool-call-service/utils.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,mBAAmB,EAAE,MAAM,wBAAwB,CAAC;AAC7D,OAAO,EAAE,eAAe,EAAE,MAAM,4BAA4B,CAAC;AAE7D,wBAAsB,4BAA4B,CAAC,OAAO,EAAE;IAC1D,SAAS,EAAE,eAAe,EAAE,CAAC;IAC7B,SAAS,EAAE,MAAM,CAAC;IAClB,aAAa,EAAE,MAAM,CAAC;IACtB,aAAa,EAAE,mBAAmB,CAAC;IACnC,UAAU,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,EAAE,CAAC;CACxB,iBA4BA"}
+{"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../src/tool-call-service/utils.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,mBAAmB,EAAE,MAAM,wBAAwB,CAAC;AAC7D,OAAO,EAAE,eAAe,EAAE,MAAM,4BAA4B,CAAC;AAE7D,wBAAgB,WAAW,CAAC,YAAY,EAAE,MAAM,EAAE,sBAMjD;AAED,wBAAsB,4BAA4B,CAChD,QAAQ,EAAE,MAAM,EAChB,OAAO,EAAE;IACP,SAAS,EAAE,eAAe,EAAE,CAAC;IAC7B,SAAS,EAAE,MAAM,CAAC;IAClB,aAAa,EAAE,MAAM,CAAC;IACtB,aAAa,EAAE,mBAAmB,CAAC;IACnC,UAAU,EAAE,MAAM,CAAC;CACpB,iBAiBF"}

package/dist/tool-call-service/utils.js

@@ -1,8 +1,16 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
+exports.getQueueUrl = getQueueUrl;
 exports.sendToolRequestToRemoteQueue = sendToolRequestToRemoteQueue;
 const client_sqs_1 = require("@aws-sdk/client-sqs");
-async function sendToolRequestToRemoteQueue(payload) {
+function getQueueUrl(featureFlags) {
+    const useToolExecutionOnEcs = featureFlags.includes("useToolExecutionOnEcs");
+    if (useToolExecutionOnEcs) {
+        return process.env.TOOL_EXECUTION_SQS_URL_ECS;
+    }
+    return process.env.TOOL_EXECUTION_SQS_URL;
+}
+async function sendToolRequestToRemoteQueue(queueUrl, payload) {
     const sqs = new client_sqs_1.SQSClient({
         region: process.env.AWS_REGION,
         credentials: {
@@ -10,14 +18,6 @@ async function sendToolRequestToRemoteQueue(payload) {
             secretAccessKey: process.env.AWS_SECRET_ACCESS_KEY,
         },
     });
-    const useToolExecutionOnEcs = payload.featureFlags.includes("useToolExecutionOnEcs");
-    let queueUrl = process.env.TOOL_EXECUTION_SQS_URL;
-    if (useToolExecutionOnEcs) {
-        queueUrl = process.env.TOOL_EXECUTION_SQS_URL_ECS;
-    }
-    if (!queueUrl) {
-        throw new Error(`${useToolExecutionOnEcs ? "TOOL_EXECUTION_SQS_URL_ECS" : "TOOL_EXECUTION_SQS_URL"} is required for remote execution.`);
-    }
     await sqs.send(new client_sqs_1.SendMessageCommand({
         QueueUrl: queueUrl,
         MessageBody: JSON.stringify(payload),

package/dist/tools/commit-and-create-pr.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"commit-and-create-pr.d.ts","sourceRoot":"","sources":["../../src/tools/commit-and-create-pr.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,4BAA4B,CAAC;AAOvD,UAAU,cAAc;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;CACd;AAED,wBAAgB,cAAc,CAAC,GAAG,EAAE,MAAM,GAAG,cAAc,CAiB1D;AA2CD,eAAO,MAAM,qBAAqB,EAAE,IA0FnC,CAAC"}
+{"version":3,"file":"commit-and-create-pr.d.ts","sourceRoot":"","sources":["../../src/tools/commit-and-create-pr.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAEV,IAAI,EAEL,MAAM,4BAA4B,CAAC;AAOpC,UAAU,cAAc;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;CACd;AAED,wBAAgB,cAAc,CAAC,GAAG,EAAE,MAAM,GAAG,cAAc,CAiB1D;AAiCD,eAAO,MAAM,qBAAqB,EAAE,IAkFnC,CAAC"}

package/dist/tools/commit-and-create-pr.js

@@ -31,16 +31,12 @@ For example, if you used the test run tool, you should include the results (and
 videos and other artifacts that help the reviewer gain more context and confidence in the changes. If tests pass, reviewer will see the video and merge the PR.
 If tests fail, reviewer will see the video and the test artifacts, and will be able to help you debug the issue.`),
 });
-function formatDescriptionWithTimestamp(description, existingBody, type = "create") {
+function descriptionWithTimestamp(description) {
     const timestamp = new Date()
         .toISOString()
         .replace("T", " ")
         .replace("Z", " UTC");
-    const timestampText = `<sup>${type === "create" ? "Created" : "Updated"} at ${timestamp}</sup>`;
-    if (existingBody) {
-        return `${existingBody}\n\n---\n${description}\n\n${timestampText}`;
-    }
-    return `${description}\n\n${timestampText}`;
+    return `${description}\n\n<sup>Updated at ${timestamp}</sup>`;
 }
 exports.createPullRequestTool = {
     schema: {
@@ -79,14 +75,10 @@ Don't ask the user for this information, just come up with it yourself.
             });
             const existingPR = existingPRs?.find((pr) => pr.head.ref === branchName);
             if (existingPR) {
-                // Append the new description to the existing PR description
-                const updatedDescription = formatDescriptionWithTimestamp(pullRequestDescription, existingPR.body, "update");
                 await (0, utils_1.callGitHubProxy)({
                     method: "PATCH",
                     url: `https://api.github.com/repos/${owner}/${repo}/pulls/${existingPR.number}`,
-                    body: {
-                        body: updatedDescription,
-                    },
+                    body: { body: descriptionWithTimestamp(pullRequestDescription) },
                     apiKey,
                 });
                 return {
@@ -94,7 +86,6 @@ Don't ask the user for this information, just come up with it yourself.
                     result: `Committed and pushed changes to existing PR: ${existingPR.html_url}`,
                 };
             }
-            const initialDescription = formatDescriptionWithTimestamp(pullRequestDescription);
             const pr = (await (0, utils_1.callGitHubProxy)({
                 method: "POST",
                 url: `https://api.github.com/repos/${owner}/${repo}/pulls`,
@@ -102,7 +93,7 @@ Don't ask the user for this information, just come up with it yourself.
                     title: pullRequestTitle,
                     head: branchName,
                     base: "main",
-                    body: initialDescription,
+                    body: descriptionWithTimestamp(pullRequestDescription),
                 },
                 apiKey,
             }));

package/dist/tools/diagnosis-fetcher.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"diagnosis-fetcher.d.ts","sourceRoot":"","sources":["../../src/tools/diagnosis-fetcher.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,4BAA4B,CAAC;AAevD,eAAO,MAAM,wBAAwB,EAAE,IAyFtC,CAAC"}
+{"version":3,"file":"diagnosis-fetcher.d.ts","sourceRoot":"","sources":["../../src/tools/diagnosis-fetcher.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAe,MAAM,4BAA4B,CAAC;AAepE,eAAO,MAAM,wBAAwB,EAAE,IAqFtC,CAAC"}

package/dist/tools/diagnosis-fetcher.js

@@ -19,7 +19,9 @@ exports.fetchDiagnosisReportTool = {
         description: "Fetch details about a test case diagnosis using its URL or slug",
         parameters: DiagnosisSchema,
     },
-    execute: async ({ input, repoPath, apiKey, }) => {
+    execute: async (params) => {
+        const { input } = params;
+        const { repoPath, apiKey } = params;
         const { diagnosisUrl } = input;
         // Extract the slug from the URL - it's the part after the last '--'
         const slug = diagnosisUrl.split("--").pop();

package/dist/tools/download-build.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"download-build.d.ts","sourceRoot":"","sources":["../../src/tools/download-build.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,4BAA4B,CAAC;AACvD,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAIxB,eAAO,MAAM,uBAAuB;;;;;;EAElC,CAAC;AAEH,MAAM,MAAM,sBAAsB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,uBAAuB,CAAC,CAAC;AAE7E,eAAO,MAAM,iBAAiB,EAAE,IAsC/B,CAAC"}
+{"version":3,"file":"download-build.d.ts","sourceRoot":"","sources":["../../src/tools/download-build.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAe,MAAM,4BAA4B,CAAC;AACpE,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAIxB,eAAO,MAAM,uBAAuB;;;;;;EAElC,CAAC;AAEH,MAAM,MAAM,sBAAsB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,uBAAuB,CAAC,CAAC;AAE7E,eAAO,MAAM,iBAAiB,EAAE,IAkC/B,CAAC"}

package/dist/tools/download-build.js

@@ -14,7 +14,9 @@ have a build URL, you can try getting the environment details with the getEnviro
 Environment details will include the build URL.`,
         parameters: exports.downloadBuildToolSchema,
     },
-    execute: async ({ input, repoPath, apiKey, }) => {
+    execute: async (params) => {
+        const { input } = params;
+        const { repoPath, apiKey } = params;
         if (!(await (0, test_build_1.hasDownloadScript)(repoPath))) {
             return {
                 isError: true,

package/dist/tools/environment-crud.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"environment-crud.d.ts","sourceRoot":"","sources":["../../src/tools/environment-crud.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,4BAA4B,CAAC;AA6DvD,eAAO,MAAM,kBAAkB,EAAE,IAiFhC,CAAC;AAGF,eAAO,MAAM,gBAAgB,EAAE,IAAI,EAAyB,CAAC"}
+{"version":3,"file":"environment-crud.d.ts","sourceRoot":"","sources":["../../src/tools/environment-crud.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAEV,IAAI,EAEL,MAAM,4BAA4B,CAAC;AA6DpC,eAAO,MAAM,kBAAkB,EAAE,IAoFhC,CAAC;AAGF,eAAO,MAAM,gBAAgB,EAAE,IAAI,EAAyB,CAAC"}

package/dist/tools/grep/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/grep/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAc,MAAM,4BAA4B,CAAC;AA8HnE,eAAO,MAAM,QAAQ,EAAE,IAuBtB,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/grep/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAEV,IAAI,EAGL,MAAM,4BAA4B,CAAC;AA8HpC,eAAO,MAAM,QAAQ,EAAE,IA2BtB,CAAC"}

package/dist/tools/str_replace_editor.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"str_replace_editor.d.ts","sourceRoot":"","sources":["../../src/tools/str_replace_editor.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,4BAA4B,CAAC;AAOnE,UAAU,eAAe;IACvB,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,UAAU,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC9B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AA6UD;;;GAGG;AACH,wBAAsB,wBAAwB,CAAC,EAC7C,KAAK,EACL,QAAQ,GACT,EAAE;IACD,KAAK,EAAE,eAAe,CAAC;IACvB,QAAQ,EAAE,MAAM,CAAC;CAClB,GAAG,OAAO,CAAC,UAAU,CAAC,CA2CtB;AAsID,eAAO,MAAM,eAAe,EAAE,IAAI,EAKjC,CAAC"}
+{"version":3,"file":"str_replace_editor.d.ts","sourceRoot":"","sources":["../../src/tools/str_replace_editor.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,IAAI,EAAe,UAAU,EAAE,MAAM,4BAA4B,CAAC;AAOhF,UAAU,eAAe;IACvB,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,UAAU,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC9B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AA6UD;;;GAGG;AACH,wBAAsB,wBAAwB,CAAC,EAC7C,KAAK,EACL,QAAQ,GACT,EAAE;IACD,KAAK,EAAE,eAAe,CAAC;IACvB,QAAQ,EAAE,MAAM,CAAC;CAClB,GAAG,OAAO,CAAC,UAAU,CAAC,CA2CtB;AA8HD,eAAO,MAAM,eAAe,EAAE,IAAI,EAKjC,CAAC"}

package/dist/tools/str_replace_editor.js

@@ -340,7 +340,9 @@ File contents are returned with line numbers, starting from 1.
             path: zod_1.z.string().describe("The path to the file or directory to view."),
         }),
     },
-    execute: async ({ input, repoPath, }) => {
+    execute: async (params) => {
+        const { input } = params;
+        const { repoPath } = params;
         return strReplaceEditorExecutor({
             input: {
                 command: "view",
@@ -362,7 +364,9 @@ directories "tests", "tests/example", and "tests/example/foo.spec.ts".`,
             file_text: zod_1.z.string().describe("The contents of the new file."),
         }),
     },
-    execute: async ({ input, repoPath, }) => {
+    execute: async (params) => {
+        const { input } = params;
+        const { repoPath } = params;
         return strReplaceEditorExecutor({
             input: {
                 command: "create",
@@ -384,7 +388,9 @@ in the file. If old_str is not unique, the tool will return an error.`,
             new_str: zod_1.z.string().describe("The string to replace old_str with."),
         }),
     },
-    execute: async ({ input, repoPath, }) => {
+    execute: async (params) => {
+        const { input } = params;
+        const { repoPath } = params;
         return strReplaceEditorExecutor({
             input: {
                 command: "str_replace",
@@ -412,7 +418,9 @@ To insert a string at the end of the file, you should use insert_line = (total l
             new_str: zod_1.z.string().describe("The string to insert."),
         }),
     },
-    execute: async ({ input, repoPath, }) => {
+    execute: async (params) => {
+        const { input } = params;
+        const { repoPath } = params;
         return strReplaceEditorExecutor({
             input: {
                 command: "insert",

package/dist/tools/test-gen-browser.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"test-gen-browser.d.ts","sourceRoot":"","sources":["../../src/tools/test-gen-browser.ts"],"names":[],"mappings":"AACA,OAAO,EAAoB,IAAI,EAAE,MAAM,4BAA4B,CAAC;AAmGpE,eAAO,MAAM,4BAA4B,EAAE,IAwI1C,CAAC"}
+{"version":3,"file":"test-gen-browser.d.ts","sourceRoot":"","sources":["../../src/tools/test-gen-browser.ts"],"names":[],"mappings":"AACA,OAAO,EAAoB,IAAI,EAAE,MAAM,4BAA4B,CAAC;AAmGpE,eAAO,MAAM,4BAA4B,EAAE,IA0I1C,CAAC"}

package/dist/tools/test-gen-browser.js

@@ -95,7 +95,7 @@ exports.generateTestWithBrowserAgent = {
         description: BROWSER_AGENT_DESCRIPTION,
         parameters: BrowserAgentSchema,
     },
-    execute: async ({ input, repoPath, trace, collectArtifacts, }) => {
+    execute: async ({ input, repoPath, trace, collectArtifacts, environmentOverrides = {}, }) => {
         const { filePath, project } = input;
         const absoluteFilePath = path_1.default.join(repoPath, filePath);
         if (!fs_1.default.existsSync(absoluteFilePath)) {
@@ -152,7 +152,7 @@ exports.generateTestWithBrowserAgent = {
         catch (err) {
             console.warn("[generateTestWithBrowserAgent] Error preparing playwright for codegen", err);
         }
-        let envOverrides = {};
+        let envOverrides = { ...environmentOverrides };
         if (input.buildUrl) {
             envOverrides = {
                 ...envOverrides,

package/dist/tools/test-run.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"test-run.d.ts","sourceRoot":"","sources":["../../src/tools/test-run.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAc,MAAM,4BAA4B,CAAC;AAsDnE,eAAO,MAAM,WAAW,EAAE,IAiFzB,CAAC"}
+{"version":3,"file":"test-run.d.ts","sourceRoot":"","sources":["../../src/tools/test-run.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAc,MAAM,4BAA4B,CAAC;AAwDnE,eAAO,MAAM,WAAW,EAAE,IAoFzB,CAAC"}

package/dist/tools/test-run.js

@@ -10,6 +10,7 @@ const path_1 = __importDefault(require("path"));
 const zod_1 = require("zod");
 const artifacts_1 = require("../artifacts");
 const utils_1 = require("../artifacts/utils");
+const utils_2 = require("../utils");
 const RunTestSchema = zod_1.z.object({
     testName: zod_1.z.string().describe("The name of the test to run"),
     suites: zod_1.z
@@ -30,13 +31,14 @@ function buildReportUrl(projectName, testRunId) {
     return `https://reports.empirical.run/${projectName}/${testRunId}/index.html`;
 }
 function buildResult({ hasTestPassed, summaryJson, reportUrl, }) {
+    const truncatedSummaryJson = (0, utils_2.truncateJsonValues)(summaryJson);
     return `
 Test run is complete. Result: ${hasTestPassed ? "Passed" : "Failed"}
 
 ${reportUrl ? `Report URL: ${reportUrl}` : ""}
 
 # Raw result (in Playwright JSON format)
-${JSON.stringify(summaryJson)}
+${JSON.stringify(truncatedSummaryJson)}
 `;
 }
 exports.runTestTool = {
@@ -45,14 +47,15 @@ exports.runTestTool = {
         description: "Run a test",
         parameters: RunTestSchema,
     },
-    execute: async ({ input, repoPath, collectArtifacts, }) => {
+    execute: async ({ input, repoPath, collectArtifacts, environmentOverrides = {}, }) => {
         let reportUrl = undefined;
-        let envOverrides = {};
+        let envOverrides = { ...environmentOverrides };
         if ((0, artifacts_1.isArtifactCollectionEnabled)()) {
             const projectName = "test-gen-chat-agent";
             const testRunId = Date.now().toString();
             reportUrl = buildReportUrl(projectName, testRunId);
             envOverrides = {
+                ...envOverrides,
                 PROJECT_NAME: projectName,
                 TEST_RUN_GITHUB_ACTION_ID: testRunId,
             };

package/dist/utils/index.d.ts

@@ -1,2 +1,3 @@
 export { getTypescriptTestBlock } from "../bin/utils/platform/web";
+export { truncateJsonValues } from "./json";
 //# sourceMappingURL=index.d.ts.map

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

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/utils/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,sBAAsB,EAAE,MAAM,2BAA2B,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/utils/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,sBAAsB,EAAE,MAAM,2BAA2B,CAAC;AACnE,OAAO,EAAE,kBAAkB,EAAE,MAAM,QAAQ,CAAC"}

package/dist/utils/index.js

@@ -1,5 +1,7 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.getTypescriptTestBlock = void 0;
+exports.truncateJsonValues = exports.getTypescriptTestBlock = void 0;
 var web_1 = require("../bin/utils/platform/web");
 Object.defineProperty(exports, "getTypescriptTestBlock", { enumerable: true, get: function () { return web_1.getTypescriptTestBlock; } });
+var json_1 = require("./json");
+Object.defineProperty(exports, "truncateJsonValues", { enumerable: true, get: function () { return json_1.truncateJsonValues; } });

package/dist/utils/json.d.ts

@@ -0,0 +1,2 @@
+export declare function truncateJsonValues(obj: any, maxLength?: number): any;
+//# sourceMappingURL=json.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"json.d.ts","sourceRoot":"","sources":["../../src/utils/json.ts"],"names":[],"mappings":"AAAA,wBAAgB,kBAAkB,CAAC,GAAG,EAAE,GAAG,EAAE,SAAS,GAAE,MAAa,GAAG,GAAG,CAuB1E"}

package/dist/utils/json.js

@@ -0,0 +1,24 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.truncateJsonValues = truncateJsonValues;
+function truncateJsonValues(obj, maxLength = 1000) {
+    if (typeof obj === "string") {
+        if (obj.length > maxLength) {
+            const suffix = "...[truncated]";
+            const truncateAt = maxLength - suffix.length;
+            return obj.substring(0, truncateAt) + suffix;
+        }
+        return obj;
+    }
+    if (Array.isArray(obj)) {
+        return obj.map((item) => truncateJsonValues(item, maxLength));
+    }
+    if (obj !== null && typeof obj === "object") {
+        const truncated = {};
+        for (const [key, value] of Object.entries(obj)) {
+            truncated[key] = truncateJsonValues(value, maxLength);
+        }
+        return truncated;
+    }
+    return obj;
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@empiricalrun/test-gen",
-  "version": "0.64.1",
+  "version": "0.64.3",
   "publishConfig": {
     "registry": "https://registry.npmjs.org/",
     "access": "public"
@@ -64,7 +64,7 @@
     "zod": "^3.23.8",
     "@empiricalrun/llm": "^0.17.3",
     "@empiricalrun/r2-uploader": "^0.3.9",
-    "@empiricalrun/test-run": "^0.10.0"
+    "@empiricalrun/test-run": "^0.10.1"
   },
   "devDependencies": {
     "@playwright/test": "1.47.1",
@@ -79,7 +79,7 @@
     "playwright": "1.47.1",
     "serve-handler": "^6.1.6",
     "ts-patch": "^3.3.0",
-    "@empiricalrun/shared-types": "0.4.1"
+    "@empiricalrun/shared-types": "0.4.2"
   },
   "scripts": {
     "dev": "tspc --build --watch",

package/tsconfig.tsbuildinfo

@@ -1 +1 @@
-{"root":["./src/index.ts","./src/actions/assert.ts","./src/actions/click.ts","./src/actions/done.ts","./src/actions/fill.ts","./src/actions/goto.ts","./src/actions/hover.ts","./src/actions/index.ts","./src/actions/next-task.ts","./src/actions/press.ts","./src/actions/skill.ts","./src/actions/text-content.ts","./src/actions/constants/index.ts","./src/actions/utils/index.ts","./src/agent/browsing/index.ts","./src/agent/browsing/run.ts","./src/agent/browsing/utils.ts","./src/agent/chat/agent-loop.ts","./src/agent/chat/exports.ts","./src/agent/chat/index.ts","./src/agent/chat/models.ts","./src/agent/chat/prompt.ts","./src/agent/chat/repo.ts","./src/agent/chat/state.ts","./src/agent/chat/types.ts","./src/agent/chat/utils.ts","./src/agent/codegen/create-test-block.ts","./src/agent/codegen/fix-ts-errors.ts","./src/agent/codegen/generate-code-apply-changes.ts","./src/agent/codegen/lexical-scoped-vars.ts","./src/agent/codegen/repo-edit.ts","./src/agent/codegen/run.ts","./src/agent/codegen/skills-retriever.ts","./src/agent/codegen/test-update-feedback.ts","./src/agent/codegen/types.ts","./src/agent/codegen/update-flow.ts","./src/agent/codegen/use-skill.ts","./src/agent/codegen/utils.ts","./src/agent/cua/computer.ts","./src/agent/cua/index.ts","./src/agent/cua/model.ts","./src/agent/cua/pw-codegen/element-from-point.ts","./src/agent/cua/pw-codegen/types.ts","./src/agent/cua/pw-codegen/pw-pause/index.ts","./src/agent/cua/pw-codegen/pw-pause/utils.ts","./src/agent/diagnosis-agent/index.ts","./src/agent/diagnosis-agent/strict-mode-violation.ts","./src/agent/enrich-prompt/index.ts","./src/agent/enrich-prompt/utils.ts","./src/agent/infer-agent/index.ts","./src/agent/master/action-tool-calls.ts","./src/agent/master/element-annotation.ts","./src/agent/master/execute-browser-action.ts","./src/agent/master/execute-skill-action.ts","./src/agent/master/next-action.ts","./src/agent/master/planner.ts","./src/agent/master/run.ts","./src/agent/master/scroller.ts","./src/agent/master/with-hints.ts","./src/agent/master/browser-tests/cua.spec.ts","./src/agent/master/browser-tests/fixtures.ts","./src/agent/master/browser-tests/index.spec.ts","./src/agent/master/browser-tests/skills.spec.ts","./src/agent/master/icon-descriptor/index.ts","./src/agent/master/icon-descriptor/normalize-svg.ts","./src/agent/planner/run-time-planner.ts","./src/agent/planner/run.ts","./src/artifacts/index.ts","./src/artifacts/utils.ts","./src/bin/index.ts","./src/bin/logger/index.ts","./src/bin/utils/context.ts","./src/bin/utils/index.ts","./src/bin/utils/fs/index.ts","./src/bin/utils/platform/web/index.ts","./src/bin/utils/platform/web/test-files/ts-path-import-validate.ts","./src/bin/utils/scenarios/index.ts","./src/browser-injected-scripts/annotate-elements.spec.ts","./src/constants/index.ts","./src/errors/index.ts","./src/evals/add-scenario-agent.evals.ts","./src/evals/append-create-test-agent.evals.ts","./src/evals/fetch-pom-skills-agent.evals.ts","./src/evals/infer-master-or-code-agent.evals.ts","./src/evals/master-agent.evals.ts","./src/evals/type.ts","./src/evals/update-scenario-agent.evals.ts","./src/file/client.ts","./src/file/server.ts","./src/human-in-the-loop/cli.ts","./src/human-in-the-loop/index.ts","./src/human-in-the-loop/ipc.ts","./src/page/index.ts","./src/prompts/lib/ts-transformer.ts","./src/reporter/index.ts","./src/reporter/lib.ts","./src/session/index.ts","./src/test-build/index.ts","./src/tool-call-service/index.ts","./src/tool-call-service/utils.ts","./src/tools/commit-and-create-pr.ts","./src/tools/diagnosis-fetcher.ts","./src/tools/download-build.ts","./src/tools/environment-crud.ts","./src/tools/str_replace_editor.ts","./src/tools/test-gen-browser.ts","./src/tools/test-run.ts","./src/tools/grep/index.ts","./src/tools/grep/ripgrep/index.ts","./src/tools/grep/ripgrep/types.ts","./src/tools/test-run-fetcher/index.ts","./src/tools/test-run-fetcher/types.ts","./src/tools/utils/index.ts","./src/types/handlebars.d.ts","./src/types/index.ts","./src/uploader/index.ts","./src/uploader/utils.ts","./src/utils/checkpoint.ts","./src/utils/env.ts","./src/utils/exec.ts","./src/utils/file-tree.ts","./src/utils/file.ts","./src/utils/git.ts","./src/utils/html.ts","./src/utils/index.ts","./src/utils/repo-tree.ts","./src/utils/slug.ts","./src/utils/string.ts","./src/utils/stripAnsi.ts"],"version":"5.8.3"}
+{"root":["./src/index.ts","./src/actions/assert.ts","./src/actions/click.ts","./src/actions/done.ts","./src/actions/fill.ts","./src/actions/goto.ts","./src/actions/hover.ts","./src/actions/index.ts","./src/actions/next-task.ts","./src/actions/press.ts","./src/actions/skill.ts","./src/actions/text-content.ts","./src/actions/constants/index.ts","./src/actions/utils/index.ts","./src/agent/browsing/index.ts","./src/agent/browsing/run.ts","./src/agent/browsing/utils.ts","./src/agent/chat/agent-loop.ts","./src/agent/chat/exports.ts","./src/agent/chat/index.ts","./src/agent/chat/models.ts","./src/agent/chat/prompt.ts","./src/agent/chat/repo.ts","./src/agent/chat/state.ts","./src/agent/chat/types.ts","./src/agent/chat/utils.ts","./src/agent/codegen/create-test-block.ts","./src/agent/codegen/fix-ts-errors.ts","./src/agent/codegen/generate-code-apply-changes.ts","./src/agent/codegen/lexical-scoped-vars.ts","./src/agent/codegen/repo-edit.ts","./src/agent/codegen/run.ts","./src/agent/codegen/skills-retriever.ts","./src/agent/codegen/test-update-feedback.ts","./src/agent/codegen/types.ts","./src/agent/codegen/update-flow.ts","./src/agent/codegen/use-skill.ts","./src/agent/codegen/utils.ts","./src/agent/cua/computer.ts","./src/agent/cua/index.ts","./src/agent/cua/model.ts","./src/agent/cua/pw-codegen/element-from-point.ts","./src/agent/cua/pw-codegen/types.ts","./src/agent/cua/pw-codegen/pw-pause/index.ts","./src/agent/cua/pw-codegen/pw-pause/utils.ts","./src/agent/diagnosis-agent/index.ts","./src/agent/diagnosis-agent/strict-mode-violation.ts","./src/agent/enrich-prompt/index.ts","./src/agent/enrich-prompt/utils.ts","./src/agent/infer-agent/index.ts","./src/agent/master/action-tool-calls.ts","./src/agent/master/element-annotation.ts","./src/agent/master/execute-browser-action.ts","./src/agent/master/execute-skill-action.ts","./src/agent/master/next-action.ts","./src/agent/master/planner.ts","./src/agent/master/run.ts","./src/agent/master/scroller.ts","./src/agent/master/with-hints.ts","./src/agent/master/browser-tests/cua.spec.ts","./src/agent/master/browser-tests/fixtures.ts","./src/agent/master/browser-tests/index.spec.ts","./src/agent/master/browser-tests/skills.spec.ts","./src/agent/master/icon-descriptor/index.ts","./src/agent/master/icon-descriptor/normalize-svg.ts","./src/agent/planner/run-time-planner.ts","./src/agent/planner/run.ts","./src/artifacts/index.ts","./src/artifacts/utils.ts","./src/bin/index.ts","./src/bin/logger/index.ts","./src/bin/utils/context.ts","./src/bin/utils/index.ts","./src/bin/utils/fs/index.ts","./src/bin/utils/platform/web/index.ts","./src/bin/utils/platform/web/test-files/ts-path-import-validate.ts","./src/bin/utils/scenarios/index.ts","./src/browser-injected-scripts/annotate-elements.spec.ts","./src/constants/index.ts","./src/errors/index.ts","./src/evals/add-scenario-agent.evals.ts","./src/evals/append-create-test-agent.evals.ts","./src/evals/fetch-pom-skills-agent.evals.ts","./src/evals/infer-master-or-code-agent.evals.ts","./src/evals/master-agent.evals.ts","./src/evals/type.ts","./src/evals/update-scenario-agent.evals.ts","./src/file/client.ts","./src/file/server.ts","./src/human-in-the-loop/cli.ts","./src/human-in-the-loop/index.ts","./src/human-in-the-loop/ipc.ts","./src/page/index.ts","./src/prompts/lib/ts-transformer.ts","./src/reporter/index.ts","./src/reporter/lib.ts","./src/session/index.ts","./src/test-build/index.ts","./src/tool-call-service/index.ts","./src/tool-call-service/utils.ts","./src/tools/commit-and-create-pr.ts","./src/tools/diagnosis-fetcher.ts","./src/tools/download-build.ts","./src/tools/environment-crud.ts","./src/tools/str_replace_editor.ts","./src/tools/test-gen-browser.ts","./src/tools/test-run.ts","./src/tools/grep/index.ts","./src/tools/grep/ripgrep/index.ts","./src/tools/grep/ripgrep/types.ts","./src/tools/test-run-fetcher/index.ts","./src/tools/test-run-fetcher/types.ts","./src/tools/utils/index.ts","./src/types/handlebars.d.ts","./src/types/index.ts","./src/uploader/index.ts","./src/uploader/utils.ts","./src/utils/checkpoint.ts","./src/utils/env.ts","./src/utils/exec.ts","./src/utils/file-tree.ts","./src/utils/file.ts","./src/utils/git.ts","./src/utils/html.ts","./src/utils/index.ts","./src/utils/json.ts","./src/utils/repo-tree.ts","./src/utils/slug.ts","./src/utils/string.ts","./src/utils/stripAnsi.ts"],"version":"5.8.3"}