veryfront 0.1.876 → 0.1.878
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/esm/deno.d.ts +3 -2
- package/esm/deno.js +4 -3
- package/esm/src/eval/agent-service/durable-run-canaries/cli-runner.d.ts.map +1 -0
- package/esm/src/{agent/testing → eval/agent-service}/durable-run-canaries/environment.d.ts +1 -1
- package/esm/src/eval/agent-service/durable-run-canaries/environment.d.ts.map +1 -0
- package/esm/src/{agent/testing → eval/agent-service}/durable-run-canaries/environment.js +1 -1
- package/esm/src/eval/agent-service/durable-run-canaries/index.d.ts.map +1 -0
- package/esm/src/eval/agent-service/durable-run-canaries/runner.d.ts.map +1 -0
- package/esm/src/eval/agent-service/durable-run-canaries/validation.d.ts.map +1 -0
- package/esm/src/eval/agent-service/live-evals/api-client.d.ts.map +1 -0
- package/esm/src/eval/agent-service/live-evals/cli-runner.d.ts.map +1 -0
- package/esm/src/{agent/testing → eval/agent-service}/live-evals/environment.d.ts +1 -1
- package/esm/src/eval/agent-service/live-evals/environment.d.ts.map +1 -0
- package/esm/src/{agent/testing → eval/agent-service}/live-evals/environment.js +1 -1
- package/esm/src/{agent/testing → eval/agent-service}/live-evals/formatting.d.ts +1 -1
- package/esm/src/eval/agent-service/live-evals/formatting.d.ts.map +1 -0
- package/esm/src/eval/agent-service/live-evals/index.d.ts.map +1 -0
- package/esm/src/eval/agent-service/live-evals/metadata.d.ts.map +1 -0
- package/esm/src/eval/agent-service/live-evals/performance.d.ts.map +1 -0
- package/esm/src/{agent/testing → eval/agent-service}/live-evals/preflight.d.ts +1 -1
- package/esm/src/eval/agent-service/live-evals/preflight.d.ts.map +1 -0
- package/esm/src/{agent/testing → eval/agent-service}/live-evals/preflight.js +1 -1
- package/esm/src/eval/agent-service/live-evals/report.d.ts.map +1 -0
- package/esm/src/eval/agent-service/live-evals/request.d.ts.map +1 -0
- package/esm/src/eval/agent-service/live-evals/result.d.ts.map +1 -0
- package/esm/src/{agent/testing → eval/agent-service}/live-evals/runner.d.ts +1 -1
- package/esm/src/eval/agent-service/live-evals/runner.d.ts.map +1 -0
- package/esm/src/{agent/testing → eval/agent-service}/live-evals/runner.js +1 -1
- package/esm/src/eval/agent-service.d.ts +90 -0
- package/esm/src/eval/agent-service.d.ts.map +1 -0
- package/esm/src/eval/agent-service.js +276 -0
- package/esm/src/eval/index.d.ts +22 -0
- package/esm/src/eval/index.d.ts.map +1 -1
- package/esm/src/eval/index.js +22 -0
- package/esm/src/utils/version-constant.d.ts +1 -1
- package/esm/src/utils/version-constant.js +1 -1
- package/package.json +5 -5
- package/esm/src/agent/testing/agent-tester.d.ts +0 -53
- package/esm/src/agent/testing/agent-tester.d.ts.map +0 -1
- package/esm/src/agent/testing/agent-tester.js +0 -132
- package/esm/src/agent/testing/durable-run-canaries/cli-runner.d.ts.map +0 -1
- package/esm/src/agent/testing/durable-run-canaries/environment.d.ts.map +0 -1
- package/esm/src/agent/testing/durable-run-canaries/index.d.ts.map +0 -1
- package/esm/src/agent/testing/durable-run-canaries/runner.d.ts.map +0 -1
- package/esm/src/agent/testing/durable-run-canaries/validation.d.ts.map +0 -1
- package/esm/src/agent/testing/index.d.ts +0 -10
- package/esm/src/agent/testing/index.d.ts.map +0 -1
- package/esm/src/agent/testing/index.js +0 -9
- package/esm/src/agent/testing/live-evals/api-client.d.ts.map +0 -1
- package/esm/src/agent/testing/live-evals/cli-runner.d.ts.map +0 -1
- package/esm/src/agent/testing/live-evals/environment.d.ts.map +0 -1
- package/esm/src/agent/testing/live-evals/formatting.d.ts.map +0 -1
- package/esm/src/agent/testing/live-evals/index.d.ts.map +0 -1
- package/esm/src/agent/testing/live-evals/metadata.d.ts.map +0 -1
- package/esm/src/agent/testing/live-evals/performance.d.ts.map +0 -1
- package/esm/src/agent/testing/live-evals/preflight.d.ts.map +0 -1
- package/esm/src/agent/testing/live-evals/report.d.ts.map +0 -1
- package/esm/src/agent/testing/live-evals/request.d.ts.map +0 -1
- package/esm/src/agent/testing/live-evals/result.d.ts.map +0 -1
- package/esm/src/agent/testing/live-evals/runner.d.ts.map +0 -1
- /package/esm/src/{agent/testing → eval/agent-service}/durable-run-canaries/cli-runner.d.ts +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/durable-run-canaries/cli-runner.js +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/durable-run-canaries/index.d.ts +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/durable-run-canaries/index.js +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/durable-run-canaries/runner.d.ts +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/durable-run-canaries/runner.js +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/durable-run-canaries/validation.d.ts +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/durable-run-canaries/validation.js +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/live-evals/api-client.d.ts +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/live-evals/api-client.js +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/live-evals/cli-runner.d.ts +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/live-evals/cli-runner.js +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/live-evals/formatting.js +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/live-evals/index.d.ts +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/live-evals/index.js +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/live-evals/metadata.d.ts +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/live-evals/metadata.js +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/live-evals/performance.d.ts +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/live-evals/performance.js +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/live-evals/report.d.ts +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/live-evals/report.js +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/live-evals/request.d.ts +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/live-evals/request.js +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/live-evals/result.d.ts +0 -0
- /package/esm/src/{agent/testing → eval/agent-service}/live-evals/result.js +0 -0
package/esm/deno.d.ts
CHANGED
|
@@ -24,8 +24,8 @@ declare namespace _default {
|
|
|
24
24
|
"./mdx": string;
|
|
25
25
|
"./agent": string;
|
|
26
26
|
"./agent/identity": string;
|
|
27
|
-
"./agent/testing": string;
|
|
28
27
|
"./eval": string;
|
|
28
|
+
"./eval/agent-service": string;
|
|
29
29
|
"./channels/control-plane": string;
|
|
30
30
|
"./channels/invoke": string;
|
|
31
31
|
"./tool": string;
|
|
@@ -93,6 +93,7 @@ declare namespace _default {
|
|
|
93
93
|
"veryfront/embedding": string;
|
|
94
94
|
"veryfront/runs": string;
|
|
95
95
|
"veryfront/eval": string;
|
|
96
|
+
"veryfront/eval/agent-service": string;
|
|
96
97
|
"veryfront/agent/identity": string;
|
|
97
98
|
"veryfront/agent/react": string;
|
|
98
99
|
"veryfront/agent/middleware": string;
|
|
@@ -138,7 +139,6 @@ declare namespace _default {
|
|
|
138
139
|
"veryfront/provider/shared": string;
|
|
139
140
|
"veryfront/tool/schema": string;
|
|
140
141
|
"veryfront/agent/composition": string;
|
|
141
|
-
"veryfront/agent/testing": string;
|
|
142
142
|
"veryfront/extensions": string;
|
|
143
143
|
"veryfront/testing": string;
|
|
144
144
|
"veryfront/testing/assert": string;
|
|
@@ -173,6 +173,7 @@ declare namespace _default {
|
|
|
173
173
|
"#veryfront/data": string;
|
|
174
174
|
"#veryfront/errors": string;
|
|
175
175
|
"#veryfront/eval": string;
|
|
176
|
+
"#veryfront/eval/agent-service": string;
|
|
176
177
|
"#veryfront/html": string;
|
|
177
178
|
"#veryfront/mcp": string;
|
|
178
179
|
"#veryfront/middleware": string;
|
package/esm/deno.js
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
export default {
|
|
2
2
|
"name": "veryfront",
|
|
3
|
-
"version": "0.1.
|
|
3
|
+
"version": "0.1.878",
|
|
4
4
|
"license": "Apache-2.0",
|
|
5
5
|
"nodeModulesDir": "auto",
|
|
6
6
|
"minimumDependencyAge": {
|
|
@@ -79,8 +79,8 @@ export default {
|
|
|
79
79
|
"./mdx": "./src/mdx/index.ts",
|
|
80
80
|
"./agent": "./src/agent/index.ts",
|
|
81
81
|
"./agent/identity": "./src/agent/identity-contracts.ts",
|
|
82
|
-
"./agent/testing": "./src/agent/testing/index.ts",
|
|
83
82
|
"./eval": "./src/eval/index.ts",
|
|
83
|
+
"./eval/agent-service": "./src/eval/agent-service.ts",
|
|
84
84
|
"./channels/control-plane": "./src/channels/control-plane.ts",
|
|
85
85
|
"./channels/invoke": "./src/channels/invoke.ts",
|
|
86
86
|
"./tool": "./src/tool/index.ts",
|
|
@@ -148,6 +148,7 @@ export default {
|
|
|
148
148
|
"veryfront/embedding": "./src/embedding/index.ts",
|
|
149
149
|
"veryfront/runs": "./src/runs/index.ts",
|
|
150
150
|
"veryfront/eval": "./src/eval/index.ts",
|
|
151
|
+
"veryfront/eval/agent-service": "./src/eval/agent-service.ts",
|
|
151
152
|
"veryfront/agent/identity": "./src/agent/identity-contracts.ts",
|
|
152
153
|
"veryfront/agent/react": "./src/agent/react/index.ts",
|
|
153
154
|
"veryfront/agent/middleware": "./src/agent/middleware/index.ts",
|
|
@@ -193,7 +194,6 @@ export default {
|
|
|
193
194
|
"veryfront/provider/shared": "./src/provider/shared/index.ts",
|
|
194
195
|
"veryfront/tool/schema": "./src/tool/schema/index.ts",
|
|
195
196
|
"veryfront/agent/composition": "./src/agent/composition/index.ts",
|
|
196
|
-
"veryfront/agent/testing": "./src/agent/testing/index.ts",
|
|
197
197
|
"veryfront/extensions": "./src/extensions/index.ts",
|
|
198
198
|
"veryfront/testing": "./src/testing/index.ts",
|
|
199
199
|
"veryfront/testing/assert": "./src/testing/assert.ts",
|
|
@@ -228,6 +228,7 @@ export default {
|
|
|
228
228
|
"#veryfront/data": "./src/data/index.ts",
|
|
229
229
|
"#veryfront/errors": "./src/errors/index.ts",
|
|
230
230
|
"#veryfront/eval": "./src/eval/index.ts",
|
|
231
|
+
"#veryfront/eval/agent-service": "./src/eval/agent-service.ts",
|
|
231
232
|
"#veryfront/html": "./src/html/index.ts",
|
|
232
233
|
"#veryfront/mcp": "./src/mcp/index.ts",
|
|
233
234
|
"#veryfront/middleware": "./src/middleware/index.ts",
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"cli-runner.d.ts","sourceRoot":"","sources":["../../../../../src/src/eval/agent-service/durable-run-canaries/cli-runner.ts"],"names":[],"mappings":"AAGA,OAAO,EAAE,KAAK,kBAAkB,EAAE,MAAM,6BAA6B,CAAC;AAEtE,OAAO,EACL,4BAA4B,EAC5B,KAAK,oBAAoB,EAEzB,KAAK,4BAA4B,EAClC,MAAM,aAAa,CAAC;AAErB,KAAK,SAAS,GAAG,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC,CAAC;AAEpD,6DAA6D;AAC7D,MAAM,WAAW,mCAAmC;IAClD,OAAO,EAAE,kBAAkB,CAAC;IAC5B,gBAAgB,EAAE,MAAM,CAAC;CAC1B;AAED,oDAAoD;AACpD,MAAM,WAAW,2BAA2B;IAC1C,GAAG,EAAE,SAAS,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,EAAE,CAAC,KAAK,EAAE,mCAAmC,KAAK,oBAAoB,EAAE,CAAC;IACpF,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,GAAG,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,IAAI,CAAC;IAChC,YAAY,CAAC,EAAE,CACb,MAAM,EAAE,4BAA4B,KACjC,UAAU,CAAC,OAAO,4BAA4B,CAAC,CAAC;CACtD;AAcD,kCAAkC;AAClC,wBAAsB,sBAAsB,CAC1C,KAAK,EAAE,2BAA2B,GACjC,OAAO,CAAC,MAAM,CAAC,CAwEjB"}
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { type AgentServiceConfigInput } from "
|
|
1
|
+
import { type AgentServiceConfigInput } from "../../../agent/service/config.js";
|
|
2
2
|
/** Public API contract for durable run canary environment. */
|
|
3
3
|
export interface DurableRunCanaryEnvironment {
|
|
4
4
|
apiUrl: string;
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"environment.d.ts","sourceRoot":"","sources":["../../../../../src/src/eval/agent-service/durable-run-canaries/environment.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,KAAK,uBAAuB,EAE7B,MAAM,kCAAkC,CAAC;AAE1C,8DAA8D;AAC9D,MAAM,WAAW,2BAA2B;IAC1C,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,gBAAgB,EAAE,MAAM,CAAC;IACzB,sBAAsB,EAAE,OAAO,CAAC;CACjC;AAED,uDAAuD;AACvD,eAAO,MAAM,qCAAqC,SAAU,CAAC;AAE7D,+CAA+C;AAC/C,wBAAgB,kCAAkC,CAChD,GAAG,GAAE,uBAA4B,GAChC,2BAA2B,CAY7B"}
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { parseAgentServiceConfig } from "
|
|
1
|
+
import { parseAgentServiceConfig, } from "../../../agent/service/config.js";
|
|
2
2
|
/** Default value for durable run canary timeout ms. */
|
|
3
3
|
export const DEFAULT_DURABLE_RUN_CANARY_TIMEOUT_MS = 240_000;
|
|
4
4
|
/** Resolves durable run canary environment. */
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../../src/src/eval/agent-service/durable-run-canaries/index.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,KAAK,mCAAmC,EACxC,sBAAsB,EACtB,KAAK,2BAA2B,GACjC,MAAM,iBAAiB,CAAC;AACzB,OAAO,EACL,qCAAqC,EACrC,KAAK,2BAA2B,EAChC,kCAAkC,GACnC,MAAM,kBAAkB,CAAC;AAC1B,OAAO,EACL,+BAA+B,EAC/B,4BAA4B,EAC5B,KAAK,yBAAyB,EAC9B,KAAK,yBAAyB,EAC9B,KAAK,oBAAoB,EACzB,KAAK,kCAAkC,EACvC,KAAK,uBAAuB,EAC5B,KAAK,4BAA4B,EACjC,KAAK,sBAAsB,EAC3B,KAAK,4BAA4B,EACjC,+BAA+B,EAC/B,KAAK,0BAA0B,EAC/B,KAAK,oCAAoC,EACzC,KAAK,6BAA6B,EAClC,gCAAgC,EAChC,+BAA+B,GAChC,MAAM,aAAa,CAAC;AAErB,OAAO,EACL,eAAe,EACf,oCAAoC,EACpC,oBAAoB,EACpB,oBAAoB,EACpB,gBAAgB,GACjB,MAAM,iBAAiB,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../../../../src/src/eval/agent-service/durable-run-canaries/runner.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,qCAAqC,CAAC;AAIvE,oDAAoD;AACpD,MAAM,WAAW,yBAAyB;IACxC,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,gBAAgB,EAAE,MAAM,CAAC;IACzB,KAAK,CAAC,EAAE,CAAC,KAAK,EAAE,MAAM,GAAG,GAAG,GAAG,OAAO,EAAE,IAAI,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,QAAQ,CAAC,CAAC;CAClF;AAED,4DAA4D;AAC5D,MAAM,WAAW,kCAAkC;IACjD,cAAc,EAAE,MAAM,CAAC;IACvB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,8DAA8D;AAC9D,MAAM,WAAW,oCAAoC;IACnD,cAAc,EAAE,MAAM,CAAC;IACvB,MAAM,EAAE,MAAM,CAAC;CAChB;AAED,sDAAsD;AACtD,MAAM,WAAW,6BAA8B,SAAQ,kCAAkC;IACvF,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,aAAa,EAAE,MAAM,CAAC;CACvB;AAED,qDAAqD;AACrD,eAAO,MAAM,gCAAgC;;;;;;;;4BAO5C,CAAC;AAEF,4CAA4C;AAC5C,MAAM,MAAM,uBAAuB,GAAG,WAAW,CAC/C,UAAU,CAAC,OAAO,gCAAgC,CAAC,CACpD,CAAC;AAEF,8DAA8D;AAC9D,MAAM,WAAW,0BAA0B;IACzC,KAAK,EAAE,MAAM,CAAC;IACd,cAAc,EAAE,MAAM,CAAC;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,aAAa,EAAE,MAAM,CAAC;IACtB,2BAA2B,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3C,iBAAiB,EAAE,MAAM,GAAG,IAAI,CAAC;IACjC,eAAe,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B,iBAAiB,EAAE,MAAM,GAAG,IAAI,CAAC;IACjC,oBAAoB,EAAE,MAAM,GAAG,IAAI,CAAC;IACpC,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;CAC3B;AA4CD,6CAA6C;AAC7C,wBAAgB,+BAA+B,CAAC,KAAK,EAAE,OAAO,GAAG,0BAA0B,CAoC1F;AAkGD,6DAA6D;AAC7D,MAAM,WAAW,yBAAyB;IACxC,oBAAoB,EAAE,CAAC,KAAK,EAAE,kCAAkC,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;IACnF,aAAa,EAAE,CAAC,KAAK,EAAE,kCAAkC,KAAK,OAAO,CAAC,0BAA0B,CAAC,CAAC;IAClG,qBAAqB,EAAE,CAAC,KAAK,EAAE;QAAE,cAAc,EAAE,MAAM,CAAA;KAAE,KAAK,OAAO,CAAC,uBAAuB,EAAE,CAAC,CAAC;IACjG,wBAAwB,EAAE,CACxB,KAAK,EAAE,oCAAoC,KACxC,OAAO,CAAC,uBAAuB,CAAC,CAAC;IACtC,eAAe,EAAE,CAAC,KAAK,EAAE,6BAA6B,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;CAC1E;AAED,4CAA4C;AAC5C,wBAAgB,+BAA+B,CAC7C,MAAM,EAAE,yBAAyB,GAChC,yBAAyB,CAgF3B;AAED,+CAA+C;AAC/C,MAAM,WAAW,sBAAsB;IACrC,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,GAAG,MAAM,CAAC;IACxB,OAAO,EAAE,MAAM,CAAC;IAChB,UAAU,EAAE,MAAM,CAAC;IACnB,cAAc,EAAE,MAAM,CAAC;IACvB,KAAK,EAAE,MAAM,CAAC;IACd,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;CAC1B;AAED,gEAAgE;AAChE,MAAM,WAAW,4BAA4B;IAC3C,aAAa,CAAC,EAAE,MAAM,EAAE,GAAG,CAAC,CAAC,KAAK,EAAE,MAAM,KAAK,MAAM,EAAE,CAAC,CAAC;IACzD,OAAO,EAAE,CAAC,KAAK,CAAC,EAAE;QAAE,KAAK,EAAE,MAAM,CAAA;KAAE,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;IACtD,cAAc,EAAE,MAAM,CAAC;IACvB,MAAM,EAAE,MAAM,CAAC;IACf,YAAY,CAAC,EAAE,MAAM,OAAO,CAAC,CAAC,MAAM,OAAO,CAAC,IAAI,CAAC,CAAC,GAAG,IAAI,CAAC,CAAC;IAC3D,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,CAAC,KAAK,EAAE;QAChB,QAAQ,EAAE,uBAAuB,EAAE,CAAC;QACpC,GAAG,EAAE,0BAA0B,CAAC;KACjC,KAAK,OAAO,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC;CAC5B;AAED,uDAAuD;AACvD,MAAM,WAAW,oBAAoB;IACnC,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,MAAM,OAAO,CAAC,4BAA4B,CAAC,CAAC;CACtD;AAED,uDAAuD;AACvD,MAAM,WAAW,4BAA6B,SAAQ,yBAAyB;IAC7E,sBAAsB,EAAE,OAAO,CAAC;CACjC;AA0DD,iBAAS,qCAAqC,CAAC,QAAQ,EAAE,uBAAuB,EAAE,GAAG,MAAM,EAAE,CAc5F;AAQD,iBAAS,mBAAmB,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAEpD;AA2CD,wCAAwC;AACxC,wBAAgB,4BAA4B,CAC1C,MAAM,EAAE,4BAA4B,EACpC,SAAS,GAAE,yBAAmE;wBAkB7C,oBAAoB,KAAG,OAAO,CAAC,sBAAsB,CAAC;EAqFxF;AAED,0DAA0D;AAC1D,eAAO,MAAM,+BAA+B;;;CAG3C,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"validation.d.ts","sourceRoot":"","sources":["../../../../../src/src/eval/agent-service/durable-run-canaries/validation.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,uBAAuB,EAAE,0BAA0B,EAAE,MAAM,aAAa,CAAC;AA8BvF,+DAA+D;AAC/D,wBAAgB,eAAe,CAAC,GAAG,EAAE,0BAA0B,GAAG,IAAI,CAQrE;AAED,wCAAwC;AACxC,wBAAgB,oBAAoB,CAClC,QAAQ,EAAE,uBAAuB,EAAE,EACnC,SAAS,EAAE,MAAM,GAChB,uBAAuB,CASzB;AAED,gCAAgC;AAChC,wBAAgB,gBAAgB,CAAC,KAAK,EAAE,OAAO,GAAG,MAAM,CAUvD;AAED,qCAAqC;AACrC,wBAAgB,oBAAoB,CAAC,OAAO,EAAE,uBAAuB,GAAG,MAAM,CAK7E;AAED,yDAAyD;AACzD,wBAAgB,oCAAoC,CAAC,QAAQ,EAAE,uBAAuB,EAAE,GAAG,IAAI,CAY9F"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"api-client.d.ts","sourceRoot":"","sources":["../../../../../src/src/eval/agent-service/live-evals/api-client.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,qCAAqC,CAAC;AAEvE,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,aAAa,CAAC;AAIvD,iCAAiC;AACjC,MAAM,WAAW,kBAAkB;IACjC,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,KAAK,CAAC,EAAE,CAAC,KAAK,EAAE,MAAM,GAAG,GAAG,GAAG,OAAO,EAAE,IAAI,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,QAAQ,CAAC,CAAC;CAClF;AAED,mDAAmD;AACnD,MAAM,WAAW,2BAA2B;IAC1C,gBAAgB,EAAE,MAAM,CAAC;CAC1B;AAED,uDAAuD;AACvD,MAAM,WAAW,+BAAgC,SAAQ,2BAA2B;IAClF,KAAK,EAAE,MAAM,CAAC;CACf;AAED,gDAAgD;AAChD,MAAM,WAAW,yBAA0B,SAAQ,2BAA2B;IAC5E,cAAc,EAAE,MAAM,CAAC;CACxB;AAED,0DAA0D;AAC1D,MAAM,WAAW,iCAAkC,SAAQ,2BAA2B;IACpF,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,IAAI,EAAE,QAAQ,GAAG,UAAU,CAAC;IAC5B,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAED,gDAAgD;AAChD,MAAM,WAAW,wBAAyB,SAAQ,2BAA2B;IAC3E,QAAQ,EAAE,MAAM,CAAC;CAClB;AAED,kDAAkD;AAClD,MAAM,WAAW,0BAA2B,SAAQ,2BAA2B;IAC7E,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAED,+DAA+D;AAC/D,MAAM,WAAW,oCAAqC,SAAQ,yBAAyB;IACrF,WAAW,EAAE,WAAW,CAAC;IACzB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,+DAA+D;AAC/D,MAAM,WAAW,2BAA2B;IAC1C,CAAC,GAAG,EAAE,MAAM,GAAG,MAAM,GAAG,OAAO,GAAG,MAAM,GAAG,IAAI,CAAC;CACjD;AAED,yDAAyD;AACzD,MAAM,WAAW,gCAAiC,SAAQ,2BAA2B;IACnF,cAAc,EAAE,MAAM,CAAC;IACvB,cAAc,EAAE,MAAM,CAAC;IACvB,MAAM,EAAE,2BAA2B,CAAC;CACrC;AAED,iDAAiD;AACjD,MAAM,WAAW,yBAA0B,SAAQ,2BAA2B;IAC5E,cAAc,EAAE,MAAM,CAAC;IACvB,cAAc,EAAE,MAAM,CAAC;CACxB;AA4BD,QAAA,MAAM,2BAA2B;;;GAKhC,CAAC;AAQF,gDAAgD;AAChD,MAAM,MAAM,0BAA0B,GAAG,WAAW,CAClD,UAAU,CAAC,OAAO,2BAA2B,CAAC,CAC/C,CAAC;AAEF,oDAAoD;AACpD,MAAM,WAAW,iBAAiB;IAChC,kBAAkB,CAAC,KAAK,EAAE,+BAA+B,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;IAC5E,kBAAkB,CAAC,KAAK,EAAE,yBAAyB,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IACpE,0BAA0B,CAAC,KAAK,EAAE,iCAAiC,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;IACtF,cAAc,CAAC,KAAK,EAAE,wBAAwB,GAAG,OAAO,CAAC,mBAAmB,GAAG,IAAI,CAAC,CAAC;IACrF,aAAa,CAAC,KAAK,EAAE,0BAA0B,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;IAClE,iBAAiB,CAAC,KAAK,EAAE,wBAAwB,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAClE,qBAAqB,CAAC,KAAK,EAAE,yBAAyB,GAAG,OAAO,CAAC,0BAA0B,EAAE,CAAC,CAAC;IAC/F,uBAAuB,CAAC,KAAK,EAAE,oCAAoC,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;IACtF,mBAAmB,CAAC,KAAK,EAAE,gCAAgC,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC5E,kBAAkB,CAAC,KAAK,EAAE,yBAAyB,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;CACrE;AAsID,mCAAmC;AACnC,wBAAgB,uBAAuB,CAAC,OAAO,EAAE,kBAAkB,GAAG,iBAAiB,CAatF;AAED,qCAAqC;AACrC,wBAAsB,0BAA0B,CAC9C,OAAO,EAAE,kBAAkB,EAC3B,KAAK,EAAE,+BAA+B,GACrC,OAAO,CAAC,MAAM,CAAC,CAuBjB;AAED,4CAA4C;AAC5C,wBAAsB,0BAA0B,CAC9C,OAAO,EAAE,kBAAkB,EAC3B,KAAK,EAAE,yBAAyB,GAC/B,OAAO,CAAC,IAAI,CAAC,CAiBf;AAED,+CAA+C;AAC/C,wBAAsB,kCAAkC,CACtD,OAAO,EAAE,kBAAkB,EAC3B,KAAK,EAAE,iCAAiC,GACvC,OAAO,CAAC,MAAM,CAAC,CAuDjB;AAED,qCAAqC;AACrC,wBAAsB,sBAAsB,CAC1C,OAAO,EAAE,kBAAkB,EAC3B,KAAK,EAAE,wBAAwB,GAC9B,OAAO,CAAC,mBAAmB,GAAG,IAAI,CAAC,CA4BrC;AAED,gCAAgC;AAChC,wBAAsB,qBAAqB,CACzC,OAAO,EAAE,kBAAkB,EAC3B,KAAK,EAAE,0BAA0B,GAChC,OAAO,CAAC,MAAM,CAAC,CA6BjB;AAED,4CAA4C;AAC5C,wBAAsB,yBAAyB,CAC7C,OAAO,EAAE,kBAAkB,EAC3B,KAAK,EAAE,wBAAwB,GAC9B,OAAO,CAAC,IAAI,CAAC,CAoBf;AAED,0CAA0C;AAC1C,wBAAsB,6BAA6B,CACjD,OAAO,EAAE,kBAAkB,EAC3B,KAAK,EAAE,yBAAyB,GAC/B,OAAO,CAAC,0BAA0B,EAAE,CAAC,CAoBvC;AAED,yDAAyD;AACzD,wBAAsB,+BAA+B,CACnD,OAAO,EAAE,kBAAkB,EAC3B,KAAK,EAAE,oCAAoC,GAC1C,OAAO,CAAC,MAAM,CAAC,CAsBjB;AAED,mDAAmD;AACnD,wBAAsB,2BAA2B,CAC/C,OAAO,EAAE,kBAAkB,EAC3B,KAAK,EAAE,gCAAgC,GACtC,OAAO,CAAC,IAAI,CAAC,CAmBf;AAED,kDAAkD;AAClD,wBAAsB,0BAA0B,CAC9C,OAAO,EAAE,kBAAkB,EAC3B,KAAK,EAAE,yBAAyB,GAC/B,OAAO,CAAC,IAAI,CAAC,CAkBf"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"cli-runner.d.ts","sourceRoot":"","sources":["../../../../../src/src/eval/agent-service/live-evals/cli-runner.ts"],"names":[],"mappings":"AAGA,OAAO,EAAkC,KAAK,eAAe,EAAE,MAAM,kBAAkB,CAAC;AAQxF,OAAO,EACL,iBAAiB,EACjB,sBAAsB,EACtB,yBAAyB,EACzB,WAAW,EACX,KAAK,YAAY,EACjB,KAAK,oBAAoB,EAC1B,MAAM,aAAa,CAAC;AAKrB,KAAK,SAAS,GAAG,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC,CAAC;AAEpD,yDAAyD;AACzD,MAAM,WAAW,qBAAqB;IACpC,aAAa,EAAE,YAAY,EAAE,CAAC;IAC9B,UAAU,EAAE,YAAY,EAAE,CAAC;IAC3B,sBAAsB,EAAE,YAAY,EAAE,CAAC;CACxC;AAED,oDAAoD;AACpD,MAAM,WAAW,2BAA2B;IAC1C,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,gBAAgB,EAAE,MAAM,CAAC;IACzB,cAAc,EAAE,OAAO,CAAC;IACxB,WAAW,EAAE,OAAO,WAAW,CAAC;IAChC,iBAAiB,EAAE,OAAO,iBAAiB,CAAC;IAC5C,sBAAsB,EAAE,OAAO,sBAAsB,CAAC;IACtD,gBAAgB,EAAE,UAAU,CAAC,OAAO,yBAAyB,CAAC,CAAC,kBAAkB,CAAC,CAAC;IACnF,SAAS,EAAE,UAAU,CAAC,OAAO,yBAAyB,CAAC,CAAC,WAAW,CAAC,CAAC;IACrE,QAAQ,EAAE,UAAU,CAAC,OAAO,yBAAyB,CAAC,CAAC,UAAU,CAAC,CAAC;CACpE;AAED,2CAA2C;AAC3C,MAAM,WAAW,mBAAmB;IAClC,GAAG,EAAE,SAAS,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,SAAS,MAAM,EAAE,CAAC,CAAC;IAC5C,WAAW,EAAE,CAAC,KAAK,EAAE,2BAA2B,KAAK,qBAAqB,CAAC;IAC3E,QAAQ,CAAC,EAAE,SAAS,eAAe,EAAE,CAAC;IACtC,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,GAAG,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,IAAI,CAAC;IAChC,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,IAAI,CAAC;IAClC,iBAAiB,CAAC,EAAE,CAClB,MAAM,EAAE,oBAAoB,KACzB,UAAU,CAAC,OAAO,yBAAyB,CAAC,CAAC;CACnD;AAuBD,yBAAyB;AACzB,wBAAsB,cAAc,CAAC,KAAK,EAAE,mBAAmB,GAAG,OAAO,CAAC,MAAM,CAAC,CAwJhF"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"environment.d.ts","sourceRoot":"","sources":["../../../../../src/src/eval/agent-service/live-evals/environment.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,KAAK,uBAAuB,EAE7B,MAAM,kCAAkC,CAAC;AAE1C,qDAAqD;AACrD,MAAM,WAAW,mBAAmB;IAClC,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,GAAG,SAAS,CAAC;IAC9B,QAAQ,EAAE,MAAM,GAAG,SAAS,CAAC;IAC7B,KAAK,EAAE,MAAM,GAAG,SAAS,CAAC;CAC3B;AAED,4CAA4C;AAC5C,eAAO,MAAM,0BAA0B,oCAAoC,CAAC;AAE5E,sCAAsC;AACtC,wBAAgB,0BAA0B,CACxC,GAAG,GAAE,uBAA4B,GAChC,mBAAmB,CAerB"}
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { parseAgentServiceConfig } from "
|
|
1
|
+
import { parseAgentServiceConfig, } from "../../../agent/service/config.js";
|
|
2
2
|
/** Default value for live eval endpoint. */
|
|
3
3
|
export const DEFAULT_LIVE_EVAL_ENDPOINT = "http://127.0.0.1:3001/api/ag-ui";
|
|
4
4
|
/** Resolves live eval environment. */
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import { Buffer } from "node:buffer";
|
|
2
|
-
import type { AgUiSseProgressSnapshot as EvalProgressSnapshot } from "
|
|
2
|
+
import type { AgUiSseProgressSnapshot as EvalProgressSnapshot } from "../../../agent/index.js";
|
|
3
3
|
/** Create plain text pdf. */
|
|
4
4
|
export declare function createPlainTextPdf(lines: string[]): Buffer;
|
|
5
5
|
/** Builds progress line. */
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"formatting.d.ts","sourceRoot":"","sources":["../../../../../src/src/eval/agent-service/live-evals/formatting.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC;AACrC,OAAO,KAAK,EAAE,uBAAuB,IAAI,oBAAoB,EAAE,MAAM,yBAAyB,CAAC;AAS/F,6BAA6B;AAC7B,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,MAAM,CA4C1D;AAED,4BAA4B;AAC5B,wBAAgB,iBAAiB,CAAC,KAAK,EAAE;IACvC,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,EAAE,oBAAoB,CAAC;CAChC,GAAG,MAAM,CAcT;AAED,6BAA6B;AAC7B,wBAAgB,kBAAkB,CAAC,QAAQ,EAAE,oBAAoB,GAAG,MAAM,CAUzE;AAED,2CAA2C;AAC3C,wBAAgB,0BAA0B,CACxC,QAAQ,EAAE,MAAM,EAAE,EAClB,MAAM,EAAE,MAAM,EAAE,GACf,OAAO,CAcT"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../../src/src/eval/agent-service/live-evals/index.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,KAAK,2BAA2B,EAChC,KAAK,qBAAqB,EAC1B,cAAc,EACd,KAAK,mBAAmB,GACzB,MAAM,iBAAiB,CAAC;AACzB,OAAO,EACL,0BAA0B,EAC1B,KAAK,mBAAmB,EACxB,0BAA0B,GAC3B,MAAM,kBAAkB,CAAC;AAC1B,OAAO,EACL,0BAA0B,EAC1B,uBAAuB,EACvB,0BAA0B,EAC1B,kCAAkC,EAClC,qBAAqB,EACrB,0BAA0B,EAC1B,yBAAyB,EACzB,sBAAsB,EACtB,6BAA6B,EAC7B,KAAK,iBAAiB,EACtB,KAAK,kBAAkB,EACvB,KAAK,yBAAyB,EAC9B,KAAK,+BAA+B,EACpC,KAAK,0BAA0B,EAC/B,KAAK,yBAAyB,EAC9B,KAAK,0BAA0B,EAC/B,KAAK,2BAA2B,EAChC,KAAK,wBAAwB,EAC7B,KAAK,iCAAiC,EACtC,KAAK,2BAA2B,EAChC,KAAK,gCAAgC,EACrC,KAAK,oCAAoC,EACzC,2BAA2B,EAC3B,+BAA+B,GAChC,MAAM,iBAAiB,CAAC;AACzB,OAAO,EACL,kBAAkB,EAClB,iBAAiB,EACjB,0BAA0B,EAC1B,kBAAkB,GACnB,MAAM,iBAAiB,CAAC;AACzB,OAAO,EACL,4BAA4B,EAC5B,+BAA+B,EAC/B,KAAK,gCAAgC,GACtC,MAAM,gBAAgB,CAAC;AACxB,OAAO,EACL,8BAA8B,EAC9B,KAAK,4BAA4B,EACjC,KAAK,eAAe,EACpB,KAAK,yBAAyB,GAC/B,MAAM,kBAAkB,CAAC;AAC1B,OAAO,EACL,yBAAyB,EACzB,KAAK,8BAA8B,EACnC,gCAAgC,EAChC,8CAA8C,EAC9C,KAAK,2BAA2B,EAChC,KAAK,mBAAmB,EACxB,KAAK,mBAAmB,EACxB,oBAAoB,GACrB,MAAM,eAAe,CAAC;AACvB,OAAO,EACL,wBAAwB,EACxB,KAAK,6BAA6B,EAClC,KAAK,mBAAmB,GACzB,MAAM,cAAc,CAAC;AACtB,OAAO,EACL,2BAA2B,EAC3B,2BAA2B,EAC3B,0BAA0B,EAC1B,mBAAmB,EACnB,KAAK,oBAAoB,EACzB,KAAK,0BAA0B,EAC/B,KAAK,uBAAuB,EAC5B,+BAA+B,EAC/B,mBAAmB,GACpB,MAAM,aAAa,CAAC;AACrB,OAAO,EACL,sBAAsB,EACtB,sBAAsB,EACtB,uBAAuB,EACvB,KAAK,oBAAoB,GAC1B,MAAM,aAAa,CAAC;AACrB,OAAO,EACL,iBAAiB,EACjB,sBAAsB,EACtB,yBAAyB,EACzB,WAAW,EACX,KAAK,YAAY,EACjB,KAAK,eAAe,EACpB,KAAK,mBAAmB,EACxB,KAAK,8BAA8B,EACnC,KAAK,oBAAoB,EACzB,uBAAuB,EACvB,KAAK,qBAAqB,GAC3B,MAAM,aAAa,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"metadata.d.ts","sourceRoot":"","sources":["../../../../../src/src/eval/agent-service/live-evals/metadata.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AACxD,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAEhD,sDAAsD;AACtD,MAAM,MAAM,mBAAmB,GAAG,WAAW,GAAG,OAAO,GAAG,cAAc,CAAC;AAEzE,uDAAuD;AACvD,MAAM,WAAW,mBAAmB;IAClC,GAAG,EAAE,MAAM,CAAC;IACZ,MAAM,CAAC,EAAE,MAAM,GAAG,SAAS,MAAM,EAAE,CAAC;IACpC,UAAU,CAAC,EAAE,MAAM,GAAG,SAAS,MAAM,EAAE,CAAC;IACxC,QAAQ,CAAC,EAAE,MAAM,GAAG,SAAS,MAAM,EAAE,CAAC;CACvC;AAED,mDAAmD;AACnD,MAAM,WAAW,2BAA2B;IAC1C,kBAAkB,CAAC,EAAE,SAAS,MAAM,EAAE,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC;IAC7D,yBAAyB,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;IAC9C,YAAY,CAAC,EAAE,SAAS,mBAAmB,EAAE,CAAC;CAC/C;AAED,uDAAuD;AACvD,MAAM,WAAW,8BAA+B,SAAQ,2BAA2B;IACjF,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,mBAAmB,CAAC;IAC7B,cAAc,EAAE,OAAO,CAAC;CACzB;AAED,gEAAgE;AAChE,eAAO,MAAM,8CAA8C,EAAE,SAAS,MAAM,EAI3E,CAAC;AAEF,kDAAkD;AAClD,eAAO,MAAM,gCAAgC,EAAE,SAAS,mBAAmB,EAqB1E,CAAC;AA+DF,sCAAsC;AACtC,wBAAgB,yBAAyB,CACvC,KAAK,EAAE,8BAA8B,GACpC,oBAAoB,CAkCtB;AAED,kCAAkC;AAClC,wBAAgB,oBAAoB,CAAC,KAAK,SAAS,YAAY,EAC7D,KAAK,EAAE,SAAS,KAAK,EAAE,EACvB,OAAO,EAAE,mBAAmB,EAC5B,OAAO,GAAE,2BAAgC,GACxC,KAAK,CAAC,KAAK,GAAG;IAAE,QAAQ,EAAE,oBAAoB,CAAA;CAAE,CAAC,CAUnD"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"performance.d.ts","sourceRoot":"","sources":["../../../../../src/src/eval/agent-service/live-evals/performance.ts"],"names":[],"mappings":"AAAA,iDAAiD;AACjD,MAAM,MAAM,eAAe,GAAG,WAAW,CAAC;AAE1C,gEAAgE;AAChE,MAAM,WAAW,4BAA4B;IAC3C,OAAO,EAAE,eAAe,CAAC;IACzB,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,2DAA2D;AAC3D,MAAM,WAAW,yBAAyB;IACxC,KAAK,EAAE,MAAM,CAAC;IACd,aAAa,EAAE,MAAM,CAAC;IACtB,aAAa,EAAE,MAAM,CAAC;IACtB,aAAa,EAAE,MAAM,CAAC;IACtB,aAAa,EAAE,MAAM,CAAC;IACtB,aAAa,EAAE,MAAM,CAAC;CACvB;AAkBD,0CAA0C;AAC1C,wBAAgB,8BAA8B,CAC5C,OAAO,EAAE,4BAA4B,EAAE,GACtC,MAAM,CAAC,eAAe,EAAE,yBAAyB,CAAC,CAiBpD"}
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { type AgentServiceConfigInput } from "
|
|
1
|
+
import { type AgentServiceConfigInput } from "../../../agent/service/config.js";
|
|
2
2
|
/** Result returned from runtime confidence preflight. */
|
|
3
3
|
export interface RuntimeConfidencePreflightResult {
|
|
4
4
|
ok: boolean;
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"preflight.d.ts","sourceRoot":"","sources":["../../../../../src/src/eval/agent-service/live-evals/preflight.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,KAAK,uBAAuB,EAE7B,MAAM,kCAAkC,CAAC;AAE1C,yDAAyD;AACzD,MAAM,WAAW,gCAAgC;IAC/C,EAAE,EAAE,OAAO,CAAC;IACZ,cAAc,EAAE,MAAM,CAAC;IACvB,QAAQ,EAAE,MAAM,EAAE,CAAC;CACpB;AAED,8CAA8C;AAC9C,wBAAgB,4BAA4B,CAC1C,GAAG,GAAE,uBAA4B,EACjC,cAAc,GAAE,MAAuD,GACtE,gCAAgC,CAoBlC;AAED,iDAAiD;AACjD,wBAAgB,+BAA+B,CAC7C,MAAM,EAAE,gCAAgC,EACxC,MAAM,GAAE,IAAI,CAAC,OAAO,EAAE,OAAO,GAAG,KAAK,CAAW,GAC/C,IAAI,CAKN"}
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { parseAgentServiceConfig } from "
|
|
1
|
+
import { parseAgentServiceConfig, } from "../../../agent/service/config.js";
|
|
2
2
|
/** Evaluate runtime confidence env helper. */
|
|
3
3
|
export function evaluateRuntimeConfidenceEnv(env = {}, resolvedApiUrl = parseAgentServiceConfig(env).VERYFRONT_API_URL) {
|
|
4
4
|
const messages = [`Resolved VERYFRONT_API_URL: ${resolvedApiUrl}`];
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"report.d.ts","sourceRoot":"","sources":["../../../../../src/src/eval/agent-service/live-evals/report.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,kBAAkB,CAAC;AAExD,uDAAuD;AACvD,MAAM,WAAW,oBAAoB;IACnC,IAAI,EAAE,SAAS,MAAM,EAAE,CAAC;CACzB;AAED,2DAA2D;AAC3D,MAAM,WAAW,uBAAuB;IACtC,OAAO,EAAE,eAAe,CAAC;IACzB,MAAM,EAAE,MAAM,GAAG,MAAM,GAAG,MAAM,CAAC;CAClC;AAED,kDAAkD;AAClD,MAAM,WAAW,0BAA0B,CAAC,KAAK,SAAS;IAAE,EAAE,EAAE,MAAM,CAAA;CAAE;IACtE,QAAQ,EAAE,KAAK,EAAE,CAAC;IAClB,aAAa,EAAE,KAAK,EAAE,CAAC;IACvB,UAAU,EAAE,KAAK,EAAE,CAAC;IACpB,sBAAsB,EAAE,KAAK,EAAE,CAAC;IAChC,gBAAgB,EAAE,GAAG,CAAC,MAAM,CAAC,CAAC;IAC9B,iBAAiB,CAAC,EAAE,GAAG,CAAC,MAAM,CAAC,CAAC;IAChC,aAAa,EAAE,OAAO,CAAC;IACvB,yBAAyB,EAAE,OAAO,CAAC;CACpC;AAED,oDAAoD;AACpD,wBAAgB,mBAAmB,CACjC,IAAI,EAAE,SAAS,MAAM,EAAE,EACvB,aAAa,EAAE,GAAG,CAAC,MAAM,CAAC,GACzB,OAAO,CAQT;AAED,yCAAyC;AACzC,wBAAgB,2BAA2B,CACzC,KAAK,EAAE,SAAS;IACd,QAAQ,CAAC,EAAE,oBAAoB,CAAC;CACjC,EAAE,GACF,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAaxB;AAED,qCAAqC;AACrC,wBAAgB,mBAAmB,CACjC,KAAK,SAAS;IAAE,EAAE,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,EAAE,oBAAoB,CAAA;CAAE,EAE7D,KAAK,EAAE,0BAA0B,CAAC,KAAK,CAAC,GACvC,KAAK,EAAE,CAoBT;AAED,6CAA6C;AAC7C,wBAAgB,+BAA+B,CAAC,KAAK,EAAE;IACrD,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,SAAS,MAAM,EAAE,CAAC,CAAC;IAC5C,gBAAgB,EAAE,GAAG,CAAC,MAAM,CAAC,CAAC;IAC9B,kBAAkB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CACpC,GAAG,GAAG,CAAC,MAAM,CAAC,CAsBd;AAED,UAAU,qBAAqB;IAC7B,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;CACjB;AAgBD,wCAAwC;AACxC,wBAAgB,2BAA2B,CACzC,QAAQ,EAAE,SAAS,eAAe,EAAE,EACpC,OAAO,EAAE,uBAAuB,EAAE,GACjC,MAAM,CAAC,eAAe,EAAE,qBAAqB,CAAC,CAShD;AAED,uCAAuC;AACvC,wBAAgB,0BAA0B,CACxC,OAAO,EAAE,uBAAuB,EAAE,GACjC;IACD,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;CACjB,CAMA"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"request.d.ts","sourceRoot":"","sources":["../../../../../src/src/eval/agent-service/live-evals/request.ts"],"names":[],"mappings":"AAGA,MAAM,WAAW,mBAAmB;IAClC,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC9B,KAAK,EAAE,OAAO,EAAE,CAAC;IACjB,OAAO,EAAE,OAAO,EAAE,CAAC;IACnB,cAAc,CAAC,EAAE;QACf,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;KACpC,CAAC;IACF,QAAQ,EAAE,KAAK,CAAC;QACd,EAAE,EAAE,MAAM,CAAC;QACX,IAAI,EAAE,MAAM,CAAC;QACb,OAAO,EAAE,MAAM,CAAC;KACjB,CAAC,CAAC;CACJ;AAED,sDAAsD;AACtD,MAAM,WAAW,6BAA6B;IAC5C,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAClC,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,cAAc,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB,qBAAqB,CAAC,EAAE,OAAO,CAAC;IAChC,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAED,qCAAqC;AACrC,wBAAgB,wBAAwB,CACtC,KAAK,EAAE,6BAA6B,GACnC,mBAAmB,CA6CrB"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"result.d.ts","sourceRoot":"","sources":["../../../../../src/src/eval/agent-service/live-evals/result.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,kBAAkB,CAAC;AAExD,yCAAyC;AACzC,MAAM,WAAW,oBAAoB;IACnC,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,eAAe,CAAC;IACzB,MAAM,EAAE,MAAM,GAAG,MAAM,GAAG,MAAM,CAAC;IACjC,OAAO,EAAE,MAAM,CAAC;IAChB,UAAU,EAAE,MAAM,CAAC;IACnB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;IACzB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,UAAU,CAAC,EAAE,MAAM,EAAE,CAAC;IACtB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AA2CD,gDAAgD;AAChD,wBAAgB,uBAAuB,CAAC,KAAK,EAAE;IAC7C,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,eAAe,CAAC;IACzB,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;CACnB,GAAG,oBAAoB,CASvB;AAED,+CAA+C;AAC/C,wBAAgB,sBAAsB,CAAC,KAAK,EAAE;IAC5C,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,eAAe,CAAC;IACzB,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;IAClB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;IACzB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,UAAU,CAAC,EAAE,MAAM,EAAE,CAAC;IACtB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB,GAAG,oBAAoB,CAEvB;AAED,+CAA+C;AAC/C,wBAAgB,sBAAsB,CAAC,KAAK,EAAE;IAC5C,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,eAAe,CAAC;IACzB,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;IAClB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;IACzB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,UAAU,CAAC,EAAE,MAAM,EAAE,CAAC;IACtB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB,GAAG,oBAAoB,CAEvB"}
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { type AgUiSseProgressSnapshot as EvalProgressSnapshot, type ParsedAgUiSseRun as ParsedRun } from "
|
|
1
|
+
import { type AgUiSseProgressSnapshot as EvalProgressSnapshot, type ParsedAgUiSseRun as ParsedRun } from "../../../agent/index.js";
|
|
2
2
|
import { type LiveEvalRuntime } from "./performance.js";
|
|
3
3
|
import { type LiveEvalCaseMetadata } from "./report.js";
|
|
4
4
|
import { type LiveEvalResultRecord } from "./result.js";
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../../../../src/src/eval/agent-service/live-evals/runner.ts"],"names":[],"mappings":"AACA,OAAO,EAEL,KAAK,uBAAuB,IAAI,oBAAoB,EAIpD,KAAK,gBAAgB,IAAI,SAAS,EACnC,MAAM,yBAAyB,CAAC;AAEjC,OAAO,EAAE,KAAK,eAAe,EAAE,MAAM,kBAAkB,CAAC;AAExD,OAAO,EAAE,KAAK,oBAAoB,EAAE,MAAM,aAAa,CAAC;AACxD,OAAO,EAIL,KAAK,oBAAoB,EAC1B,MAAM,aAAa,CAAC;AAErB,4CAA4C;AAC5C,MAAM,WAAW,qBAAqB;IACpC,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAClC,mBAAmB,CAAC,EAAE,eAAe,CAAC;IACtC,OAAO,CAAC,EAAE,MAAM,OAAO,CAAC,IAAI,CAAC,CAAC;IAC9B,YAAY,CAAC,EAAE,MAAM,OAAO,CAAC,CAAC,MAAM,OAAO,CAAC,IAAI,CAAC,CAAC,GAAG,IAAI,CAAC,CAAC;CAC5D;AAED,6BAA6B;AAC7B,MAAM,WAAW,eAAe;IAC9B,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;CAC1B;AAED,8CAA8C;AAC9C,MAAM,WAAW,YAAY;IAC3B,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IACpB,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM,CAAC;IACzB,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB,qBAAqB,CAAC,EAAE,OAAO,CAAC;IAChC,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,wBAAwB,CAAC,EAAE,MAAM,EAAE,CAAC;IACpC,QAAQ,CAAC,EAAE,oBAAoB,CAAC;IAChC,OAAO,CAAC,EAAE,CAAC,OAAO,EAAE,eAAe,KAAK,OAAO,CAAC,qBAAqB,CAAC,CAAC;IACvE,MAAM,EAAE,CACN,GAAG,EAAE,SAAS,EACd,QAAQ,EAAE,qBAAqB,GAAG,IAAI,KACnC,MAAM,GAAG,IAAI,GAAG,OAAO,CAAC,MAAM,GAAG,IAAI,CAAC,CAAC;CAC7C;AAED,UAAU,cAAc;IACtB,QAAQ,EAAE,MAAM,CAAC;IACjB,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;IAC3B,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAED,sDAAsD;AACtD,MAAM,WAAW,mBAAmB;IAClC,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,uDAAuD;AACvD,MAAM,WAAW,8BAA8B;IAC7C,QAAQ,EAAE,MAAM,CAAC;IACjB,gBAAgB,EAAE,MAAM,CAAC;CAC1B;AAED,8CAA8C;AAC9C,MAAM,WAAW,oBAAoB;IACnC,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,gBAAgB,EAAE,MAAM,CAAC;IACzB,qBAAqB,EAAE,MAAM,CAAC;IAC9B,cAAc,EAAE,OAAO,CAAC;IACxB,KAAK,CAAC,EAAE,CAAC,KAAK,EAAE,MAAM,GAAG,GAAG,GAAG,OAAO,EAAE,IAAI,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,QAAQ,CAAC,CAAC;IACjF,GAAG,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,IAAI,CAAC;IAChC,eAAe,CAAC,EAAE,CAAC,KAAK,EAAE,8BAA8B,KAAK,OAAO,CAAC,mBAAmB,GAAG,IAAI,CAAC,CAAC;CAClG;AAED,UAAU,kBAAkB;IAC1B,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;CAClB;AAED,UAAU,oBAAqB,SAAQ,kBAAkB;IACvD,MAAM,EAAE,MAAM,CAAC;CAChB;AAED,UAAU,mBAAmB;IAC3B,IAAI,EAAE,OAAO,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;CAChB;AAyJD,iBAAS,4BAA4B,CAAC,QAAQ,EAAE,qBAAqB,GAAG,IAAI,GAAG,MAAM,EAAE,CAYtF;AASD,UAAU,qBAAqB;IAC7B,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,eAAe,CAAC;IACzB,SAAS,EAAE,MAAM,CAAC;IAClB,cAAc,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;CAC1B;AAED,UAAU,yBAAyB;IACjC,GAAG,EAAE,SAAS,CAAC;IACf,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,cAAc,EAAE,MAAM,CAAC;CACxB;AAED,UAAU,oBAAoB;IAC5B,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,cAAc,EAAE,MAAM,CAAC;IACvB,UAAU,EAAE,MAAM,EAAE,CAAC;IACrB,eAAe,EAAE,MAAM,CAAC;IACxB,WAAW,EAAE,MAAM,CAAC;CACrB;AAED,iBAAS,0BAA0B,CAAC,KAAK,EAAE,yBAAyB,GAAG,oBAAoB,CAQ1F;AAED,iBAAS,yBAAyB,CAAC,KAAK,EAAE;IACxC,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,EAAE,qBAAqB,CAAC;IAC/B,YAAY,EAAE,oBAAoB,CAAC;CACpC,GAAG,oBAAoB,CAevB;AAED,iBAAS,yBAAyB,CAAC,KAAK,EAAE;IACxC,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,EAAE,qBAAqB,CAAC;IAC/B,YAAY,EAAE,oBAAoB,CAAC;CACpC,GAAG,oBAAoB,CAevB;AAED,iBAAS,gCAAgC,CAAC,KAAK,EAAE;IAC/C,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,EAAE,qBAAqB,CAAC;IAC/B,QAAQ,EAAE,oBAAoB,CAAC;CAChC,GAAG,oBAAoB,CAcvB;AA+FD,iBAAS,YAAY,CAAC,GAAG,EAAE,SAAS,GAAG,MAAM,GAAG,IAAI,CASnD;AAED,yCAAyC;AACzC,wBAAgB,WAAW,CAAC,GAAG,EAAE,SAAS,GAAG,OAAO,CAEnD;AAED,kCAAkC;AAClC,wBAAgB,iBAAiB,CAAC,GAAG,EAAE,SAAS,EAAE,OAAO,EAAE,MAAM,GAAG,OAAO,CAE1E;AAED,wCAAwC;AACxC,wBAAgB,sBAAsB,CAAC,GAAG,EAAE,SAAS,GAAG,MAAM,CAE7D;AAED,qCAAqC;AACrC,wBAAgB,yBAAyB,CAAC,MAAM,EAAE,oBAAoB,GAAG;IACvE,OAAO,EAAE,CAAC,QAAQ,EAAE,YAAY,EAAE,OAAO,EAAE,eAAe,KAAK,OAAO,CAAC,oBAAoB,CAAC,CAAC;IAC7F,gBAAgB,EAAE,CAAC,KAAK,EAAE,cAAc,KAAK,OAAO,CAAC,MAAM,GAAG,IAAI,CAAC,CAAC;IACpE,SAAS,EAAE,CACT,gBAAgB,EAAE,CAAC,GAAG,EAAE,SAAS,KAAK,MAAM,GAAG,IAAI,EACnD,UAAU,EAAE,kBAAkB,KAC3B,CAAC,GAAG,EAAE,SAAS,KAAK,OAAO,CAAC,MAAM,GAAG,IAAI,CAAC,CAAC;IAChD,QAAQ,EAAE,CAAC,KAAK,EAAE,oBAAoB,KAAK,OAAO,CAAC,mBAAmB,CAAC,CAAC;CACzE,CAoIA;AAED,wDAAwD;AACxD,eAAO,MAAM,uBAAuB;;;;;;;CAOnC,CAAC"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import * as dntShim from "../../../../_dnt.shims.js";
|
|
2
|
-
import { agUiSseEventTypes, buildAgUiSseTraceSignature as buildTraceSignature, getAgUiSseStringField as getStringField, parseAgUiSseResponse as parseSseResponse, } from "
|
|
2
|
+
import { agUiSseEventTypes, buildAgUiSseTraceSignature as buildTraceSignature, getAgUiSseStringField as getStringField, parseAgUiSseResponse as parseSseResponse, } from "../../../agent/index.js";
|
|
3
3
|
import { buildFailureSuffix, buildProgressLine, containsOrderedSubsequence } from "./formatting.js";
|
|
4
4
|
import { buildLiveEvalRequestBody } from "./request.js";
|
|
5
5
|
import { createFailedEvalResult, createPassedEvalResult, createSkippedEvalResult, } from "./result.js";
|
|
@@ -0,0 +1,90 @@
|
|
|
1
|
+
import "../../_dnt.polyfills.js";
|
|
2
|
+
import { type AgUiSseProgressSnapshot } from "../agent/index.js";
|
|
3
|
+
import type { EvalAgentAdapter, EvalAgentAdapterContext } from "./types.js";
|
|
4
|
+
export * from "./agent-service/live-evals/index.js";
|
|
5
|
+
export * from "./agent-service/durable-run-canaries/index.js";
|
|
6
|
+
/** Default local AG-UI endpoint used by agent-service evals. */
|
|
7
|
+
export declare const DEFAULT_AGENT_SERVICE_EVAL_ENDPOINT = "http://127.0.0.1:3001/api/ag-ui";
|
|
8
|
+
/** Environment input accepted by agent-service eval helpers. */
|
|
9
|
+
export type AgentServiceEvalEnvironmentInput = Record<string, string | number | undefined>;
|
|
10
|
+
/** Resolved environment values for live agent-service evals. */
|
|
11
|
+
export interface AgentServiceEvalEnvironment {
|
|
12
|
+
endpoint: string;
|
|
13
|
+
authToken: string;
|
|
14
|
+
apiUrl: string;
|
|
15
|
+
projectId?: string;
|
|
16
|
+
branchId?: string;
|
|
17
|
+
model?: string;
|
|
18
|
+
}
|
|
19
|
+
/** Preflight result for a live agent-service eval environment. */
|
|
20
|
+
export interface AgentServiceEvalEnvironmentPreflightResult {
|
|
21
|
+
ok: boolean;
|
|
22
|
+
resolvedApiUrl: string;
|
|
23
|
+
messages: string[];
|
|
24
|
+
}
|
|
25
|
+
/** Veryfront forwarded props included in an AG-UI eval request. */
|
|
26
|
+
export interface AgentServiceEvalForwardedProps {
|
|
27
|
+
projectId?: string;
|
|
28
|
+
conversationId?: string;
|
|
29
|
+
branchId?: string;
|
|
30
|
+
model?: string;
|
|
31
|
+
runtimeOverrides?: {
|
|
32
|
+
allowedTools?: string[];
|
|
33
|
+
maxSteps?: number;
|
|
34
|
+
};
|
|
35
|
+
}
|
|
36
|
+
/** Input accepted by `buildAgentServiceEvalRequestBody`. */
|
|
37
|
+
export interface BuildAgentServiceEvalRequestBodyInput {
|
|
38
|
+
exampleId: string;
|
|
39
|
+
input: unknown;
|
|
40
|
+
metadata?: Record<string, unknown>;
|
|
41
|
+
projectId?: string | null;
|
|
42
|
+
conversationId?: string | null;
|
|
43
|
+
branchId?: string | null;
|
|
44
|
+
model?: string | null;
|
|
45
|
+
allowedTools?: string[];
|
|
46
|
+
forceRuntimeOverrides?: boolean;
|
|
47
|
+
maxSteps?: number;
|
|
48
|
+
}
|
|
49
|
+
/** AG-UI request body sent to an agent-service endpoint. */
|
|
50
|
+
export interface AgentServiceEvalRequestBody {
|
|
51
|
+
threadId: string;
|
|
52
|
+
runId: string;
|
|
53
|
+
state: Record<string, unknown>;
|
|
54
|
+
tools: [];
|
|
55
|
+
context: [];
|
|
56
|
+
forwardedProps?: {
|
|
57
|
+
veryfront: AgentServiceEvalForwardedProps;
|
|
58
|
+
};
|
|
59
|
+
messages: Array<{
|
|
60
|
+
id: string;
|
|
61
|
+
role: "user";
|
|
62
|
+
content: string;
|
|
63
|
+
}>;
|
|
64
|
+
}
|
|
65
|
+
/** Configuration for the live agent-service eval adapter. */
|
|
66
|
+
export interface AgentServiceEvalAdapterConfig {
|
|
67
|
+
endpoint?: string;
|
|
68
|
+
authToken: string;
|
|
69
|
+
projectId?: string | null;
|
|
70
|
+
conversationId?: string | null;
|
|
71
|
+
branchId?: string | null;
|
|
72
|
+
model?: string | null;
|
|
73
|
+
allowedTools?: string[];
|
|
74
|
+
forceRuntimeOverrides?: boolean;
|
|
75
|
+
maxSteps?: number;
|
|
76
|
+
requestTimeoutMs?: number;
|
|
77
|
+
progressThrottleMs?: number;
|
|
78
|
+
fetch?: (input: string | URL | Request, init?: RequestInit) => Promise<Response>;
|
|
79
|
+
now?: () => number;
|
|
80
|
+
onProgress?: (snapshot: AgUiSseProgressSnapshot, context: EvalAgentAdapterContext) => void;
|
|
81
|
+
}
|
|
82
|
+
/** Resolve environment values for live agent-service eval execution. */
|
|
83
|
+
export declare function resolveAgentServiceEvalEnvironment(env?: AgentServiceEvalEnvironmentInput): AgentServiceEvalEnvironment;
|
|
84
|
+
/** Evaluate whether the required live agent-service eval environment is present. */
|
|
85
|
+
export declare function evaluateAgentServiceEvalEnvironment(env?: AgentServiceEvalEnvironmentInput, resolvedApiUrl?: string): AgentServiceEvalEnvironmentPreflightResult;
|
|
86
|
+
/** Build the AG-UI request body for a single eval example. */
|
|
87
|
+
export declare function buildAgentServiceEvalRequestBody(input: BuildAgentServiceEvalRequestBodyInput): AgentServiceEvalRequestBody;
|
|
88
|
+
/** Create an `EvalAgentAdapter` that executes examples against an AG-UI agent-service endpoint. */
|
|
89
|
+
export declare function createAgentServiceEvalAdapter(config: AgentServiceEvalAdapterConfig): EvalAgentAdapter;
|
|
90
|
+
//# sourceMappingURL=agent-service.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"agent-service.d.ts","sourceRoot":"","sources":["../../../src/src/eval/agent-service.ts"],"names":[],"mappings":"AAAA,OAAO,yBAAyB,CAAC;AAEjC,OAAO,EAEL,KAAK,uBAAuB,EAK7B,MAAM,mBAAmB,CAAC;AAC3B,OAAO,KAAK,EACV,gBAAgB,EAChB,uBAAuB,EAGxB,MAAM,YAAY,CAAC;AAEpB,cAAc,qCAAqC,CAAC;AACpD,cAAc,+CAA+C,CAAC;AAE9D,gEAAgE;AAChE,eAAO,MAAM,mCAAmC,oCAAoC,CAAC;AAErF,gEAAgE;AAChE,MAAM,MAAM,gCAAgC,GAAG,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAAC,CAAC;AAE3F,gEAAgE;AAChE,MAAM,WAAW,2BAA2B;IAC1C,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,kEAAkE;AAClE,MAAM,WAAW,0CAA0C;IACzD,EAAE,EAAE,OAAO,CAAC;IACZ,cAAc,EAAE,MAAM,CAAC;IACvB,QAAQ,EAAE,MAAM,EAAE,CAAC;CACpB;AAED,mEAAmE;AACnE,MAAM,WAAW,8BAA8B;IAC7C,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,gBAAgB,CAAC,EAAE;QACjB,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;QACxB,QAAQ,CAAC,EAAE,MAAM,CAAC;KACnB,CAAC;CACH;AAED,4DAA4D;AAC5D,MAAM,WAAW,qCAAqC;IACpD,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,EAAE,OAAO,CAAC;IACf,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACnC,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,cAAc,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,KAAK,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB,qBAAqB,CAAC,EAAE,OAAO,CAAC;IAChC,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAED,4DAA4D;AAC5D,MAAM,WAAW,2BAA2B;IAC1C,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAC/B,KAAK,EAAE,EAAE,CAAC;IACV,OAAO,EAAE,EAAE,CAAC;IACZ,cAAc,CAAC,EAAE;QACf,SAAS,EAAE,8BAA8B,CAAC;KAC3C,CAAC;IACF,QAAQ,EAAE,KAAK,CAAC;QACd,EAAE,EAAE,MAAM,CAAC;QACX,IAAI,EAAE,MAAM,CAAC;QACb,OAAO,EAAE,MAAM,CAAC;KACjB,CAAC,CAAC;CACJ;AAED,6DAA6D;AAC7D,MAAM,WAAW,6BAA6B;IAC5C,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,cAAc,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,KAAK,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB,qBAAqB,CAAC,EAAE,OAAO,CAAC;IAChC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,KAAK,CAAC,EAAE,CAAC,KAAK,EAAE,MAAM,GAAG,GAAG,GAAG,OAAO,EAAE,IAAI,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,QAAQ,CAAC,CAAC;IACjF,GAAG,CAAC,EAAE,MAAM,MAAM,CAAC;IACnB,UAAU,CAAC,EAAE,CAAC,QAAQ,EAAE,uBAAuB,EAAE,OAAO,EAAE,uBAAuB,KAAK,IAAI,CAAC;CAC5F;AAmKD,wEAAwE;AACxE,wBAAgB,kCAAkC,CAChD,GAAG,GAAE,gCAAqC,GACzC,2BAA2B,CAe7B;AAED,oFAAoF;AACpF,wBAAgB,mCAAmC,CACjD,GAAG,GAAE,gCAAqC,EAC1C,cAAc,SAAiD,GAC9D,0CAA0C,CAe5C;AAED,8DAA8D;AAC9D,wBAAgB,gCAAgC,CAC9C,KAAK,EAAE,qCAAqC,GAC3C,2BAA2B,CAyB7B;AAED,mGAAmG;AACnG,wBAAgB,6BAA6B,CAC3C,MAAM,EAAE,6BAA6B,GACpC,gBAAgB,CAoElB"}
|
|
@@ -0,0 +1,276 @@
|
|
|
1
|
+
import "../../_dnt.polyfills.js";
|
|
2
|
+
import * as dntShim from "../../_dnt.shims.js";
|
|
3
|
+
import { agUiSseEventTypes, getAgUiSseStringField, parseAgentServiceConfig, parseAgUiSseResponse, } from "../agent/index.js";
|
|
4
|
+
export * from "./agent-service/live-evals/index.js";
|
|
5
|
+
export * from "./agent-service/durable-run-canaries/index.js";
|
|
6
|
+
/** Default local AG-UI endpoint used by agent-service evals. */
|
|
7
|
+
export const DEFAULT_AGENT_SERVICE_EVAL_ENDPOINT = "http://127.0.0.1:3001/api/ag-ui";
|
|
8
|
+
function readString(value) {
|
|
9
|
+
return typeof value === "string" && value.length > 0 ? value : undefined;
|
|
10
|
+
}
|
|
11
|
+
function readNumber(value) {
|
|
12
|
+
return typeof value === "number" && Number.isFinite(value) ? value : undefined;
|
|
13
|
+
}
|
|
14
|
+
function readStringArray(value) {
|
|
15
|
+
return Array.isArray(value) && value.every((entry) => typeof entry === "string")
|
|
16
|
+
? [...value]
|
|
17
|
+
: undefined;
|
|
18
|
+
}
|
|
19
|
+
function isRecord(value) {
|
|
20
|
+
return value !== null && typeof value === "object" && !Array.isArray(value);
|
|
21
|
+
}
|
|
22
|
+
function stringifyPromptInput(input) {
|
|
23
|
+
if (typeof input === "string")
|
|
24
|
+
return input;
|
|
25
|
+
if (isRecord(input)) {
|
|
26
|
+
const prompt = readString(input.prompt) ?? readString(input.message) ?? readString(input.input);
|
|
27
|
+
if (prompt)
|
|
28
|
+
return prompt;
|
|
29
|
+
}
|
|
30
|
+
return JSON.stringify(input);
|
|
31
|
+
}
|
|
32
|
+
function getInputMetadata(input) {
|
|
33
|
+
if (!isRecord(input) || !isRecord(input.metadata))
|
|
34
|
+
return {};
|
|
35
|
+
return { ...input.metadata };
|
|
36
|
+
}
|
|
37
|
+
function getRequestOverrides(input) {
|
|
38
|
+
const record = isRecord(input.input) ? input.input : {};
|
|
39
|
+
return {
|
|
40
|
+
projectId: input.projectId ?? readString(record.projectId) ?? null,
|
|
41
|
+
conversationId: input.conversationId ?? readString(record.conversationId) ?? null,
|
|
42
|
+
branchId: input.branchId ?? readString(record.branchId) ?? null,
|
|
43
|
+
model: input.model ?? readString(record.model) ?? null,
|
|
44
|
+
allowedTools: input.allowedTools ?? readStringArray(record.allowedTools),
|
|
45
|
+
forceRuntimeOverrides: input.forceRuntimeOverrides ?? record.forceRuntimeOverrides === true,
|
|
46
|
+
maxSteps: input.maxSteps ?? readNumber(record.maxSteps),
|
|
47
|
+
};
|
|
48
|
+
}
|
|
49
|
+
function createVeryfrontForwardedProps(input) {
|
|
50
|
+
const overrides = getRequestOverrides(input);
|
|
51
|
+
const veryfront = {};
|
|
52
|
+
if (overrides.projectId)
|
|
53
|
+
veryfront.projectId = overrides.projectId;
|
|
54
|
+
if (overrides.conversationId)
|
|
55
|
+
veryfront.conversationId = overrides.conversationId;
|
|
56
|
+
if (overrides.branchId)
|
|
57
|
+
veryfront.branchId = overrides.branchId;
|
|
58
|
+
if (overrides.model)
|
|
59
|
+
veryfront.model = overrides.model;
|
|
60
|
+
const shouldForwardRuntimeOverrides = !!overrides.allowedTools ||
|
|
61
|
+
overrides.forceRuntimeOverrides ||
|
|
62
|
+
overrides.maxSteps !== undefined;
|
|
63
|
+
if (shouldForwardRuntimeOverrides) {
|
|
64
|
+
veryfront.runtimeOverrides = {
|
|
65
|
+
...(overrides.allowedTools ? { allowedTools: overrides.allowedTools } : {}),
|
|
66
|
+
...(overrides.maxSteps !== undefined ? { maxSteps: overrides.maxSteps } : {}),
|
|
67
|
+
};
|
|
68
|
+
}
|
|
69
|
+
return Object.keys(veryfront).length > 0 ? veryfront : null;
|
|
70
|
+
}
|
|
71
|
+
function createHeaders(authToken) {
|
|
72
|
+
return {
|
|
73
|
+
"Content-Type": "application/json",
|
|
74
|
+
Authorization: `Bearer ${authToken}`,
|
|
75
|
+
};
|
|
76
|
+
}
|
|
77
|
+
function getNow(config) {
|
|
78
|
+
return config.now?.() ?? Date.now();
|
|
79
|
+
}
|
|
80
|
+
function stringifyError(value) {
|
|
81
|
+
if (typeof value === "string" && value.length > 0)
|
|
82
|
+
return value;
|
|
83
|
+
if (isRecord(value)) {
|
|
84
|
+
const message = readString(value.message) ?? readString(value.error);
|
|
85
|
+
if (message)
|
|
86
|
+
return message;
|
|
87
|
+
}
|
|
88
|
+
try {
|
|
89
|
+
return JSON.stringify(value);
|
|
90
|
+
}
|
|
91
|
+
catch {
|
|
92
|
+
return String(value);
|
|
93
|
+
}
|
|
94
|
+
}
|
|
95
|
+
function getToolResultError(event) {
|
|
96
|
+
return stringifyError(event.result) ?? stringifyError(event.content) ?? "Tool call failed";
|
|
97
|
+
}
|
|
98
|
+
function createToolCalls(events) {
|
|
99
|
+
const toolCalls = new Map();
|
|
100
|
+
for (const [index, event] of events.entries()) {
|
|
101
|
+
const type = getAgUiSseStringField(event, "type");
|
|
102
|
+
if (type === agUiSseEventTypes.toolCallStart) {
|
|
103
|
+
const name = getAgUiSseStringField(event, "toolCallName");
|
|
104
|
+
if (!name)
|
|
105
|
+
continue;
|
|
106
|
+
const id = getAgUiSseStringField(event, "toolCallId");
|
|
107
|
+
const key = id ?? `name:${name}`;
|
|
108
|
+
const existing = toolCalls.get(key);
|
|
109
|
+
toolCalls.set(key, {
|
|
110
|
+
name,
|
|
111
|
+
status: existing?.status ?? "ok",
|
|
112
|
+
...(existing?.error ? { error: existing.error } : {}),
|
|
113
|
+
...(existing?.metadata ? { metadata: existing.metadata } : {}),
|
|
114
|
+
});
|
|
115
|
+
continue;
|
|
116
|
+
}
|
|
117
|
+
if (type === agUiSseEventTypes.toolCallResult) {
|
|
118
|
+
const id = getAgUiSseStringField(event, "toolCallId");
|
|
119
|
+
const toolName = getAgUiSseStringField(event, "toolCallName");
|
|
120
|
+
const key = id ?? (toolName ? `name:${toolName}` : `result:${index}`);
|
|
121
|
+
const existing = toolCalls.get(key);
|
|
122
|
+
const failed = event.isError === true;
|
|
123
|
+
toolCalls.set(key, {
|
|
124
|
+
name: existing?.name ?? toolName ?? "tool",
|
|
125
|
+
status: failed ? "error" : existing?.status ?? "ok",
|
|
126
|
+
...(failed ? { error: getToolResultError(event) } : {}),
|
|
127
|
+
...(existing?.metadata ? { metadata: existing.metadata } : {}),
|
|
128
|
+
});
|
|
129
|
+
}
|
|
130
|
+
}
|
|
131
|
+
return [...toolCalls.values()];
|
|
132
|
+
}
|
|
133
|
+
function createRunOutput(run) {
|
|
134
|
+
return {
|
|
135
|
+
text: run.text,
|
|
136
|
+
agUi: {
|
|
137
|
+
responseStatus: run.responseStatus,
|
|
138
|
+
eventTypes: run.eventTypes,
|
|
139
|
+
runError: run.runError,
|
|
140
|
+
},
|
|
141
|
+
};
|
|
142
|
+
}
|
|
143
|
+
function createRequestInit(config, body) {
|
|
144
|
+
return {
|
|
145
|
+
method: "POST",
|
|
146
|
+
headers: createHeaders(config.authToken),
|
|
147
|
+
body: JSON.stringify(body),
|
|
148
|
+
...(config.requestTimeoutMs ? { signal: AbortSignal.timeout(config.requestTimeoutMs) } : {}),
|
|
149
|
+
};
|
|
150
|
+
}
|
|
151
|
+
/** Resolve environment values for live agent-service eval execution. */
|
|
152
|
+
export function resolveAgentServiceEvalEnvironment(env = {}) {
|
|
153
|
+
return {
|
|
154
|
+
endpoint: typeof env.AG_UI_EVAL_ENDPOINT === "string"
|
|
155
|
+
? env.AG_UI_EVAL_ENDPOINT
|
|
156
|
+
: DEFAULT_AGENT_SERVICE_EVAL_ENDPOINT,
|
|
157
|
+
authToken: typeof env.VERYFRONT_TOKEN === "string" ? env.VERYFRONT_TOKEN : "",
|
|
158
|
+
apiUrl: typeof env.VERYFRONT_API_URL === "string"
|
|
159
|
+
? env.VERYFRONT_API_URL
|
|
160
|
+
: parseAgentServiceConfig(env).VERYFRONT_API_URL,
|
|
161
|
+
...(typeof env.AG_UI_EVAL_PROJECT_ID === "string"
|
|
162
|
+
? { projectId: env.AG_UI_EVAL_PROJECT_ID }
|
|
163
|
+
: {}),
|
|
164
|
+
...(typeof env.AG_UI_EVAL_BRANCH_ID === "string" ? { branchId: env.AG_UI_EVAL_BRANCH_ID } : {}),
|
|
165
|
+
...(typeof env.AG_UI_EVAL_MODEL === "string" ? { model: env.AG_UI_EVAL_MODEL } : {}),
|
|
166
|
+
};
|
|
167
|
+
}
|
|
168
|
+
/** Evaluate whether the required live agent-service eval environment is present. */
|
|
169
|
+
export function evaluateAgentServiceEvalEnvironment(env = {}, resolvedApiUrl = parseAgentServiceConfig(env).VERYFRONT_API_URL) {
|
|
170
|
+
const messages = [`Resolved VERYFRONT_API_URL: ${resolvedApiUrl}`];
|
|
171
|
+
let hasBlockers = false;
|
|
172
|
+
if (typeof env.VERYFRONT_TOKEN !== "string" || env.VERYFRONT_TOKEN.length === 0) {
|
|
173
|
+
hasBlockers = true;
|
|
174
|
+
messages.push("BLOCKER: VERYFRONT_TOKEN is missing");
|
|
175
|
+
}
|
|
176
|
+
if (typeof env.AG_UI_EVAL_PROJECT_ID !== "string" || env.AG_UI_EVAL_PROJECT_ID.length === 0) {
|
|
177
|
+
hasBlockers = true;
|
|
178
|
+
messages.push("BLOCKER: AG_UI_EVAL_PROJECT_ID is missing");
|
|
179
|
+
}
|
|
180
|
+
messages.push(`Agent-service eval preflight: ${hasBlockers ? "FAIL" : "PASS"}`);
|
|
181
|
+
return { ok: !hasBlockers, resolvedApiUrl, messages };
|
|
182
|
+
}
|
|
183
|
+
/** Build the AG-UI request body for a single eval example. */
|
|
184
|
+
export function buildAgentServiceEvalRequestBody(input) {
|
|
185
|
+
const veryfront = createVeryfrontForwardedProps(input);
|
|
186
|
+
const metadata = {
|
|
187
|
+
...getInputMetadata(input.input),
|
|
188
|
+
...(input.metadata ?? {}),
|
|
189
|
+
};
|
|
190
|
+
return {
|
|
191
|
+
threadId: dntShim.crypto.randomUUID(),
|
|
192
|
+
runId: `eval-run-${dntShim.crypto.randomUUID()}`,
|
|
193
|
+
state: {
|
|
194
|
+
evalCase: input.exampleId,
|
|
195
|
+
...metadata,
|
|
196
|
+
},
|
|
197
|
+
tools: [],
|
|
198
|
+
context: [],
|
|
199
|
+
...(veryfront ? { forwardedProps: { veryfront } } : {}),
|
|
200
|
+
messages: [
|
|
201
|
+
{
|
|
202
|
+
id: dntShim.crypto.randomUUID(),
|
|
203
|
+
role: "user",
|
|
204
|
+
content: stringifyPromptInput(input.input),
|
|
205
|
+
},
|
|
206
|
+
],
|
|
207
|
+
};
|
|
208
|
+
}
|
|
209
|
+
/** Create an `EvalAgentAdapter` that executes examples against an AG-UI agent-service endpoint. */
|
|
210
|
+
export function createAgentServiceEvalAdapter(config) {
|
|
211
|
+
const requestFetch = config.fetch ?? fetch;
|
|
212
|
+
const endpoint = config.endpoint ?? DEFAULT_AGENT_SERVICE_EVAL_ENDPOINT;
|
|
213
|
+
return async (context) => {
|
|
214
|
+
const started = getNow(config);
|
|
215
|
+
const body = buildAgentServiceEvalRequestBody({
|
|
216
|
+
exampleId: context.example.id,
|
|
217
|
+
input: context.example.input,
|
|
218
|
+
metadata: context.example.metadata,
|
|
219
|
+
projectId: config.projectId,
|
|
220
|
+
conversationId: config.conversationId,
|
|
221
|
+
branchId: config.branchId,
|
|
222
|
+
model: config.model,
|
|
223
|
+
allowedTools: config.allowedTools,
|
|
224
|
+
forceRuntimeOverrides: config.forceRuntimeOverrides,
|
|
225
|
+
maxSteps: config.maxSteps,
|
|
226
|
+
});
|
|
227
|
+
try {
|
|
228
|
+
const parseOptions = {
|
|
229
|
+
...(config.progressThrottleMs !== undefined
|
|
230
|
+
? { progressThrottleMs: config.progressThrottleMs }
|
|
231
|
+
: {}),
|
|
232
|
+
...(config.onProgress
|
|
233
|
+
? { onProgress: (snapshot) => config.onProgress?.(snapshot, context) }
|
|
234
|
+
: {}),
|
|
235
|
+
};
|
|
236
|
+
const response = await requestFetch(endpoint, createRequestInit(config, body));
|
|
237
|
+
const run = await parseAgUiSseResponse(response, parseOptions);
|
|
238
|
+
const completed = response.ok && run.runError === null &&
|
|
239
|
+
run.eventTypes.includes(agUiSseEventTypes.runFinished);
|
|
240
|
+
const output = createRunOutput(run);
|
|
241
|
+
return {
|
|
242
|
+
text: run.text,
|
|
243
|
+
output,
|
|
244
|
+
trace: {
|
|
245
|
+
events: run.events,
|
|
246
|
+
toolCalls: createToolCalls(run.events),
|
|
247
|
+
},
|
|
248
|
+
durationMs: getNow(config) - started,
|
|
249
|
+
completed,
|
|
250
|
+
...(!completed
|
|
251
|
+
? { error: run.runError ?? `AG-UI response failed with status ${response.status}` }
|
|
252
|
+
: {}),
|
|
253
|
+
};
|
|
254
|
+
}
|
|
255
|
+
catch (error) {
|
|
256
|
+
return {
|
|
257
|
+
text: "",
|
|
258
|
+
output: {
|
|
259
|
+
text: "",
|
|
260
|
+
agUi: {
|
|
261
|
+
responseStatus: 0,
|
|
262
|
+
eventTypes: [],
|
|
263
|
+
runError: error instanceof Error ? error.message : String(error),
|
|
264
|
+
},
|
|
265
|
+
},
|
|
266
|
+
trace: {
|
|
267
|
+
events: [],
|
|
268
|
+
toolCalls: [],
|
|
269
|
+
},
|
|
270
|
+
durationMs: getNow(config) - started,
|
|
271
|
+
completed: false,
|
|
272
|
+
error: error instanceof Error ? error.message : String(error),
|
|
273
|
+
};
|
|
274
|
+
}
|
|
275
|
+
};
|
|
276
|
+
}
|