@empiricalrun/test-gen 0.33.10 → 0.33.11
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 +6 -0
- package/dist/agent/browsing/index.d.ts.map +1 -1
- package/dist/agent/browsing/index.js +2 -4
- package/dist/agent/master/run.d.ts.map +1 -1
- package/dist/agent/master/run.js +9 -0
- package/dist/bin/index.js +3 -0
- package/dist/session/index.d.ts +1 -0
- package/dist/session/index.d.ts.map +1 -1
- package/dist/session/index.js +6 -1
- package/package.json +3 -3
package/CHANGELOG.md
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,GAAG,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AACrD,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAElC,OAAO,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAClD,OAAO,EAAE,YAAY,EAAE,MAAM,kBAAkB,CAAC;AAIhD,OAAO,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAMnD,MAAM,MAAM,oBAAoB,GAAG,OAAO,CAAC,oBAAoB,CAAC,GAAG;IACjE,YAAY,CAAC,EAAE;QACb,iBAAiB,CAAC,EAAE,MAAM,EAAE,CAAC;KAC9B,CAAC;CACH,CAAC;AAEF,wBAAsB,6BAA6B,CAAC,EAClD,KAAK,EACL,MAAM,EACN,MAAM,EACN,IAAI,EACJ,OAAO,EACP,GAAG,EACH,OAAO,GACR,EAAE;IACD,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,WAAW,CAAC;IACnB,MAAM,EAAE,YAAY,CAAC;IACrB,IAAI,EAAE,IAAI,CAAC;IACX,OAAO,EAAE,oBAAoB,CAAC;IAC9B,GAAG,EAAE,GAAG,CAAC;IACT,OAAO,EAAE,iBAAiB,CAAC;CAC5B,
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,GAAG,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AACrD,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAElC,OAAO,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAClD,OAAO,EAAE,YAAY,EAAE,MAAM,kBAAkB,CAAC;AAIhD,OAAO,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAMnD,MAAM,MAAM,oBAAoB,GAAG,OAAO,CAAC,oBAAoB,CAAC,GAAG;IACjE,YAAY,CAAC,EAAE;QACb,iBAAiB,CAAC,EAAE,MAAM,EAAE,CAAC;KAC9B,CAAC;CACH,CAAC;AAEF,wBAAsB,6BAA6B,CAAC,EAClD,KAAK,EACL,MAAM,EACN,MAAM,EACN,IAAI,EACJ,OAAO,EACP,GAAG,EACH,OAAO,GACR,EAAE;IACD,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,WAAW,CAAC;IACnB,MAAM,EAAE,YAAY,CAAC;IACrB,IAAI,EAAE,IAAI,CAAC;IACX,OAAO,EAAE,oBAAoB,CAAC;IAC9B,GAAG,EAAE,GAAG,CAAC;IACT,OAAO,EAAE,iBAAiB,CAAC;CAC5B,iBAyIA"}
|
|
@@ -18,8 +18,7 @@ async function executeTaskUsingBrowsingAgent({ trace, action, logger, page, opti
|
|
|
18
18
|
const browsingAgentSpan = trace.span({
|
|
19
19
|
name: `browsing-agent`,
|
|
20
20
|
});
|
|
21
|
-
|
|
22
|
-
if (sessionState === "request_complete") {
|
|
21
|
+
if (await (0, session_1.shouldStopSession)()) {
|
|
23
22
|
break;
|
|
24
23
|
}
|
|
25
24
|
const pageContentSpan = browsingAgentSpan.span({
|
|
@@ -96,8 +95,7 @@ async function executeTaskUsingBrowsingAgent({ trace, action, logger, page, opti
|
|
|
96
95
|
const toolCallsSpan = browsingAgentSpan.span({ name: "tool-calls" });
|
|
97
96
|
for (const i in toolCalls) {
|
|
98
97
|
const toolCall = toolCalls[i];
|
|
99
|
-
|
|
100
|
-
if (sessionState === "request_complete") {
|
|
98
|
+
if (await (0, session_1.shouldStopSession)()) {
|
|
101
99
|
break;
|
|
102
100
|
}
|
|
103
101
|
try {
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/master/run.ts"],"names":[],"mappings":"AAAA,OAAO,EAGL,GAAG,EACH,WAAW,EACZ,MAAM,mBAAmB,CAAC;AAG3B,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAElC,OAAO,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAYlD,OAAO,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AACvC,OAAO,EACL,oBAAoB,EAErB,MAAM,aAAa,CAAC;AAOrB,wBAAsB,aAAa,CAAC,EAClC,IAAI,EACJ,eAAe,EACf,aAAa,EACb,IAAI,EACJ,KAAK,EACL,GAAG,EACH,OAAO,EACP,cAAc,EACd,OAAO,EACP,aAAa,GACd,EAAE;IACD,IAAI,EAAE,MAAM,CAAC;IACb,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,aAAa,EAAE,GAAG,EAAE,CAAC;IACrB,IAAI,EAAE,IAAI,CAAC;IACX,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,GAAG,EAAE,GAAG,CAAC;IACT,OAAO,EAAE,oBAAoB,CAAC;IAC9B,cAAc,EAAE,MAAM,CAAC;IACvB,OAAO,EAAE,iBAAiB,CAAC;IAC3B,aAAa,EAAE,OAAO,CAAC;CACxB,2FAwDA;AAED,wBAAsB,0BAA0B,CAAC,EAC/C,IAAI,EACJ,IAAI,EACJ,QAAQ,EACR,OAAO,GACR,EAAE;IACD,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,IAAI,CAAC;IACX,QAAQ,EAAE,QAAQ,CAAC;IACnB,OAAO,EAAE,oBAAoB,CAAC;CAC/B;;;
|
|
1
|
+
{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/master/run.ts"],"names":[],"mappings":"AAAA,OAAO,EAGL,GAAG,EACH,WAAW,EACZ,MAAM,mBAAmB,CAAC;AAG3B,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAElC,OAAO,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAYlD,OAAO,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AACvC,OAAO,EACL,oBAAoB,EAErB,MAAM,aAAa,CAAC;AAOrB,wBAAsB,aAAa,CAAC,EAClC,IAAI,EACJ,eAAe,EACf,aAAa,EACb,IAAI,EACJ,KAAK,EACL,GAAG,EACH,OAAO,EACP,cAAc,EACd,OAAO,EACP,aAAa,GACd,EAAE;IACD,IAAI,EAAE,MAAM,CAAC;IACb,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,aAAa,EAAE,GAAG,EAAE,CAAC;IACrB,IAAI,EAAE,IAAI,CAAC;IACX,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,GAAG,EAAE,GAAG,CAAC;IACT,OAAO,EAAE,oBAAoB,CAAC;IAC9B,cAAc,EAAE,MAAM,CAAC;IACvB,OAAO,EAAE,iBAAiB,CAAC;IAC3B,aAAa,EAAE,OAAO,CAAC;CACxB,2FAwDA;AAED,wBAAsB,0BAA0B,CAAC,EAC/C,IAAI,EACJ,IAAI,EACJ,QAAQ,EACR,OAAO,GACR,EAAE;IACD,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,IAAI,CAAC;IACX,QAAQ,EAAE,QAAQ,CAAC;IACnB,OAAO,EAAE,oBAAoB,CAAC;CAC/B;;;GAkMA"}
|
package/dist/agent/master/run.js
CHANGED
|
@@ -108,6 +108,9 @@ async function createTestUsingMasterAgent({ task, page, testCase, options, }) {
|
|
|
108
108
|
let failedActions = [];
|
|
109
109
|
let disableSkills = false;
|
|
110
110
|
while (!isGivenTaskDone) {
|
|
111
|
+
if (await (0, session_1.shouldStopSession)()) {
|
|
112
|
+
break;
|
|
113
|
+
}
|
|
111
114
|
const masterAgentSpan = trace.span({
|
|
112
115
|
name: "master-agent",
|
|
113
116
|
input: {
|
|
@@ -134,6 +137,9 @@ async function createTestUsingMasterAgent({ task, page, testCase, options, }) {
|
|
|
134
137
|
const testGenSnapshotUpdatePromise = testGenReporter.sendCurrentView(buffer);
|
|
135
138
|
const pageScreenshot = buffer.toString("base64");
|
|
136
139
|
let output;
|
|
140
|
+
if (await (0, session_1.shouldStopSession)()) {
|
|
141
|
+
break;
|
|
142
|
+
}
|
|
137
143
|
const toolCall = await getNextAction({
|
|
138
144
|
task,
|
|
139
145
|
executedActions: masterAgentActions,
|
|
@@ -146,6 +152,9 @@ async function createTestUsingMasterAgent({ task, page, testCase, options, }) {
|
|
|
146
152
|
actions,
|
|
147
153
|
disableSkills,
|
|
148
154
|
});
|
|
155
|
+
if (await (0, session_1.shouldStopSession)()) {
|
|
156
|
+
break;
|
|
157
|
+
}
|
|
149
158
|
if (toolCall) {
|
|
150
159
|
const args = JSON.parse(toolCall.function.arguments);
|
|
151
160
|
const masterAgentActionSpan = masterAgentSpan.span({
|
package/dist/bin/index.js
CHANGED
|
@@ -33,6 +33,9 @@ async function runAgent(testGenConfig) {
|
|
|
33
33
|
console.warn("Failed to send log url to test gen update", e);
|
|
34
34
|
}
|
|
35
35
|
}
|
|
36
|
+
if (await (0, session_1.shouldStopSession)()) {
|
|
37
|
+
return;
|
|
38
|
+
}
|
|
36
39
|
if (testGenConfig.options?.agent !== "code") {
|
|
37
40
|
// this assumes we have only one scenario in test config
|
|
38
41
|
logger.success(`Generating test using ${testGenConfig.options?.agent} agent. ${process.env.LOG_URL ? `[view log](${process.env.LOG_URL})` : ""}`);
|
package/dist/session/index.d.ts
CHANGED
|
@@ -7,6 +7,7 @@ export declare function setSessionDetails({ sessionId, generationId, }: {
|
|
|
7
7
|
sessionId: number;
|
|
8
8
|
generationId: number;
|
|
9
9
|
}): void;
|
|
10
|
+
export declare function shouldStopSession(): Promise<boolean>;
|
|
10
11
|
export declare function getSessionState(): Promise<"started" | "completed" | "request_complete">;
|
|
11
12
|
export declare function endSession(): Promise<void>;
|
|
12
13
|
export { getSessionDetails };
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/session/index.ts"],"names":[],"mappings":"AAeA,iBAAS,iBAAiB;;;;EAMzB;AAED,wBAAgB,iBAAiB,CAAC,EAChC,SAAS,EACT,YAAY,GACb,EAAE;IACD,SAAS,EAAE,MAAM,CAAC;IAClB,YAAY,EAAE,MAAM,CAAC;CACtB,QAGA;AAED,wBAAsB,eAAe,0DAkBpC;AAED,wBAAsB,UAAU,kBAkB/B;AAED,OAAO,EAAE,iBAAiB,EAAE,CAAC"}
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/session/index.ts"],"names":[],"mappings":"AAeA,iBAAS,iBAAiB;;;;EAMzB;AAED,wBAAgB,iBAAiB,CAAC,EAChC,SAAS,EACT,YAAY,GACb,EAAE;IACD,SAAS,EAAE,MAAM,CAAC;IAClB,YAAY,EAAE,MAAM,CAAC;CACtB,QAGA;AAED,wBAAsB,iBAAiB,qBAGtC;AAED,wBAAsB,eAAe,0DAkBpC;AAED,wBAAsB,UAAU,kBAkB/B;AAED,OAAO,EAAE,iBAAiB,EAAE,CAAC"}
|
package/dist/session/index.js
CHANGED
|
@@ -3,7 +3,7 @@ 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.getSessionDetails = exports.endSession = exports.getSessionState = exports.setSessionDetails = void 0;
|
|
6
|
+
exports.getSessionDetails = exports.endSession = exports.getSessionState = exports.shouldStopSession = exports.setSessionDetails = void 0;
|
|
7
7
|
const package_json_1 = __importDefault(require("../../package.json"));
|
|
8
8
|
const sessionDetails = {
|
|
9
9
|
sessionId: undefined,
|
|
@@ -24,6 +24,11 @@ function setSessionDetails({ sessionId, generationId, }) {
|
|
|
24
24
|
sessionDetails.generationId = generationId;
|
|
25
25
|
}
|
|
26
26
|
exports.setSessionDetails = setSessionDetails;
|
|
27
|
+
async function shouldStopSession() {
|
|
28
|
+
const sessionState = await getSessionState();
|
|
29
|
+
return sessionState !== "started";
|
|
30
|
+
}
|
|
31
|
+
exports.shouldStopSession = shouldStopSession;
|
|
27
32
|
async function getSessionState() {
|
|
28
33
|
const apiPath = `${DASHBOARD_DOMAIN}/api/sessions/${sessionDetails.sessionId}/generations/${sessionDetails.generationId}/state`;
|
|
29
34
|
const response = await fetch(apiPath, {
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@empiricalrun/test-gen",
|
|
3
|
-
"version": "0.33.
|
|
3
|
+
"version": "0.33.11",
|
|
4
4
|
"publishConfig": {
|
|
5
5
|
"registry": "https://registry.npmjs.org/",
|
|
6
6
|
"access": "public"
|
|
@@ -45,8 +45,8 @@
|
|
|
45
45
|
"tsx": "^4.16.2",
|
|
46
46
|
"typescript": "^5.3.3",
|
|
47
47
|
"@empiricalrun/llm": "^0.9.13",
|
|
48
|
-
"@empiricalrun/
|
|
49
|
-
"@empiricalrun/
|
|
48
|
+
"@empiricalrun/reporter": "^0.21.1",
|
|
49
|
+
"@empiricalrun/r2-uploader": "^0.3.6"
|
|
50
50
|
},
|
|
51
51
|
"devDependencies": {
|
|
52
52
|
"@types/detect-port": "^1.3.5",
|