@effect-agent/testing 0.1.0-beta.26 → 0.1.0-beta.28
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/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@effect-agent/testing",
|
|
3
|
-
"version": "0.1.0-beta.
|
|
3
|
+
"version": "0.1.0-beta.28",
|
|
4
4
|
"exports": {
|
|
5
5
|
".": {
|
|
6
6
|
"types": "./dist/index.d.mts",
|
|
@@ -8,14 +8,14 @@
|
|
|
8
8
|
}
|
|
9
9
|
},
|
|
10
10
|
"dependencies": {
|
|
11
|
-
"@effect-agent/capabilities": "0.1.0-beta.
|
|
12
|
-
"@effect-agent/core": "0.1.0-beta.
|
|
13
|
-
"@effect-agent/engine": "0.1.0-beta.
|
|
14
|
-
"@effect-agent/platform-node": "0.1.0-beta.
|
|
15
|
-
"@effect-agent/sandbox": "0.1.0-beta.
|
|
16
|
-
"@effect-agent/session": "0.1.0-beta.
|
|
17
|
-
"@effect-agent/storage-memory": "0.1.0-beta.
|
|
18
|
-
"@effect-agent/storage-sqlite": "0.1.0-beta.
|
|
11
|
+
"@effect-agent/capabilities": "0.1.0-beta.28",
|
|
12
|
+
"@effect-agent/core": "0.1.0-beta.28",
|
|
13
|
+
"@effect-agent/engine": "0.1.0-beta.28",
|
|
14
|
+
"@effect-agent/platform-node": "0.1.0-beta.28",
|
|
15
|
+
"@effect-agent/sandbox": "0.1.0-beta.28",
|
|
16
|
+
"@effect-agent/session": "0.1.0-beta.28",
|
|
17
|
+
"@effect-agent/storage-memory": "0.1.0-beta.28",
|
|
18
|
+
"@effect-agent/storage-sqlite": "0.1.0-beta.28",
|
|
19
19
|
"effect": "4.0.0-rc.110"
|
|
20
20
|
},
|
|
21
21
|
"description": "Scripted models, deterministic fixtures, and adapter conformance kits for testing Effect Agent applications.",
|
|
@@ -86,7 +86,7 @@ export const TravelPlannerPhase2 = Agent.define("travel-planner-phase-2", {
|
|
|
86
86
|
maxToolCalls: 4,
|
|
87
87
|
maxDuration: "30 seconds",
|
|
88
88
|
toolConcurrency: 3,
|
|
89
|
-
tokenBudget:
|
|
89
|
+
tokenBudget: 4_096,
|
|
90
90
|
}),
|
|
91
91
|
description:
|
|
92
92
|
"Build a review-only itinerary and require approval before creating a temporary hold.",
|
|
@@ -719,6 +719,25 @@ export const phase6TravelPlannerGoldenEvidence: Schema.Json = [
|
|
|
719
719
|
turn: 1,
|
|
720
720
|
inputTokens: 128,
|
|
721
721
|
outputTokens: 96,
|
|
722
|
+
runScopedPrefixLength: 2,
|
|
723
|
+
modelUsage: [
|
|
724
|
+
{
|
|
725
|
+
provider: "scripted",
|
|
726
|
+
model: "travel-planner-phase-4",
|
|
727
|
+
inputTokens: {
|
|
728
|
+
total: 128,
|
|
729
|
+
uncached: 128,
|
|
730
|
+
cacheRead: 0,
|
|
731
|
+
cacheWrite: 0,
|
|
732
|
+
},
|
|
733
|
+
outputTokens: {
|
|
734
|
+
total: 96,
|
|
735
|
+
text: 96,
|
|
736
|
+
reasoning: 0,
|
|
737
|
+
},
|
|
738
|
+
costMicrousd: 0,
|
|
739
|
+
},
|
|
740
|
+
],
|
|
722
741
|
messages: {
|
|
723
742
|
content: [
|
|
724
743
|
{
|
|
@@ -868,6 +887,24 @@ export const phase6TravelPlannerGoldenEvidence: Schema.Json = [
|
|
|
868
887
|
turn: 2,
|
|
869
888
|
inputTokens: 128,
|
|
870
889
|
outputTokens: 96,
|
|
890
|
+
modelUsage: [
|
|
891
|
+
{
|
|
892
|
+
provider: "scripted",
|
|
893
|
+
model: "travel-planner-phase-4",
|
|
894
|
+
inputTokens: {
|
|
895
|
+
total: 128,
|
|
896
|
+
uncached: 128,
|
|
897
|
+
cacheRead: 0,
|
|
898
|
+
cacheWrite: 0,
|
|
899
|
+
},
|
|
900
|
+
outputTokens: {
|
|
901
|
+
total: 96,
|
|
902
|
+
text: 96,
|
|
903
|
+
reasoning: 0,
|
|
904
|
+
},
|
|
905
|
+
costMicrousd: 0,
|
|
906
|
+
},
|
|
907
|
+
],
|
|
871
908
|
messages: {
|
|
872
909
|
content: [
|
|
873
910
|
{
|
|
@@ -883,8 +920,46 @@ export const phase6TravelPlannerGoldenEvidence: Schema.Json = [
|
|
|
883
920
|
},
|
|
884
921
|
},
|
|
885
922
|
{
|
|
886
|
-
batchId: "
|
|
923
|
+
batchId: "turn:run:{submissionId}:2",
|
|
887
924
|
sequence: 8,
|
|
925
|
+
record: {
|
|
926
|
+
recordId: "run-completed:run:{submissionId}",
|
|
927
|
+
family: "conversation",
|
|
928
|
+
schemaVersion: 1,
|
|
929
|
+
createdAt: "{timestamp}",
|
|
930
|
+
deploymentId: "{deploymentId}",
|
|
931
|
+
payload: {
|
|
932
|
+
_tag: "RunCompleted",
|
|
933
|
+
runId: "run:{submissionId}",
|
|
934
|
+
output: {
|
|
935
|
+
itineraries: [
|
|
936
|
+
{
|
|
937
|
+
title: "Westward light, eastbound overnight",
|
|
938
|
+
route: "San Francisco → London",
|
|
939
|
+
dates: "14–19 September 2026",
|
|
940
|
+
flight: "EA 218 · nonstop · SFO 18:40 → LHR 13:05+1",
|
|
941
|
+
lodging: "Bloomsbury House · refundable studio · 4 nights",
|
|
942
|
+
activities: ["British Museum timed entry", "Thames evening walk"],
|
|
943
|
+
estimatedTotalCents: 284000,
|
|
944
|
+
currency: "USD",
|
|
945
|
+
quoteId: "quote-sfo-lhr-001",
|
|
946
|
+
assumptions: [
|
|
947
|
+
"Two travelers sharing one studio",
|
|
948
|
+
"Quote is read-only availability, not a reservation",
|
|
949
|
+
],
|
|
950
|
+
unresolvedConstraints: [
|
|
951
|
+
"Traveler names and accessibility requests are intentionally omitted",
|
|
952
|
+
],
|
|
953
|
+
nextAction: "review",
|
|
954
|
+
},
|
|
955
|
+
],
|
|
956
|
+
},
|
|
957
|
+
},
|
|
958
|
+
},
|
|
959
|
+
},
|
|
960
|
+
{
|
|
961
|
+
batchId: "submission-settlement:{submissionId}",
|
|
962
|
+
sequence: 9,
|
|
888
963
|
record: {
|
|
889
964
|
recordId: "settlement:{submissionId}",
|
|
890
965
|
family: "conversation",
|
|
@@ -921,6 +996,40 @@ export const phase6TravelPlannerGoldenEvidence: Schema.Json = [
|
|
|
921
996
|
},
|
|
922
997
|
],
|
|
923
998
|
},
|
|
999
|
+
usageSummary: {
|
|
1000
|
+
modelCalls: 2,
|
|
1001
|
+
inputTokens: {
|
|
1002
|
+
total: 256,
|
|
1003
|
+
uncached: 256,
|
|
1004
|
+
cacheRead: 0,
|
|
1005
|
+
cacheWrite: 0,
|
|
1006
|
+
},
|
|
1007
|
+
outputTokens: {
|
|
1008
|
+
total: 192,
|
|
1009
|
+
text: 192,
|
|
1010
|
+
reasoning: 0,
|
|
1011
|
+
},
|
|
1012
|
+
costMicrousd: 0,
|
|
1013
|
+
byModel: [
|
|
1014
|
+
{
|
|
1015
|
+
provider: "scripted",
|
|
1016
|
+
model: "travel-planner-phase-4",
|
|
1017
|
+
modelCalls: 2,
|
|
1018
|
+
inputTokens: {
|
|
1019
|
+
total: 256,
|
|
1020
|
+
uncached: 256,
|
|
1021
|
+
cacheRead: 0,
|
|
1022
|
+
cacheWrite: 0,
|
|
1023
|
+
},
|
|
1024
|
+
outputTokens: {
|
|
1025
|
+
total: 192,
|
|
1026
|
+
text: 192,
|
|
1027
|
+
reasoning: 0,
|
|
1028
|
+
},
|
|
1029
|
+
costMicrousd: 0,
|
|
1030
|
+
},
|
|
1031
|
+
],
|
|
1032
|
+
},
|
|
924
1033
|
},
|
|
925
1034
|
},
|
|
926
1035
|
},
|