@empiricalrun/test-gen 0.62.0 → 0.63.0

package/CHANGELOG.md

@@ -1,5 +1,24 @@
 # @empiricalrun/test-gen
 
+## 0.63.0
+
+### Minor Changes
+
+- f926e40: feat: consume repoPath in test run tool
+- 55c7913: feat: artifact collection is fire-and-forget at tool level
+- 2f1ee31: chore: consume repoPath and apikey in tools requiring Dashboard api call
+
+### Patch Changes
+
+- c491cdd: fix: insert_line description in text editor tools
+- 0fea8f1: test: add a multiline old_str replacement test
+- 115a023: chore: allow grep tool to consume repoPath for the scope of search
+- 6c7740b: feat: consume repoPath for all file edit tools
+- Updated dependencies [f926e40]
+- Updated dependencies [2f1ee31]
+  - @empiricalrun/test-run@0.9.4
+  - @empiricalrun/llm@0.17.1
+
 ## 0.62.0
 
 ### Minor 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,iBAmHA;AAuBD,wBAAsB,wBAAwB,CAAC,EAC7C,aAAa,EACb,aAAa,GACd,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,aAAa,EAAE,MAAM,CAAC;CACvB,iBAyDA"}
+{"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,iBAsHA;AAuBD,wBAAsB,wBAAwB,CAAC,EAC7C,aAAa,EACb,aAAa,GACd,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,aAAa,EAAE,MAAM,CAAC;CACvB,iBA4DA"}

package/dist/agent/chat/index.js

@@ -29,7 +29,7 @@ async function runChatAgentForCLI({ useDiskForChatState, selectedModel, initialP
     // TODO: Store branch name in chat state so that we don't recreate it every time
     const randomId = crypto.randomUUID().substring(0, 8);
     const branchName = `branch-${randomId}`;
-    await (0, git_1.checkoutBranch)(branchName);
+    await (0, git_1.checkoutBranch)(branchName, process.cwd());
     let messagesLoadedFromDisk = chatState?.messages || [];
     let chatModel = (0, chat_1.createChatModel)(messagesLoadedFromDisk, selectedModel);
     if (initialPromptContent && chatModel.messages.length === 0) {
@@ -100,6 +100,9 @@ async function runChatAgentForCLI({ useDiskForChatState, selectedModel, initialP
         }
         else {
             // TODO: Should we pass a loader function? That would allow us to show a spinner
+            if (!process.env.EMPIRICALRUN_API_KEY) {
+                throw new Error("EMPIRICALRUN_API_KEY is not set");
+            }
             const toolCallService = new tool_call_service_1.ToolCallService({
                 chatSessionId: null,
                 selectedModel,
@@ -159,6 +162,9 @@ async function runChatAgentForDashboard({ chatSessionId, selectedModel, }) {
             chatSessionId,
         },
     });
+    if (!process.env.EMPIRICALRUN_API_KEY) {
+        throw new Error("EMPIRICALRUN_API_KEY is not set");
+    }
     const toolCallService = new tool_call_service_1.ToolCallService({
         chatSessionId,
         selectedModel,
@@ -167,7 +173,7 @@ async function runChatAgentForDashboard({ chatSessionId, selectedModel, }) {
         apiKey: process.env.EMPIRICALRUN_API_KEY,
         trace,
     });
-    await (0, git_1.checkoutBranch)(branchName);
+    await (0, git_1.checkoutBranch)(branchName, process.cwd());
     let chatModel = (0, chat_1.createChatModel)(chatState.messages, selectedModel);
     let reporterFunc = async (chatState, latest) => {
         const response = await fetch(`${DASHBOARD_DOMAIN}/api/chat-sessions/${chatSessionId}`, {

package/dist/artifacts/index.d.ts

@@ -40,4 +40,13 @@ export declare function isArtifactCollectionEnabled(): string | undefined;
  * // Throws: "Invalid path: relative/path/test.png..."
  */
 export declare function collectArtifacts(inputs: ArtifactInput[], repoDir: string, toolCallId: string): Promise<Artifact[]>;
+export declare class UploadArtifactsQueue {
+    private toolCallId;
+    private baseRepoPath;
+    private artifactResults;
+    private uploadPromise;
+    constructor(baseRepoPath: string, toolCallId: string);
+    addTask(artifacts: ArtifactInput[]): Promise<void>;
+    waitForCompletion(): Promise<Artifact[]>;
+}
 //# sourceMappingURL=index.d.ts.map

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

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/artifacts/index.ts"],"names":[],"mappings":"AAIA,OAAO,EACL,QAAQ,EACR,aAAa,EAGd,MAAM,4BAA4B,CAAC;AAkBpC,wBAAgB,2BAA2B,uBAM1C;AA6HD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAsCG;AACH,wBAAsB,gBAAgB,CACpC,MAAM,EAAE,aAAa,EAAE,EACvB,OAAO,EAAE,MAAM,EACf,UAAU,EAAE,MAAM,GACjB,OAAO,CAAC,QAAQ,EAAE,CAAC,CA6DrB"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/artifacts/index.ts"],"names":[],"mappings":"AAIA,OAAO,EACL,QAAQ,EACR,aAAa,EAGd,MAAM,4BAA4B,CAAC;AAkBpC,wBAAgB,2BAA2B,uBAM1C;AA6HD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAsCG;AACH,wBAAsB,gBAAgB,CACpC,MAAM,EAAE,aAAa,EAAE,EACvB,OAAO,EAAE,MAAM,EACf,UAAU,EAAE,MAAM,GACjB,OAAO,CAAC,QAAQ,EAAE,CAAC,CA6DrB;AAED,qBAAa,oBAAoB;IAC/B,OAAO,CAAC,UAAU,CAAS;IAC3B,OAAO,CAAC,YAAY,CAAS;IAC7B,OAAO,CAAC,eAAe,CAAkB;IACzC,OAAO,CAAC,aAAa,CAA8B;gBAEvC,YAAY,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM;IAKvC,OAAO,CAAC,SAAS,EAAE,aAAa,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC;IAelD,iBAAiB,IAAI,OAAO,CAAC,QAAQ,EAAE,CAAC;CAMtD"}

package/dist/artifacts/index.js

@@ -3,6 +3,7 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
     return (mod && mod.__esModule) ? mod : { "default": mod };
 };
 Object.defineProperty(exports, "__esModule", { value: true });
+exports.UploadArtifactsQueue = void 0;
 exports.isArtifactCollectionEnabled = isArtifactCollectionEnabled;
 exports.collectArtifacts = collectArtifacts;
 const r2_uploader_1 = require("@empiricalrun/r2-uploader");
@@ -207,3 +208,30 @@ async function collectArtifacts(inputs, repoDir, toolCallId) {
         return [];
     }
 }
+class UploadArtifactsQueue {
+    toolCallId;
+    baseRepoPath;
+    artifactResults = [];
+    uploadPromise = null;
+    constructor(baseRepoPath, toolCallId) {
+        this.baseRepoPath = baseRepoPath;
+        this.toolCallId = toolCallId;
+    }
+    async addTask(artifacts) {
+        this.uploadPromise = collectArtifacts(artifacts, this.baseRepoPath, this.toolCallId)
+            .then((results) => {
+            this.artifactResults.push(...results);
+        })
+            .catch((error) => {
+            console.error("Error processing artifact upload task:", error);
+            throw error;
+        });
+    }
+    async waitForCompletion() {
+        if (this.uploadPromise) {
+            await this.uploadPromise;
+        }
+        return this.artifactResults;
+    }
+}
+exports.UploadArtifactsQueue = UploadArtifactsQueue;

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,IAAI,EACJ,WAAW,EACX,UAAU,EACX,MAAM,wBAAwB,CAAC;AAiBhC,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;gBACR,EACV,aAAa,EACb,aAAa,EACb,UAAU,EACV,QAAQ,EACR,MAAM,EACN,KAAK,GACN,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;KACrB;IAiCK,WAAW,CAAC,SAAS,EAAE,eAAe,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC;IAcxD,OAAO,CAAC,SAAS,EAAE,eAAe,EAAE,GAAG,OAAO,CAAC,UAAU,EAAE,CAAC;CAmDnE"}
+{"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,IAAI,EACJ,WAAW,EACX,UAAU,EACX,MAAM,wBAAwB,CAAC;AAkBhC,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;gBACR,EACV,aAAa,EACb,aAAa,EACb,UAAU,EACV,QAAQ,EACR,MAAM,EACN,KAAK,GACN,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;KACrB;IAiCK,WAAW,CAAC,SAAS,EAAE,eAAe,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC;IAcxD,OAAO,CAAC,SAAS,EAAE,eAAe,EAAE,GAAG,OAAO,CAAC,UAAU,EAAE,CAAC;CAsEnE"}

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

@@ -79,6 +79,13 @@ class ToolCallService {
                 name: `tool: ${toolCall.name}`,
                 input: toolCall.input,
             });
+            const uploadArtifactsQueue = new artifacts_1.UploadArtifactsQueue(this.repoPath, toolCall.id);
+            const collectArtifactsFn = (artifactsInput) => {
+                uploadArtifactsQueue.addTask(artifactsInput).catch((error) => {
+                    console.error("Error collecting artifacts:", error);
+                });
+                return;
+            };
             const toolExecutor = this.toolExecutors[toolCall.name];
             if (!toolExecutor) {
                 const errorResult = {
@@ -90,9 +97,13 @@ class ToolCallService {
                 span?.end({ output: errorResult });
                 continue;
             }
+            if (!this.apiKey) {
+                throw new Error("API key is required for tool execution");
+            }
             try {
-                const result = await toolExecutor(toolCall.input, this.repoPath, this.apiKey, this.trace, artifacts_1.collectArtifacts);
-                toolResults.push(result);
+                const result = await toolExecutor(toolCall.input, this.repoPath, this.apiKey, this.trace, collectArtifactsFn);
+                const artifacts = await uploadArtifactsQueue.waitForCompletion();
+                toolResults.push({ ...result, artifacts });
                 span?.end({ output: result });
             }
             catch (error) {
@@ -105,7 +116,11 @@ class ToolCallService {
                 span?.end({ output: errorResult });
             }
         }
-        await (0, checkpoint_1.createCheckpoint)(toolCalls, this.branchName);
+        await (0, checkpoint_1.createCheckpoint)({
+            toolCalls,
+            branchName: this.branchName,
+            repoPath: this.repoPath,
+        });
         executeSpan?.end({ output: { toolResults } });
         return toolResults;
     }

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,wBAAwB,CAAC;AAOnD,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,IA2EnC,CAAC"}
+{"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,wBAAwB,CAAC;AAOnD,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,IAsFnC,CAAC"}

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

@@ -55,11 +55,13 @@ Don't ask the user for this information, just come up with it yourself.
 `,
         parameters: createPullRequestSchema,
     },
-    execute: async (input) => {
+    execute: async (input, repoPath, apiKey) => {
         try {
             const { pullRequestTitle, pullRequestDescription } = input;
-            const branchName = await (0, git_1.getCurrentBranchName)();
-            const repoUrl = (0, child_process_1.execSync)("git config --get remote.origin.url")
+            const branchName = await (0, git_1.getCurrentBranchName)(repoPath);
+            const repoUrl = (0, child_process_1.execSync)("git config --get remote.origin.url", {
+                cwd: repoPath,
+            })
                 .toString()
                 .trim();
             const { owner, repo } = parseGitHubUrl(repoUrl);
@@ -70,8 +72,11 @@ Don't ask the user for this information, just come up with it yourself.
                     head: `${owner}:${branchName}`,
                     state: "open",
                 },
+                apiKey,
             }));
-            (0, child_process_1.execSync)(`git push origin ${branchName} --set-upstream`);
+            (0, child_process_1.execSync)(`git push origin ${branchName} --set-upstream`, {
+                cwd: repoPath,
+            });
             const existingPR = existingPRs?.find((pr) => pr.head.ref === branchName);
             if (existingPR) {
                 // Append the new description to the existing PR description
@@ -82,6 +87,7 @@ Don't ask the user for this information, just come up with it yourself.
                     body: {
                         body: updatedDescription,
                     },
+                    apiKey,
                 });
                 return {
                     isError: false,
@@ -98,6 +104,7 @@ Don't ask the user for this information, just come up with it yourself.
                     base: "main",
                     body: initialDescription,
                 },
+                apiKey,
             }));
             return {
                 isError: false,

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,wBAAwB,CAAC;AAenD,eAAO,MAAM,wBAAwB,EAAE,IAgFtC,CAAC"}
+{"version":3,"file":"diagnosis-fetcher.d.ts","sourceRoot":"","sources":["../../src/tools/diagnosis-fetcher.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAenD,eAAO,MAAM,wBAAwB,EAAE,IAqFtC,CAAC"}

package/dist/tools/diagnosis-fetcher.js

@@ -19,7 +19,7 @@ exports.fetchDiagnosisReportTool = {
         description: "Fetch details about a test case diagnosis using its URL or slug",
         parameters: DiagnosisSchema,
     },
-    execute: async (input) => {
+    execute: async (input, repoPath, apiKey) => {
         const { diagnosisUrl } = input;
         // Extract the slug from the URL - it's the part after the last '--'
         const slug = diagnosisUrl.split("--").pop();
@@ -33,6 +33,7 @@ exports.fetchDiagnosisReportTool = {
         try {
             data = await (0, utils_1.makeDashboardRequest)({
                 path: `/api/diagnosis/${slug}/detailed`,
+                apiKey,
             });
         }
         catch (error) {
@@ -43,8 +44,8 @@ exports.fetchDiagnosisReportTool = {
         }
         const { test_case, diagnosis } = data.data;
         const project = diagnosis?.test_project || "unknown";
-        const sourceContext = await promises_1.default.readFile(path_1.default.join("tests", test_case.file_path), "utf-8");
-        const repoName = path_1.default.basename(process.cwd());
+        const sourceContext = await promises_1.default.readFile(path_1.default.join(repoPath, "tests", test_case.file_path), "utf-8");
+        const repoName = path_1.default.basename(repoPath);
         const cleanErrorStack = diagnosis?.failed_run_metadata?.stack?.replace(`"/runner/_work/${repoName}/${repoName}/source-repo/"`, "");
         // Format the response as markdown
         const markdownResponse = `

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,wBAAwB,CAAC;AAyDnD,eAAO,MAAM,kBAAkB,EAAE,IAwEhC,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,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AA6DnD,eAAO,MAAM,kBAAkB,EAAE,IA6EhC,CAAC;AAGF,eAAO,MAAM,gBAAgB,EAAE,IAAI,EAAyB,CAAC"}

package/dist/tools/environment-crud.js

@@ -5,10 +5,11 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.environmentTools = exports.getEnvironmentTool = void 0;
 const fs_1 = __importDefault(require("fs"));
+const path_1 = __importDefault(require("path"));
 const zod_1 = require("zod");
 const utils_1 = require("./utils");
-const getProjectRepoName = () => {
-    const packageJson = fs_1.default.readFileSync("package.json", "utf8");
+const getProjectRepoName = (repoPath) => {
+    const packageJson = fs_1.default.readFileSync(path_1.default.join(repoPath, "package.json"), "utf8");
     if (!packageJson) {
         throw new Error("Could not find or read package.json file");
     }
@@ -30,11 +31,11 @@ exports.getEnvironmentTool = {
         description: "Fetch details of an existing environment",
         parameters: GetEnvironmentSchema,
     },
-    execute: async (input) => {
+    execute: async (input, repoPath, apiKey) => {
         // Get project repo name
         let projectRepoName;
         try {
-            projectRepoName = getProjectRepoName();
+            projectRepoName = getProjectRepoName(repoPath);
         }
         catch (error) {
             return {
@@ -52,6 +53,7 @@ exports.getEnvironmentTool = {
             response = await (0, utils_1.makeDashboardRequest)({
                 path: `/api/environments?${queryParams.toString()}`,
                 method: "GET",
+                apiKey,
             });
         }
         catch (error) {

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,wBAAwB,CAAC;AA2H/D,eAAO,MAAM,QAAQ,EAAE,IAmBtB,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/grep/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAc,MAAM,wBAAwB,CAAC;AA6H/D,eAAO,MAAM,QAAQ,EAAE,IAoBtB,CAAC"}

package/dist/tools/grep/index.js

@@ -23,9 +23,9 @@ const GrepInputSchema = zod_1.z.object({
         .optional()
         .describe("File pattern to search in (e.g., '*.ts' for TypeScript files)"),
 });
-async function usingSystemGrep(input) {
+async function usingSystemGrep(input, repoPath) {
     try {
-        const dir = input.directory || process.cwd();
+        const dir = path_1.default.join(repoPath, input.directory || "");
         // Create exclude pattern for grep
         const excludePatterns = repo_tree_1.DEFAULT_EXCLUDE.map((pattern) => typeof pattern === "string" ? pattern : pattern.source)
             .map((pattern) => `--exclude-dir="${pattern}"`)
@@ -71,9 +71,9 @@ async function usingSystemGrep(input) {
         };
     }
 }
-async function usingRipgrep(input) {
+async function usingRipgrep(input, repoPath) {
     try {
-        const dir = path_1.default.join(process.cwd(), input.directory || "");
+        const dir = path_1.default.join(repoPath, input.directory || "");
         const escapedPattern = input.pattern
             .replace(/\s+/g, "\\ ")
             .replace(/([[\]()])/g, "\\$1");
@@ -86,7 +86,7 @@ async function usingRipgrep(input) {
             // Can add submatches and offset info to the summary if needed
             return {
                 lines: result.lines.text,
-                path: path_1.default.relative(process.cwd(), result.path.text),
+                path: path_1.default.relative(repoPath, result.path.text),
                 // line number is 1-indexed
                 line_number: result.line_number,
             };
@@ -98,7 +98,7 @@ ${result.path}:${result.line_number}
 ${result.lines}\`\`\`
 `;
         });
-        const relDir = path_1.default.relative(process.cwd(), dir);
+        const relDir = path_1.default.relative(repoPath, dir);
         const header = `Found ${resultsSummary.length} results for "${input.pattern}" in "${relDir}".
 All paths are relative to the current working directory.`;
         return {
@@ -123,13 +123,13 @@ If ripgrep is not available, it will fall back to using system grep.
 Search is case insensitive and regex patterns are not supported.`,
         parameters: GrepInputSchema,
     },
-    execute: async (input) => {
+    execute: async (input, repoPath) => {
         if ((0, ripgrep_1.isRgAvailable)()) {
-            return usingRipgrep(input);
+            return usingRipgrep(input, repoPath);
         }
         else {
             console.warn("ripgrep is not available, falling back to system grep.");
-            return usingSystemGrep(input);
+            return usingSystemGrep(input, repoPath);
         }
     },
 };

package/dist/tools/str_replace_editor.d.ts

@@ -17,7 +17,7 @@ export declare function cleanupBackupFiles(repoDir: string): number;
  * Our implementation of Claude's built-in text editor tool
  * https://docs.anthropic.com/en/docs/build-with-claude/tool-use/text-editor-tool
  */
-export declare function strReplaceEditorExecutor(input: StrReplaceInput): Promise<ToolResult>;
+export declare function strReplaceEditorExecutor(input: StrReplaceInput, repoPath: string): Promise<ToolResult>;
 export declare const textEditorTools: Tool[];
 export {};
 //# sourceMappingURL=str_replace_editor.d.ts.map

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,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AA2B1D,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;AAqED;;;GAGG;AACH,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,CAwC1D;AAMD;;;GAGG;AACH,wBAAsB,wBAAwB,CAC5C,KAAK,EAAE,eAAe,GACrB,OAAO,CAAC,UAAU,CAAC,CAiNrB;AAiGD,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,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AA2B1D,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;AAqED;;;GAGG;AACH,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,CAwC1D;AAMD;;;GAGG;AACH,wBAAsB,wBAAwB,CAC5C,KAAK,EAAE,eAAe,EACtB,QAAQ,EAAE,MAAM,GACf,OAAO,CAAC,UAAU,CAAC,CAkNrB;AAuHD,eAAO,MAAM,eAAe,EAAE,IAAI,EAKjC,CAAC"}

package/dist/tools/str_replace_editor.js

@@ -139,24 +139,25 @@ function escapeRegExp(text) {
  * Our implementation of Claude's built-in text editor tool
  * https://docs.anthropic.com/en/docs/build-with-claude/tool-use/text-editor-tool
  */
-async function strReplaceEditorExecutor(input) {
-    const repoDir = process.cwd();
+async function strReplaceEditorExecutor(input, repoPath) {
+    const repoDir = repoPath;
     const { path: filePath } = input;
+    const absoluteFilePath = path_1.default.join(repoDir, filePath);
     try {
         let content;
         let lines;
         let newContent;
         switch (input.command) {
             case "view":
-                if (!fs_1.default.existsSync(filePath)) {
+                if (!fs_1.default.existsSync(absoluteFilePath)) {
                     return {
                         result: "Error: File not found",
                         isError: true,
                     };
                 }
                 // Handle directory view
-                if (fs_1.default.statSync(filePath).isDirectory()) {
-                    const files = fs_1.default.readdirSync(filePath);
+                if (fs_1.default.statSync(absoluteFilePath).isDirectory()) {
+                    const files = fs_1.default.readdirSync(absoluteFilePath);
                     return {
                         result: files.join("\n"),
                         isError: false,
@@ -165,7 +166,7 @@ async function strReplaceEditorExecutor(input) {
                 else {
                     // Check if file is binary, which is not supported
                     const { isBinary } = await import("istextorbinary");
-                    const binary = isBinary(filePath);
+                    const binary = isBinary(absoluteFilePath);
                     if (binary) {
                         return {
                             result: "Error: File is binary, which is not supported",
@@ -174,7 +175,7 @@ async function strReplaceEditorExecutor(input) {
                     }
                 }
                 // Handle file view
-                content = fs_1.default.readFileSync(filePath, "utf8");
+                content = fs_1.default.readFileSync(absoluteFilePath, "utf8");
                 lines = content.split("\n");
                 if (input.view_range) {
                     const [start, end] = input.view_range;
@@ -201,12 +202,12 @@ async function strReplaceEditorExecutor(input) {
                 if (filePath.endsWith("spec.ts") && !filePath.startsWith("tests/")) {
                     throw new Error("Creating spec.ts files is not allowed outside tests/ directory");
                 }
-                const parentDir = path_1.default.dirname(filePath);
+                const parentDir = path_1.default.dirname(absoluteFilePath);
                 if (parentDir !== "." && !fs_1.default.existsSync(parentDir)) {
                     // Ensure parent directory exists
                     fs_1.default.mkdirSync(parentDir, { recursive: true });
                 }
-                fs_1.default.writeFileSync(filePath, input.file_text);
+                fs_1.default.writeFileSync(absoluteFilePath, input.file_text);
                 let createTypescriptResult = await (0, web_1.runTypescriptCompiler)(repoDir);
                 if (!createTypescriptResult.success) {
                     return {
@@ -226,8 +227,8 @@ async function strReplaceEditorExecutor(input) {
                     // "" is valid as new_str, so we check for nullish -- not falsy
                     throw new Error("new_str is required for str_replace command");
                 }
-                createBackup(filePath);
-                content = fs_1.default.readFileSync(filePath, "utf8");
+                createBackup(absoluteFilePath);
+                content = fs_1.default.readFileSync(absoluteFilePath, "utf8");
                 // Normalize newlines in both the content and search string
                 const normalizedContent = content.replace(/\r\n/g, "\n");
                 const normalizedOldStr = input.old_str
@@ -259,7 +260,7 @@ async function strReplaceEditorExecutor(input) {
                         };
                     }
                     newContent = normalizedContent.replace(normalizedOldStr, input.new_str);
-                    fs_1.default.writeFileSync(filePath, newContent);
+                    fs_1.default.writeFileSync(absoluteFilePath, newContent);
                     let strReplaceTypescriptResult = await (0, web_1.runTypescriptCompiler)(repoDir);
                     if (!strReplaceTypescriptResult.success) {
                         return {
@@ -278,8 +279,8 @@ async function strReplaceEditorExecutor(input) {
                 if (input.insert_line === undefined || !input.new_str) {
                     throw new Error("insert_line and new_str are required for insert command");
                 }
-                createBackup(filePath);
-                content = fs_1.default.readFileSync(filePath, "utf8");
+                createBackup(absoluteFilePath);
+                content = fs_1.default.readFileSync(absoluteFilePath, "utf8");
                 lines = content.split("\n");
                 if (input.insert_line < 1) {
                     throw new Error("insert_line must be greater than or equal to 1 (line numbers are 1-indexed).");
@@ -288,7 +289,7 @@ async function strReplaceEditorExecutor(input) {
                     throw new Error(`The file at ${filePath} has only ${lines.length} lines, so insert_line must be less than or equal to ${lines.length + 1}. At the maximum value of ${lines.length + 1}, you can insert at the end of the file.`);
                 }
                 lines.splice(input.insert_line - 1, 0, input.new_str);
-                fs_1.default.writeFileSync(filePath, lines.join("\n"));
+                fs_1.default.writeFileSync(absoluteFilePath, lines.join("\n"));
                 let insertTypescriptResult = await (0, web_1.runTypescriptCompiler)(repoDir);
                 if (!insertTypescriptResult.success) {
                     return {
@@ -303,8 +304,8 @@ async function strReplaceEditorExecutor(input) {
                     };
                 }
             case "undo_edit":
-                if (hasBackup(filePath)) {
-                    restoreBackup(filePath);
+                if (hasBackup(absoluteFilePath)) {
+                    restoreBackup(absoluteFilePath);
                     return {
                         result: `Successfully restored ${filePath} from backup`,
                         isError: false,
@@ -342,11 +343,11 @@ 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) => {
+    execute: async (input, repoPath) => {
         return strReplaceEditorExecutor({
             command: "view",
             path: input.path,
-        });
+        }, repoPath);
     },
 };
 const fileCreateTool = {
@@ -358,12 +359,12 @@ const fileCreateTool = {
             file_text: zod_1.z.string().describe("The contents of the new file."),
         }),
     },
-    execute: async (input) => {
+    execute: async (input, repoPath) => {
         return strReplaceEditorExecutor({
             command: "create",
             path: input.path,
             file_text: input.file_text,
-        });
+        }, repoPath);
     },
 };
 const stringReplaceTool = {
@@ -377,13 +378,13 @@ 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) => {
+    execute: async (input, repoPath) => {
         return strReplaceEditorExecutor({
             command: "str_replace",
             path: input.path,
             old_str: input.old_str,
             new_str: input.new_str,
-        });
+        }, repoPath);
     },
 };
 const stringInsertTool = {
@@ -396,17 +397,19 @@ const stringInsertTool = {
                 .number()
                 .int()
                 .min(1)
-                .describe("The line number after which to insert the text (1 for beginning of file)."),
+                .describe(`The line number on which to insert the text (1 for beginning of file).
+To insert a string at the beginning of the file, you should use insert_line = 1.
+To insert a string at the end of the file, you should use insert_line = (total lines + 1).`),
             new_str: zod_1.z.string().describe("The string to insert."),
         }),
     },
-    execute: async (input) => {
+    execute: async (input, repoPath) => {
         return strReplaceEditorExecutor({
             command: "insert",
             path: input.path,
             insert_line: input.insert_line,
             new_str: input.new_str,
-        });
+        }, repoPath);
     },
 };
 exports.textEditorTools = [

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,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AA4FnD,eAAO,MAAM,4BAA4B,EAAE,IAoH1C,CAAC"}
+{"version":3,"file":"test-gen-browser.d.ts","sourceRoot":"","sources":["../../src/tools/test-gen-browser.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AA2FnD,eAAO,MAAM,4BAA4B,EAAE,IA8G1C,CAAC"}

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

@@ -12,7 +12,6 @@ const utils_1 = require("../agent/browsing/utils");
 const pw_pause_1 = require("../agent/cua/pw-codegen/pw-pause");
 const web_1 = require("../bin/utils/platform/web");
 const scenarios_1 = require("../bin/utils/scenarios");
-const slug_1 = require("../utils/slug");
 const BrowserAgentSchema = zod_1.z.object({
     testName: zod_1.z.string().describe("The name of the test to create or modify"),
     testSuites: zod_1.z
@@ -153,9 +152,8 @@ exports.generateTestWithBrowserAgent = {
         await (0, pw_pause_1.revertToOriginalPwCode)(repoDir);
         await promises_1.default.writeFile(fileName, fileBackup, "utf-8");
         const { isError, error, actionsSummary, artifacts } = toolResult;
-        let collectedArtifacts;
         if (artifacts) {
-            collectedArtifacts = await collectArtifacts?.(artifacts, repoDir, (0, slug_1.slugify)(testName));
+            void collectArtifacts?.(artifacts);
         }
         if (!isError) {
             return {
@@ -165,7 +163,6 @@ and the generated Playwright code:
 
 ${actionsSummary}
 `,
-                artifacts: collectedArtifacts,
             };
         }
         else {

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

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/test-run-fetcher/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAcnD,wBAAgB,0BAA0B,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAOnE;AAED,eAAO,MAAM,sBAAsB,EAAE,IAwHpC,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/test-run-fetcher/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAcnD,wBAAgB,0BAA0B,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAOnE;AAED,eAAO,MAAM,sBAAsB,EAAE,IA6HpC,CAAC"}

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

@@ -24,7 +24,7 @@ exports.fetchTestRunReportTool = {
         description: "Fetch details about a test run using its URL",
         parameters: TestRunSchema,
     },
-    execute: async (input) => {
+    execute: async (input, repoPath, apiKey) => {
         const { testRunUrl } = input;
         // Remove query parameters if they exist
         const urlWithoutParams = testRunUrl.split("?")[0] || testRunUrl;
@@ -42,6 +42,7 @@ exports.fetchTestRunReportTool = {
         try {
             data = await (0, utils_1.makeDashboardRequest)({
                 path: `/api/test-runs/${runId}?repo_name=${repoName}`,
+                apiKey,
             });
         }
         catch (error) {

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":"AACA,OAAO,KAAK,EAAE,IAAI,EAAc,MAAM,wBAAwB,CAAC;AA4C/D,eAAO,MAAM,WAAW,EAAE,IAsEzB,CAAC"}
+{"version":3,"file":"test-run.d.ts","sourceRoot":"","sources":["../../src/tools/test-run.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,IAAI,EAAc,MAAM,wBAAwB,CAAC;AA4C/D,eAAO,MAAM,WAAW,EAAE,IAiEzB,CAAC"}

package/dist/tools/test-run.js

@@ -50,7 +50,7 @@ exports.runTestTool = {
             console.warn("R2 credentials not found: report artifacts will not be uploaded");
         }
         const { testName, suites, fileName, project } = input;
-        const repoDir = process.cwd();
+        const repoDir = repoPath;
         try {
             const result = await (0, test_run_1.runSingleTest)({
                 testName,
@@ -61,7 +61,7 @@ exports.runTestTool = {
                 repoDir,
             });
             const attachments = (0, utils_1.extractAttachmentsFromPlaywrightJSONReport)(result.summaryJson);
-            const artifacts = await collectArtifacts?.(attachments, repoDir, Date.now().toString());
+            void collectArtifacts?.(attachments);
             return {
                 isError: false,
                 result: buildResult({
@@ -69,7 +69,6 @@ exports.runTestTool = {
                     summaryJson: result.summaryJson,
                     reportUrl: reportUrl,
                 }),
-                artifacts,
             };
         }
         catch (error) {

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

@@ -1,13 +1,15 @@
 import { JSONReport as PlaywrightJSONReport } from "@playwright/test/reporter";
-export declare function makeDashboardRequest<T>({ path, method, body, }: {
+export declare function makeDashboardRequest<T>({ path, method, body, apiKey, }: {
     path: string;
     method?: string;
     body?: any;
+    apiKey: string;
 }): Promise<T>;
-export declare function callGitHubProxy({ method, url, body, }: {
+export declare function callGitHubProxy({ method, url, body, apiKey, }: {
     method: string;
     url: string;
     body?: any;
+    apiKey: string;
 }): Promise<unknown>;
 type AttachmentInfo = {
     name: string;

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

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/utils/index.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,UAAU,IAAI,oBAAoB,EAEnC,MAAM,2BAA2B,CAAC;AAEnC,wBAAsB,oBAAoB,CAAC,CAAC,EAAE,EAC5C,IAAI,EACJ,MAAc,EACd,IAAI,GACL,EAAE;IACD,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,GAAG,CAAC;CACZ,GAAG,OAAO,CAAC,CAAC,CAAC,CAuBb;AAED,wBAAsB,eAAe,CAAC,EACpC,MAAM,EACN,GAAG,EACH,IAAI,GACL,EAAE;IACD,MAAM,EAAE,MAAM,CAAC;IACf,GAAG,EAAE,MAAM,CAAC;IACZ,IAAI,CAAC,EAAE,GAAG,CAAC;CACZ,oBAWA;AAED,KAAK,cAAc,GAAG;IACpB,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;CACrB,CAAC;AAEF;;;;GAIG;AACH,wBAAgB,0CAA0C,CACxD,MAAM,EAAE,oBAAoB,GAC3B,cAAc,EAAE,CA4ClB"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/utils/index.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,UAAU,IAAI,oBAAoB,EAEnC,MAAM,2BAA2B,CAAC;AAEnC,wBAAsB,oBAAoB,CAAC,CAAC,EAAE,EAC5C,IAAI,EACJ,MAAc,EACd,IAAI,EACJ,MAAM,GACP,EAAE;IACD,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,GAAG,CAAC;IACX,MAAM,EAAE,MAAM,CAAC;CAChB,GAAG,OAAO,CAAC,CAAC,CAAC,CAmBb;AAED,wBAAsB,eAAe,CAAC,EACpC,MAAM,EACN,GAAG,EACH,IAAI,EACJ,MAAM,GACP,EAAE;IACD,MAAM,EAAE,MAAM,CAAC;IACf,GAAG,EAAE,MAAM,CAAC;IACZ,IAAI,CAAC,EAAE,GAAG,CAAC;IACX,MAAM,EAAE,MAAM,CAAC;CAChB,oBAYA;AAED,KAAK,cAAc,GAAG;IACpB,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;CACrB,CAAC;AAEF;;;;GAIG;AACH,wBAAgB,0CAA0C,CACxD,MAAM,EAAE,oBAAoB,GAC3B,cAAc,EAAE,CA4ClB"}

package/dist/tools/utils/index.js

@@ -3,14 +3,10 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.makeDashboardRequest = makeDashboardRequest;
 exports.callGitHubProxy = callGitHubProxy;
 exports.extractAttachmentsFromPlaywrightJSONReport = extractAttachmentsFromPlaywrightJSONReport;
-async function makeDashboardRequest({ path, method = "GET", body, }) {
-    const API_KEY = process.env.EMPIRICALRUN_API_KEY;
-    if (!API_KEY) {
-        throw new Error("EMPIRICALRUN_API_KEY is not found");
-    }
+async function makeDashboardRequest({ path, method = "GET", body, apiKey, }) {
     const requestHeaders = {
         "Content-Type": "application/json",
-        Authorization: `Bearer ${API_KEY}`,
+        Authorization: `Bearer ${apiKey}`,
         "User-Agent": "empiricalrun/test-gen",
     };
     const baseUrl = process.env.DASHBOARD_DOMAIN || "https://dash.empirical.run";
@@ -25,7 +21,7 @@ async function makeDashboardRequest({ path, method = "GET", body, }) {
     }
     return await response.json();
 }
-async function callGitHubProxy({ method, url, body, }) {
+async function callGitHubProxy({ method, url, body, apiKey, }) {
     const githubApiPath = url.replace("https://api.github.com", "");
     return makeDashboardRequest({
         path: "/api/github/proxy",
@@ -35,6 +31,7 @@ async function callGitHubProxy({ method, url, body, }) {
             url: githubApiPath,
             body,
         },
+        apiKey,
     });
 }
 /**

package/dist/utils/checkpoint.d.ts

@@ -1,3 +1,7 @@
 import { PendingToolCall } from "@empiricalrun/llm/chat";
-export declare function createCheckpoint(toolCalls: PendingToolCall[], branchName: string): Promise<void>;
+export declare function createCheckpoint({ toolCalls, branchName, repoPath, }: {
+    toolCalls: PendingToolCall[];
+    branchName: string;
+    repoPath: string;
+}): Promise<void>;
 //# sourceMappingURL=checkpoint.d.ts.map

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

@@ -1 +1 @@
-{"version":3,"file":"checkpoint.d.ts","sourceRoot":"","sources":["../../src/utils/checkpoint.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAC;AAIzD,wBAAsB,gBAAgB,CACpC,SAAS,EAAE,eAAe,EAAE,EAC5B,UAAU,EAAE,MAAM,iBAqBnB"}
+{"version":3,"file":"checkpoint.d.ts","sourceRoot":"","sources":["../../src/utils/checkpoint.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAC;AAIzD,wBAAsB,gBAAgB,CAAC,EACrC,SAAS,EACT,UAAU,EACV,QAAQ,GACT,EAAE;IACD,SAAS,EAAE,eAAe,EAAE,CAAC;IAC7B,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;CAClB,iBAyBA"}

package/dist/utils/checkpoint.js

@@ -2,8 +2,8 @@
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.createCheckpoint = createCheckpoint;
 const git_1 = require("./git");
-async function createCheckpoint(toolCalls, branchName) {
-    const filesChanged = await (0, git_1.getFilesChanged)();
+async function createCheckpoint({ toolCalls, branchName, repoPath, }) {
+    const filesChanged = await (0, git_1.getFilesChanged)(repoPath);
     const toolsWithUpdatedFiles = toolCalls
         .filter((tc) => "path" in tc.input &&
         tc.input.path &&
@@ -15,6 +15,11 @@ async function createCheckpoint(toolCalls, branchName) {
     const filesToCommit = toolsWithUpdatedFiles.map((tool) => tool.path);
     if (toolsWithUpdatedFiles.length > 0) {
         let commitMessage = `${toolsWithUpdatedFiles.map((tool) => `${tool.name} on ${tool.path}`).join(", ")}`;
-        await (0, git_1.commitFilesAndPushBranch)(commitMessage, branchName, filesToCommit);
+        await (0, git_1.commitFilesAndPushBranch)({
+            commitMessage,
+            branchName,
+            files: filesToCommit,
+            repoPath,
+        });
     }
 }

package/dist/utils/git.d.ts

@@ -1,8 +1,13 @@
-export declare function getGitDiff(filepath: string): string;
-export declare function checkoutBranch(branchName: string): Promise<void>;
-export declare function commitAsBotUser(commitMessage: string): Promise<void>;
-export declare function getCurrentBranchName(): Promise<string>;
-export declare function pullBranch(branchName: string): Promise<void>;
-export declare function commitFilesAndPushBranch(commitMessage: string, branchName: string, files: string[]): Promise<void>;
-export declare function getFilesChanged(): Promise<string[]>;
+export declare function getGitDiff(filepath: string, cwd: string): string;
+export declare function checkoutBranch(branchName: string, cwd: string): Promise<void>;
+export declare function commitAsBotUser(commitMessage: string, cwd: string): Promise<void>;
+export declare function getCurrentBranchName(repoPath: string): Promise<string>;
+export declare function pullBranch(branchName: string, cwd: string): Promise<void>;
+export declare function commitFilesAndPushBranch({ commitMessage, branchName, files, repoPath, }: {
+    commitMessage: string;
+    branchName: string;
+    files: string[];
+    repoPath: string;
+}): Promise<void>;
+export declare function getFilesChanged(cwd: string): Promise<string[]>;
 //# sourceMappingURL=git.d.ts.map

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

@@ -1 +1 @@
-{"version":3,"file":"git.d.ts","sourceRoot":"","sources":["../../src/utils/git.ts"],"names":[],"mappings":"AAKA,wBAAgB,UAAU,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAKnD;AAED,wBAAsB,cAAc,CAAC,UAAU,EAAE,MAAM,iBAQtD;AAED,wBAAsB,eAAe,CAAC,aAAa,EAAE,MAAM,iBAS1D;AAED,wBAAsB,oBAAoB,oBAGzC;AAED,wBAAsB,UAAU,CAAC,UAAU,EAAE,MAAM,iBAElD;AAED,wBAAsB,wBAAwB,CAC5C,aAAa,EAAE,MAAM,EACrB,UAAU,EAAE,MAAM,EAClB,KAAK,EAAE,MAAM,EAAE,iBAKhB;AAED,wBAAsB,eAAe,sBAUpC"}
+{"version":3,"file":"git.d.ts","sourceRoot":"","sources":["../../src/utils/git.ts"],"names":[],"mappings":"AAKA,wBAAgB,UAAU,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,EAAE,MAAM,GAAG,MAAM,CAMhE;AAED,wBAAsB,cAAc,CAAC,UAAU,EAAE,MAAM,EAAE,GAAG,EAAE,MAAM,iBAQnE;AAED,wBAAsB,eAAe,CAAC,aAAa,EAAE,MAAM,EAAE,GAAG,EAAE,MAAM,iBAavE;AAED,wBAAsB,oBAAoB,CAAC,QAAQ,EAAE,MAAM,mBAO1D;AAED,wBAAsB,UAAU,CAAC,UAAU,EAAE,MAAM,EAAE,GAAG,EAAE,MAAM,iBAE/D;AAED,wBAAsB,wBAAwB,CAAC,EAC7C,aAAa,EACb,UAAU,EACV,KAAK,EACL,QAAQ,GACT,EAAE;IACD,aAAa,EAAE,MAAM,CAAC;IACtB,UAAU,EAAE,MAAM,CAAC;IACnB,KAAK,EAAE,MAAM,EAAE,CAAC;IAChB,QAAQ,EAAE,MAAM,CAAC;CAClB,iBAIA;AAED,wBAAsB,eAAe,CAAC,GAAG,EAAE,MAAM,qBAUhD"}

package/dist/utils/git.js

@@ -10,44 +10,54 @@ exports.getFilesChanged = getFilesChanged;
 const child_process_1 = require("child_process");
 const GIT_USER_NAME = "empiricalrun[bot]";
 const GIT_USER_EMAIL = "180257021+empiricalrun[bot]@users.noreply.github.com";
-function getGitDiff(filepath) {
+function getGitDiff(filepath, cwd) {
     const diff = (0, child_process_1.execSync)(`git diff ${filepath}`, {
         encoding: "utf-8",
+        cwd,
     });
     return diff;
 }
-async function checkoutBranch(branchName) {
+async function checkoutBranch(branchName, cwd) {
     // TODO: This assumes repoDir is process.cwd()
     try {
-        (0, child_process_1.execSync)(`git checkout ${branchName}`);
+        (0, child_process_1.execSync)(`git checkout ${branchName}`, { cwd });
     }
     catch (e) {
         // If branch doesn't exist, create it
-        (0, child_process_1.execSync)(`git checkout -b ${branchName}`);
+        (0, child_process_1.execSync)(`git checkout -b ${branchName}`, { cwd });
     }
 }
-async function commitAsBotUser(commitMessage) {
-    const status = (0, child_process_1.execSync)("git status --porcelain", { encoding: "utf-8" });
+async function commitAsBotUser(commitMessage, cwd) {
+    const status = (0, child_process_1.execSync)("git status --porcelain", {
+        encoding: "utf-8",
+        cwd,
+    });
     if (!status) {
         return; // Nothing to commit
     }
     const commitMessageWithSkipCi = `${commitMessage} [skip ci]`;
-    (0, child_process_1.execSync)(`git -c user.name="${GIT_USER_NAME}" -c user.email="${GIT_USER_EMAIL}" commit -m "${commitMessageWithSkipCi}"`);
+    (0, child_process_1.execSync)(`git -c user.name="${GIT_USER_NAME}" -c user.email="${GIT_USER_EMAIL}" commit -m "${commitMessageWithSkipCi}"`, { cwd });
 }
-async function getCurrentBranchName() {
-    const branchName = (0, child_process_1.execSync)("git branch --show-current").toString().trim();
+async function getCurrentBranchName(repoPath) {
+    const branchName = (0, child_process_1.execSync)("git branch --show-current", {
+        cwd: repoPath,
+    })
+        .toString()
+        .trim();
     return branchName;
 }
-async function pullBranch(branchName) {
-    (0, child_process_1.execSync)(`git pull origin ${branchName}`);
+async function pullBranch(branchName, cwd) {
+    (0, child_process_1.execSync)(`git pull origin ${branchName}`, { cwd });
 }
-async function commitFilesAndPushBranch(commitMessage, branchName, files) {
-    (0, child_process_1.execSync)(`git add ${files.join(" ")}`);
-    await commitAsBotUser(commitMessage);
-    (0, child_process_1.execSync)(`git push origin ${branchName}`);
+async function commitFilesAndPushBranch({ commitMessage, branchName, files, repoPath, }) {
+    (0, child_process_1.execSync)(`git add ${files.join(" ")}`, { cwd: repoPath });
+    await commitAsBotUser(commitMessage, repoPath);
+    (0, child_process_1.execSync)(`git push origin ${branchName}`, { cwd: repoPath });
 }
-async function getFilesChanged() {
-    const output = (0, child_process_1.execSync)("git status --porcelain --untracked-files").toString();
+async function getFilesChanged(cwd) {
+    const output = (0, child_process_1.execSync)("git status --porcelain --untracked-files", {
+        cwd,
+    }).toString();
     let filesChanged = output
         .split("\n")
         .map((line) => line.trim().split(" ").pop())

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@empiricalrun/test-gen",
-  "version": "0.62.0",
+  "version": "0.63.0",
   "publishConfig": {
     "registry": "https://registry.npmjs.org/",
     "access": "public"
@@ -62,9 +62,9 @@
     "tsx": "^4.16.2",
     "typescript": "^5.3.3",
     "zod": "^3.23.8",
-    "@empiricalrun/llm": "^0.17.0",
+    "@empiricalrun/llm": "^0.17.1",
     "@empiricalrun/r2-uploader": "^0.3.9",
-    "@empiricalrun/test-run": "^0.9.3"
+    "@empiricalrun/test-run": "^0.9.4"
   },
   "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": "0.3.0"
+    "@empiricalrun/shared-types": "0.4.0"
   },
   "scripts": {
     "dev": "tspc --build --watch",