@effect-agent/testing 0.1.0-beta.6 → 0.1.0-beta.8
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.d.mts +39 -11
- package/dist/index.mjs +5 -0
- package/dist/index.mjs.map +1 -1
- package/package.json +18 -18
- package/src/certification.ts +5 -0
- package/src/fixtures/travel-planner/phase6.ts +4 -0
package/dist/index.d.mts
CHANGED
|
@@ -141,7 +141,7 @@ declare const ChaosPlan_base: Schema.Class<ChaosPlan, Schema.Struct<{
|
|
|
141
141
|
readonly lanes: Schema.Int;
|
|
142
142
|
readonly submissions: Schema.NonEmptyArray<typeof ChaosSubmissionSpec>;
|
|
143
143
|
/** Coordinator failpoint arms, consumed one per round (each fails every hit that round). */
|
|
144
|
-
readonly failpointArms: Schema.$Array<Schema.Literals<readonly ["submit:after-admit", "submit:after-materialize", "claim:after-claim", "input:after-canonical-append", "turn:after-canonical-append", "turn:after-response-append", "turn:after-results-append", "tools:after-prepared-append", "step:after-step-append", "approval:after-request-append", "approval:after-suspend", "join:after-claim", "join:after-canonical-append", "resolve:after-intent", "terminalize:after-reserve", "terminalize:after-canonical-append", "abort:after-intent", "subagent:after-reserve", "subagent:after-request-append", "subagent:after-admit", "subagent:after-child-ready", "subagent:after-start-append", "subagent:after-sibling-settle", "subagent:after-suspend", "subagent:after-child-abort-intent", "subagent:after-join-append", "subagent:after-release-pending", "subagent:after-release"]>>;
|
|
144
|
+
readonly failpointArms: Schema.$Array<Schema.Literals<readonly ["submit:after-admit", "submit:after-materialize", "claim:after-claim", "input:after-canonical-append", "turn:after-canonical-append", "turn:after-response-append", "turn:after-results-append", "compaction:after-canonical-append", "tools:after-prepared-append", "step:after-step-append", "approval:after-request-append", "approval:after-suspend", "join:after-claim", "join:after-canonical-append", "resolve:after-intent", "terminalize:after-reserve", "terminalize:after-canonical-append", "abort:after-intent", "subagent:after-reserve", "subagent:after-request-append", "subagent:after-admit", "subagent:after-child-ready", "subagent:after-start-append", "subagent:after-sibling-settle", "subagent:after-suspend", "subagent:after-child-abort-intent", "subagent:after-join-append", "subagent:after-release-pending", "subagent:after-release"]>>;
|
|
145
145
|
/** Adapter-owned failpoint arms (e.g. SQLite `ledger:*`/`append:*` locations). */
|
|
146
146
|
readonly adapterArms: Schema.$Array<Schema.String>;
|
|
147
147
|
/** Flattened submission indices to abort mid-plan (modulo the submission count). */
|
|
@@ -365,7 +365,7 @@ declare const delegateDocumentSummary: import("@effect-agent/capabilities").Suba
|
|
|
365
365
|
readonly failure: typeof DocumentUnavailable;
|
|
366
366
|
readonly failureMode: "error";
|
|
367
367
|
}, DocumentLibrary>;
|
|
368
|
-
}, typeof SummaryRequest, typeof SummaryFinding, typeof DocumentSummaryFailed, never, never>;
|
|
368
|
+
}, typeof SummaryRequest, typeof SummaryFinding, typeof DocumentSummaryFailed, never, never, "error">;
|
|
369
369
|
/** Total mapping from every expected child Run failure to the declared Tool failure (SUB-028). */
|
|
370
370
|
declare const mapSummaryChildFailure: (failure: {
|
|
371
371
|
readonly _tag: string;
|
|
@@ -401,10 +401,20 @@ declare class ResearchDigest extends ResearchDigest_base {}
|
|
|
401
401
|
declare const docsCoordinatorConfidentialMarker = "docs-coordinator-vault-19x";
|
|
402
402
|
declare const docsMissionConfidentialMarker = "docs-mission-dossier-42f";
|
|
403
403
|
declare const DocsResearcherToolkit: Toolkit.Toolkit<{
|
|
404
|
-
readonly delegate_document_summary:
|
|
404
|
+
readonly delegate_document_summary: Tool.Tool<"delegate_document_summary", {
|
|
405
|
+
readonly parameters: typeof SummaryRequest;
|
|
406
|
+
readonly success: typeof SummaryFinding;
|
|
407
|
+
readonly failure: import("@effect-agent/capabilities").SubagentToolFailure<typeof DocumentSummaryFailed>;
|
|
408
|
+
readonly failureMode: "error";
|
|
409
|
+
}, import("@effect-agent/engine").AgentSpawner | import("@effect-agent/core").IdGenerator | import("@effect-agent/engine").RunEventSink | import("@effect-agent/engine").SubagentDurability>;
|
|
405
410
|
}>;
|
|
406
411
|
declare const DocsResearcher: import("@effect-agent/core").Definition<typeof ResearchRequest, typeof ResearchDigest, string, Toolkit.Toolkit<{
|
|
407
|
-
readonly delegate_document_summary:
|
|
412
|
+
readonly delegate_document_summary: Tool.Tool<"delegate_document_summary", {
|
|
413
|
+
readonly parameters: typeof SummaryRequest;
|
|
414
|
+
readonly success: typeof SummaryFinding;
|
|
415
|
+
readonly failure: import("@effect-agent/capabilities").SubagentToolFailure<typeof DocumentSummaryFailed>;
|
|
416
|
+
readonly failureMode: "error";
|
|
417
|
+
}, import("@effect-agent/engine").AgentSpawner | import("@effect-agent/core").IdGenerator | import("@effect-agent/engine").RunEventSink | import("@effect-agent/engine").SubagentDurability>;
|
|
408
418
|
}>>;
|
|
409
419
|
/** The default two-document research mission. */
|
|
410
420
|
declare const researchMissionRequest: ResearchRequest;
|
|
@@ -1043,6 +1053,8 @@ declare const phase3TravelPlannerEncodedFixture: readonly {
|
|
|
1043
1053
|
readonly turn: number;
|
|
1044
1054
|
readonly messages: Schema.Json;
|
|
1045
1055
|
readonly messagesDigest: string;
|
|
1056
|
+
readonly inputTokens?: number | undefined;
|
|
1057
|
+
readonly outputTokens?: number | undefined;
|
|
1046
1058
|
} | {
|
|
1047
1059
|
readonly _tag: "ToolCallSettled";
|
|
1048
1060
|
readonly runId: string;
|
|
@@ -1074,8 +1086,10 @@ declare const phase3TravelPlannerEncodedFixture: readonly {
|
|
|
1074
1086
|
} | {
|
|
1075
1087
|
readonly _tag: "CompactionCreated";
|
|
1076
1088
|
readonly runId: string;
|
|
1077
|
-
readonly
|
|
1078
|
-
readonly
|
|
1089
|
+
readonly turn: number;
|
|
1090
|
+
readonly kind: "clear-tool-results" | "summarize";
|
|
1091
|
+
readonly coversThrough: number;
|
|
1092
|
+
readonly summary?: string | undefined;
|
|
1079
1093
|
} | {
|
|
1080
1094
|
readonly _tag: "RunFailed";
|
|
1081
1095
|
readonly runId: string;
|
|
@@ -1224,6 +1238,8 @@ declare const phase3TravelPlannerEncodedFixture: readonly {
|
|
|
1224
1238
|
readonly turn: number;
|
|
1225
1239
|
readonly messages: Schema.Json;
|
|
1226
1240
|
readonly messagesDigest: string;
|
|
1241
|
+
readonly inputTokens?: number | undefined;
|
|
1242
|
+
readonly outputTokens?: number | undefined;
|
|
1227
1243
|
} | {
|
|
1228
1244
|
readonly _tag: "ToolCallSettled";
|
|
1229
1245
|
readonly runId: string;
|
|
@@ -1255,8 +1271,10 @@ declare const phase3TravelPlannerEncodedFixture: readonly {
|
|
|
1255
1271
|
} | {
|
|
1256
1272
|
readonly _tag: "CompactionCreated";
|
|
1257
1273
|
readonly runId: string;
|
|
1258
|
-
readonly
|
|
1259
|
-
readonly
|
|
1274
|
+
readonly turn: number;
|
|
1275
|
+
readonly kind: "clear-tool-results" | "summarize";
|
|
1276
|
+
readonly coversThrough: number;
|
|
1277
|
+
readonly summary?: string | undefined;
|
|
1260
1278
|
} | {
|
|
1261
1279
|
readonly _tag: "RunFailed";
|
|
1262
1280
|
readonly runId: string;
|
|
@@ -2774,7 +2792,7 @@ declare const destinationResearchDelegation: import("@effect-agent/capabilities"
|
|
|
2774
2792
|
readonly failure: typeof DestinationGuideUnavailable;
|
|
2775
2793
|
readonly failureMode: "error";
|
|
2776
2794
|
}, DestinationGuide>;
|
|
2777
|
-
}, typeof DestinationResearchRequest, typeof DestinationResearchFindings, typeof DestinationResearchFailed, ResearchDispatchGate, never>;
|
|
2795
|
+
}, typeof DestinationResearchRequest, typeof DestinationResearchFindings, typeof DestinationResearchFailed, ResearchDispatchGate, never, "error">;
|
|
2778
2796
|
/** Total mapping from every expected child Run failure to the declared Tool failure (SUB-028). */
|
|
2779
2797
|
declare const mapResearchChildFailure: (failure: {
|
|
2780
2798
|
readonly _tag: string;
|
|
@@ -2809,10 +2827,20 @@ declare class DestinationShortlist extends DestinationShortlist_base {}
|
|
|
2809
2827
|
declare const coordinatorConfidentialMarker = "coordinator-vault-7q42";
|
|
2810
2828
|
declare const missionConfidentialMarker = "traveler-dossier-19f";
|
|
2811
2829
|
declare const TravelCoordinatorToolkit: Toolkit.Toolkit<{
|
|
2812
|
-
readonly delegate_destination_research:
|
|
2830
|
+
readonly delegate_destination_research: Tool.Tool<"delegate_destination_research", {
|
|
2831
|
+
readonly parameters: typeof DestinationResearchRequest;
|
|
2832
|
+
readonly success: typeof DestinationResearchFindings;
|
|
2833
|
+
readonly failure: import("@effect-agent/capabilities").SubagentToolFailure<typeof DestinationResearchFailed>;
|
|
2834
|
+
readonly failureMode: "error";
|
|
2835
|
+
}, import("@effect-agent/engine").AgentSpawner | import("@effect-agent/core").IdGenerator | import("@effect-agent/engine").RunEventSink | import("@effect-agent/engine").SubagentDurability>;
|
|
2813
2836
|
}>;
|
|
2814
2837
|
declare const TravelCoordinator: import("@effect-agent/core").Definition<typeof ResearchMission, typeof DestinationShortlist, string, Toolkit.Toolkit<{
|
|
2815
|
-
readonly delegate_destination_research:
|
|
2838
|
+
readonly delegate_destination_research: Tool.Tool<"delegate_destination_research", {
|
|
2839
|
+
readonly parameters: typeof DestinationResearchRequest;
|
|
2840
|
+
readonly success: typeof DestinationResearchFindings;
|
|
2841
|
+
readonly failure: import("@effect-agent/capabilities").SubagentToolFailure<typeof DestinationResearchFailed>;
|
|
2842
|
+
readonly failureMode: "error";
|
|
2843
|
+
}, import("@effect-agent/engine").AgentSpawner | import("@effect-agent/core").IdGenerator | import("@effect-agent/engine").RunEventSink | import("@effect-agent/engine").SubagentDurability>;
|
|
2816
2844
|
}>>;
|
|
2817
2845
|
declare const researchMission: ResearchMission;
|
|
2818
2846
|
declare const expectedDestinationShortlist: DestinationShortlist;
|
package/dist/index.mjs
CHANGED
|
@@ -29,6 +29,7 @@ const CERTIFICATION_SCENARIOS = [
|
|
|
29
29
|
*/
|
|
30
30
|
const TIER2_UNREACHED_LOCATIONS = [
|
|
31
31
|
"abort:after-intent",
|
|
32
|
+
"compaction:after-canonical-append",
|
|
32
33
|
"resolve:after-intent",
|
|
33
34
|
"subagent:after-child-abort-intent"
|
|
34
35
|
];
|
|
@@ -4519,6 +4520,8 @@ const phase6TravelPlannerGoldenEvidence = [
|
|
|
4519
4520
|
runId: "run:{submissionId}",
|
|
4520
4521
|
turnId: "turn:run:{submissionId}:1",
|
|
4521
4522
|
turn: 1,
|
|
4523
|
+
inputTokens: 128,
|
|
4524
|
+
outputTokens: 96,
|
|
4522
4525
|
messages: { content: [
|
|
4523
4526
|
{
|
|
4524
4527
|
options: {},
|
|
@@ -4660,6 +4663,8 @@ const phase6TravelPlannerGoldenEvidence = [
|
|
|
4660
4663
|
runId: "run:{submissionId}",
|
|
4661
4664
|
turnId: "turn:run:{submissionId}:2",
|
|
4662
4665
|
turn: 2,
|
|
4666
|
+
inputTokens: 128,
|
|
4667
|
+
outputTokens: 96,
|
|
4663
4668
|
messages: { content: [{
|
|
4664
4669
|
options: {},
|
|
4665
4670
|
role: "assistant",
|