@empiricalrun/test-gen 0.46.9 → 0.46.10
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 +7 -0
- package/dist/actions/goto.d.ts.map +1 -1
- package/dist/actions/goto.js +1 -1
- package/dist/agent/browsing/index.d.ts.map +1 -1
- package/dist/agent/browsing/index.js +2 -1
- package/dist/agent/master/execute-browser-action.d.ts +5 -0
- package/dist/agent/master/execute-browser-action.d.ts.map +1 -1
- package/dist/agent/master/execute-browser-action.js +1 -0
- package/dist/agent/master/run.d.ts.map +1 -1
- package/dist/agent/master/run.js +4 -2
- package/dist/agent/planner/run-time-planner.d.ts +3 -2
- package/dist/agent/planner/run-time-planner.d.ts.map +1 -1
- package/dist/agent/planner/run-time-planner.js +4 -1
- package/package.json +3 -3
package/CHANGELOG.md
CHANGED
|
@@ -1,5 +1,12 @@
|
|
|
1
1
|
# @empiricalrun/test-gen
|
|
2
2
|
|
|
3
|
+
## 0.46.10
|
|
4
|
+
|
|
5
|
+
### Patch Changes
|
|
6
|
+
|
|
7
|
+
- b3ca74f: fix: bubble up browsing agent execution feedback for runtime planner
|
|
8
|
+
- b484292: fix: pages summary for runtime planner works without scoped vars
|
|
9
|
+
|
|
3
10
|
## 0.46.9
|
|
4
11
|
|
|
5
12
|
### Patch Changes
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"goto.d.ts","sourceRoot":"","sources":["../../src/actions/goto.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,cAAc,EAAE,yBAAyB,EAAE,MAAM,UAAU,CAAC;AAGrE,eAAO,MAAM,2BAA2B,cAAc,CAAC;AAEvD,eAAO,MAAM,mBAAmB,EAAE,yBAAyB,CAAC,cAAc,
|
|
1
|
+
{"version":3,"file":"goto.d.ts","sourceRoot":"","sources":["../../src/actions/goto.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,cAAc,EAAE,yBAAyB,EAAE,MAAM,UAAU,CAAC;AAGrE,eAAO,MAAM,2BAA2B,cAAc,CAAC;AAEvD,eAAO,MAAM,mBAAmB,EAAE,yBAAyB,CAAC,cAAc,CA0CzE,CAAC"}
|
package/dist/actions/goto.js
CHANGED
|
@@ -35,7 +35,7 @@ const gotoActionGenerator = (page) => {
|
|
|
35
35
|
},
|
|
36
36
|
url: {
|
|
37
37
|
type: "string",
|
|
38
|
-
description: "URL to navigate page to",
|
|
38
|
+
description: "URL to navigate page to. Must be prefixed with http or https (preferably https)",
|
|
39
39
|
},
|
|
40
40
|
},
|
|
41
41
|
required: ["reason", "url"],
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAiB,GAAG,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AACpE,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAElC,OAAO,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAOlD,MAAM,MAAM,kBAAkB,GAAG;IAC/B,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,GAAG,SAAS,CAAC;CAC1B,CAAC;AAEF,wBAAsB,6BAA6B,CAAC,EAClD,MAAM,EACN,IAAI,EACJ,OAAO,EACP,GAAG,EACH,KAAK,GACN,EAAE;IACD,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,IAAI,CAAC;IACX,OAAO,EAAE,iBAAiB,CAAC;IAC3B,GAAG,CAAC,EAAE,GAAG,CAAC;IACV,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,GAAG,OAAO,CAAC,kBAAkB,GAAG,SAAS,CAAC,
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAiB,GAAG,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AACpE,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAElC,OAAO,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAOlD,MAAM,MAAM,kBAAkB,GAAG;IAC/B,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,GAAG,SAAS,CAAC;CAC1B,CAAC;AAEF,wBAAsB,6BAA6B,CAAC,EAClD,MAAM,EACN,IAAI,EACJ,OAAO,EACP,GAAG,EACH,KAAK,GACN,EAAE;IACD,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,IAAI,CAAC;IACX,OAAO,EAAE,iBAAiB,CAAC;IAC3B,GAAG,CAAC,EAAE,GAAG,CAAC;IACV,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,GAAG,OAAO,CAAC,kBAAkB,GAAG,SAAS,CAAC,CA8D1C"}
|
|
@@ -59,9 +59,10 @@ async function executeTaskUsingBrowsingAgent({ action, page, actions, llm, trace
|
|
|
59
59
|
}
|
|
60
60
|
catch (e) {
|
|
61
61
|
void testgenUpdatesReporter.sendMessage(e.message);
|
|
62
|
+
throw e;
|
|
62
63
|
}
|
|
63
64
|
}
|
|
64
|
-
toolCallsSpan?.end({ output: { toolCalls } });
|
|
65
|
+
toolCallsSpan?.end({ output: { toolCalls, generatedCodeSteps } });
|
|
65
66
|
return generatedCodeSteps;
|
|
66
67
|
}
|
|
67
68
|
exports.executeTaskUsingBrowsingAgent = executeTaskUsingBrowsingAgent;
|
|
@@ -15,5 +15,10 @@ export declare function executeBrowserAction({ page, nextAction, flags, actions,
|
|
|
15
15
|
llm: LLM;
|
|
16
16
|
}): Promise<{
|
|
17
17
|
generatedCodeSteps: string[];
|
|
18
|
+
output: {
|
|
19
|
+
action: string;
|
|
20
|
+
reason: string;
|
|
21
|
+
elementAnnotation?: string | undefined;
|
|
22
|
+
};
|
|
18
23
|
}>;
|
|
19
24
|
//# sourceMappingURL=execute-browser-action.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"execute-browser-action.d.ts","sourceRoot":"","sources":["../../../src/agent/master/execute-browser-action.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,GAAG,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AACrD,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,kBAAkB,CAAC;AAE7C,OAAO,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAYlD,wBAAsB,oBAAoB,CAAC,EACzC,IAAI,EACJ,UAAU,EACV,KAAK,EACL,OAAO,EACP,GAAG,EACH,KAAK,GACN,EAAE;IACD,IAAI,EAAE,IAAI,CAAC;IACX,UAAU,EAAE;QACV,UAAU,EAAE,MAAM,CAAC;QACnB,YAAY,EAAE,MAAM,CAAC;KACtB,CAAC;IACF,KAAK,EAAE;QACL,4BAA4B,EAAE,OAAO,CAAC;KACvC,CAAC;IACF,OAAO,EAAE,iBAAiB,CAAC;IAC3B,KAAK,EAAE,WAAW,GAAG,SAAS,CAAC;IAC/B,GAAG,EAAE,GAAG,CAAC;CACV
|
|
1
|
+
{"version":3,"file":"execute-browser-action.d.ts","sourceRoot":"","sources":["../../../src/agent/master/execute-browser-action.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,GAAG,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AACrD,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,kBAAkB,CAAC;AAE7C,OAAO,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAYlD,wBAAsB,oBAAoB,CAAC,EACzC,IAAI,EACJ,UAAU,EACV,KAAK,EACL,OAAO,EACP,GAAG,EACH,KAAK,GACN,EAAE;IACD,IAAI,EAAE,IAAI,CAAC;IACX,UAAU,EAAE;QACV,UAAU,EAAE,MAAM,CAAC;QACnB,YAAY,EAAE,MAAM,CAAC;KACtB,CAAC;IACF,KAAK,EAAE;QACL,4BAA4B,EAAE,OAAO,CAAC;KACvC,CAAC;IACF,OAAO,EAAE,iBAAiB,CAAC;IAC3B,KAAK,EAAE,WAAW,GAAG,SAAS,CAAC;IAC/B,GAAG,EAAE,GAAG,CAAC;CACV;;;gBAIW,MAAM;gBACN,MAAM;;;GAkIjB"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/master/run.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EACV,QAAQ,EACR,oBAAoB,EACrB,MAAM,4BAA4B,CAAC;AACpC,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAelC,OAAO,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AAgCxC,wBAAsB,0BAA0B,CAAC,EAC/C,IAAI,EACJ,IAAI,EACJ,QAAQ,EACR,QAAQ,EACR,OAAO,EACP,SAAS,GACV,EAAE;IACD,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,IAAI,CAAC;IACX,QAAQ,CAAC,EAAE,QAAQ,CAAC;IACpB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,OAAO,CAAC,oBAAoB,CAAC,CAAC;IACvC,SAAS,CAAC,EAAE,SAAS,CAAC;CACvB;;;
|
|
1
|
+
{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/master/run.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EACV,QAAQ,EACR,oBAAoB,EACrB,MAAM,4BAA4B,CAAC;AACpC,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAelC,OAAO,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AAgCxC,wBAAsB,0BAA0B,CAAC,EAC/C,IAAI,EACJ,IAAI,EACJ,QAAQ,EACR,QAAQ,EACR,OAAO,EACP,SAAS,GACV,EAAE;IACD,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,IAAI,CAAC;IACX,QAAQ,CAAC,EAAE,QAAQ,CAAC;IACpB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,OAAO,CAAC,oBAAoB,CAAC,CAAC;IACvC,SAAS,CAAC,EAAE,SAAS,CAAC;CACvB;;;GAwRA"}
|
package/dist/agent/master/run.js
CHANGED
|
@@ -115,7 +115,7 @@ async function createTestUsingMasterAgent({ task, page, testCase, specPath, opti
|
|
|
115
115
|
task,
|
|
116
116
|
successfulActions: [...masterAgentActions],
|
|
117
117
|
pages: getPageVariables(actions.getStateVariables()),
|
|
118
|
-
currentPage: testGenPage
|
|
118
|
+
currentPage: testGenPage,
|
|
119
119
|
});
|
|
120
120
|
isGivenTaskDone = plannerResp.isDone;
|
|
121
121
|
if (isGivenTaskDone) {
|
|
@@ -191,7 +191,7 @@ async function createTestUsingMasterAgent({ task, page, testCase, specPath, opti
|
|
|
191
191
|
break;
|
|
192
192
|
}
|
|
193
193
|
default: {
|
|
194
|
-
const
|
|
194
|
+
const result = await (0, execute_browser_action_1.executeBrowserAction)({
|
|
195
195
|
page,
|
|
196
196
|
nextAction,
|
|
197
197
|
flags: {
|
|
@@ -201,7 +201,9 @@ async function createTestUsingMasterAgent({ task, page, testCase, specPath, opti
|
|
|
201
201
|
llm,
|
|
202
202
|
trace: masterAgentActionSpan,
|
|
203
203
|
});
|
|
204
|
+
const { generatedCodeSteps: codeFromExecuteAction, output: outputFromExecuteAction, } = result;
|
|
204
205
|
generatedCodeSteps.push(...codeFromExecuteAction);
|
|
206
|
+
output = outputFromExecuteAction;
|
|
205
207
|
}
|
|
206
208
|
}
|
|
207
209
|
// resetting error count as there is a successful action
|
|
@@ -1,11 +1,12 @@
|
|
|
1
1
|
import { TraceClient } from "@empiricalrun/llm";
|
|
2
2
|
import type { Page } from "playwright";
|
|
3
|
+
import { TestGenPage } from "../../types";
|
|
3
4
|
export declare function runtimePlanner({ trace, task, successfulActions, pages, currentPage, }: {
|
|
4
5
|
trace?: TraceClient;
|
|
5
6
|
successfulActions: string[];
|
|
6
7
|
task: string;
|
|
7
|
-
pages
|
|
8
|
-
currentPage
|
|
8
|
+
pages: Record<string, Page>;
|
|
9
|
+
currentPage: TestGenPage;
|
|
9
10
|
}): Promise<{
|
|
10
11
|
pageName: string;
|
|
11
12
|
isDone: boolean;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"run-time-planner.d.ts","sourceRoot":"","sources":["../../../src/agent/planner/run-time-planner.ts"],"names":[],"mappings":"AAAA,OAAO,EAAsB,WAAW,EAAE,MAAM,mBAAmB,CAAC;AACpE,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;
|
|
1
|
+
{"version":3,"file":"run-time-planner.d.ts","sourceRoot":"","sources":["../../../src/agent/planner/run-time-planner.ts"],"names":[],"mappings":"AAAA,OAAO,EAAsB,WAAW,EAAE,MAAM,mBAAmB,CAAC;AACpE,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAGvC,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAG1C,wBAAsB,cAAc,CAAC,EACnC,KAAK,EACL,IAAI,EACJ,iBAAiB,EACjB,KAAK,EACL,WAAW,GACZ,EAAE;IACD,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,iBAAiB,EAAE,MAAM,EAAE,CAAC;IAC5B,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC;IAC5B,WAAW,EAAE,WAAW,CAAC;CAC1B;;;;GA+FA"}
|
|
@@ -14,9 +14,12 @@ async function runtimePlanner({ trace, task, successfulActions, pages, currentPa
|
|
|
14
14
|
},
|
|
15
15
|
});
|
|
16
16
|
const llm = new llm_1.LLM({ provider: "openai" });
|
|
17
|
+
// Add the current page to the page state variables
|
|
18
|
+
pages[currentPage.name] = currentPage.pwPageInstance;
|
|
17
19
|
const pagesSummary = pages
|
|
18
20
|
? Object.entries(pages).map(([pageName, page]) => `${pageName}: Currently on ${page.url()}`)
|
|
19
21
|
: [];
|
|
22
|
+
const pageNamesEnum = Object.keys(pages);
|
|
20
23
|
const response = await llm.createChatCompletion({
|
|
21
24
|
trace: runTimePlannerSpan,
|
|
22
25
|
traceName: "runtime-planner-llm",
|
|
@@ -53,7 +56,7 @@ async function runtimePlanner({ trace, task, successfulActions, pages, currentPa
|
|
|
53
56
|
},
|
|
54
57
|
pageName: {
|
|
55
58
|
type: "string",
|
|
56
|
-
enum:
|
|
59
|
+
enum: pageNamesEnum,
|
|
57
60
|
description: "page name for the next action.",
|
|
58
61
|
},
|
|
59
62
|
isDone: {
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@empiricalrun/test-gen",
|
|
3
|
-
"version": "0.46.
|
|
3
|
+
"version": "0.46.10",
|
|
4
4
|
"publishConfig": {
|
|
5
5
|
"registry": "https://registry.npmjs.org/",
|
|
6
6
|
"access": "public"
|
|
@@ -73,9 +73,9 @@
|
|
|
73
73
|
"ts-morph": "^23.0.0",
|
|
74
74
|
"tsx": "^4.16.2",
|
|
75
75
|
"typescript": "^5.3.3",
|
|
76
|
-
"@empiricalrun/llm": "^0.9.35",
|
|
77
76
|
"@empiricalrun/r2-uploader": "^0.3.8",
|
|
78
|
-
"@empiricalrun/reporter": "^0.23.1"
|
|
77
|
+
"@empiricalrun/reporter": "^0.23.1",
|
|
78
|
+
"@empiricalrun/llm": "^0.9.35"
|
|
79
79
|
},
|
|
80
80
|
"devDependencies": {
|
|
81
81
|
"@playwright/test": "1.47.1",
|