agentv 3.3.0 → 3.4.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +1 -1
- package/dist/{chunk-6LP5Z5Y4.js → chunk-A7ZDUB46.js} +233 -115
- package/dist/chunk-A7ZDUB46.js.map +1 -0
- package/dist/{chunk-5M3K2DMV.js → chunk-GOZV2HN2.js} +167 -80
- package/dist/chunk-GOZV2HN2.js.map +1 -0
- package/dist/{chunk-4ZMSAQWS.js → chunk-RE5I3U2S.js} +156 -46
- package/dist/chunk-RE5I3U2S.js.map +1 -0
- package/dist/cli.js +3 -3
- package/dist/{dist-OC53WD3P.js → dist-AFDYFH6Y.js} +2 -2
- package/dist/index.js +3 -3
- package/dist/{interactive-NA6SAIAG.js → interactive-WXXTZ7PD.js} +44 -4
- package/dist/interactive-WXXTZ7PD.js.map +1 -0
- package/dist/templates/.agentv/.env.example +11 -9
- package/dist/templates/.agentv/config.yaml +5 -0
- package/dist/templates/.agentv/targets.yaml +0 -16
- package/package.json +2 -2
- package/dist/chunk-4ZMSAQWS.js.map +0 -1
- package/dist/chunk-5M3K2DMV.js.map +0 -1
- package/dist/chunk-6LP5Z5Y4.js.map +0 -1
- package/dist/interactive-NA6SAIAG.js.map +0 -1
- /package/dist/{dist-OC53WD3P.js.map → dist-AFDYFH6Y.js.map} +0 -0
|
@@ -301,7 +301,7 @@ var require_dist = __commonJS({
|
|
|
301
301
|
}
|
|
302
302
|
});
|
|
303
303
|
|
|
304
|
-
// ../../packages/core/dist/chunk-
|
|
304
|
+
// ../../packages/core/dist/chunk-JO4HIAEF.js
|
|
305
305
|
import { constants } from "node:fs";
|
|
306
306
|
import { access, readFile } from "node:fs/promises";
|
|
307
307
|
import path from "node:path";
|
|
@@ -419,7 +419,7 @@ __export(external_exports2, {
|
|
|
419
419
|
void: () => voidType
|
|
420
420
|
});
|
|
421
421
|
|
|
422
|
-
// ../../packages/core/dist/chunk-
|
|
422
|
+
// ../../packages/core/dist/chunk-JO4HIAEF.js
|
|
423
423
|
var TEST_MESSAGE_ROLE_VALUES = ["system", "user", "assistant", "tool"];
|
|
424
424
|
var TEST_MESSAGE_ROLES = TEST_MESSAGE_ROLE_VALUES;
|
|
425
425
|
var TEST_MESSAGE_ROLE_SET = new Set(TEST_MESSAGE_ROLE_VALUES);
|
|
@@ -1810,6 +1810,7 @@ var AGENT_PROVIDER_KINDS = [
|
|
|
1810
1810
|
"copilot-sdk",
|
|
1811
1811
|
"copilot-cli",
|
|
1812
1812
|
"pi-coding-agent",
|
|
1813
|
+
"pi-agent-sdk",
|
|
1813
1814
|
"claude",
|
|
1814
1815
|
"claude-cli",
|
|
1815
1816
|
"claude-sdk",
|
|
@@ -20912,6 +20913,29 @@ var MockProvider = class {
|
|
|
20912
20913
|
return this.delayMs;
|
|
20913
20914
|
}
|
|
20914
20915
|
};
|
|
20916
|
+
function extractPiTextContent(content) {
|
|
20917
|
+
if (typeof content === "string") {
|
|
20918
|
+
return content;
|
|
20919
|
+
}
|
|
20920
|
+
if (!Array.isArray(content)) {
|
|
20921
|
+
return void 0;
|
|
20922
|
+
}
|
|
20923
|
+
const textParts = [];
|
|
20924
|
+
for (const part of content) {
|
|
20925
|
+
if (!part || typeof part !== "object") {
|
|
20926
|
+
continue;
|
|
20927
|
+
}
|
|
20928
|
+
const p = part;
|
|
20929
|
+
if (p.type === "text" && typeof p.text === "string") {
|
|
20930
|
+
textParts.push(p.text);
|
|
20931
|
+
}
|
|
20932
|
+
}
|
|
20933
|
+
return textParts.length > 0 ? textParts.join("\n") : void 0;
|
|
20934
|
+
}
|
|
20935
|
+
function toFiniteNumber(value) {
|
|
20936
|
+
if (typeof value === "number" && Number.isFinite(value)) return value;
|
|
20937
|
+
return void 0;
|
|
20938
|
+
}
|
|
20915
20939
|
var piAgentModule = null;
|
|
20916
20940
|
var piAiModule = null;
|
|
20917
20941
|
async function loadPiModules() {
|
|
@@ -20952,7 +20976,8 @@ var PiAgentSdkProvider = class {
|
|
|
20952
20976
|
throw new Error("Pi agent SDK request was aborted before execution");
|
|
20953
20977
|
}
|
|
20954
20978
|
const { Agent, getModel, getEnvApiKey } = await loadPiModules();
|
|
20955
|
-
const
|
|
20979
|
+
const startTimeIso = (/* @__PURE__ */ new Date()).toISOString();
|
|
20980
|
+
const startMs = Date.now();
|
|
20956
20981
|
const providerName = this.config.provider ?? "anthropic";
|
|
20957
20982
|
const modelId = this.config.model ?? "claude-sonnet-4-20250514";
|
|
20958
20983
|
const model = getModel(providerName, modelId);
|
|
@@ -20969,16 +20994,73 @@ var PiAgentSdkProvider = class {
|
|
|
20969
20994
|
return this.config.apiKey ?? getEnvApiKey(provider) ?? void 0;
|
|
20970
20995
|
}
|
|
20971
20996
|
});
|
|
20972
|
-
|
|
20973
|
-
let
|
|
20997
|
+
let tokenUsage;
|
|
20998
|
+
let costUsd;
|
|
20999
|
+
const toolTrackers = /* @__PURE__ */ new Map();
|
|
21000
|
+
const completedToolResults = /* @__PURE__ */ new Map();
|
|
20974
21001
|
const unsubscribe = agent.subscribe((event) => {
|
|
20975
|
-
|
|
20976
|
-
|
|
20977
|
-
|
|
20978
|
-
|
|
20979
|
-
|
|
20980
|
-
|
|
21002
|
+
switch (event.type) {
|
|
21003
|
+
case "message_end": {
|
|
21004
|
+
const msg = event.message;
|
|
21005
|
+
if (msg && typeof msg === "object" && "role" in msg && msg.role === "assistant" && "usage" in msg) {
|
|
21006
|
+
const usage = msg.usage;
|
|
21007
|
+
if (usage && typeof usage === "object") {
|
|
21008
|
+
const u = usage;
|
|
21009
|
+
const input = toFiniteNumber(u.input);
|
|
21010
|
+
const output = toFiniteNumber(u.output);
|
|
21011
|
+
const cached = toFiniteNumber(u.cacheRead);
|
|
21012
|
+
let callDelta;
|
|
21013
|
+
if (input !== void 0 || output !== void 0) {
|
|
21014
|
+
callDelta = {
|
|
21015
|
+
input: input ?? 0,
|
|
21016
|
+
output: output ?? 0,
|
|
21017
|
+
...cached !== void 0 ? { cached } : {}
|
|
21018
|
+
};
|
|
21019
|
+
tokenUsage = {
|
|
21020
|
+
input: (tokenUsage?.input ?? 0) + callDelta.input,
|
|
21021
|
+
output: (tokenUsage?.output ?? 0) + callDelta.output,
|
|
21022
|
+
...cached !== void 0 ? { cached: (tokenUsage?.cached ?? 0) + cached } : tokenUsage?.cached !== void 0 ? { cached: tokenUsage.cached } : {}
|
|
21023
|
+
};
|
|
21024
|
+
}
|
|
21025
|
+
const cost = u.cost;
|
|
21026
|
+
if (cost && typeof cost === "object") {
|
|
21027
|
+
const total = toFiniteNumber(cost.total);
|
|
21028
|
+
if (total !== void 0) {
|
|
21029
|
+
costUsd = (costUsd ?? 0) + total;
|
|
21030
|
+
}
|
|
21031
|
+
}
|
|
21032
|
+
request.streamCallbacks?.onLlmCallEnd?.(modelId, callDelta);
|
|
21033
|
+
}
|
|
20981
21034
|
}
|
|
21035
|
+
break;
|
|
21036
|
+
}
|
|
21037
|
+
case "tool_execution_start": {
|
|
21038
|
+
toolTrackers.set(event.toolCallId, {
|
|
21039
|
+
toolCallId: event.toolCallId,
|
|
21040
|
+
toolName: event.toolName,
|
|
21041
|
+
args: event.args,
|
|
21042
|
+
startMs: Date.now(),
|
|
21043
|
+
startTime: (/* @__PURE__ */ new Date()).toISOString()
|
|
21044
|
+
});
|
|
21045
|
+
request.streamCallbacks?.onToolCallStart?.(event.toolName, event.toolCallId);
|
|
21046
|
+
break;
|
|
21047
|
+
}
|
|
21048
|
+
case "tool_execution_end": {
|
|
21049
|
+
const tracker = toolTrackers.get(event.toolCallId);
|
|
21050
|
+
const durationMs = tracker ? Date.now() - tracker.startMs : 0;
|
|
21051
|
+
completedToolResults.set(event.toolCallId, {
|
|
21052
|
+
output: event.result,
|
|
21053
|
+
durationMs
|
|
21054
|
+
});
|
|
21055
|
+
request.streamCallbacks?.onToolCallEnd?.(
|
|
21056
|
+
event.toolName,
|
|
21057
|
+
tracker?.args,
|
|
21058
|
+
event.result,
|
|
21059
|
+
durationMs,
|
|
21060
|
+
event.toolCallId
|
|
21061
|
+
);
|
|
21062
|
+
toolTrackers.delete(event.toolCallId);
|
|
21063
|
+
break;
|
|
20982
21064
|
}
|
|
20983
21065
|
}
|
|
20984
21066
|
});
|
|
@@ -20997,10 +21079,12 @@ var PiAgentSdkProvider = class {
|
|
|
20997
21079
|
}
|
|
20998
21080
|
await agent.waitForIdle();
|
|
20999
21081
|
const agentMessages = agent.state.messages;
|
|
21082
|
+
const output = [];
|
|
21000
21083
|
for (const msg of agentMessages) {
|
|
21001
|
-
output.push(convertAgentMessage(msg));
|
|
21084
|
+
output.push(convertAgentMessage(msg, toolTrackers, completedToolResults));
|
|
21002
21085
|
}
|
|
21003
|
-
const
|
|
21086
|
+
const endTimeIso = (/* @__PURE__ */ new Date()).toISOString();
|
|
21087
|
+
const durationMs = Date.now() - startMs;
|
|
21004
21088
|
return {
|
|
21005
21089
|
raw: {
|
|
21006
21090
|
messages: agentMessages,
|
|
@@ -21009,49 +21093,54 @@ var PiAgentSdkProvider = class {
|
|
|
21009
21093
|
provider: this.config.provider
|
|
21010
21094
|
},
|
|
21011
21095
|
output,
|
|
21012
|
-
|
|
21096
|
+
tokenUsage,
|
|
21097
|
+
costUsd,
|
|
21098
|
+
durationMs,
|
|
21099
|
+
startTime: startTimeIso,
|
|
21100
|
+
endTime: endTimeIso
|
|
21013
21101
|
};
|
|
21014
21102
|
} finally {
|
|
21015
21103
|
unsubscribe();
|
|
21016
21104
|
}
|
|
21017
21105
|
}
|
|
21018
21106
|
};
|
|
21019
|
-
function
|
|
21020
|
-
if (typeof content === "string") {
|
|
21021
|
-
return content;
|
|
21022
|
-
}
|
|
21023
|
-
if (!Array.isArray(content)) {
|
|
21024
|
-
return void 0;
|
|
21025
|
-
}
|
|
21026
|
-
const textParts = [];
|
|
21027
|
-
for (const part of content) {
|
|
21028
|
-
if (!part || typeof part !== "object") {
|
|
21029
|
-
continue;
|
|
21030
|
-
}
|
|
21031
|
-
const p = part;
|
|
21032
|
-
if (p.type === "text" && typeof p.text === "string") {
|
|
21033
|
-
textParts.push(p.text);
|
|
21034
|
-
}
|
|
21035
|
-
}
|
|
21036
|
-
return textParts.length > 0 ? textParts.join("\n") : void 0;
|
|
21037
|
-
}
|
|
21038
|
-
function convertAgentMessage(message) {
|
|
21107
|
+
function convertAgentMessage(message, toolTrackers, completedToolResults) {
|
|
21039
21108
|
if (!message || typeof message !== "object") {
|
|
21040
21109
|
return { role: "unknown", content: String(message) };
|
|
21041
21110
|
}
|
|
21042
21111
|
const msg = message;
|
|
21043
21112
|
const role = typeof msg.role === "string" ? msg.role : "unknown";
|
|
21044
|
-
const content =
|
|
21045
|
-
const toolCalls = extractToolCalls3(msg.content);
|
|
21113
|
+
const content = extractPiTextContent(msg.content);
|
|
21114
|
+
const toolCalls = extractToolCalls3(msg.content, toolTrackers, completedToolResults);
|
|
21046
21115
|
const startTime = typeof msg.timestamp === "number" ? new Date(msg.timestamp).toISOString() : typeof msg.timestamp === "string" ? msg.timestamp : void 0;
|
|
21116
|
+
let msgTokenUsage;
|
|
21117
|
+
if (msg.usage && typeof msg.usage === "object") {
|
|
21118
|
+
const u = msg.usage;
|
|
21119
|
+
const input = toFiniteNumber(u.input);
|
|
21120
|
+
const output = toFiniteNumber(u.output);
|
|
21121
|
+
if (input !== void 0 || output !== void 0) {
|
|
21122
|
+
msgTokenUsage = {
|
|
21123
|
+
input: input ?? 0,
|
|
21124
|
+
output: output ?? 0,
|
|
21125
|
+
...toFiniteNumber(u.cacheRead) !== void 0 ? { cached: toFiniteNumber(u.cacheRead) } : {}
|
|
21126
|
+
};
|
|
21127
|
+
}
|
|
21128
|
+
}
|
|
21129
|
+
const metadata = {};
|
|
21130
|
+
if (msg.api) metadata.api = msg.api;
|
|
21131
|
+
if (msg.provider) metadata.provider = msg.provider;
|
|
21132
|
+
if (msg.model) metadata.model = msg.model;
|
|
21133
|
+
if (msg.stopReason) metadata.stopReason = msg.stopReason;
|
|
21047
21134
|
return {
|
|
21048
21135
|
role,
|
|
21049
21136
|
content,
|
|
21050
21137
|
toolCalls: toolCalls.length > 0 ? toolCalls : void 0,
|
|
21051
|
-
startTime
|
|
21138
|
+
startTime,
|
|
21139
|
+
metadata: Object.keys(metadata).length > 0 ? metadata : void 0,
|
|
21140
|
+
tokenUsage: msgTokenUsage
|
|
21052
21141
|
};
|
|
21053
21142
|
}
|
|
21054
|
-
function extractToolCalls3(content) {
|
|
21143
|
+
function extractToolCalls3(content, toolTrackers, completedToolResults) {
|
|
21055
21144
|
if (!Array.isArray(content)) {
|
|
21056
21145
|
return [];
|
|
21057
21146
|
}
|
|
@@ -21062,10 +21151,17 @@ function extractToolCalls3(content) {
|
|
|
21062
21151
|
}
|
|
21063
21152
|
const p = part;
|
|
21064
21153
|
if (p.type === "toolCall" && typeof p.name === "string") {
|
|
21154
|
+
const id = typeof p.id === "string" ? p.id : void 0;
|
|
21155
|
+
const tracker = id ? toolTrackers.get(id) : void 0;
|
|
21156
|
+
const completed = id ? completedToolResults.get(id) : void 0;
|
|
21065
21157
|
toolCalls.push({
|
|
21066
21158
|
tool: p.name,
|
|
21067
21159
|
input: p.arguments,
|
|
21068
|
-
id
|
|
21160
|
+
id,
|
|
21161
|
+
output: completed?.output,
|
|
21162
|
+
durationMs: completed?.durationMs,
|
|
21163
|
+
startTime: tracker?.startTime,
|
|
21164
|
+
endTime: tracker?.startTime && completed?.durationMs !== void 0 ? new Date(new Date(tracker.startTime).getTime() + completed.durationMs).toISOString() : void 0
|
|
21069
21165
|
});
|
|
21070
21166
|
}
|
|
21071
21167
|
}
|
|
@@ -21597,14 +21693,14 @@ function extractTokenUsage(events) {
|
|
|
21597
21693
|
const usage = record.usage;
|
|
21598
21694
|
if (usage && typeof usage === "object") {
|
|
21599
21695
|
const u = usage;
|
|
21600
|
-
const input =
|
|
21601
|
-
const output =
|
|
21696
|
+
const input = toFiniteNumber(u.input_tokens ?? u.inputTokens ?? u.input);
|
|
21697
|
+
const output = toFiniteNumber(u.output_tokens ?? u.outputTokens ?? u.output);
|
|
21602
21698
|
if (input !== void 0 || output !== void 0) {
|
|
21603
21699
|
const result = {
|
|
21604
21700
|
input: input ?? 0,
|
|
21605
21701
|
output: output ?? 0
|
|
21606
21702
|
};
|
|
21607
|
-
const cached =
|
|
21703
|
+
const cached = toFiniteNumber(u.cache_read_input_tokens ?? u.cached ?? u.cachedTokens);
|
|
21608
21704
|
if (cached !== void 0) {
|
|
21609
21705
|
return { ...result, cached };
|
|
21610
21706
|
}
|
|
@@ -21629,13 +21725,13 @@ function aggregateUsageFromMessages(messages) {
|
|
|
21629
21725
|
const usage = m.usage;
|
|
21630
21726
|
if (!usage || typeof usage !== "object") continue;
|
|
21631
21727
|
const u = usage;
|
|
21632
|
-
const input =
|
|
21633
|
-
const output =
|
|
21728
|
+
const input = toFiniteNumber(u.input_tokens ?? u.inputTokens ?? u.input);
|
|
21729
|
+
const output = toFiniteNumber(u.output_tokens ?? u.outputTokens ?? u.output);
|
|
21634
21730
|
if (input !== void 0 || output !== void 0) {
|
|
21635
21731
|
found = true;
|
|
21636
21732
|
totalInput += input ?? 0;
|
|
21637
21733
|
totalOutput += output ?? 0;
|
|
21638
|
-
const cached =
|
|
21734
|
+
const cached = toFiniteNumber(u.cache_read_input_tokens ?? u.cached ?? u.cachedTokens);
|
|
21639
21735
|
if (cached !== void 0) {
|
|
21640
21736
|
totalCached = (totalCached ?? 0) + cached;
|
|
21641
21737
|
}
|
|
@@ -21648,10 +21744,6 @@ function aggregateUsageFromMessages(messages) {
|
|
|
21648
21744
|
}
|
|
21649
21745
|
return result;
|
|
21650
21746
|
}
|
|
21651
|
-
function toNumber(value) {
|
|
21652
|
-
if (typeof value === "number" && Number.isFinite(value)) return value;
|
|
21653
|
-
return void 0;
|
|
21654
|
-
}
|
|
21655
21747
|
function convertPiMessage(message) {
|
|
21656
21748
|
if (!message || typeof message !== "object") {
|
|
21657
21749
|
return void 0;
|
|
@@ -21661,7 +21753,7 @@ function convertPiMessage(message) {
|
|
|
21661
21753
|
if (typeof role !== "string") {
|
|
21662
21754
|
return void 0;
|
|
21663
21755
|
}
|
|
21664
|
-
const content =
|
|
21756
|
+
const content = extractPiTextContent(msg.content);
|
|
21665
21757
|
const toolCalls = extractToolCalls4(msg.content);
|
|
21666
21758
|
const startTime = typeof msg.timestamp === "number" ? new Date(msg.timestamp).toISOString() : typeof msg.timestamp === "string" ? msg.timestamp : void 0;
|
|
21667
21759
|
const metadata = {};
|
|
@@ -21678,25 +21770,6 @@ function convertPiMessage(message) {
|
|
|
21678
21770
|
metadata: Object.keys(metadata).length > 0 ? metadata : void 0
|
|
21679
21771
|
};
|
|
21680
21772
|
}
|
|
21681
|
-
function extractTextContent4(content) {
|
|
21682
|
-
if (typeof content === "string") {
|
|
21683
|
-
return content;
|
|
21684
|
-
}
|
|
21685
|
-
if (!Array.isArray(content)) {
|
|
21686
|
-
return void 0;
|
|
21687
|
-
}
|
|
21688
|
-
const textParts = [];
|
|
21689
|
-
for (const part of content) {
|
|
21690
|
-
if (!part || typeof part !== "object") {
|
|
21691
|
-
continue;
|
|
21692
|
-
}
|
|
21693
|
-
const p = part;
|
|
21694
|
-
if (p.type === "text" && typeof p.text === "string") {
|
|
21695
|
-
textParts.push(p.text);
|
|
21696
|
-
}
|
|
21697
|
-
}
|
|
21698
|
-
return textParts.length > 0 ? textParts.join("\n") : void 0;
|
|
21699
|
-
}
|
|
21700
21773
|
function extractToolCalls4(content) {
|
|
21701
21774
|
if (!Array.isArray(content)) {
|
|
21702
21775
|
return [];
|
|
@@ -25709,8 +25782,8 @@ var FieldAccuracyEvaluator = class {
|
|
|
25709
25782
|
*/
|
|
25710
25783
|
compareNumericTolerance(path46, candidateValue, expectedValue, fieldConfig, weight) {
|
|
25711
25784
|
const { tolerance = 0, relative = false } = fieldConfig;
|
|
25712
|
-
const candidateNum =
|
|
25713
|
-
const expectedNum =
|
|
25785
|
+
const candidateNum = toNumber(candidateValue);
|
|
25786
|
+
const expectedNum = toNumber(expectedValue);
|
|
25714
25787
|
if (candidateNum === null || expectedNum === null) {
|
|
25715
25788
|
return {
|
|
25716
25789
|
path: path46,
|
|
@@ -25856,7 +25929,7 @@ function resolvePath(obj, path46) {
|
|
|
25856
25929
|
}
|
|
25857
25930
|
return current;
|
|
25858
25931
|
}
|
|
25859
|
-
function
|
|
25932
|
+
function toNumber(value) {
|
|
25860
25933
|
if (typeof value === "number") {
|
|
25861
25934
|
return value;
|
|
25862
25935
|
}
|
|
@@ -28744,7 +28817,9 @@ async function runEvaluation(options) {
|
|
|
28744
28817
|
testId: evalCase.id,
|
|
28745
28818
|
status: "failed",
|
|
28746
28819
|
completedAt: Date.now(),
|
|
28747
|
-
error: budgetResult.error
|
|
28820
|
+
error: budgetResult.error,
|
|
28821
|
+
score: budgetResult.score,
|
|
28822
|
+
executionStatus: budgetResult.executionStatus
|
|
28748
28823
|
});
|
|
28749
28824
|
}
|
|
28750
28825
|
if (onResult) {
|
|
@@ -28775,7 +28850,9 @@ async function runEvaluation(options) {
|
|
|
28775
28850
|
testId: evalCase.id,
|
|
28776
28851
|
status: "failed",
|
|
28777
28852
|
completedAt: Date.now(),
|
|
28778
|
-
error: haltResult.error
|
|
28853
|
+
error: haltResult.error,
|
|
28854
|
+
score: haltResult.score,
|
|
28855
|
+
executionStatus: haltResult.executionStatus
|
|
28779
28856
|
});
|
|
28780
28857
|
}
|
|
28781
28858
|
if (onResult) {
|
|
@@ -28855,7 +28932,9 @@ async function runEvaluation(options) {
|
|
|
28855
28932
|
startedAt: 0,
|
|
28856
28933
|
// Not used for completed status
|
|
28857
28934
|
completedAt: Date.now(),
|
|
28858
|
-
error: result.error
|
|
28935
|
+
error: result.error,
|
|
28936
|
+
score: result.score,
|
|
28937
|
+
executionStatus: result.executionStatus
|
|
28859
28938
|
});
|
|
28860
28939
|
}
|
|
28861
28940
|
if (onResult) {
|
|
@@ -29026,7 +29105,9 @@ async function runBatchEvaluation(options) {
|
|
|
29026
29105
|
const merged = computed ? mergeExecutionMetrics(computed, {
|
|
29027
29106
|
tokenUsage: providerResponse.tokenUsage,
|
|
29028
29107
|
costUsd: providerResponse.costUsd,
|
|
29029
|
-
durationMs: providerResponse.durationMs
|
|
29108
|
+
durationMs: providerResponse.durationMs,
|
|
29109
|
+
startTime: providerResponse.startTime,
|
|
29110
|
+
endTime: providerResponse.endTime
|
|
29030
29111
|
}) : void 0;
|
|
29031
29112
|
const trace2 = merged?.trace;
|
|
29032
29113
|
const costUsd = merged?.costUsd;
|
|
@@ -29091,7 +29172,9 @@ async function runBatchEvaluation(options) {
|
|
|
29091
29172
|
testId: evalCase.id,
|
|
29092
29173
|
status: "failed",
|
|
29093
29174
|
completedAt: Date.now(),
|
|
29094
|
-
error: error instanceof Error ? error.message : String(error)
|
|
29175
|
+
error: error instanceof Error ? error.message : String(error),
|
|
29176
|
+
score: errorResult.score,
|
|
29177
|
+
executionStatus: errorResult.executionStatus
|
|
29095
29178
|
});
|
|
29096
29179
|
}
|
|
29097
29180
|
continue;
|
|
@@ -29107,7 +29190,9 @@ async function runBatchEvaluation(options) {
|
|
|
29107
29190
|
status: result.error ? "failed" : "completed",
|
|
29108
29191
|
startedAt: 0,
|
|
29109
29192
|
completedAt: Date.now(),
|
|
29110
|
-
error: result.error
|
|
29193
|
+
error: result.error,
|
|
29194
|
+
score: result.score,
|
|
29195
|
+
executionStatus: result.executionStatus
|
|
29111
29196
|
});
|
|
29112
29197
|
}
|
|
29113
29198
|
}
|
|
@@ -29417,7 +29502,9 @@ async function runEvalCase(options) {
|
|
|
29417
29502
|
const merged = computed ? mergeExecutionMetrics(computed, {
|
|
29418
29503
|
tokenUsage: providerResponse.tokenUsage,
|
|
29419
29504
|
costUsd: providerResponse.costUsd,
|
|
29420
|
-
durationMs: providerResponse.durationMs
|
|
29505
|
+
durationMs: providerResponse.durationMs,
|
|
29506
|
+
startTime: providerResponse.startTime,
|
|
29507
|
+
endTime: providerResponse.endTime
|
|
29421
29508
|
}) : void 0;
|
|
29422
29509
|
const trace2 = merged?.trace;
|
|
29423
29510
|
const costUsd = merged?.costUsd;
|
|
@@ -31203,4 +31290,4 @@ export {
|
|
|
31203
31290
|
OtelStreamingObserver,
|
|
31204
31291
|
createAgentKernel
|
|
31205
31292
|
};
|
|
31206
|
-
//# sourceMappingURL=chunk-
|
|
31293
|
+
//# sourceMappingURL=chunk-GOZV2HN2.js.map
|