@empiricalrun/test-gen 0.54.1 → 0.56.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (36) hide show
  1. package/CHANGELOG.md +31 -0
  2. package/dist/agent/browsing/utils.d.ts.map +1 -1
  3. package/dist/agent/browsing/utils.js +20 -2
  4. package/dist/agent/chat/agent-loop.d.ts +3 -1
  5. package/dist/agent/chat/agent-loop.d.ts.map +1 -1
  6. package/dist/agent/chat/agent-loop.js +10 -58
  7. package/dist/agent/chat/exports.d.ts +8 -0
  8. package/dist/agent/chat/exports.d.ts.map +1 -0
  9. package/dist/agent/chat/exports.js +7 -0
  10. package/dist/agent/chat/index.d.ts.map +1 -1
  11. package/dist/agent/chat/index.js +5 -0
  12. package/dist/agent/chat/model.d.ts.map +1 -1
  13. package/dist/agent/chat/model.js +3 -0
  14. package/dist/agent/chat/types.d.ts +1 -1
  15. package/dist/agent/chat/types.d.ts.map +1 -1
  16. package/dist/bin/index.js +7 -4
  17. package/dist/bin/utils/index.d.ts +1 -1
  18. package/dist/bin/utils/index.d.ts.map +1 -1
  19. package/dist/test-build/index.d.ts +2 -9
  20. package/dist/test-build/index.d.ts.map +1 -1
  21. package/dist/test-build/index.js +16 -15
  22. package/dist/tool-call-service/index.d.ts +18 -0
  23. package/dist/tool-call-service/index.d.ts.map +1 -0
  24. package/dist/tool-call-service/index.js +101 -0
  25. package/dist/tools/download-build.d.ts +3 -0
  26. package/dist/tools/download-build.d.ts.map +1 -0
  27. package/dist/tools/download-build.js +39 -0
  28. package/dist/tools/environment-crud.d.ts +4 -0
  29. package/dist/tools/environment-crud.d.ts.map +1 -0
  30. package/dist/tools/environment-crud.js +98 -0
  31. package/dist/tools/str_replace_editor.d.ts +1 -1
  32. package/dist/tools/str_replace_editor.d.ts.map +1 -1
  33. package/dist/tools/str_replace_editor.js +20 -25
  34. package/dist/tools/utils/index.d.ts.map +1 -1
  35. package/dist/tools/utils/index.js +6 -2
  36. package/package.json +12 -3
package/CHANGELOG.md CHANGED
@@ -1,5 +1,36 @@
1
1
  # @empiricalrun/test-gen
2
2
 
3
+ ## 0.56.0
4
+
5
+ ### Minor Changes
6
+
7
+ - b073084: feat: new APIs and refactor to introduce tool execute service
8
+
9
+ ### Patch Changes
10
+
11
+ - 93d7a0b: feat: add tool call for downloading builds
12
+ - a58ac3f: feat: add dashboard agent integration and optimize tsx handling
13
+ - Updated dependencies [87af227]
14
+ - Updated dependencies [3831109]
15
+ - Updated dependencies [a58ac3f]
16
+ - Updated dependencies [f77e33d]
17
+ - @empiricalrun/llm@0.15.1
18
+ - @empiricalrun/test-run@0.8.1
19
+
20
+ ## 0.55.0
21
+
22
+ ### Minor Changes
23
+
24
+ - 508565d: feat: add support for openai chat model
25
+ - 8da022c: feat: add environment fetching tool
26
+
27
+ ### Patch Changes
28
+
29
+ - 99e4e6e: refactor: decoupled agentloop and toolcall
30
+ - Updated dependencies [cc4cb5e]
31
+ - Updated dependencies [508565d]
32
+ - @empiricalrun/llm@0.15.0
33
+
3
34
  ## 0.54.1
4
35
 
5
36
  ### Patch Changes
@@ -1 +1 @@
1
- {"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/utils.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,KAAK,EAAe,QAAQ,EAAE,MAAM,4BAA4B,CAAC;AAIxE,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAClC,OAAO,EAAE,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AAsBvD,wBAAgB,QAAQ,CAAC,GAAG,EAAE,GAAG,GAAG,GAAG,IAAI,MAAM,CAKhD;AAED,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,MAAM,EAAE,UAIvD;AAiFD,wBAAsB,yBAAyB,CAAC,EAC9C,YAAY,EACZ,YAAY,EACZ,cAAc,GACf,EAAE;IACD,YAAY,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,MAAM,EAAE,CAAC;CAC1B,iBA2BA;AAED,wBAAsB,cAAc,CAAC,EACnC,YAAY,EACZ,cAAc,EACd,QAAQ,GACT,EAAE;IACD,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,MAAM,EAAE,CAAC;IACzB,QAAQ,EAAE,MAAM,CAAC;CAClB,iBAoBA;AAED,wBAAsB,yBAAyB,CAAC,EAC9C,QAAQ,EACR,QAAQ,EACR,KAAK,GACN,EAAE;IACD,QAAQ,EAAE,QAAQ,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,GAAG,OAAO,CAAC,MAAM,CAAC,CAyDlB;AAyBD,wBAAsB,wBAAwB,CAAC,IAAI,EAAE,IAAI,iBA2HxD;AAED;;;GAGG;AACH,wBAAsB,oBAAoB,CACxC,OAAO,EAAE,MAAM,GACd,OAAO,CAAC,oBAAoB,CAAC,CAM/B;AAWD,wBAAsB,oBAAoB,CACxC,gBAAgB,EAAE,oBAAoB,GACrC,OAAO,CAAC,MAAM,EAAE,CAAC,CAQnB;AAED;;;;;GAKG;AACH,wBAAsB,iBAAiB,CACrC,YAAY,EAAE,MAAM,EACpB,gBAAgB,EAAE,oBAAoB,EACtC,gBAAgB,GAAE,MAAM,EAAU,GACjC,OAAO,CAAC,MAAM,CAAC,CA+CjB;AAED,qBAAa,eAAe;IACd,OAAO,CAAC,SAAS;gBAAT,SAAS,EAAE,MAAM;IACrC,OAAO,CAAC,aAAa,CAAqB;YAE5B,mBAAmB;YAUnB,gBAAgB;IAsBjB,OAAO;IAoBb,SAAS;CAKjB"}
1
+ {"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/utils.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,KAAK,EAAe,QAAQ,EAAE,MAAM,4BAA4B,CAAC;AAIxE,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAClC,OAAO,EAAE,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AAwBvD,wBAAgB,QAAQ,CAAC,GAAG,EAAE,GAAG,GAAG,GAAG,IAAI,MAAM,CAKhD;AAED,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,MAAM,EAAE,UAIvD;AAiFD,wBAAsB,yBAAyB,CAAC,EAC9C,YAAY,EACZ,YAAY,EACZ,cAAc,GACf,EAAE;IACD,YAAY,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,MAAM,EAAE,CAAC;CAC1B,iBA2BA;AAED,wBAAsB,cAAc,CAAC,EACnC,YAAY,EACZ,cAAc,EACd,QAAQ,GACT,EAAE;IACD,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,MAAM,EAAE,CAAC;IACzB,QAAQ,EAAE,MAAM,CAAC;CAClB,iBAoBA;AAED,wBAAsB,yBAAyB,CAAC,EAC9C,QAAQ,EACR,QAAQ,EACR,KAAK,GACN,EAAE;IACD,QAAQ,EAAE,QAAQ,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,GAAG,OAAO,CAAC,MAAM,CAAC,CAyDlB;AAyBD,wBAAsB,wBAAwB,CAAC,IAAI,EAAE,IAAI,iBA2HxD;AAED;;;GAGG;AACH,wBAAsB,oBAAoB,CACxC,OAAO,EAAE,MAAM,GACd,OAAO,CAAC,oBAAoB,CAAC,CA0B/B;AAWD,wBAAsB,oBAAoB,CACxC,gBAAgB,EAAE,oBAAoB,GACrC,OAAO,CAAC,MAAM,EAAE,CAAC,CAQnB;AAED;;;;;GAKG;AACH,wBAAsB,iBAAiB,CACrC,YAAY,EAAE,MAAM,EACpB,gBAAgB,EAAE,oBAAoB,EACtC,gBAAgB,GAAE,MAAM,EAAU,GACjC,OAAO,CAAC,MAAM,CAAC,CA+CjB;AAED,qBAAa,eAAe;IACd,OAAO,CAAC,SAAS;gBAAT,SAAS,EAAE,MAAM;IACrC,OAAO,CAAC,aAAa,CAAqB;YAE5B,mBAAmB;YAUnB,gBAAgB;IAsBjB,OAAO;IAoBb,SAAS;CAKjB"}
@@ -8,7 +8,8 @@ const fs_extra_1 = __importDefault(require("fs-extra"));
8
8
  const minimatch_1 = require("minimatch");
9
9
  const path_1 = __importDefault(require("path"));
10
10
  const ts_morph_1 = require("ts-morph");
11
- const api_1 = __importDefault(require("tsx/cjs/api"));
11
+ // For TypeScript type safety
12
+ let tsxImport = null;
12
13
  const logger_1 = require("../../bin/logger");
13
14
  const context_1 = require("../../bin/utils/context");
14
15
  const fs_1 = require("../../bin/utils/fs");
@@ -295,8 +296,25 @@ exports.injectPwLocatorGenerator = injectPwLocatorGenerator;
295
296
  * @return {*} {Promise<PlaywrightTestConfig>}
296
297
  */
297
298
  async function readPlaywrightConfig(repoDir) {
299
+ if (typeof window !== "undefined") {
300
+ throw new Error("readPlaywrightConfig cannot be used in browser environments");
301
+ }
302
+ else {
303
+ // Only initialize on server side
304
+ // This will only execute on the server
305
+ await import("tsx/cjs/api")
306
+ .then((module) => {
307
+ tsxImport = module;
308
+ })
309
+ .catch((err) => {
310
+ console.error("Failed to import tsx:", err);
311
+ });
312
+ }
313
+ if (!tsxImport) {
314
+ throw new Error("tsx module not available");
315
+ }
298
316
  const [lastDir] = repoDir.split("/").reverse();
299
- const playwrightConfig = (await api_1.default.require("./playwright.config.ts", `${repoDir}/${lastDir}`)).default;
317
+ const playwrightConfig = (await tsxImport.require("./playwright.config.ts", `${repoDir}/${lastDir}`)).default;
300
318
  return playwrightConfig;
301
319
  }
302
320
  exports.readPlaywrightConfig = readPlaywrightConfig;
@@ -1,9 +1,11 @@
1
1
  import { TraceClient } from "@empiricalrun/llm";
2
2
  import { IChatModel } from "@empiricalrun/llm/chat";
3
+ import { ToolCallService } from "../../tool-call-service";
3
4
  import { ReporterFunction, SupportedChatModels } from "./types";
4
- export declare function chatAgentLoop({ chatModel, selectedModel, reporter, trace, }: {
5
+ export declare function chatAgentLoop({ chatModel, selectedModel, reporter, trace, toolCallService, }: {
5
6
  chatModel: IChatModel<any>;
6
7
  selectedModel: SupportedChatModels;
8
+ toolCallService: ToolCallService;
7
9
  reporter: ReporterFunction;
8
10
  trace?: TraceClient;
9
11
  }): Promise<void>;
@@ -1 +1 @@
1
- {"version":3,"file":"agent-loop.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/agent-loop.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,EACL,UAAU,EAIX,MAAM,wBAAwB,CAAC;AAiBhC,OAAO,EAAE,gBAAgB,EAAE,mBAAmB,EAAE,MAAM,SAAS,CAAC;AAyChE,wBAAsB,aAAa,CAAC,EAClC,SAAS,EACT,aAAa,EACb,QAAQ,EACR,KAAK,GACN,EAAE;IACD,SAAS,EAAE,UAAU,CAAC,GAAG,CAAC,CAAC;IAC3B,aAAa,EAAE,mBAAmB,CAAC;IACnC,QAAQ,EAAE,gBAAgB,CAAC;IAC3B,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,iBAiDA"}
1
+ {"version":3,"file":"agent-loop.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/agent-loop.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,EACL,UAAU,EAIX,MAAM,wBAAwB,CAAC;AAGhC,OAAO,EAAE,eAAe,EAAE,MAAM,yBAAyB,CAAC;AAG1D,OAAO,EAAE,gBAAgB,EAAE,mBAAmB,EAAE,MAAM,SAAS,CAAC;AAgBhE,wBAAsB,aAAa,CAAC,EAClC,SAAS,EACT,aAAa,EACb,QAAQ,EACR,KAAK,EACL,eAAe,GAChB,EAAE;IACD,SAAS,EAAE,UAAU,CAAC,GAAG,CAAC,CAAC;IAC3B,aAAa,EAAE,mBAAmB,CAAC;IACnC,eAAe,EAAE,eAAe,CAAC;IACjC,QAAQ,EAAE,gBAAgB,CAAC;IAC3B,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,iBAoCA"}
@@ -3,80 +3,33 @@ Object.defineProperty(exports, "__esModule", { value: true });
3
3
  exports.chatAgentLoop = void 0;
4
4
  const chat_1 = require("@empiricalrun/llm/chat");
5
5
  const picocolors_1 = require("picocolors");
6
- const web_1 = require("../../bin/utils/platform/web");
7
- const commit_and_create_pr_1 = require("../../tools/commit-and-create-pr");
8
- const diagnosis_fetcher_1 = require("../../tools/diagnosis-fetcher");
9
- const grep_1 = require("../../tools/grep");
10
- const str_replace_editor_1 = require("../../tools/str_replace_editor");
11
- const test_gen_browser_1 = require("../../tools/test-gen-browser");
12
- const test_run_1 = require("../../tools/test-run");
13
- const test_run_fetcher_1 = require("../../tools/test-run-fetcher");
14
6
  const prompt_1 = require("./prompt");
15
7
  const state_1 = require("./state");
16
- function getTools(selectedModel) {
17
- let tools = [
18
- grep_1.grepTool,
19
- test_run_1.runTestTool,
20
- test_run_fetcher_1.fetchTestRunReportTool,
21
- diagnosis_fetcher_1.fetchDiagnosisReportTool,
22
- test_gen_browser_1.generateTestWithBrowserAgent,
23
- commit_and_create_pr_1.commitAndPushChangesTool,
24
- ];
25
- if (selectedModel.startsWith("gemini")) {
26
- // Claude will have its own built-in text editor tools
27
- str_replace_editor_1.textEditorTools.forEach((tool) => {
28
- const originalExecute = tool.execute;
29
- tool.execute = (input) => originalExecute(input, web_1.validateTypescript);
30
- });
31
- tools.push(...str_replace_editor_1.textEditorTools);
32
- }
33
- const toolExecutors = {
34
- ...Object.fromEntries(tools.map((tool) => [tool.schema.name, tool.execute])),
35
- };
36
- if (selectedModel.startsWith("claude")) {
37
- toolExecutors.str_replace_editor = (input) => (0, str_replace_editor_1.strReplaceEditorExecutor)(input, web_1.validateTypescript);
38
- }
39
- return { tools, toolExecutors };
40
- }
41
8
  function getModelName(model) {
42
9
  if (model.startsWith("claude"))
43
10
  return "Claude";
44
11
  if (model.startsWith("gemini"))
45
12
  return "Gemini";
13
+ if (model.startsWith("o4"))
14
+ return "o4";
46
15
  return "AI";
47
16
  }
48
17
  const log = (...args) => {
49
18
  console.log((0, picocolors_1.gray)(args.join(" ")));
50
19
  };
51
- async function chatAgentLoop({ chatModel, selectedModel, reporter, trace, }) {
20
+ const isRemote = process.env.TOOL_EXECUTION_IS_REMOTE === "true" || false;
21
+ async function chatAgentLoop({ chatModel, selectedModel, reporter, trace, toolCallService, }) {
52
22
  const systemPrompt = await (0, prompt_1.buildSystemPrompt)();
53
23
  trace?.update({ input: { systemPrompt } });
54
- const { tools, toolExecutors } = getTools(selectedModel);
24
+ const { tools } = await toolCallService.getTools();
55
25
  while (!chatModel.askUserForInput) {
56
26
  const toolCalls = chatModel.getPendingToolCalls();
27
+ console.log("toolCalls", toolCalls);
57
28
  if (toolCalls.length) {
58
- const toolResults = [];
59
- for (const call of toolCalls) {
60
- const args = JSON.stringify(call.input);
61
- log(`Executing tool ${call.name} with args: ${args}`);
62
- const toolExecutor = toolExecutors[call.name];
63
- let callResponse;
64
- if (!toolExecutor) {
65
- callResponse = {
66
- isError: true,
67
- result: `Invalid function/tool call: ${call.name} not found`,
68
- };
69
- }
70
- else {
71
- callResponse = await toolExecutor(call.input);
72
- if (callResponse.isError) {
73
- log(`Tool ${call.name} failed: ${callResponse.result}`);
74
- }
75
- else {
76
- log(`Tool ${call.name} completed`);
77
- }
78
- }
79
- toolResults.push(callResponse);
29
+ const toolResults = await toolCallService.execute(toolCalls, isRemote);
30
+ if (isRemote) {
31
+ log(`Tool call remote execution in progress`);
32
+ break;
80
33
  }
81
34
  chatModel.pushToolResultsMessage(toolCalls, toolResults);
82
35
  }
@@ -94,6 +47,5 @@ async function chatAgentLoop({ chatModel, selectedModel, reporter, trace, }) {
94
47
  const latest = chatModel.getHumanReadableLatestMessage();
95
48
  await reporter((0, state_1.chatStateFromModel)(chatModel, selectedModel), latest);
96
49
  }
97
- (0, str_replace_editor_1.cleanupBackupFiles)(process.cwd());
98
50
  }
99
51
  exports.chatAgentLoop = chatAgentLoop;
@@ -0,0 +1,8 @@
1
+ import { IChatModel } from "@empiricalrun/llm/chat";
2
+ import { chatAgentLoop } from "./agent-loop";
3
+ import { createChatModel } from "./model";
4
+ import { ChatStateOnDisk } from "./state";
5
+ import { ReporterFunction, SupportedChatModels } from "./types";
6
+ export { chatAgentLoop, createChatModel };
7
+ export type { ChatStateOnDisk, IChatModel, ReporterFunction, SupportedChatModels, };
8
+ //# sourceMappingURL=exports.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"exports.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/exports.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAEpD,OAAO,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAC7C,OAAO,EAAE,eAAe,EAAE,MAAM,SAAS,CAAC;AAC1C,OAAO,EAAE,eAAe,EAAE,MAAM,SAAS,CAAC;AAC1C,OAAO,EAAE,gBAAgB,EAAE,mBAAmB,EAAE,MAAM,SAAS,CAAC;AAEhE,OAAO,EAAE,aAAa,EAAE,eAAe,EAAE,CAAC;AAE1C,YAAY,EACV,eAAe,EACf,UAAU,EACV,gBAAgB,EAChB,mBAAmB,GACpB,CAAC"}
@@ -0,0 +1,7 @@
1
+ "use strict";
2
+ Object.defineProperty(exports, "__esModule", { value: true });
3
+ exports.createChatModel = exports.chatAgentLoop = void 0;
4
+ const agent_loop_1 = require("./agent-loop");
5
+ Object.defineProperty(exports, "chatAgentLoop", { enumerable: true, get: function () { return agent_loop_1.chatAgentLoop; } });
6
+ const model_1 = require("./model");
7
+ Object.defineProperty(exports, "createChatModel", { enumerable: true, get: function () { return model_1.createChatModel; } });
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/index.ts"],"names":[],"mappings":"AAYA,OAAO,EAAoB,mBAAmB,EAAE,MAAM,SAAS,CAAC;AAiBhE,wBAAsB,kBAAkB,CAAC,EACvC,mBAAmB,EACnB,aAAa,EACb,oBAAoB,GACrB,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,mBAAmB,EAAE,OAAO,CAAC;IAC7B,oBAAoB,EAAE,MAAM,GAAG,SAAS,CAAC;CAC1C,iBAoFA;AAuBD,wBAAsB,wBAAwB,CAAC,EAC7C,aAAa,EACb,aAAa,GACd,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,aAAa,EAAE,MAAM,CAAC;CACvB,iBAyCA"}
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/index.ts"],"names":[],"mappings":"AAaA,OAAO,EAAoB,mBAAmB,EAAE,MAAM,SAAS,CAAC;AAiBhE,wBAAsB,kBAAkB,CAAC,EACvC,mBAAmB,EACnB,aAAa,EACb,oBAAoB,GACrB,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,mBAAmB,EAAE,OAAO,CAAC;IAC7B,oBAAoB,EAAE,MAAM,GAAG,SAAS,CAAC;CAC1C,iBAsFA;AAuBD,wBAAsB,wBAAwB,CAAC,EAC7C,aAAa,EACb,aAAa,GACd,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,aAAa,EAAE,MAAM,CAAC;CACvB,iBA2CA"}
@@ -4,6 +4,7 @@ exports.runChatAgentForDashboard = exports.runChatAgentForCLI = void 0;
4
4
  const llm_1 = require("@empiricalrun/llm");
5
5
  const picocolors_1 = require("picocolors");
6
6
  const human_in_the_loop_1 = require("../../human-in-the-loop");
7
+ const tool_call_service_1 = require("../../tool-call-service");
7
8
  const git_1 = require("../../utils/git");
8
9
  const agent_loop_1 = require("./agent-loop");
9
10
  const model_1 = require("./model");
@@ -83,11 +84,13 @@ async function runChatAgentForCLI({ useDiskForChatState, selectedModel, initialP
83
84
  }
84
85
  else {
85
86
  // TODO: Should we pass a loader function? That would allow us to show a spinner
87
+ const toolCallService = new tool_call_service_1.ToolCallService(null, selectedModel);
86
88
  await (0, agent_loop_1.chatAgentLoop)({
87
89
  chatModel,
88
90
  selectedModel,
89
91
  reporter: reporterFunc,
90
92
  trace,
93
+ toolCallService,
91
94
  });
92
95
  }
93
96
  }
@@ -125,6 +128,7 @@ async function runChatAgentForDashboard({ chatSessionId, selectedModel, }) {
125
128
  chatSessionId,
126
129
  },
127
130
  });
131
+ const toolCallService = new tool_call_service_1.ToolCallService(chatSessionId, selectedModel);
128
132
  await (0, git_1.checkoutBranch)(branchName);
129
133
  let chatModel = (0, model_1.createChatModel)(chatState.messages, selectedModel);
130
134
  let reporterFunc = async (chatState, latest) => {
@@ -147,6 +151,7 @@ async function runChatAgentForDashboard({ chatSessionId, selectedModel, }) {
147
151
  selectedModel,
148
152
  reporter: reporterFunc,
149
153
  trace,
154
+ toolCallService,
150
155
  });
151
156
  await (0, git_1.commitLocalAndPushBranchToRemote)(branchName);
152
157
  }
@@ -1 +1 @@
1
- {"version":3,"file":"model.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/model.ts"],"names":[],"mappings":"AAAA,OAAO,EAGL,UAAU,EACX,MAAM,wBAAwB,CAAC;AAEhC,OAAO,EAAE,mBAAmB,EAAE,MAAM,SAAS,CAAC;AAE9C,wBAAgB,eAAe,CAC7B,QAAQ,EAAE,GAAG,EAAE,EACf,aAAa,EAAE,mBAAmB,GACjC,UAAU,CAAC,GAAG,CAAC,CAQjB"}
1
+ {"version":3,"file":"model.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/model.ts"],"names":[],"mappings":"AAAA,OAAO,EAGL,UAAU,EAEX,MAAM,wBAAwB,CAAC;AAEhC,OAAO,EAAE,mBAAmB,EAAE,MAAM,SAAS,CAAC;AAE9C,wBAAgB,eAAe,CAC7B,QAAQ,EAAE,GAAG,EAAE,EACf,aAAa,EAAE,mBAAmB,GACjC,UAAU,CAAC,GAAG,CAAC,CAWjB"}
@@ -9,6 +9,9 @@ function createChatModel(messages, selectedModel) {
9
9
  if (selectedModel.startsWith("gemini")) {
10
10
  return new chat_1.GeminiChatModel(messages);
11
11
  }
12
+ if (selectedModel.startsWith("o4")) {
13
+ return new chat_1.OpenAIChatModel(messages);
14
+ }
12
15
  throw new Error(`Unsupported model: ${selectedModel}`);
13
16
  }
14
17
  exports.createChatModel = createChatModel;
@@ -1,5 +1,5 @@
1
1
  import { ChatStateOnDisk } from "./state";
2
- export type SupportedChatModels = "claude-3-7-sonnet-20250219" | "claude-3-5-sonnet-20241022" | "gemini-2.5-pro-preview-03-25";
2
+ export type SupportedChatModels = "claude-3-7-sonnet-20250219" | "claude-3-5-sonnet-20241022" | "gemini-2.5-pro-preview-03-25" | "o4-mini-2025-04-16";
3
3
  type LatestMessage = {
4
4
  role: string;
5
5
  textMessage: string;
@@ -1 +1 @@
1
- {"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/types.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,SAAS,CAAC;AAE1C,MAAM,MAAM,mBAAmB,GAC3B,4BAA4B,GAC5B,4BAA4B,GAC5B,8BAA8B,CAAC;AAEnC,KAAK,aAAa,GAAG;IACnB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;CACrB,CAAC;AAEF,MAAM,MAAM,gBAAgB,GAAG,CAC7B,KAAK,EAAE,eAAe,CAAC,GAAG,CAAC,EAC3B,0BAA0B,EAAE,aAAa,GAAG,SAAS,KAClD,OAAO,CAAC,IAAI,CAAC,CAAC"}
1
+ {"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/types.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,SAAS,CAAC;AAE1C,MAAM,MAAM,mBAAmB,GAC3B,4BAA4B,GAC5B,4BAA4B,GAC5B,8BAA8B,GAC9B,oBAAoB,CAAC;AAEzB,KAAK,aAAa,GAAG;IACnB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;CACrB,CAAC;AAEF,MAAM,MAAM,gBAAgB,GAAG,CAC7B,KAAK,EAAE,eAAe,CAAC,GAAG,CAAC,EAC3B,0BAA0B,EAAE,aAAa,GAAG,SAAS,KAClD,OAAO,CAAC,IAAI,CAAC,CAAC"}
package/dist/bin/index.js CHANGED
@@ -38,10 +38,11 @@ function setupProcessListeners(cleanup) {
38
38
  async function runChatAgent({ modelInput, chatSessionId, useDiskForChatState, initialPromptPath, }) {
39
39
  const MODEL_MAPPING = {
40
40
  "claude-3-7": "claude-3-7-sonnet-20250219",
41
- "3-7": "claude-3-7-sonnet-20250219",
42
41
  "claude-3-5": "claude-3-5-sonnet-20241022",
43
- "3-5": "claude-3-5-sonnet-20241022",
44
42
  "gemini-2.5-pro-preview-03-25": "gemini-2.5-pro-preview-03-25",
43
+ "gemini-2.5-pro": "gemini-2.5-pro-preview-03-25",
44
+ "o4-mini": "o4-mini-2025-04-16",
45
+ "o4-mini-2025-04-16": "o4-mini-2025-04-16",
45
46
  };
46
47
  if (modelInput && !MODEL_MAPPING[modelInput]) {
47
48
  throw new Error(`Invalid chat model: ${modelInput}`);
@@ -232,8 +233,10 @@ async function main() {
232
233
  generationId: testGenConfig.options?.metadata.generationId,
233
234
  projectRepoName: testGenConfig.options?.metadata.projectRepoName,
234
235
  });
235
- // Download the build if repo has a download script
236
- await (0, test_build_1.downloadBuild)(testGenConfig.build || {});
236
+ if (testGenConfig.build?.url) {
237
+ // Download the build if repo has a download script
238
+ await (0, test_build_1.downloadBuild)(testGenConfig.build.url);
239
+ }
237
240
  if (completedOptions.useChat) {
238
241
  await runChatAgent({
239
242
  chatSessionId: completedOptions.chatSessionId,
@@ -8,7 +8,7 @@ export interface CliOptions {
8
8
  useDiskForChatState?: boolean;
9
9
  initialPrompt?: string;
10
10
  chatSessionId?: string;
11
- chatModel?: "claude-3-7" | "3-7" | "claude-3-5" | "3-5" | "claude-3-7-sonnet-20250219" | "claude-3-5-sonnet-20241022" | "gemini-2.5-pro-preview-03-25";
11
+ chatModel?: "claude-3-7" | "claude-3-5" | "claude-3-7-sonnet-20250219" | "claude-3-5-sonnet-20241022" | "gemini-2.5-pro" | "gemini-2.5-pro-preview-03-25" | "o4-mini" | "o4-mini-2025-04-16";
12
12
  }
13
13
  export declare function validateAndCompleteCliOptions(options: CliOptions): Promise<CliOptions>;
14
14
  export declare function printBanner(): void;
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/bin/utils/index.ts"],"names":[],"mappings":"AAEA,MAAM,WAAW,UAAU;IACzB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,SAAS,CAAC,EACN,YAAY,GACZ,KAAK,GACL,YAAY,GACZ,KAAK,GACL,4BAA4B,GAC5B,4BAA4B,GAC5B,8BAA8B,CAAC;CACpC;AAQD,wBAAsB,6BAA6B,CACjD,OAAO,EAAE,UAAU,GAClB,OAAO,CAAC,UAAU,CAAC,CAyDrB;AAED,wBAAgB,WAAW,SAgC1B"}
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/bin/utils/index.ts"],"names":[],"mappings":"AAEA,MAAM,WAAW,UAAU;IACzB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,SAAS,CAAC,EACN,YAAY,GACZ,YAAY,GACZ,4BAA4B,GAC5B,4BAA4B,GAC5B,gBAAgB,GAChB,8BAA8B,GAC9B,SAAS,GACT,oBAAoB,CAAC;CAC1B;AAQD,wBAAsB,6BAA6B,CACjD,OAAO,EAAE,UAAU,GAClB,OAAO,CAAC,UAAU,CAAC,CAyDrB;AAED,wBAAgB,WAAW,SAgC1B"}
@@ -1,10 +1,3 @@
1
- import { Build } from "@empiricalrun/shared-types";
2
- /**
3
- * method to download the build from the URL provided in the build object
4
- * this is only used in cases like chrome extension as of yet.
5
- * @export
6
- * @param {Build} build
7
- * @return {*} {Promise<void>}
8
- */
9
- export declare function downloadBuild(build: Build): Promise<void>;
1
+ export declare function hasDownloadScript(): Promise<boolean>;
2
+ export declare function downloadBuild(buildUrl: string): Promise<void>;
10
3
  //# sourceMappingURL=index.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/test-build/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,MAAM,4BAA4B,CAAC;AAUnD;;;;;;GAMG;AACH,wBAAsB,aAAa,CAAC,KAAK,EAAE,KAAK,GAAG,OAAO,CAAC,IAAI,CAAC,CAY/D"}
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/test-build/index.ts"],"names":[],"mappings":"AAeA,wBAAsB,iBAAiB,IAAI,OAAO,CAAC,OAAO,CAAC,CAG1D;AAED,wBAAsB,aAAa,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAUnE"}
@@ -3,26 +3,27 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
3
3
  return (mod && mod.__esModule) ? mod : { "default": mod };
4
4
  };
5
5
  Object.defineProperty(exports, "__esModule", { value: true });
6
- exports.downloadBuild = void 0;
6
+ exports.downloadBuild = exports.hasDownloadScript = void 0;
7
7
  const fs_extra_1 = __importDefault(require("fs-extra"));
8
8
  const logger_1 = require("../bin/logger");
9
9
  const exec_1 = require("../utils/exec");
10
- /**
11
- * method to download the build from the URL provided in the build object
12
- * this is only used in cases like chrome extension as of yet.
13
- * @export
14
- * @param {Build} build
15
- * @return {*} {Promise<void>}
16
- */
17
- async function downloadBuild(build) {
18
- const logger = new logger_1.CustomLogger({ useReporter: false });
10
+ async function getPackageJSON() {
19
11
  const packageJSONPath = "package.json";
20
12
  const packageJsonStr = await fs_extra_1.default.readFile(packageJSONPath, "utf-8");
21
- const packageJSONData = JSON.parse(packageJsonStr);
22
- const buildDownloadScript = packageJSONData.scripts["download"];
23
- if (buildDownloadScript && build?.url) {
24
- logger.log(`Downloading build from ${build.url}`);
25
- await (0, exec_1.cmd)(`npm run download ${build.url}`.split(" "), {
13
+ return JSON.parse(packageJsonStr);
14
+ }
15
+ async function hasDownloadScript() {
16
+ const packageJSON = await getPackageJSON();
17
+ return !!packageJSON.scripts["download"];
18
+ }
19
+ exports.hasDownloadScript = hasDownloadScript;
20
+ async function downloadBuild(buildUrl) {
21
+ const logger = new logger_1.CustomLogger({ useReporter: false });
22
+ const packageJSON = await getPackageJSON();
23
+ const buildDownloadScript = packageJSON.scripts["download"];
24
+ if (buildDownloadScript && buildUrl) {
25
+ logger.log(`Downloading build from ${buildUrl}`);
26
+ await (0, exec_1.cmd)(`npm run download ${buildUrl}`.split(" "), {
26
27
  env: { ...Object(process.env) },
27
28
  });
28
29
  }
@@ -0,0 +1,18 @@
1
+ import { PendingToolCall, Tool, ToolResult } from "@empiricalrun/llm/chat";
2
+ import { SupportedChatModels } from "../agent/chat/types";
3
+ export type { SupportedChatModels };
4
+ type ToolExecutors = {
5
+ [key: string]: (input: any) => Promise<ToolResult>;
6
+ };
7
+ export declare class ToolCallService {
8
+ tools: Tool[];
9
+ toolExecutors: ToolExecutors;
10
+ chatSessionId: number | null;
11
+ selectedModel: SupportedChatModels;
12
+ constructor(chatSessionId: number | null, selectedModel: SupportedChatModels);
13
+ getTools(): Promise<{
14
+ tools: Tool[];
15
+ }>;
16
+ execute(toolCalls: PendingToolCall[], isRemote: boolean): Promise<ToolResult[]>;
17
+ }
18
+ //# sourceMappingURL=index.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/tool-call-service/index.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,eAAe,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAE3E,OAAO,EAAE,mBAAmB,EAAE,MAAM,qBAAqB,CAAC;AAa1D,YAAY,EAAE,mBAAmB,EAAE,CAAC;AAEpC,KAAK,aAAa,GAAG;IACnB,CAAC,GAAG,EAAE,MAAM,GAAG,CAAC,KAAK,EAAE,GAAG,KAAK,OAAO,CAAC,UAAU,CAAC,CAAC;CACpD,CAAC;AAyBF,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;gBAGjC,aAAa,EAAE,MAAM,GAAG,IAAI,EAC5B,aAAa,EAAE,mBAAmB;IAgB9B,QAAQ;;;IAaR,OAAO,CACX,SAAS,EAAE,eAAe,EAAE,EAC5B,QAAQ,EAAE,OAAO,GAChB,OAAO,CAAC,UAAU,EAAE,CAAC;CAqCzB"}
@@ -0,0 +1,101 @@
1
+ "use strict";
2
+ Object.defineProperty(exports, "__esModule", { value: true });
3
+ exports.ToolCallService = void 0;
4
+ const client_sqs_1 = require("@aws-sdk/client-sqs");
5
+ const commit_and_create_pr_1 = require("../tools/commit-and-create-pr");
6
+ const diagnosis_fetcher_1 = require("../tools/diagnosis-fetcher");
7
+ const download_build_1 = require("../tools/download-build");
8
+ const environment_crud_1 = require("../tools/environment-crud");
9
+ const grep_1 = require("../tools/grep");
10
+ const str_replace_editor_1 = require("../tools/str_replace_editor");
11
+ const test_gen_browser_1 = require("../tools/test-gen-browser");
12
+ const test_run_1 = require("../tools/test-run");
13
+ const test_run_fetcher_1 = require("../tools/test-run-fetcher");
14
+ async function sendToolRequestToRemoteQueue(payload) {
15
+ const sqs = new client_sqs_1.SQSClient({
16
+ region: process.env.AWS_REGION,
17
+ credentials: {
18
+ accessKeyId: process.env.AWS_ACCESS_KEY_ID,
19
+ secretAccessKey: process.env.AWS_SECRET_ACCESS_KEY,
20
+ },
21
+ });
22
+ const queueUrl = process.env.TOOL_REQUEST_QUEUE_URL;
23
+ await sqs.send(new client_sqs_1.SendMessageCommand({
24
+ QueueUrl: queueUrl,
25
+ MessageBody: JSON.stringify(payload),
26
+ MessageGroupId: payload.requestId,
27
+ MessageDeduplicationId: payload.requestId, // unique id for the tool request
28
+ }));
29
+ }
30
+ class ToolCallService {
31
+ tools = [];
32
+ toolExecutors = {};
33
+ chatSessionId;
34
+ selectedModel;
35
+ constructor(chatSessionId, selectedModel) {
36
+ this.chatSessionId = chatSessionId;
37
+ this.selectedModel = selectedModel;
38
+ this.tools = [
39
+ grep_1.grepTool,
40
+ test_run_1.runTestTool,
41
+ test_run_fetcher_1.fetchTestRunReportTool,
42
+ diagnosis_fetcher_1.fetchDiagnosisReportTool,
43
+ test_gen_browser_1.generateTestWithBrowserAgent,
44
+ commit_and_create_pr_1.commitAndPushChangesTool,
45
+ environment_crud_1.getEnvironmentTool,
46
+ download_build_1.downloadBuildTool,
47
+ ];
48
+ }
49
+ async getTools() {
50
+ if (!this.selectedModel.startsWith("claude")) {
51
+ this.tools.push(...str_replace_editor_1.textEditorTools);
52
+ }
53
+ this.tools.forEach((tool) => {
54
+ this.toolExecutors[tool.schema.name] = tool.execute;
55
+ });
56
+ if (this.selectedModel.startsWith("claude")) {
57
+ this.toolExecutors["str_replace_editor"] = str_replace_editor_1.strReplaceEditorExecutor;
58
+ }
59
+ return { tools: this.tools };
60
+ }
61
+ async execute(toolCalls, isRemote) {
62
+ if (isRemote && this.chatSessionId) {
63
+ console.log("Executing tool remotely", toolCalls);
64
+ await sendToolRequestToRemoteQueue({
65
+ toolCalls,
66
+ requestId: crypto.randomUUID(),
67
+ chatSessionId: this.chatSessionId,
68
+ selectedModel: this.selectedModel,
69
+ });
70
+ return toolCalls.map(() => ({
71
+ isError: false,
72
+ result: `Tool request sent to remote queue to execute.`,
73
+ }));
74
+ }
75
+ else {
76
+ const toolResults = [];
77
+ for (const toolCall of toolCalls) {
78
+ const toolExecutor = this.toolExecutors[toolCall.name];
79
+ if (!toolExecutor) {
80
+ toolResults.push({
81
+ isError: true,
82
+ result: `Invalid function/tool call: invalid_tool_call not found`,
83
+ });
84
+ continue;
85
+ }
86
+ try {
87
+ const result = await toolExecutor(toolCall.input);
88
+ toolResults.push(result);
89
+ }
90
+ catch (error) {
91
+ toolResults.push({
92
+ isError: true,
93
+ result: error instanceof Error ? error.message : String(error),
94
+ });
95
+ }
96
+ }
97
+ return toolResults;
98
+ }
99
+ }
100
+ }
101
+ exports.ToolCallService = ToolCallService;
@@ -0,0 +1,3 @@
1
+ import type { Tool } from "@empiricalrun/llm/chat";
2
+ export declare const downloadBuildTool: Tool;
3
+ //# sourceMappingURL=download-build.d.ts.map
@@ -0,0 +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,wBAAwB,CAAC;AAKnD,eAAO,MAAM,iBAAiB,EAAE,IAkC/B,CAAC"}
@@ -0,0 +1,39 @@
1
+ "use strict";
2
+ Object.defineProperty(exports, "__esModule", { value: true });
3
+ exports.downloadBuildTool = void 0;
4
+ const zod_1 = require("zod");
5
+ const test_build_1 = require("../test-build");
6
+ exports.downloadBuildTool = {
7
+ schema: {
8
+ name: "downloadBuild",
9
+ description: `Download a build from a build URL. If you do not have
10
+ have a build URL, you can try getting the environment details with the getEnvironment tool.
11
+ Environment details will include the build URL.`,
12
+ parameters: zod_1.z.object({
13
+ buildUrl: zod_1.z.string().describe("The URL of the build to download"),
14
+ }),
15
+ },
16
+ execute: async (input) => {
17
+ if (!(await (0, test_build_1.hasDownloadScript)())) {
18
+ return {
19
+ isError: true,
20
+ result: `This repo does not have a download script in package.json.
21
+ You probably don't need to worry about this, since it means this repo does not have a build to download.`,
22
+ };
23
+ }
24
+ const { buildUrl } = input;
25
+ try {
26
+ await (0, test_build_1.downloadBuild)(buildUrl);
27
+ return {
28
+ isError: false,
29
+ result: "Build downloaded successfully",
30
+ };
31
+ }
32
+ catch (error) {
33
+ return {
34
+ isError: true,
35
+ result: `Failed to download build: ${error}`,
36
+ };
37
+ }
38
+ },
39
+ };
@@ -0,0 +1,4 @@
1
+ import type { Tool } from "@empiricalrun/llm/chat";
2
+ export declare const getEnvironmentTool: Tool;
3
+ export declare const environmentTools: Tool[];
4
+ //# sourceMappingURL=environment-crud.d.ts.map
@@ -0,0 +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"}
@@ -0,0 +1,98 @@
1
+ "use strict";
2
+ var __importDefault = (this && this.__importDefault) || function (mod) {
3
+ return (mod && mod.__esModule) ? mod : { "default": mod };
4
+ };
5
+ Object.defineProperty(exports, "__esModule", { value: true });
6
+ exports.environmentTools = exports.getEnvironmentTool = void 0;
7
+ const fs_1 = __importDefault(require("fs"));
8
+ const zod_1 = require("zod");
9
+ const utils_1 = require("./utils");
10
+ const getProjectRepoName = () => {
11
+ const packageJson = fs_1.default.readFileSync("package.json", "utf8");
12
+ if (!packageJson) {
13
+ throw new Error("Could not find or read package.json file");
14
+ }
15
+ const packageJsonData = JSON.parse(packageJson);
16
+ if (!packageJsonData.name) {
17
+ throw new Error("package.json does not contain a name field");
18
+ }
19
+ return packageJsonData.name;
20
+ };
21
+ const GetEnvironmentSchema = zod_1.z.object({
22
+ environment_slug: zod_1.z
23
+ .string()
24
+ .describe("The unique identifier (slug) for the environment you want to fetch details for. This is typically a URL-friendly version of the environment name."),
25
+ });
26
+ // Get Environment Tool
27
+ exports.getEnvironmentTool = {
28
+ schema: {
29
+ name: "getEnvironment",
30
+ description: "Fetch details of an existing environment",
31
+ parameters: GetEnvironmentSchema,
32
+ },
33
+ execute: async (input) => {
34
+ // Get project repo name
35
+ let projectRepoName;
36
+ try {
37
+ projectRepoName = getProjectRepoName();
38
+ }
39
+ catch (error) {
40
+ return {
41
+ isError: true,
42
+ result: `Failed to get project repository name from package.json: ${error instanceof Error ? error.message : String(error)}`,
43
+ };
44
+ }
45
+ // Make API request
46
+ let response;
47
+ try {
48
+ const queryParams = new URLSearchParams({
49
+ project_repo_name: projectRepoName,
50
+ environment_slug: input.environment_slug,
51
+ });
52
+ response = await (0, utils_1.makeDashboardRequest)({
53
+ path: `/api/environments?${queryParams.toString()}`,
54
+ method: "GET",
55
+ });
56
+ }
57
+ catch (error) {
58
+ return {
59
+ isError: true,
60
+ result: `Failed to make API request to fetch environment: ${error instanceof Error ? error.message : String(error)}`,
61
+ };
62
+ }
63
+ // Parse and validate response
64
+ try {
65
+ const data = response;
66
+ if (data.error) {
67
+ return {
68
+ isError: true,
69
+ result: `API returned error while fetching environment: ${data.error.message}`,
70
+ data: data.error,
71
+ };
72
+ }
73
+ if (!data.data?.environment) {
74
+ return {
75
+ isError: true,
76
+ result: `Environment "${input.environment_slug}" not found in project "${projectRepoName}"`,
77
+ data: data.data,
78
+ };
79
+ }
80
+ return {
81
+ result: JSON.stringify({
82
+ message: `Successfully fetched environment "${input.environment_slug}"`,
83
+ data: data.data,
84
+ }),
85
+ isError: false,
86
+ data: data.data,
87
+ };
88
+ }
89
+ catch (error) {
90
+ return {
91
+ isError: true,
92
+ result: `Failed to parse or validate environment response data: ${error instanceof Error ? error.message : String(error)}`,
93
+ };
94
+ }
95
+ },
96
+ };
97
+ // Export all environment tools as a single array
98
+ exports.environmentTools = [exports.getEnvironmentTool];
@@ -17,7 +17,7 @@ export declare function cleanupBackupFiles(repoDir: string): number;
17
17
  * Our implementation of Claude's built-in text editor tool
18
18
  * https://docs.anthropic.com/en/docs/build-with-claude/tool-use/text-editor-tool
19
19
  */
20
- export declare function strReplaceEditorExecutor(input: StrReplaceInput, typeChecker?: (filePath: string) => string[]): Promise<ToolResult>;
20
+ export declare function strReplaceEditorExecutor(input: StrReplaceInput): Promise<ToolResult>;
21
21
  export declare const textEditorTools: Tool[];
22
22
  export {};
23
23
  //# sourceMappingURL=str_replace_editor.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"str_replace_editor.d.ts","sourceRoot":"","sources":["../../src/tools/str_replace_editor.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAyB1D,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;AAED;;;GAGG;AACH,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,CAwC1D;AAMD;;;GAGG;AACH,wBAAsB,wBAAwB,CAC5C,KAAK,EAAE,eAAe,EACtB,WAAW,CAAC,EAAE,CAAC,QAAQ,EAAE,MAAM,KAAK,MAAM,EAAE,GAC3C,OAAO,CAAC,UAAU,CAAC,CA8IrB;AA6GD,eAAO,MAAM,eAAe,EAAE,IAAI,EAKjC,CAAC"}
1
+ {"version":3,"file":"str_replace_editor.d.ts","sourceRoot":"","sources":["../../src/tools/str_replace_editor.ts"],"names":[],"mappings":"AAAA,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;AAED;;;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,CAwIrB;AA+FD,eAAO,MAAM,eAAe,EAAE,IAAI,EAKjC,CAAC"}
@@ -7,6 +7,7 @@ exports.textEditorTools = exports.strReplaceEditorExecutor = exports.cleanupBack
7
7
  const fs_1 = __importDefault(require("fs"));
8
8
  const path_1 = __importDefault(require("path"));
9
9
  const zod_1 = require("zod");
10
+ const web_1 = require("../bin/utils/platform/web");
10
11
  function createBackup(filePath) {
11
12
  const backupPath = `${filePath}.bak`;
12
13
  if (fs_1.default.existsSync(filePath)) {
@@ -79,7 +80,7 @@ function escapeRegExp(text) {
79
80
  * Our implementation of Claude's built-in text editor tool
80
81
  * https://docs.anthropic.com/en/docs/build-with-claude/tool-use/text-editor-tool
81
82
  */
82
- async function strReplaceEditorExecutor(input, typeChecker) {
83
+ async function strReplaceEditorExecutor(input) {
83
84
  const { path: filePath } = input;
84
85
  try {
85
86
  let content;
@@ -156,25 +157,19 @@ async function strReplaceEditorExecutor(input, typeChecker) {
156
157
  }
157
158
  newContent = content.replace(input.old_str, input.new_str);
158
159
  fs_1.default.writeFileSync(filePath, newContent);
159
- if (typeChecker) {
160
- const errors = typeChecker(filePath);
161
- if (errors.length > 0) {
162
- return {
163
- result: `Edits to file ${filePath} have been applied. However, type checks are failing with errors:\n${errors.join("\n")}`,
164
- isError: true,
165
- };
166
- }
167
- else {
168
- return {
169
- result: `Edits to file ${filePath} have been applied. Type checks have also passed.`,
170
- isError: false,
171
- };
172
- }
160
+ const errors = (0, web_1.validateTypescript)(filePath);
161
+ if (errors.length > 0) {
162
+ return {
163
+ result: `Edits to file ${filePath} have been applied. However, type checks are failing with errors:\n${errors.join("\n")}`,
164
+ isError: true,
165
+ };
166
+ }
167
+ else {
168
+ return {
169
+ result: `Edits to file ${filePath} have been applied. Type checks have also passed.`,
170
+ isError: false,
171
+ };
173
172
  }
174
- return {
175
- result: `Edits to file ${filePath} have been applied.`,
176
- isError: false,
177
- };
178
173
  }
179
174
  case "insert":
180
175
  if (input.insert_line === undefined || !input.new_str) {
@@ -245,12 +240,12 @@ const fileCreateTool = {
245
240
  file_text: zod_1.z.string().describe("The contents of the new file."),
246
241
  }),
247
242
  },
248
- execute: async (input, typeChecker) => {
243
+ execute: async (input) => {
249
244
  return strReplaceEditorExecutor({
250
245
  command: "create",
251
246
  path: input.path,
252
247
  file_text: input.file_text,
253
- }, typeChecker);
248
+ });
254
249
  },
255
250
  };
256
251
  const stringReplaceTool = {
@@ -264,13 +259,13 @@ in the file. If old_str is not unique, the tool will return an error.`,
264
259
  new_str: zod_1.z.string().describe("The string to replace old_str with."),
265
260
  }),
266
261
  },
267
- execute: async (input, typeChecker) => {
262
+ execute: async (input) => {
268
263
  return strReplaceEditorExecutor({
269
264
  command: "str_replace",
270
265
  path: input.path,
271
266
  old_str: input.old_str,
272
267
  new_str: input.new_str,
273
- }, typeChecker);
268
+ });
274
269
  },
275
270
  };
276
271
  const stringInsertTool = {
@@ -287,13 +282,13 @@ const stringInsertTool = {
287
282
  new_str: zod_1.z.string().describe("The string to insert."),
288
283
  }),
289
284
  },
290
- execute: async (input, typeChecker) => {
285
+ execute: async (input) => {
291
286
  return strReplaceEditorExecutor({
292
287
  command: "insert",
293
288
  path: input.path,
294
289
  insert_line: input.insert_line,
295
290
  new_str: input.new_str,
296
- }, typeChecker);
291
+ });
297
292
  },
298
293
  };
299
294
  exports.textEditorTools = [
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/utils/index.ts"],"names":[],"mappings":"AAAA,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,CAmBb;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"}
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/utils/index.ts"],"names":[],"mappings":"AAAA,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"}
@@ -2,12 +2,16 @@
2
2
  Object.defineProperty(exports, "__esModule", { value: true });
3
3
  exports.callGitHubProxy = exports.makeDashboardRequest = void 0;
4
4
  async function makeDashboardRequest({ path, method = "GET", body, }) {
5
+ const API_KEY = process.env.EMPIRICALRUN_API_KEY;
6
+ if (!API_KEY) {
7
+ throw new Error("EMPIRICALRUN_API_KEY is not found");
8
+ }
5
9
  const requestHeaders = {
6
10
  "Content-Type": "application/json",
7
- Authorization: `Bearer ${process.env.EMPIRICALRUN_API_KEY}`,
11
+ Authorization: `Bearer ${API_KEY}`,
8
12
  "User-Agent": "empiricalrun/test-gen",
9
13
  };
10
- const baseUrl = "https://dash.empirical.run";
14
+ const baseUrl = process.env.DASHBOARD_DOMAIN || "https://dash.empirical.run";
11
15
  const response = await fetch(`${baseUrl}${path}`, {
12
16
  method,
13
17
  headers: requestHeaders,
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@empiricalrun/test-gen",
3
- "version": "0.54.1",
3
+ "version": "0.56.0",
4
4
  "publishConfig": {
5
5
  "registry": "https://registry.npmjs.org/",
6
6
  "access": "public"
@@ -18,6 +18,10 @@
18
18
  "types": "./dist/agent/chat/state.d.ts",
19
19
  "default": "./dist/agent/chat/state.js"
20
20
  },
21
+ "./chat": {
22
+ "types": "./dist/agent/chat/exports.d.ts",
23
+ "default": "./dist/agent/chat/exports.js"
24
+ },
21
25
  "./utils": {
22
26
  "types": "./dist/utils/index.d.ts",
23
27
  "default": "./dist/utils/index.js"
@@ -25,6 +29,10 @@
25
29
  ".": {
26
30
  "types": "./dist/index.d.ts",
27
31
  "default": "./dist/index.js"
32
+ },
33
+ "./tool-call-service": {
34
+ "types": "./dist/tool-call-service/index.d.ts",
35
+ "default": "./dist/tool-call-service/index.js"
28
36
  }
29
37
  },
30
38
  "repository": {
@@ -33,6 +41,7 @@
33
41
  },
34
42
  "author": "Empirical Team <hey@empirical.run>",
35
43
  "dependencies": {
44
+ "@aws-sdk/client-sqs": "^3.787.0",
36
45
  "@babel/parser": "^7.26.3",
37
46
  "@types/sanitize-html": "^2.11.0",
38
47
  "async-retry": "^1.3.3",
@@ -56,9 +65,9 @@
56
65
  "tsx": "^4.16.2",
57
66
  "typescript": "^5.3.3",
58
67
  "zod": "^3.23.8",
59
- "@empiricalrun/llm": "^0.14.8",
68
+ "@empiricalrun/llm": "^0.15.1",
60
69
  "@empiricalrun/r2-uploader": "^0.3.8",
61
- "@empiricalrun/test-run": "^0.8.0"
70
+ "@empiricalrun/test-run": "^0.8.1"
62
71
  },
63
72
  "devDependencies": {
64
73
  "@playwright/test": "1.47.1",