@effect-agent/testing 0.1.0-beta.25 → 0.1.0-beta.27
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/dist/index.mjs
CHANGED
|
@@ -2698,7 +2698,7 @@ const TravelPlannerPhase2 = Agent.define("travel-planner-phase-2", {
|
|
|
2698
2698
|
maxToolCalls: 4,
|
|
2699
2699
|
maxDuration: "30 seconds",
|
|
2700
2700
|
toolConcurrency: 3,
|
|
2701
|
-
tokenBudget:
|
|
2701
|
+
tokenBudget: 4096
|
|
2702
2702
|
}),
|
|
2703
2703
|
description: "Build a review-only itinerary and require approval before creating a temporary hold.",
|
|
2704
2704
|
metadata: {
|
|
@@ -4554,6 +4554,23 @@ const phase6TravelPlannerGoldenEvidence = [
|
|
|
4554
4554
|
turn: 1,
|
|
4555
4555
|
inputTokens: 128,
|
|
4556
4556
|
outputTokens: 96,
|
|
4557
|
+
runScopedPrefixLength: 2,
|
|
4558
|
+
modelUsage: [{
|
|
4559
|
+
provider: "scripted",
|
|
4560
|
+
model: "travel-planner-phase-4",
|
|
4561
|
+
inputTokens: {
|
|
4562
|
+
total: 128,
|
|
4563
|
+
uncached: 128,
|
|
4564
|
+
cacheRead: 0,
|
|
4565
|
+
cacheWrite: 0
|
|
4566
|
+
},
|
|
4567
|
+
outputTokens: {
|
|
4568
|
+
total: 96,
|
|
4569
|
+
text: 96,
|
|
4570
|
+
reasoning: 0
|
|
4571
|
+
},
|
|
4572
|
+
costMicrousd: 0
|
|
4573
|
+
}],
|
|
4557
4574
|
messages: { content: [
|
|
4558
4575
|
{
|
|
4559
4576
|
options: {},
|
|
@@ -4697,6 +4714,22 @@ const phase6TravelPlannerGoldenEvidence = [
|
|
|
4697
4714
|
turn: 2,
|
|
4698
4715
|
inputTokens: 128,
|
|
4699
4716
|
outputTokens: 96,
|
|
4717
|
+
modelUsage: [{
|
|
4718
|
+
provider: "scripted",
|
|
4719
|
+
model: "travel-planner-phase-4",
|
|
4720
|
+
inputTokens: {
|
|
4721
|
+
total: 128,
|
|
4722
|
+
uncached: 128,
|
|
4723
|
+
cacheRead: 0,
|
|
4724
|
+
cacheWrite: 0
|
|
4725
|
+
},
|
|
4726
|
+
outputTokens: {
|
|
4727
|
+
total: 96,
|
|
4728
|
+
text: 96,
|
|
4729
|
+
reasoning: 0
|
|
4730
|
+
},
|
|
4731
|
+
costMicrousd: 0
|
|
4732
|
+
}],
|
|
4700
4733
|
messages: { content: [{
|
|
4701
4734
|
options: {},
|
|
4702
4735
|
role: "assistant",
|
|
@@ -4707,8 +4740,37 @@ const phase6TravelPlannerGoldenEvidence = [
|
|
|
4707
4740
|
}
|
|
4708
4741
|
},
|
|
4709
4742
|
{
|
|
4710
|
-
batchId: "
|
|
4743
|
+
batchId: "turn:run:{submissionId}:2",
|
|
4711
4744
|
sequence: 8,
|
|
4745
|
+
record: {
|
|
4746
|
+
recordId: "run-completed:run:{submissionId}",
|
|
4747
|
+
family: "conversation",
|
|
4748
|
+
schemaVersion: 1,
|
|
4749
|
+
createdAt: "{timestamp}",
|
|
4750
|
+
deploymentId: "{deploymentId}",
|
|
4751
|
+
payload: {
|
|
4752
|
+
_tag: "RunCompleted",
|
|
4753
|
+
runId: "run:{submissionId}",
|
|
4754
|
+
output: { itineraries: [{
|
|
4755
|
+
title: "Westward light, eastbound overnight",
|
|
4756
|
+
route: "San Francisco → London",
|
|
4757
|
+
dates: "14–19 September 2026",
|
|
4758
|
+
flight: "EA 218 · nonstop · SFO 18:40 → LHR 13:05+1",
|
|
4759
|
+
lodging: "Bloomsbury House · refundable studio · 4 nights",
|
|
4760
|
+
activities: ["British Museum timed entry", "Thames evening walk"],
|
|
4761
|
+
estimatedTotalCents: 284e3,
|
|
4762
|
+
currency: "USD",
|
|
4763
|
+
quoteId: "quote-sfo-lhr-001",
|
|
4764
|
+
assumptions: ["Two travelers sharing one studio", "Quote is read-only availability, not a reservation"],
|
|
4765
|
+
unresolvedConstraints: ["Traveler names and accessibility requests are intentionally omitted"],
|
|
4766
|
+
nextAction: "review"
|
|
4767
|
+
}] }
|
|
4768
|
+
}
|
|
4769
|
+
}
|
|
4770
|
+
},
|
|
4771
|
+
{
|
|
4772
|
+
batchId: "submission-settlement:{submissionId}",
|
|
4773
|
+
sequence: 9,
|
|
4712
4774
|
record: {
|
|
4713
4775
|
recordId: "settlement:{submissionId}",
|
|
4714
4776
|
family: "conversation",
|
|
@@ -4735,7 +4797,39 @@ const phase6TravelPlannerGoldenEvidence = [
|
|
|
4735
4797
|
assumptions: ["Two travelers sharing one studio", "Quote is read-only availability, not a reservation"],
|
|
4736
4798
|
unresolvedConstraints: ["Traveler names and accessibility requests are intentionally omitted"],
|
|
4737
4799
|
nextAction: "review"
|
|
4738
|
-
}] }
|
|
4800
|
+
}] },
|
|
4801
|
+
usageSummary: {
|
|
4802
|
+
modelCalls: 2,
|
|
4803
|
+
inputTokens: {
|
|
4804
|
+
total: 256,
|
|
4805
|
+
uncached: 256,
|
|
4806
|
+
cacheRead: 0,
|
|
4807
|
+
cacheWrite: 0
|
|
4808
|
+
},
|
|
4809
|
+
outputTokens: {
|
|
4810
|
+
total: 192,
|
|
4811
|
+
text: 192,
|
|
4812
|
+
reasoning: 0
|
|
4813
|
+
},
|
|
4814
|
+
costMicrousd: 0,
|
|
4815
|
+
byModel: [{
|
|
4816
|
+
provider: "scripted",
|
|
4817
|
+
model: "travel-planner-phase-4",
|
|
4818
|
+
modelCalls: 2,
|
|
4819
|
+
inputTokens: {
|
|
4820
|
+
total: 256,
|
|
4821
|
+
uncached: 256,
|
|
4822
|
+
cacheRead: 0,
|
|
4823
|
+
cacheWrite: 0
|
|
4824
|
+
},
|
|
4825
|
+
outputTokens: {
|
|
4826
|
+
total: 192,
|
|
4827
|
+
text: 192,
|
|
4828
|
+
reasoning: 0
|
|
4829
|
+
},
|
|
4830
|
+
costMicrousd: 0
|
|
4831
|
+
}]
|
|
4832
|
+
}
|
|
4739
4833
|
}
|
|
4740
4834
|
}
|
|
4741
4835
|
}
|