@empiricalrun/test-gen 0.78.7 → 0.79.1
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.
- package/CHANGELOG.md +20 -0
- package/dist/agent/chat/exports.d.ts +1 -1
- package/dist/agent/chat/exports.d.ts.map +1 -1
- package/dist/agent/chat/exports.js +2 -1
- package/dist/agent/code-review/index.d.ts +1 -0
- package/dist/agent/code-review/index.d.ts.map +1 -1
- package/dist/agent/code-review/index.js +3 -1
- package/dist/agent/cua/index.d.ts.map +1 -1
- package/dist/agent/cua/index.js +24 -7
- package/dist/agent/cua/model.d.ts +10 -3
- package/dist/agent/cua/model.d.ts.map +1 -1
- package/dist/agent/cua/model.js +9 -9
- package/dist/bin/utils/platform/web/index.js +2 -2
- package/dist/tools/test-run-fetcher/types.d.ts +0 -5
- package/dist/tools/test-run-fetcher/types.d.ts.map +1 -1
- package/package.json +4 -4
package/CHANGELOG.md
CHANGED
|
@@ -1,5 +1,25 @@
|
|
|
1
1
|
# @empiricalrun/test-gen
|
|
2
2
|
|
|
3
|
+
## 0.79.1
|
|
4
|
+
|
|
5
|
+
### Patch Changes
|
|
6
|
+
|
|
7
|
+
- 48126ea: feat: cost tracking for overlay dismissals
|
|
8
|
+
- Updated dependencies [48126ea]
|
|
9
|
+
- @empiricalrun/llm@0.25.2
|
|
10
|
+
- videostil@0.3.4
|
|
11
|
+
|
|
12
|
+
## 0.79.0
|
|
13
|
+
|
|
14
|
+
### Minor Changes
|
|
15
|
+
|
|
16
|
+
- 0dfc150: fix: exports config between packages
|
|
17
|
+
|
|
18
|
+
### Patch Changes
|
|
19
|
+
|
|
20
|
+
- Updated dependencies [0dfc150]
|
|
21
|
+
- @empiricalrun/test-run@0.14.0
|
|
22
|
+
|
|
3
23
|
## 0.78.7
|
|
4
24
|
|
|
5
25
|
### Patch Changes
|
|
@@ -5,7 +5,7 @@ export { getFileInfoFromGitHub, viewFileUsingGitHub, } from "../../file-info/ada
|
|
|
5
5
|
export type { AgentParams } from "../base";
|
|
6
6
|
export { BaseAgent } from "../base";
|
|
7
7
|
export type { CodeReviewResultV0, CodeReviewResultV1, CodeReviewResultV2, CodeReviewVersionedResult, } from "../code-review";
|
|
8
|
-
export { CodeReviewAgent } from "../code-review";
|
|
8
|
+
export { CodeReviewAgent, convertXmlToV2Format } from "../code-review";
|
|
9
9
|
export { TriageAgent } from "../triage";
|
|
10
10
|
export { ChatAgent } from "./index";
|
|
11
11
|
export { fetchToolCallFromId, getLatestDownloadBuildUrl, LATEST_CHAT_STATE_VERSION, } from "./state";
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"exports.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/exports.ts"],"names":[],"mappings":"AAAA,YAAY,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AACzD,OAAO,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAC;AACzD,OAAO,EAAE,qBAAqB,EAAE,MAAM,kCAAkC,CAAC;AACzE,OAAO,EACL,qBAAqB,EACrB,mBAAmB,GACpB,MAAM,iCAAiC,CAAC;AAEzC,YAAY,EAAE,WAAW,EAAE,MAAM,SAAS,CAAC;AAC3C,OAAO,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AACpC,YAAY,EACV,kBAAkB,EAClB,kBAAkB,EAClB,kBAAkB,EAClB,yBAAyB,GAC1B,MAAM,gBAAgB,CAAC;AACxB,OAAO,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC;
|
|
1
|
+
{"version":3,"file":"exports.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/exports.ts"],"names":[],"mappings":"AAAA,YAAY,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AACzD,OAAO,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAC;AACzD,OAAO,EAAE,qBAAqB,EAAE,MAAM,kCAAkC,CAAC;AACzE,OAAO,EACL,qBAAqB,EACrB,mBAAmB,GACpB,MAAM,iCAAiC,CAAC;AAEzC,YAAY,EAAE,WAAW,EAAE,MAAM,SAAS,CAAC;AAC3C,OAAO,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AACpC,YAAY,EACV,kBAAkB,EAClB,kBAAkB,EAClB,kBAAkB,EAClB,yBAAyB,GAC1B,MAAM,gBAAgB,CAAC;AACxB,OAAO,EAAE,eAAe,EAAE,oBAAoB,EAAE,MAAM,gBAAgB,CAAC;AACvE,OAAO,EAAE,WAAW,EAAE,MAAM,WAAW,CAAC;AACxC,OAAO,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AACpC,OAAO,EACL,mBAAmB,EACnB,yBAAyB,EACzB,yBAAyB,GAC1B,MAAM,SAAS,CAAC;AACjB,OAAO,EAAE,kBAAkB,EAAE,MAAM,SAAS,CAAC"}
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
"use strict";
|
|
2
2
|
Object.defineProperty(exports, "__esModule", { value: true });
|
|
3
|
-
exports.extractAttachments = exports.LATEST_CHAT_STATE_VERSION = exports.getLatestDownloadBuildUrl = exports.fetchToolCallFromId = exports.ChatAgent = exports.TriageAgent = exports.CodeReviewAgent = exports.BaseAgent = exports.viewFileUsingGitHub = exports.getFileInfoFromGitHub = exports.SUPPORTED_CHAT_MODELS = exports.createChatModel = void 0;
|
|
3
|
+
exports.extractAttachments = exports.LATEST_CHAT_STATE_VERSION = exports.getLatestDownloadBuildUrl = exports.fetchToolCallFromId = exports.ChatAgent = exports.TriageAgent = exports.convertXmlToV2Format = exports.CodeReviewAgent = exports.BaseAgent = exports.viewFileUsingGitHub = exports.getFileInfoFromGitHub = exports.SUPPORTED_CHAT_MODELS = exports.createChatModel = void 0;
|
|
4
4
|
var chat_1 = require("@empiricalrun/llm/chat");
|
|
5
5
|
Object.defineProperty(exports, "createChatModel", { enumerable: true, get: function () { return chat_1.createChatModel; } });
|
|
6
6
|
var constants_1 = require("@empiricalrun/llm/chat/constants");
|
|
@@ -12,6 +12,7 @@ var base_1 = require("../base");
|
|
|
12
12
|
Object.defineProperty(exports, "BaseAgent", { enumerable: true, get: function () { return base_1.BaseAgent; } });
|
|
13
13
|
var code_review_1 = require("../code-review");
|
|
14
14
|
Object.defineProperty(exports, "CodeReviewAgent", { enumerable: true, get: function () { return code_review_1.CodeReviewAgent; } });
|
|
15
|
+
Object.defineProperty(exports, "convertXmlToV2Format", { enumerable: true, get: function () { return code_review_1.convertXmlToV2Format; } });
|
|
15
16
|
var triage_1 = require("../triage");
|
|
16
17
|
Object.defineProperty(exports, "TriageAgent", { enumerable: true, get: function () { return triage_1.TriageAgent; } });
|
|
17
18
|
var index_1 = require("./index");
|
|
@@ -4,6 +4,7 @@ import { BaseAgent } from "../base";
|
|
|
4
4
|
import { type CodeReviewResultV0, type CodeReviewResultV1, type CodeReviewResultV2, CodeReviewSeverity, CodeReviewVerdict } from "./types";
|
|
5
5
|
export type { CodeReviewResultV1, CodeReviewResultV0, CodeReviewResultV2 };
|
|
6
6
|
export { CodeReviewVerdict, CodeReviewSeverity };
|
|
7
|
+
export { convertXmlToV2Format } from "./parser";
|
|
7
8
|
export type CodeReviewVersionedResult = CodeReviewResultV0 | CodeReviewResultV1 | CodeReviewResultV2;
|
|
8
9
|
export declare class CodeReviewAgent<T> extends BaseAgent<T> {
|
|
9
10
|
protected getTools(): ToolsForLLM;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/code-review/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAEV,WAAW,EACZ,MAAM,uCAAuC,CAAC;AAC/C,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,qCAAqC,CAAC;AAGpE,OAAO,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AAGpC,OAAO,EACL,KAAK,kBAAkB,EACvB,KAAK,kBAAkB,EACvB,KAAK,kBAAkB,EACvB,kBAAkB,EAClB,iBAAiB,EAClB,MAAM,SAAS,CAAC;AAEjB,YAAY,EAAE,kBAAkB,EAAE,kBAAkB,EAAE,kBAAkB,EAAE,CAAC;AAC3E,OAAO,EAAE,iBAAiB,EAAE,kBAAkB,EAAE,CAAC;
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/code-review/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAEV,WAAW,EACZ,MAAM,uCAAuC,CAAC;AAC/C,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,qCAAqC,CAAC;AAGpE,OAAO,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AAGpC,OAAO,EACL,KAAK,kBAAkB,EACvB,KAAK,kBAAkB,EACvB,KAAK,kBAAkB,EACvB,kBAAkB,EAClB,iBAAiB,EAClB,MAAM,SAAS,CAAC;AAEjB,YAAY,EAAE,kBAAkB,EAAE,kBAAkB,EAAE,kBAAkB,EAAE,CAAC;AAC3E,OAAO,EAAE,iBAAiB,EAAE,kBAAkB,EAAE,CAAC;AACjD,OAAO,EAAE,oBAAoB,EAAE,MAAM,UAAU,CAAC;AAEhD,MAAM,MAAM,yBAAyB,GACjC,kBAAkB,GAClB,kBAAkB,GAClB,kBAAkB,CAAC;AAEvB,qBAAa,eAAe,CAAC,CAAC,CAAE,SAAQ,SAAS,CAAC,CAAC,CAAC;IAClD,SAAS,CAAC,QAAQ,IAAI,WAAW;IAWjC,SAAS,IAAI,yBAAyB,GAAG,SAAS;cAmBlC,iBAAiB,CAC/B,eAAe,EAAE,MAAM,OAAO,CAAC,QAAQ,CAAC,GACvC,OAAO,CAAC,MAAM,CAAC;CAiInB"}
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
"use strict";
|
|
2
2
|
Object.defineProperty(exports, "__esModule", { value: true });
|
|
3
|
-
exports.CodeReviewAgent = exports.CodeReviewSeverity = exports.CodeReviewVerdict = void 0;
|
|
3
|
+
exports.CodeReviewAgent = exports.convertXmlToV2Format = exports.CodeReviewSeverity = exports.CodeReviewVerdict = void 0;
|
|
4
4
|
const tools_1 = require("../../tools");
|
|
5
5
|
const fetch_session_diff_1 = require("../../tools/fetch-session-diff");
|
|
6
6
|
const base_1 = require("../base");
|
|
@@ -9,6 +9,8 @@ const parser_1 = require("./parser");
|
|
|
9
9
|
const types_1 = require("./types");
|
|
10
10
|
Object.defineProperty(exports, "CodeReviewSeverity", { enumerable: true, get: function () { return types_1.CodeReviewSeverity; } });
|
|
11
11
|
Object.defineProperty(exports, "CodeReviewVerdict", { enumerable: true, get: function () { return types_1.CodeReviewVerdict; } });
|
|
12
|
+
var parser_2 = require("./parser");
|
|
13
|
+
Object.defineProperty(exports, "convertXmlToV2Format", { enumerable: true, get: function () { return parser_2.convertXmlToV2Format; } });
|
|
12
14
|
class CodeReviewAgent extends base_1.BaseAgent {
|
|
13
15
|
getTools() {
|
|
14
16
|
const custom = [
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/cua/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAiB,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAC/D,OAAO,EACL,iBAAiB,EACjB,KAAK,EACN,MAAM,uCAAuC,CAAC;AAS/C,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAmElC,MAAM,MAAM,kBAAkB,GAAG,KAAK,CAClC;IACE,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;CACd,GACD;IACE,IAAI,EAAE,YAAY,CAAC;IACnB,UAAU,EAAE,iBAAiB,CAAC;CAC/B,CACJ,CAAC;AAEF,wBAAsB,+BAA+B,CAAC,EACpD,IAAI,EACJ,IAAI,EACJ,KAAK,EACL,8BAAsC,GACvC,EAAE;IACD,IAAI,EAAE,IAAI,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,8BAA8B,CAAC,EAAE,OAAO,CAAC;CAC1C,GAAG,OAAO,CAAC;IACV,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,EAAE,CAAC;IACtB,MAAM,EAAE,kBAAkB,CAAC;IAC3B,KAAK,EAAE,KAAK,CAAC;CACd,CAAC,
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/cua/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAiB,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAC/D,OAAO,EACL,iBAAiB,EACjB,KAAK,EACN,MAAM,uCAAuC,CAAC;AAS/C,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAmElC,MAAM,MAAM,kBAAkB,GAAG,KAAK,CAClC;IACE,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;CACd,GACD;IACE,IAAI,EAAE,YAAY,CAAC;IACnB,UAAU,EAAE,iBAAiB,CAAC;CAC/B,CACJ,CAAC;AAEF,wBAAsB,+BAA+B,CAAC,EACpD,IAAI,EACJ,IAAI,EACJ,KAAK,EACL,8BAAsC,GACvC,EAAE;IACD,IAAI,EAAE,IAAI,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,8BAA8B,CAAC,EAAE,OAAO,CAAC;CAC1C,GAAG,OAAO,CAAC;IACV,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,EAAE,CAAC;IACtB,MAAM,EAAE,kBAAkB,CAAC;IAC3B,KAAK,EAAE,KAAK,CAAC;CACd,CAAC,CAuOD"}
|
package/dist/agent/cua/index.js
CHANGED
|
@@ -63,7 +63,12 @@ async function createTestUsingComputerUseAgent({ page, task, trace, prefersEleme
|
|
|
63
63
|
let result = [
|
|
64
64
|
{ type: "screenshot", screenshot: initialArtifact },
|
|
65
65
|
];
|
|
66
|
-
let tokensUsed = {
|
|
66
|
+
let tokensUsed = {
|
|
67
|
+
input_tokens: 0,
|
|
68
|
+
output_tokens: 0,
|
|
69
|
+
input_tokens_details: { cached_tokens: 0 },
|
|
70
|
+
output_tokens_details: { reasoning_tokens: 0 },
|
|
71
|
+
};
|
|
67
72
|
const viewport = page.viewportSize();
|
|
68
73
|
let screenWidth = viewport?.width || 1280;
|
|
69
74
|
let screenHeight = viewport?.height || 720;
|
|
@@ -94,8 +99,12 @@ async function createTestUsingComputerUseAgent({ page, task, trace, prefersEleme
|
|
|
94
99
|
screenHeight,
|
|
95
100
|
openAIClient,
|
|
96
101
|
});
|
|
97
|
-
tokensUsed.
|
|
98
|
-
tokensUsed.
|
|
102
|
+
tokensUsed.input_tokens += response.usage?.input_tokens || 0;
|
|
103
|
+
tokensUsed.output_tokens += response.usage?.output_tokens || 0;
|
|
104
|
+
tokensUsed.input_tokens_details.cached_tokens +=
|
|
105
|
+
response.usage?.input_tokens_details?.cached_tokens || 0;
|
|
106
|
+
tokensUsed.output_tokens_details.reasoning_tokens +=
|
|
107
|
+
response.usage?.output_tokens_details?.reasoning_tokens || 0;
|
|
99
108
|
let isTaskDone = false;
|
|
100
109
|
let generatedCode = "";
|
|
101
110
|
let iterationIndex = 0;
|
|
@@ -213,8 +222,12 @@ async function createTestUsingComputerUseAgent({ page, task, trace, prefersEleme
|
|
|
213
222
|
screenHeight,
|
|
214
223
|
openAIClient,
|
|
215
224
|
});
|
|
216
|
-
tokensUsed.
|
|
217
|
-
tokensUsed.
|
|
225
|
+
tokensUsed.input_tokens += response.usage?.input_tokens || 0;
|
|
226
|
+
tokensUsed.output_tokens += response.usage?.output_tokens || 0;
|
|
227
|
+
tokensUsed.input_tokens_details.cached_tokens +=
|
|
228
|
+
response.usage?.input_tokens_details?.cached_tokens || 0;
|
|
229
|
+
tokensUsed.output_tokens_details.reasoning_tokens +=
|
|
230
|
+
response.usage?.output_tokens_details?.reasoning_tokens || 0;
|
|
218
231
|
}
|
|
219
232
|
if (!isTaskDone) {
|
|
220
233
|
const logMessage = `Max iteration limit hit: Task not done after ${MAX_ITERATIONS} iterations`;
|
|
@@ -225,13 +238,17 @@ async function createTestUsingComputerUseAgent({ page, task, trace, prefersEleme
|
|
|
225
238
|
});
|
|
226
239
|
}
|
|
227
240
|
trace?.update({ output: { result } });
|
|
241
|
+
const cost = (0, model_1.tokensToCost)(tokensUsed);
|
|
228
242
|
return {
|
|
229
243
|
result,
|
|
230
244
|
code: generatedCode,
|
|
231
245
|
importPaths: [],
|
|
232
246
|
usage: {
|
|
233
|
-
tokens:
|
|
234
|
-
|
|
247
|
+
tokens: {
|
|
248
|
+
input: tokensUsed.input_tokens,
|
|
249
|
+
output: tokensUsed.output_tokens,
|
|
250
|
+
},
|
|
251
|
+
cost,
|
|
235
252
|
},
|
|
236
253
|
};
|
|
237
254
|
}
|
|
@@ -7,11 +7,18 @@ export declare function callComputerUseModel({ input, previousResponseId, screen
|
|
|
7
7
|
screenHeight: number;
|
|
8
8
|
openAIClient: OpenAI;
|
|
9
9
|
}): Promise<Response>;
|
|
10
|
-
export declare function tokensToCost(
|
|
11
|
-
|
|
12
|
-
|
|
10
|
+
export declare function tokensToCost(usage: {
|
|
11
|
+
input_tokens: number;
|
|
12
|
+
output_tokens: number;
|
|
13
|
+
input_tokens_details: {
|
|
14
|
+
cached_tokens: number;
|
|
15
|
+
};
|
|
16
|
+
output_tokens_details: {
|
|
17
|
+
reasoning_tokens: number;
|
|
18
|
+
};
|
|
13
19
|
}): {
|
|
14
20
|
input: number;
|
|
15
21
|
output: number;
|
|
22
|
+
total: number;
|
|
16
23
|
};
|
|
17
24
|
//# sourceMappingURL=model.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"model.d.ts","sourceRoot":"","sources":["../../../src/agent/cua/model.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"model.d.ts","sourceRoot":"","sources":["../../../src/agent/cua/model.ts"],"names":[],"mappings":"AACA,OAAO,MAAM,MAAM,QAAQ,CAAC;AAC5B,OAAO,EAEL,QAAQ,EACR,iBAAiB,EAClB,MAAM,0CAA0C,CAAC;AAmClD,wBAAsB,oBAAoB,CAAC,EACzC,KAAK,EACL,kBAAkB,EAClB,WAAW,EACX,YAAY,EACZ,YAAY,GACb,EAAE;IACD,KAAK,EAAE,iBAAiB,EAAE,CAAC;IAC3B,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC;CACtB,GAAG,OAAO,CAAC,QAAQ,CAAC,CAuBpB;AAED,wBAAgB,YAAY,CAAC,KAAK,EAAE;IAClC,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,oBAAoB,EAAE;QAAE,aAAa,EAAE,MAAM,CAAA;KAAE,CAAC;IAChD,qBAAqB,EAAE;QAAE,gBAAgB,EAAE,MAAM,CAAA;KAAE,CAAC;CACrD;;;;EAMA"}
|
package/dist/agent/cua/model.js
CHANGED
|
@@ -2,6 +2,8 @@
|
|
|
2
2
|
Object.defineProperty(exports, "__esModule", { value: true });
|
|
3
3
|
exports.callComputerUseModel = callComputerUseModel;
|
|
4
4
|
exports.tokensToCost = tokensToCost;
|
|
5
|
+
const llm_1 = require("@empiricalrun/llm");
|
|
6
|
+
const CUA_MODEL = "computer-use-preview-2025-03-11";
|
|
5
7
|
const INSTRUCTIONS = `You will be asked to execute some actions in a browser context.
|
|
6
8
|
Don't ask the user for confirmations - just execute the actions.
|
|
7
9
|
|
|
@@ -32,7 +34,7 @@ const pageGotoTool = {
|
|
|
32
34
|
};
|
|
33
35
|
async function callComputerUseModel({ input, previousResponseId, screenWidth, screenHeight, openAIClient, }) {
|
|
34
36
|
const response = await openAIClient.responses.create({
|
|
35
|
-
model:
|
|
37
|
+
model: CUA_MODEL,
|
|
36
38
|
previous_response_id: previousResponseId,
|
|
37
39
|
parallel_tool_calls: false,
|
|
38
40
|
tools: [
|
|
@@ -54,12 +56,10 @@ async function callComputerUseModel({ input, previousResponseId, screenWidth, sc
|
|
|
54
56
|
});
|
|
55
57
|
return response;
|
|
56
58
|
}
|
|
57
|
-
function tokensToCost(
|
|
58
|
-
|
|
59
|
-
|
|
60
|
-
|
|
61
|
-
|
|
62
|
-
|
|
63
|
-
const outputCost = (tokens.output / 1_000_000) * outputUsdFor1MTokens;
|
|
64
|
-
return { input: inputCost, output: outputCost };
|
|
59
|
+
function tokensToCost(usage) {
|
|
60
|
+
const fullUsage = {
|
|
61
|
+
...usage,
|
|
62
|
+
total_tokens: usage.input_tokens + usage.output_tokens,
|
|
63
|
+
};
|
|
64
|
+
return (0, llm_1.calculateOpenAITokenCosts)({ usage: fullUsage, model: CUA_MODEL });
|
|
65
65
|
}
|
|
@@ -25,7 +25,7 @@ exports.buildTestNamePrompt = buildTestNamePrompt;
|
|
|
25
25
|
exports.getVariableDeclarationsFromCode = getVariableDeclarationsFromCode;
|
|
26
26
|
exports.isSyntaxValid = isSyntaxValid;
|
|
27
27
|
const parser_1 = require("@babel/parser");
|
|
28
|
-
const
|
|
28
|
+
const cmd_1 = require("@empiricalrun/test-run/cmd");
|
|
29
29
|
const eslint_1 = require("eslint");
|
|
30
30
|
const fs_1 = __importDefault(require("fs"));
|
|
31
31
|
const path_1 = __importDefault(require("path"));
|
|
@@ -203,7 +203,7 @@ function appendToTestBlock(testBlock, content) {
|
|
|
203
203
|
return updateTestBlock;
|
|
204
204
|
}
|
|
205
205
|
async function runTypescriptCompiler(repoDir) {
|
|
206
|
-
let { code, output } = await (0,
|
|
206
|
+
let { code, output } = await (0, cmd_1.spawnCmd)("npx", ["tsc", "--incremental", "--noEmit", "--pretty"], {
|
|
207
207
|
cwd: repoDir,
|
|
208
208
|
envOverrides: {},
|
|
209
209
|
captureOutput: true,
|
|
@@ -78,11 +78,6 @@ type TestCaseSummaryDetail = {
|
|
|
78
78
|
merged_summary?: {
|
|
79
79
|
content: string;
|
|
80
80
|
} | null;
|
|
81
|
-
successful_run_metadata: {
|
|
82
|
-
videos: string[];
|
|
83
|
-
test_id?: string;
|
|
84
|
-
test_run_id?: string;
|
|
85
|
-
} | null;
|
|
86
81
|
failed_run_metadata: FailedRunMetadata;
|
|
87
82
|
visual_diff_summary: {
|
|
88
83
|
failedStep: string;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../../src/tools/test-run-fetcher/types.ts"],"names":[],"mappings":"AACA,KAAK,OAAO,GAAG;IACb,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC;CACnB,CAAC;AAEF,KAAK,OAAO,GAAG;IACb,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,MAAM,CAAC;IACvB,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,eAAe,EAAE,MAAM,CAAC;IACxB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,cAAc,EAAE,MAAM,CAAC;IACvB,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAC9B,iBAAiB,EAAE,MAAM,CAAC;CAC3B,CAAC;AAEF,KAAK,eAAe,GAAG,OAAO,GAAG;IAC/B,gBAAgB,EAAE,MAAM,GAAG,IAAI,CAAC;IAChC,gBAAgB,EAAE,MAAM,GAAG,IAAI,CAAC;IAChC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;CAC1B,CAAC;AAEF,KAAK,iBAAiB,GAAG;IACvB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,QAAQ,EAAE;QACR,IAAI,EAAE,MAAM,CAAC;QACb,IAAI,EAAE,MAAM,CAAC;QACb,MAAM,EAAE,MAAM,CAAC;KAChB,CAAC;IACF,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,eAAe,EAAE,OAAO,CAAC;CAC1B,CAAC;AAEF,KAAK,sBAAsB,GAAG;IAC5B,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,iCAAiC,EAAE,MAAM,CAAC;IAC1C,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE;QACP,OAAO,EAAE;YAAE,CAAC,GAAG,EAAE,MAAM,GAAG,MAAM,CAAA;SAAE,EAAE,CAAC;KACtC,CAAC;IACF,QAAQ,EAAE;QACR,OAAO,EAAE;YAAE,CAAC,GAAG,EAAE,MAAM,GAAG,MAAM,CAAA;SAAE,EAAE,CAAC;KACtC,CAAC;CACH,CAAC;AAEF,KAAK,qBAAqB,GAAG;IAC3B,EAAE,EAAE,MAAM,CAAC;IACX,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,MAAM,EAAE,MAAM,CAAC;IACf,YAAY,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC;IACrB,wBAAwB,EAAE,MAAM,CAAC;IACjC,iCAAiC,EAAE,MAAM,CAAC;IAC1C,cAAc,CAAC,EAAE;QACf,OAAO,EAAE,MAAM,CAAC;KACjB,GAAG,IAAI,CAAC;IACT,
|
|
1
|
+
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../../src/tools/test-run-fetcher/types.ts"],"names":[],"mappings":"AACA,KAAK,OAAO,GAAG;IACb,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC;CACnB,CAAC;AAEF,KAAK,OAAO,GAAG;IACb,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,MAAM,CAAC;IACvB,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,eAAe,EAAE,MAAM,CAAC;IACxB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,cAAc,EAAE,MAAM,CAAC;IACvB,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAC9B,iBAAiB,EAAE,MAAM,CAAC;CAC3B,CAAC;AAEF,KAAK,eAAe,GAAG,OAAO,GAAG;IAC/B,gBAAgB,EAAE,MAAM,GAAG,IAAI,CAAC;IAChC,gBAAgB,EAAE,MAAM,GAAG,IAAI,CAAC;IAChC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;CAC1B,CAAC;AAEF,KAAK,iBAAiB,GAAG;IACvB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,QAAQ,EAAE;QACR,IAAI,EAAE,MAAM,CAAC;QACb,IAAI,EAAE,MAAM,CAAC;QACb,MAAM,EAAE,MAAM,CAAC;KAChB,CAAC;IACF,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,eAAe,EAAE,OAAO,CAAC;CAC1B,CAAC;AAEF,KAAK,sBAAsB,GAAG;IAC5B,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,iCAAiC,EAAE,MAAM,CAAC;IAC1C,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE;QACP,OAAO,EAAE;YAAE,CAAC,GAAG,EAAE,MAAM,GAAG,MAAM,CAAA;SAAE,EAAE,CAAC;KACtC,CAAC;IACF,QAAQ,EAAE;QACR,OAAO,EAAE;YAAE,CAAC,GAAG,EAAE,MAAM,GAAG,MAAM,CAAA;SAAE,EAAE,CAAC;KACtC,CAAC;CACH,CAAC;AAEF,KAAK,qBAAqB,GAAG;IAC3B,EAAE,EAAE,MAAM,CAAC;IACX,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,MAAM,EAAE,MAAM,CAAC;IACf,YAAY,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC;IACrB,wBAAwB,EAAE,MAAM,CAAC;IACjC,iCAAiC,EAAE,MAAM,CAAC;IAC1C,cAAc,CAAC,EAAE;QACf,OAAO,EAAE,MAAM,CAAC;KACjB,GAAG,IAAI,CAAC;IACT,mBAAmB,EAAE,iBAAiB,CAAC;IACvC,mBAAmB,EAAE;QACnB,UAAU,EAAE,MAAM,CAAC;QACnB,MAAM,EAAE,MAAM,CAAC;QACf,OAAO,EAAE,MAAM,CAAC;QAChB,4BAA4B,EAAE;YAC5B,OAAO,EAAE,MAAM,CAAC;YAChB,OAAO,EAAE,MAAM,CAAC;SACjB,CAAC;KACH,GAAG,IAAI,CAAC;IACT,mBAAmB,CAAC,EAAE;QACpB,OAAO,EAAE,MAAM,CAAC;KACjB,GAAG,IAAI,CAAC;IACT,gBAAgB,EAAE;QAChB,YAAY,EAAE,sBAAsB,EAAE,CAAC;QACvC,8BAA8B,EAAE,sBAAsB,EAAE,CAAC;KAC1D,GAAG,IAAI,CAAC;IACT,UAAU,EAAE,MAAM,CAAC;IACnB,IAAI,EAAE,MAAM,CAAC;IACb,YAAY,EAAE,MAAM,CAAC;CACtB,CAAC;AAEF,MAAM,MAAM,qBAAqB,GAAG;IAClC,IAAI,EAAE;QACJ,QAAQ,EAAE;YACR,OAAO,EAAE,OAAO,CAAC;YACjB,OAAO,EAAE,eAAe,CAAC;YACzB,uBAAuB,EAAE,qBAAqB,EAAE,CAAC;SAClD,CAAC;KACH,GAAG,IAAI,CAAC;IACT,KAAK,CAAC,EAAE;QACN,OAAO,EAAE,MAAM,CAAC;KACjB,CAAC;CACH,CAAC"}
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@empiricalrun/test-gen",
|
|
3
|
-
"version": "0.
|
|
3
|
+
"version": "0.79.1",
|
|
4
4
|
"publishConfig": {
|
|
5
5
|
"registry": "https://registry.npmjs.org/",
|
|
6
6
|
"access": "public"
|
|
@@ -88,10 +88,10 @@
|
|
|
88
88
|
"yauzl": "^3.1.3",
|
|
89
89
|
"zod": "^4.0.1",
|
|
90
90
|
"@empiricalrun/ast-parser": "^0.0.10",
|
|
91
|
-
"@empiricalrun/llm": "^0.25.
|
|
91
|
+
"@empiricalrun/llm": "^0.25.2",
|
|
92
92
|
"@empiricalrun/r2-uploader": "^0.9.0",
|
|
93
|
-
"@empiricalrun/test-run": "^0.
|
|
94
|
-
"videostil": "0.3.
|
|
93
|
+
"@empiricalrun/test-run": "^0.14.0",
|
|
94
|
+
"videostil": "0.3.4"
|
|
95
95
|
},
|
|
96
96
|
"devDependencies": {
|
|
97
97
|
"@playwright/test": "1.53.2",
|