@tangle-network/agent-eval 0.145.7 → 0.145.9

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.
Files changed (36) hide show
  1. package/CHANGELOG.md +18 -0
  2. package/dist/analyst/index.d.ts +1 -1
  3. package/dist/analyst/index.js +1 -1
  4. package/dist/{benchmark-command-C7L0rQsD.js → benchmark-command-k6kI1D8H.js} +2 -2
  5. package/dist/{benchmark-command-C7L0rQsD.js.map → benchmark-command-k6kI1D8H.js.map} +1 -1
  6. package/dist/benchmarks/index.d.ts +2 -2
  7. package/dist/benchmarks/index.js +2 -2
  8. package/dist/campaign/index.d.ts +4 -4
  9. package/dist/campaign/index.js +4 -4
  10. package/dist/{campaign-DRcGtxEf.js → campaign-B3GJpJ4h.js} +9 -6
  11. package/dist/campaign-B3GJpJ4h.js.map +1 -0
  12. package/dist/cli.js +1 -1
  13. package/dist/contract/index.d.ts +3 -3
  14. package/dist/contract/index.js +3 -3
  15. package/dist/{define-agent-eval-DBOpAM_g.d.ts → define-agent-eval-BgG5DIOS.d.ts} +2 -2
  16. package/dist/{define-agent-eval-DBOpAM_g.d.ts.map → define-agent-eval-BgG5DIOS.d.ts.map} +1 -1
  17. package/dist/{define-agent-eval-m5lM7XGE.js → define-agent-eval-cSMEFzVu.js} +2 -2
  18. package/dist/{define-agent-eval-m5lM7XGE.js.map → define-agent-eval-cSMEFzVu.js.map} +1 -1
  19. package/dist/{index-D7gl974A.d.ts → index-CM-e2LiV.d.ts} +3 -3
  20. package/dist/{index-D7gl974A.d.ts.map → index-CM-e2LiV.d.ts.map} +1 -1
  21. package/dist/index.d.ts +2 -2
  22. package/dist/index.js +3 -3
  23. package/dist/{llm-judge-BO1LGjdn.js → llm-judge-DhiJqSjB.js} +2 -2
  24. package/dist/{llm-judge-BO1LGjdn.js.map → llm-judge-DhiJqSjB.js.map} +1 -1
  25. package/dist/openapi.json +1 -1
  26. package/dist/{produced-state-C6vSgOpE.js → produced-state-BNyyud4g.js} +2 -2
  27. package/dist/{produced-state-C6vSgOpE.js.map → produced-state-BNyyud4g.js.map} +1 -1
  28. package/dist/{provenance-BOtMtoiZ.d.ts → provenance-AACfvbUR.d.ts} +120 -104
  29. package/dist/provenance-AACfvbUR.d.ts.map +1 -0
  30. package/dist/{skillopt-optimization-method-Ceic2bfU.js → skillopt-optimization-method-C6JSfYxT.js} +2 -2
  31. package/dist/{skillopt-optimization-method-Ceic2bfU.js.map → skillopt-optimization-method-C6JSfYxT.js.map} +1 -1
  32. package/dist/{skillopt-optimization-method-BPQwXkkY.d.ts → skillopt-optimization-method-DO6OFgcz.d.ts} +2 -2
  33. package/dist/{skillopt-optimization-method-BPQwXkkY.d.ts.map → skillopt-optimization-method-DO6OFgcz.d.ts.map} +1 -1
  34. package/package.json +1 -1
  35. package/dist/campaign-DRcGtxEf.js.map +0 -1
  36. package/dist/provenance-BOtMtoiZ.d.ts.map +0 -1
@@ -3,7 +3,7 @@ import { a as RunRecord } from "./run-record-DVV82Gwh.js";
3
3
  import { E as LlmCallMetadata, p as ChatClient } from "./types-Cx3YUh2r.js";
4
4
  import { _ as ExternalTextEvaluationRequest, c as ExternalOptimizerModelBudget, g as ExternalTextCandidate, h as ExternalOptimizerRunnerCommand, l as ExternalOptimizerModelCall, m as ExternalOptimizerResumeMode, r as ExternalOptimizerCallbackLimits } from "./external-optimizer-contracts-lixrOZdX.js";
5
5
  import { R as Scenario, S as JudgeConfig, i as CampaignCostMeter, j as MutableSurface, o as CampaignScenarioIdentity, p as Gate } from "./types-BjsNDR49.js";
6
- import { Q as OptimizationPackageSource, W as OptimizationMethod, k as RedTeamCase, z as CanaryOptions } from "./provenance-BOtMtoiZ.js";
6
+ import { Q as OptimizationPackageSource, W as OptimizationMethod, k as RedTeamCase, z as CanaryOptions } from "./provenance-AACfvbUR.js";
7
7
  import { t as DetectRewardHackingInput } from "./reward-hacking-Bu8ev6PR.js";
8
8
  //#region src/reference-equivalence-judge.d.ts
9
9
  declare const REFERENCE_EQUIVALENCE_JUDGE_VERSION = "reference-equivalence-judge-v1-2026-07-13";
@@ -431,4 +431,4 @@ interface SkillOptOptimizationMethodConfig<TScenario extends Scenario, TArtifact
431
431
  declare function skillOptOptimizationMethod<TScenario extends Scenario, TArtifact>(config: SkillOptOptimizationMethodConfig<TScenario, TArtifact>): OptimizationMethod<TScenario, TArtifact>;
432
432
  //#endregion
433
433
  export { campaignScenarioIdentity as A, runReferenceEquivalenceJudge as B, ExternalTextOptimizerContext as C, decodeExternalTextCandidate as D, ExternalTextEvaluationResponse as E, ReferenceEquivalenceJudgeInput as F, ReferenceEquivalenceJudgeOptions as I, ReferenceEquivalenceJudgeResult as L, campaignSplitDigestFromIdentities as M, REFERENCE_EQUIVALENCE_INPUT_LIMITS as N, assertCampaignDesign as O, REFERENCE_EQUIVALENCE_JUDGE_VERSION as P, ReferenceEquivalenceScenario as R, ExternalTextOptimizationMethodConfig as S, ExternalOptimizationExample as T, DefaultProductionGateOptions as _, GepaAdaptiveEngineRun as a, composeGate as b, GepaOptimizationMethodConfig as c, gepaOptimizationMethod as d, OpenAICompatibleOptimizerModel as f, DefaultProductionGateCheck as g, heldOutGate as h, skillOptOptimizationMethod as i, campaignSplitDigest as j, assertCampaignSplitIdentity as k, GepaOptimizationRecipe as l, HeldOutGateOptions as m, SkillOptRunnerCommand as n, GepaEngineOptions as o, OptimizerModelBudget as p, SkillOptTrainerConfig as r, GepaEngineRun as s, SkillOptOptimizationMethodConfig as t, GepaRunnerCommand as u, DefaultProductionRewardHackingOptions as v, ExternalTextOptimizerResult as w, externalTextOptimizationMethod as x, defaultProductionGate as y, createReferenceEquivalenceJudge as z };
434
- //# sourceMappingURL=skillopt-optimization-method-BPQwXkkY.d.ts.map
434
+ //# sourceMappingURL=skillopt-optimization-method-DO6OFgcz.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"skillopt-optimization-method-BPQwXkkY.d.ts","names":[],"sources":["../src/reference-equivalence-judge.ts","../src/campaign/coverage.ts","../src/campaign/external-text-evaluation.ts","../src/campaign/external-text-optimization-contract.ts","../src/campaign/external-text-optimization.ts","../src/campaign/gates/compose.ts","../src/campaign/gates/default-production-gate.ts","../src/campaign/gates/heldout-gate.ts","../src/campaign/optimizer-model.ts","../src/campaign/gepa-optimization-method.ts","../src/campaign/skillopt-optimization-method.ts"],"mappings":";;;;;;;;cAOa;cAEA;WACX;WACA;WACA;;UAGe,qCAAqC;EACpD;EACA;;UAGe;EACf;EACA;EACA;;UAGe;;EAEf,MAAM;;EAEN;;EAEA,SAAS;;EAET,aAAa;;UAGE,wCAAwC;EACvD;EACA;EACA;EACA;;;iBA0Bc,gCACd,SAAS,mCACR,oBAAoB;;iBA2CD,6BACpB,OAAO,gCACP,SAAS,mCACR,QAAQ;;;;iBCxGK,qBAAqB,kBAAkB,UACrD,oBAAoB,aACpB;;iBA2Bc,yBAAyB,kBAAkB,UACzD,UAAU,YACT,2BAA2B,KAAK;;iBAUnB,kCACd,oBAAoB,4BACpB;;iBAgBc,oBAAoB,kBAAkB,UACpD,oBAAoB,aACpB;;iBAOc,4BACd,oBAAoB,4BACpB,cACA;;;UCpEe;EACf;EACA;;UAGe;EACf;EACA;IACE;IACA,YAAY;IACZ;IACA;;;IAGA;IACA;MACE;MACA;;;;iBAwHU,4BAA4B,WAAW,wBAAwB;;;UC3I9D;WACN;WACA;WACA;WACA;WACA;WACA,eAAe;WACf,mBAAmB;WACnB,uBAAuB;WACvB;WACA;;WAEA;WACA;WACA;WACA,QAAQ;;WAER,MAAM;WACN,WACP,SAFa,kCAGV,QAAQ;;UAGE;EACf,eAAe;EACf;EACA;IACM;;IACA;;IACA;IAAkB;;;;;;;;;UAqBT,qCACf,kBAAkB,UAClB;EAEA;EACA,QAAQ,KAAK;EACb;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA,SAAS;EACT;EACA;EACA,oBAAoB,UAAU;EAC9B,oBAAoB,UAAU,WAAW,UAAU;EACnD,MAAM,SAAS,iCAAiC,QAAQ;;;;;;;;;;iBCN1C,+BAA+B,kBAAkB,UAAU,WACzE,QAAQ,qCAAqC,WAAW,aACvD,mBAAmB,WAAW;;;;;;iBClEjB,YAAY,qBAAqB,kBAAkB,WAAW,aACzE,OAAO,MAAM,KAAK,WAAW,cAC/B,KAAK,WAAW;;;KCaP;KAOA,wCAAwC,KAClD;EAGA,SAAS,YAAY;;UAGN;;;EAGf,kBAAkB;;;;;EAKlB;;EAEA;;EAEA;;EAEA;;;;EAIA;;;;EAIA;;;;;EAKA;;;;EAIA;;;EAGA;;;;EAIA,iBAAiB;;;EAGjB,aAAa;;EAEb,gBAAgB;;EAEhB,SAAS;;;;;EAKT,iBAAiB;;;;;iBAMH,sBAAsB,WAAW,kBAAkB,UACjE,SAAS,+BACR,KAAK,WAAW;;;UC/EF,mBAAmB,kBAAkB,WAAW;EAC/D,WAAW;;;EAGX;;EAEA;;;;EAIA;;EAEA;;EAEA;;;;;;;;iBASc,YAAY,WAAW,kBAAkB,UACvD,SAAS,mBAAmB,aAC3B,KAAK,WAAW;;;KCvCP,uBAAuB;;UAGlB;EACf;EACA,QAAQ;;EAER,MAAM;;EAEN;;;;;UCmDe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;;;;EAKA,eAAe;;;UAIA,sBAAsB;;EAErC;;;KAIU,wBAAwB;;;;;;;KAQxB;EAEN;EACA,KAAK;;EAGL;EACA,eAAe;;EAGf;EACA,eAAe;;EAEf;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;;EAGA;EACA,eAAe;EACf;;EAGA;EACA,eAAe;EACf;;EAGA;EACA,kBAAkB;EAClB,cAAc;EACd;;;KAIM,oBAAoB;UAEf,6BAA6B,kBAAkB,UAAU;;EAExE;;EAEA,QAAQ;;EAER;;EAEA;;EAEA;;;;;EAKA;;EAEA;;EAEA;;EAEA,2BAA2B,QAAQ;;EAEnC;;;;;;EAMA,YAAY;;;;;;EAMZ,oBAAoB,UAAU;;EAE9B,oBAAoB,UAAU,WAAW,UAAU;;EAEnD,SAAS;;;;;EAKT;EACA,SAAS;;;;;;;;;;iBAWK,uBAAuB,kBAAkB,UAAU,WACjE,QAAQ,6BAA6B,WAAW,aAC/C,mBAAmB,WAAW;;;UC5IhB;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;;;;EAKA,YAAY;;KAGF,wBAAwB;UAEnB,iCAAiC,kBAAkB,UAAU;EAC5E;;EAEA;EACA;;EAEA;EACA,SAAS;;;;;EAKT,WAAW;;EAEX;;EAEA;EACA;;EAEA;;EAEA,2BAA2B,QAAQ;EACnC;EACA,oBAAoB,UAAU;EAC9B,oBAAoB,UAAU,WAAW,UAAU;EACnD,SAAS;EACT,SAAS;;;iBAIK,2BAA2B,kBAAkB,UAAU,WACrE,QAAQ,iCAAiC,WAAW,aACnD,mBAAmB,WAAW"}
1
+ {"version":3,"file":"skillopt-optimization-method-DO6OFgcz.d.ts","names":[],"sources":["../src/reference-equivalence-judge.ts","../src/campaign/coverage.ts","../src/campaign/external-text-evaluation.ts","../src/campaign/external-text-optimization-contract.ts","../src/campaign/external-text-optimization.ts","../src/campaign/gates/compose.ts","../src/campaign/gates/default-production-gate.ts","../src/campaign/gates/heldout-gate.ts","../src/campaign/optimizer-model.ts","../src/campaign/gepa-optimization-method.ts","../src/campaign/skillopt-optimization-method.ts"],"mappings":";;;;;;;;cAOa;cAEA;WACX;WACA;WACA;;UAGe,qCAAqC;EACpD;EACA;;UAGe;EACf;EACA;EACA;;UAGe;;EAEf,MAAM;;EAEN;;EAEA,SAAS;;EAET,aAAa;;UAGE,wCAAwC;EACvD;EACA;EACA;EACA;;;iBA0Bc,gCACd,SAAS,mCACR,oBAAoB;;iBA2CD,6BACpB,OAAO,gCACP,SAAS,mCACR,QAAQ;;;;iBCxGK,qBAAqB,kBAAkB,UACrD,oBAAoB,aACpB;;iBA2Bc,yBAAyB,kBAAkB,UACzD,UAAU,YACT,2BAA2B,KAAK;;iBAUnB,kCACd,oBAAoB,4BACpB;;iBAgBc,oBAAoB,kBAAkB,UACpD,oBAAoB,aACpB;;iBAOc,4BACd,oBAAoB,4BACpB,cACA;;;UCpEe;EACf;EACA;;UAGe;EACf;EACA;IACE;IACA,YAAY;IACZ;IACA;;;IAGA;IACA;MACE;MACA;;;;iBAwHU,4BAA4B,WAAW,wBAAwB;;;UC3I9D;WACN;WACA;WACA;WACA;WACA;WACA,eAAe;WACf,mBAAmB;WACnB,uBAAuB;WACvB;WACA;;WAEA;WACA;WACA;WACA,QAAQ;;WAER,MAAM;WACN,WACP,SAFa,kCAGV,QAAQ;;UAGE;EACf,eAAe;EACf;EACA;IACM;;IACA;;IACA;IAAkB;;;;;;;;;UAqBT,qCACf,kBAAkB,UAClB;EAEA;EACA,QAAQ,KAAK;EACb;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA,SAAS;EACT;EACA;EACA,oBAAoB,UAAU;EAC9B,oBAAoB,UAAU,WAAW,UAAU;EACnD,MAAM,SAAS,iCAAiC,QAAQ;;;;;;;;;;iBCN1C,+BAA+B,kBAAkB,UAAU,WACzE,QAAQ,qCAAqC,WAAW,aACvD,mBAAmB,WAAW;;;;;;iBClEjB,YAAY,qBAAqB,kBAAkB,WAAW,aACzE,OAAO,MAAM,KAAK,WAAW,cAC/B,KAAK,WAAW;;;KCaP;KAOA,wCAAwC,KAClD;EAGA,SAAS,YAAY;;UAGN;;;EAGf,kBAAkB;;;;;EAKlB;;EAEA;;EAEA;;EAEA;;;;EAIA;;;;EAIA;;;;;EAKA;;;;EAIA;;;EAGA;;;;EAIA,iBAAiB;;;EAGjB,aAAa;;EAEb,gBAAgB;;EAEhB,SAAS;;;;;EAKT,iBAAiB;;;;;iBAMH,sBAAsB,WAAW,kBAAkB,UACjE,SAAS,+BACR,KAAK,WAAW;;;UC/EF,mBAAmB,kBAAkB,WAAW;EAC/D,WAAW;;;EAGX;;EAEA;;;;EAIA;;EAEA;;EAEA;;;;;;;;iBASc,YAAY,WAAW,kBAAkB,UACvD,SAAS,mBAAmB,aAC3B,KAAK,WAAW;;;KCvCP,uBAAuB;;UAGlB;EACf;EACA,QAAQ;;EAER,MAAM;;EAEN;;;;;UCmDe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;;;;EAKA,eAAe;;;UAIA,sBAAsB;;EAErC;;;KAIU,wBAAwB;;;;;;;KAQxB;EAEN;EACA,KAAK;;EAGL;EACA,eAAe;;EAGf;EACA,eAAe;;EAEf;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;;EAGA;EACA,eAAe;EACf;;EAGA;EACA,eAAe;EACf;;EAGA;EACA,kBAAkB;EAClB,cAAc;EACd;;;KAIM,oBAAoB;UAEf,6BAA6B,kBAAkB,UAAU;;EAExE;;EAEA,QAAQ;;EAER;;EAEA;;EAEA;;;;;EAKA;;EAEA;;EAEA;;EAEA,2BAA2B,QAAQ;;EAEnC;;;;;;EAMA,YAAY;;;;;;EAMZ,oBAAoB,UAAU;;EAE9B,oBAAoB,UAAU,WAAW,UAAU;;EAEnD,SAAS;;;;;EAKT;EACA,SAAS;;;;;;;;;;iBAWK,uBAAuB,kBAAkB,UAAU,WACjE,QAAQ,6BAA6B,WAAW,aAC/C,mBAAmB,WAAW;;;UC5IhB;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;;;;EAKA,YAAY;;KAGF,wBAAwB;UAEnB,iCAAiC,kBAAkB,UAAU;EAC5E;;EAEA;EACA;;EAEA;EACA,SAAS;;;;;EAKT,WAAW;;EAEX;;EAEA;EACA;;EAEA;;EAEA,2BAA2B,QAAQ;EACnC;EACA,oBAAoB,UAAU;EAC9B,oBAAoB,UAAU,WAAW,UAAU;EACnD,SAAS;EACT,SAAS;;;iBAIK,2BAA2B,kBAAkB,UAAU,WACrE,QAAQ,iCAAiC,WAAW,aACnD,mBAAmB,WAAW"}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@tangle-network/agent-eval",
3
- "version": "0.145.7",
3
+ "version": "0.145.9",
4
4
  "description": "Evaluate and improve AI agents from runs, traces, judges, and feedback. Compare candidates, cluster failures, measure lift, and gate releases.",
5
5
  "homepage": "https://github.com/tangle-network/agent-eval#readme",
6
6
  "repository": {