@evolvingmachines/evolve 0.0.65 → 0.0.67-train-harness-wave-2026-09-25.20260926.e00a8a6d

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 (153) hide show
  1. package/dist/{chunk-DK2MGUPB.js → chunk-DI6LPW2B.js} +2 -2
  2. package/dist/cli/index.cjs +29 -29
  3. package/dist/cli/index.d.cts +3 -3
  4. package/dist/cli/index.d.ts +3 -3
  5. package/dist/cli/index.js +18 -18
  6. package/dist/index.cjs +64 -60
  7. package/dist/index.d.cts +297 -18
  8. package/dist/index.d.ts +297 -18
  9. package/dist/index.js +48 -44
  10. package/dist/{types-B_qmzlMk.d.cts → types-Dj0iMl5V.d.cts} +70 -90
  11. package/dist/{types-B_qmzlMk.d.ts → types-Dj0iMl5V.d.ts} +70 -90
  12. package/harness-capabilities.json +285 -0
  13. package/package.json +9 -6
  14. package/skills/create-adapter/SKILL.md +71 -26
  15. package/skills/create-task/SKILL.md +124 -66
  16. package/skills/evolve/SKILL.md +37 -20
  17. package/skills/evolve-evals/SKILL.md +142 -69
  18. package/skills/evolve-evals/references/cli-reference/agent.md +76 -0
  19. package/skills/evolve-evals/references/cli-reference/analysis.md +78 -0
  20. package/skills/evolve-evals/references/cli-reference/analyze.md +97 -0
  21. package/skills/evolve-evals/references/cli-reference/auth.md +96 -0
  22. package/skills/evolve-evals/references/cli-reference/check.md +157 -0
  23. package/skills/evolve-evals/references/cli-reference/dataset.md +154 -0
  24. package/skills/evolve-evals/references/cli-reference/filesystem.md +168 -0
  25. package/skills/evolve-evals/references/cli-reference/index.md +149 -0
  26. package/skills/evolve-evals/references/cli-reference/job.md +198 -0
  27. package/skills/evolve-evals/references/cli-reference/run.md +188 -0
  28. package/skills/evolve-evals/references/cli-reference/secrets.md +77 -0
  29. package/skills/evolve-evals/references/cli-reference/session.md +56 -0
  30. package/skills/evolve-evals/references/cli-reference/skill.md +52 -0
  31. package/skills/evolve-evals/references/cli-reference/skills.md +85 -0
  32. package/skills/evolve-evals/references/cli-reference/trial.md +115 -0
  33. package/skills/evolve-evals/references/cli-reference/upload.md +72 -0
  34. package/skills/evolve-evals/references/core-concepts/agents.md +101 -0
  35. package/skills/evolve-evals/references/core-concepts/analyze.md +143 -0
  36. package/skills/evolve-evals/references/core-concepts/check.md +116 -0
  37. package/skills/evolve-evals/references/core-concepts/compatibility.md +39 -0
  38. package/skills/evolve-evals/references/core-concepts/compose.md +103 -0
  39. package/skills/evolve-evals/references/core-concepts/datasets.md +234 -0
  40. package/skills/evolve-evals/references/core-concepts/index.md +52 -0
  41. package/skills/evolve-evals/references/core-concepts/jobs.md +223 -0
  42. package/skills/evolve-evals/references/core-concepts/models.md +73 -0
  43. package/skills/evolve-evals/references/core-concepts/multi-step.md +93 -0
  44. package/skills/evolve-evals/references/core-concepts/rewardkit.md +341 -0
  45. package/skills/evolve-evals/references/core-concepts/sandboxes.md +73 -0
  46. package/skills/evolve-evals/references/core-concepts/secrets.md +86 -0
  47. package/skills/evolve-evals/references/core-concepts/sharing.md +81 -0
  48. package/skills/evolve-evals/references/core-concepts/skills.md +100 -0
  49. package/skills/evolve-evals/references/core-concepts/task-config.md +169 -0
  50. package/skills/evolve-evals/references/core-concepts/task-environment.md +127 -0
  51. package/skills/evolve-evals/references/core-concepts/task-reference.md +191 -0
  52. package/skills/evolve-evals/references/core-concepts/task-verifiers.md +209 -0
  53. package/skills/evolve-evals/references/core-concepts/tasks.md +90 -0
  54. package/skills/evolve-evals/references/core-concepts/teams.md +128 -0
  55. package/skills/evolve-evals/references/core-concepts/trial-outputs.md +225 -0
  56. package/skills/evolve-evals/references/core-concepts/trials.md +129 -0
  57. package/skills/evolve-evals/references/core-concepts/upload-sdk-session.md +240 -0
  58. package/skills/evolve-evals/references/core-concepts/upload.md +166 -0
  59. package/skills/evolve-evals/references/dashboard/checks.md +54 -0
  60. package/skills/evolve-evals/references/dashboard/datasets.md +65 -0
  61. package/skills/evolve-evals/references/dashboard/jobs.md +106 -0
  62. package/skills/evolve-evals/references/dashboard/traces.md +54 -0
  63. package/skills/evolve-evals/references/dashboard/trial-viewer.md +77 -0
  64. package/skills/evolve-evals/references/getting-started/first-task.md +228 -0
  65. package/skills/evolve-evals/references/getting-started/installation.md +79 -0
  66. package/skills/evolve-evals/references/getting-started/quick-start.md +100 -0
  67. package/skills/evolve-evals/references/index.md +58 -0
  68. package/skills/evolve-evals/references/sdk/http.md +110 -0
  69. package/skills/evolve-evals/references/sdk/python.md +108 -0
  70. package/skills/evolve-evals/references/sdk/typescript.md +104 -0
  71. package/skills/evolve-evals/references/sdk-reference/agents.md +75 -0
  72. package/skills/evolve-evals/references/sdk-reference/analyses.md +125 -0
  73. package/skills/evolve-evals/references/sdk-reference/auth.md +171 -0
  74. package/skills/evolve-evals/references/sdk-reference/checks.md +146 -0
  75. package/skills/evolve-evals/references/sdk-reference/datasets.md +170 -0
  76. package/skills/evolve-evals/references/sdk-reference/error-codes.md +231 -0
  77. package/skills/evolve-evals/references/sdk-reference/errors.md +98 -0
  78. package/skills/evolve-evals/references/sdk-reference/filesystem.md +166 -0
  79. package/skills/evolve-evals/references/sdk-reference/index.md +161 -0
  80. package/skills/evolve-evals/references/sdk-reference/jobs.md +250 -0
  81. package/skills/evolve-evals/references/sdk-reference/meta.md +54 -0
  82. package/skills/evolve-evals/references/sdk-reference/methods/agents.md +154 -0
  83. package/skills/evolve-evals/references/sdk-reference/methods/analyses.md +211 -0
  84. package/skills/evolve-evals/references/sdk-reference/methods/auth.md +257 -0
  85. package/skills/evolve-evals/references/sdk-reference/methods/checks.md +462 -0
  86. package/skills/evolve-evals/references/sdk-reference/methods/datasets.md +687 -0
  87. package/skills/evolve-evals/references/sdk-reference/methods/filesystem.md +527 -0
  88. package/skills/evolve-evals/references/sdk-reference/methods/jobs.md +1023 -0
  89. package/skills/evolve-evals/references/sdk-reference/methods/meta.md +213 -0
  90. package/skills/evolve-evals/references/sdk-reference/methods/secrets.md +157 -0
  91. package/skills/evolve-evals/references/sdk-reference/methods/skills.md +125 -0
  92. package/skills/evolve-evals/references/sdk-reference/methods/trials.md +371 -0
  93. package/skills/evolve-evals/references/sdk-reference/secrets.md +104 -0
  94. package/skills/evolve-evals/references/sdk-reference/skills.md +66 -0
  95. package/skills/evolve-evals/references/sdk-reference/trials.md +164 -0
  96. package/skills/evolve-evals/references/sdk-reference/types.md +1255 -0
  97. package/skills/publish/SKILL.md +103 -79
  98. package/skills/rewardkit/SKILL.md +78 -23
  99. package/spec/openapi.yaml +164 -160
  100. package/skills/evolve-evals/references/cli-reference/agent.mdx +0 -79
  101. package/skills/evolve-evals/references/cli-reference/analysis.mdx +0 -160
  102. package/skills/evolve-evals/references/cli-reference/analyze.mdx +0 -84
  103. package/skills/evolve-evals/references/cli-reference/auth.mdx +0 -133
  104. package/skills/evolve-evals/references/cli-reference/check.mdx +0 -263
  105. package/skills/evolve-evals/references/cli-reference/dataset.mdx +0 -151
  106. package/skills/evolve-evals/references/cli-reference/index.mdx +0 -108
  107. package/skills/evolve-evals/references/cli-reference/job.mdx +0 -278
  108. package/skills/evolve-evals/references/cli-reference/run.mdx +0 -204
  109. package/skills/evolve-evals/references/cli-reference/secrets.mdx +0 -95
  110. package/skills/evolve-evals/references/cli-reference/skill.mdx +0 -61
  111. package/skills/evolve-evals/references/cli-reference/skills.mdx +0 -81
  112. package/skills/evolve-evals/references/cli-reference/trial.mdx +0 -181
  113. package/skills/evolve-evals/references/cli-reference/upload.mdx +0 -36
  114. package/skills/evolve-evals/references/core-concepts/agents.mdx +0 -106
  115. package/skills/evolve-evals/references/core-concepts/analyze.mdx +0 -86
  116. package/skills/evolve-evals/references/core-concepts/check.mdx +0 -64
  117. package/skills/evolve-evals/references/core-concepts/datasets.mdx +0 -138
  118. package/skills/evolve-evals/references/core-concepts/index.mdx +0 -42
  119. package/skills/evolve-evals/references/core-concepts/jobs.mdx +0 -151
  120. package/skills/evolve-evals/references/core-concepts/models.mdx +0 -68
  121. package/skills/evolve-evals/references/core-concepts/sandboxes.mdx +0 -52
  122. package/skills/evolve-evals/references/core-concepts/secrets.mdx +0 -41
  123. package/skills/evolve-evals/references/core-concepts/sharing.mdx +0 -35
  124. package/skills/evolve-evals/references/core-concepts/skills.mdx +0 -42
  125. package/skills/evolve-evals/references/core-concepts/tasks.mdx +0 -216
  126. package/skills/evolve-evals/references/core-concepts/teams.mdx +0 -36
  127. package/skills/evolve-evals/references/core-concepts/trial-outputs.mdx +0 -117
  128. package/skills/evolve-evals/references/core-concepts/trials.mdx +0 -103
  129. package/skills/evolve-evals/references/core-concepts/upload.mdx +0 -93
  130. package/skills/evolve-evals/references/dashboard/checks.mdx +0 -21
  131. package/skills/evolve-evals/references/dashboard/datasets.mdx +0 -24
  132. package/skills/evolve-evals/references/dashboard/jobs.mdx +0 -65
  133. package/skills/evolve-evals/references/dashboard/traces.mdx +0 -17
  134. package/skills/evolve-evals/references/dashboard/trial-viewer.mdx +0 -30
  135. package/skills/evolve-evals/references/getting-started/installation.mdx +0 -56
  136. package/skills/evolve-evals/references/getting-started/quick-start.mdx +0 -94
  137. package/skills/evolve-evals/references/index.mdx +0 -48
  138. package/skills/evolve-evals/references/sdk/python.mdx +0 -161
  139. package/skills/evolve-evals/references/sdk/typescript.mdx +0 -163
  140. package/skills/evolve-evals/references/sdk-reference/agents.mdx +0 -108
  141. package/skills/evolve-evals/references/sdk-reference/analyses.mdx +0 -222
  142. package/skills/evolve-evals/references/sdk-reference/auth.mdx +0 -116
  143. package/skills/evolve-evals/references/sdk-reference/checks.mdx +0 -300
  144. package/skills/evolve-evals/references/sdk-reference/datasets.mdx +0 -346
  145. package/skills/evolve-evals/references/sdk-reference/errors.mdx +0 -210
  146. package/skills/evolve-evals/references/sdk-reference/index.mdx +0 -60
  147. package/skills/evolve-evals/references/sdk-reference/jobs.mdx +0 -822
  148. package/skills/evolve-evals/references/sdk-reference/meta.mdx +0 -52
  149. package/skills/evolve-evals/references/sdk-reference/skills.mdx +0 -69
  150. package/skills/evolve-evals/references/sdk-reference/trials.mdx +0 -302
  151. package/skills/evolve-evals/references/sdk-reference/types.mdx +0 -274
  152. package/skills/evolve-evals/references/snippets/global-options.mdx +0 -13
  153. package/skills/evolve-evals/references/snippets/list-flags.mdx +0 -23
package/dist/index.d.ts CHANGED
@@ -4,8 +4,8 @@ import { ZodType, z } from 'zod';
4
4
  export { E2BConfig, E2BProvider, createE2BProvider } from '@evolvingmachines/e2b';
5
5
  export { DaytonaConfig, DaytonaProvider, createDaytonaProvider } from '@evolvingmachines/daytona';
6
6
  export { ModalConfig, ModalProvider, createModalProvider } from '@evolvingmachines/modal';
7
- import { U as UsageReading, i as JobVisibility, G as GatewayUsageEvent, j as JobShareRequest, k as JobShares, T as Trial, l as Job, h as TraceEvent, a as TrialAnalysis, C as Check, m as TaskCheck, H as HostedErrorCode, n as DatasetsClient, o as AgentsClient, p as JobsClient, S as SkillsClient, q as TrialsClient, r as AnalysesClient, s as ChecksClient, O as OrgsClient, t as CapabilityDocument, u as HostedClientConfig, v as AuthClient } from './types-B_qmzlMk.js';
8
- export { w as AGENT_EFFORT_SUPPORT_VALUES, x as ANALYSIS_ARTIFACT_STREAMS, y as ANALYSIS_STATUSES, z as ActiveDataset, B as AgentArm, E as AgentArmInput, F as AgentCapability, I as AgentDatasetStats, K as AgentEffortSupport, L as AgentInfo, A as AgentInput, M as AgentList, N as AgentModelOption, Q as AgentPage, V as AgentResult, W as AgentSource, X as AgentSourceInput, Y as AgentUpsertInput, Z as AnalysisArtifactStream, _ as AnalysisCheck, $ as AnalysisEvidence, a0 as AnalysisFailure, a1 as AnalysisLabel, a2 as AnalysisList, a3 as AnalysisPage, a4 as AnalysisStatus, a5 as AnalysisTranscript, a6 as AnalysisTranscriptOptions, a7 as AnalyzeConfig, a8 as AnalyzeConfigInput, a9 as AnalyzeDefaults, aa as ApiKey, ab as AttemptPhase, ac as AuthStatus, ad as Awaitable, ae as CHECK_STATUSES, af as CheckDefaults, ag as CheckLabel, ah as CheckRow, ai as CheckRowList, aj as CheckRowPage, ak as CheckTaskTally, al as CompareCell, am as CompareCoverage, an as CompareJobAggregate, ao as CompareResponse, ap as CompareTaskRow, aq as Dataset, b as DatasetImport, ar as DatasetImportFailure, as as DatasetImportList, at as DatasetImportPage, e as DatasetImportProgress, au as DatasetImportStatus, av as DatasetList, aw as DatasetPage, ax as DatasetPatch, ay as DatasetPreflight, az as DatasetRef, aA as DatasetSelector, aB as DatasetSource, D as DatasetVersion, aC as DatasetVersionArchiveSource, aD as DatasetVersionArchiveUrlSource, aE as DatasetVersionGitSource, aF as DatasetVersionHubSource, aG as DatasetVersionSource, aH as DatasetVersionState, aI as DownloadDatasetOptions, aJ as DownloadJobOptions, aK as EVAL_SANDBOX_PROVIDERS, aL as EvalAgent, aM as EvalModelInfo, aN as EvalSandboxProvider, aO as EvalStepResult, aP as ExceptionInfo, aQ as FilesystemArchiveOptions, aR as FilesystemBox, aS as FilesystemCapture, aT as FilesystemChange, aU as FilesystemChanges, aV as FilesystemChangesOptions, aW as FilesystemEntry, aX as FilesystemListOptions, aY as FilesystemListing, aZ as FilesystemReadOptions, a_ as FilesystemSearchHit, a$ as FilesystemSearchOptions, b0 as FilesystemSearchResult, b1 as FilesystemSource, b2 as FilesystemState, b3 as FilesystemStatus, b4 as FilesystemStreamEvent, b5 as FilesystemStreamOptions, b6 as FilesystemWatchResult, b7 as GATEWAY_TRACE_SEQ_BASE, b8 as GatewayUsage, b9 as GetDatasetOptions, ba as GrepJobOptions, bb as HOSTED_ERROR_CODES, bc as ImportPhase, bd as ImportPhaseProgress, be as ImportWarning, bf as InfraFailureSignature, bg as JOB_LIST_KINDS, bh as JOB_LIST_SCOPES, bi as JobAnalysisStats, J as JobCreate, bj as JobDeleteResult, c as JobEvent, bk as JobFailure, bl as JobGrepGroup, bm as JobGrepPage, bn as JobImport, bo as JobImportFailure, bp as JobImportList, bq as JobImportPage, br as JobImportPhaseName, bs as JobImportPhaseProgress, bt as JobImportProgress, bu as JobImportSkippedTrial, bv as JobImportSource, bw as JobList, bx as JobListItem, by as JobListItemList, bz as JobListItemPage, bA as JobListKind, bB as JobListScope, bC as JobPage, bD as JobShareEmail, bE as JobShareLink, bF as JobStats, bG as JobStatus, bH as JobTaskLink, bI as JobTaskRollup, bJ as JobTaskRollupList, bK as JobTaskRollupPage, bL as JobViewer, bM as JobWatch, bN as JudgeResult, bO as ListAgentsOptions, bP as ListAnalysesOptions, bQ as ListDatasetsOptions, bR as ListImportsOptions, bS as ListJobImportsOptions, bT as ListJobTasksOptions, bU as ListJobsOptions, bV as ListSkillsOptions, bW as ListTrialFilesOptions, bX as ListTrialsOptions, bY as ManagedProviderCapability, bZ as OrgInvite, b_ as OrgInviteCreated, b$ as OrgJoined, c0 as OrgMember, c1 as OrgQuota, c2 as OrgRole, c3 as OrgUsage, c4 as Organization, c5 as OrganizationDetail, c6 as Page, c7 as PageOptions, c8 as PassAtKGroup, c9 as PassAtKPoint, ca as PreflightDatasetInput, cb as PreflightDeferredCheck, cc as PreflightManifestVerdict, cd as PreflightTaskVerdict, ce as ProviderCapability, P as PublishDatasetInput, cf as PublishDatasetOptions, cg as RegradeRequest, ch as ResumeRequest, ci as RetryRequest, R as Rubric, cj as RubricCriterion, ck as RunFilesystem, cl as SANDBOX_LOG_STREAMS, cm as SandboxLogEvent, cn as SandboxLogLine, co as SandboxLogLines, cp as SandboxLogOptions, cq as SandboxLogStream, cr as SandboxProcs, cs as SkillLock, ct as SkillUpload, cu as SkillUploadList, cv as SkillUploadPage, cw as SourceJob, cx as SpendSource, cy as StartJobOptions, cz as StatusVocabulary, cA as StopResponse, cB as TASK_LINKED_BY, cC as TASK_LINK_REASONS, cD as TRIAL_ARTIFACT_STREAMS, cE as TRIAL_STATUSES, cF as Task, cG as TaskCheckTranscript, cH as TaskLinkReason, cI as TaskLinkedBy, cJ as TaskNote, cK as TaskPackageFiles, cL as TaskPackageFilesystemStatus, cM as TaskProviderVerdict, cN as TimingInfo, cO as TraceEventPage, cP as TraceOptions, cQ as TrialArtifactStream, cR as TrialCounts, cS as TrialFile, cT as TrialFilePage, cU as TrialFileRange, cV as TrialGpuCost, cW as TrialList, cX as TrialPage, cY as TrialStatus, cZ as TrialStatusTally, c_ as TrialTaskLink, c$ as TrialUploadProvenance, d0 as UploadJobOptions, d1 as UploadProvenance, d2 as UploadSkillOptions, d3 as UpstreamStatus, d4 as VerifierEnvironmentMode, d5 as VerifierResult, d6 as WatchAnalysisOptions, d7 as WatchImportOptions, d8 as WatchJobImportOptions, d9 as WatchJobOptions, da as gatewayUsageOf, db as isHostedErrorCode, dc as passAtK } from './types-B_qmzlMk.js';
7
+ import { U as UsageReading, i as JobVisibility, G as GatewayUsageEvent, j as JobShareRequest, k as JobShares, T as Trial, l as Job, h as TraceEvent, a as TrialAnalysis, C as Check, m as TaskCheck, H as HostedErrorCode, n as DatasetsClient, o as AgentsClient, p as JobsClient, S as SkillsClient, q as TrialsClient, r as AnalysesClient, s as ChecksClient, O as OrgsClient, t as CapabilityDocument, u as HostedClientConfig, v as AuthClient } from './types-Dj0iMl5V.js';
8
+ export { w as AGENT_EFFORT_SUPPORT_VALUES, x as ANALYSIS_ARTIFACT_STREAMS, y as ANALYSIS_STATUSES, z as ActiveDataset, B as AgentArm, E as AgentArmInput, F as AgentCapability, I as AgentDatasetStats, K as AgentEffortSupport, L as AgentInfo, A as AgentInput, M as AgentList, N as AgentModelOption, Q as AgentPage, V as AgentResult, W as AgentSource, X as AgentSourceInput, Y as AgentUpsertInput, Z as AnalysisArtifactStream, _ as AnalysisCheck, $ as AnalysisEvidence, a0 as AnalysisFailure, a1 as AnalysisList, a2 as AnalysisPage, a3 as AnalysisStatus, a4 as AnalysisTranscript, a5 as AnalysisTranscriptOptions, a6 as AnalyzeConfig, a7 as AnalyzeConfigInput, a8 as AnalyzeDefaults, a9 as ApiKey, aa as AttemptPhase, ab as AuthStatus, ac as Awaitable, ad as CHECK_STATUSES, ae as CheckDefaults, af as CheckRow, ag as CheckRowList, ah as CheckRowPage, ai as CheckTaskTally, aj as CompareCell, ak as CompareCoverage, al as CompareJobAggregate, am as CompareResponse, an as CompareTaskRow, ao as Dataset, b as DatasetImport, ap as DatasetImportFailure, aq as DatasetImportList, ar as DatasetImportPage, e as DatasetImportProgress, as as DatasetImportStatus, at as DatasetList, au as DatasetPage, av as DatasetPatch, aw as DatasetPreflight, ax as DatasetRef, ay as DatasetSelector, az as DatasetSource, D as DatasetVersion, aA as DatasetVersionArchiveSource, aB as DatasetVersionArchiveUrlSource, aC as DatasetVersionGitSource, aD as DatasetVersionHubSource, aE as DatasetVersionSource, aF as DatasetVersionState, aG as DownloadDatasetOptions, aH as DownloadJobOptions, aI as EVAL_SANDBOX_PROVIDERS, aJ as EvalAgent, aK as EvalModelInfo, aL as EvalSandboxProvider, aM as EvalStepResult, aN as ExceptionInfo, aO as FilesystemArchiveOptions, aP as FilesystemBox, aQ as FilesystemCapture, aR as FilesystemChange, aS as FilesystemChanges, aT as FilesystemChangesOptions, aU as FilesystemEntry, aV as FilesystemListOptions, aW as FilesystemListing, aX as FilesystemReadOptions, aY as FilesystemSearchHit, aZ as FilesystemSearchOptions, a_ as FilesystemSearchResult, a$ as FilesystemSource, b0 as FilesystemState, b1 as FilesystemStatus, b2 as FilesystemStreamEvent, b3 as FilesystemStreamOptions, b4 as FilesystemWatchResult, b5 as GATEWAY_TRACE_SEQ_BASE, b6 as GatewayUsage, b7 as GetDatasetOptions, b8 as GrepJobOptions, b9 as HOSTED_ERROR_CODES, ba as ImportPhase, bb as ImportPhaseProgress, bc as ImportWarning, bd as InfraFailureSignature, be as JOB_LIST_KINDS, bf as JOB_LIST_SCOPES, bg as JobAnalysisStats, J as JobCreate, bh as JobDeleteResult, c as JobEvent, bi as JobFailure, bj as JobGrepGroup, bk as JobGrepPage, bl as JobImport, bm as JobImportFailure, bn as JobImportList, bo as JobImportPage, bp as JobImportPhaseName, bq as JobImportPhaseProgress, br as JobImportProgress, bs as JobImportSkippedTrial, bt as JobImportSource, bu as JobList, bv as JobListItem, bw as JobListItemList, bx as JobListItemPage, by as JobListKind, bz as JobListScope, bA as JobPage, bB as JobShareEmail, bC as JobShareLink, bD as JobStats, bE as JobStatus, bF as JobTaskLink, bG as JobTaskRollup, bH as JobTaskRollupList, bI as JobTaskRollupPage, bJ as JobViewer, bK as JobWatch, bL as JudgeResult, bM as ListAgentsOptions, bN as ListAnalysesOptions, bO as ListDatasetsOptions, bP as ListImportsOptions, bQ as ListJobImportsOptions, bR as ListJobTasksOptions, bS as ListJobsOptions, bT as ListSkillsOptions, bU as ListTrialFilesOptions, bV as ListTrialsOptions, bW as ManagedProviderCapability, bX as OrgInvite, bY as OrgInviteCreated, bZ as OrgJoined, b_ as OrgMember, b$ as OrgQuota, c0 as OrgRole, c1 as OrgUsage, c2 as Organization, c3 as OrganizationDetail, c4 as Page, c5 as PageOptions, c6 as PassAtKGroup, c7 as PassAtKPoint, c8 as PreflightDatasetInput, c9 as PreflightDeferredCheck, ca as PreflightManifestVerdict, cb as PreflightTaskVerdict, cc as ProviderCapability, P as PublishDatasetInput, cd as PublishDatasetOptions, ce as RegradeRequest, cf as ResumeRequest, cg as RetryRequest, R as Rubric, ch as RubricCriterion, ci as RunFilesystem, cj as SANDBOX_LOG_STREAMS, ck as SandboxLogEvent, cl as SandboxLogLine, cm as SandboxLogLines, cn as SandboxLogOptions, co as SandboxLogStream, cp as SandboxProcs, cq as SkillLock, cr as SkillUpload, cs as SkillUploadList, ct as SkillUploadPage, cu as SourceJob, cv as SpendSource, cw as StartJobOptions, cx as StatusVocabulary, cy as StopResponse, cz as TASK_LINKED_BY, cA as TASK_LINK_REASONS, cB as TRIAL_ARTIFACT_STREAMS, cC as TRIAL_STATUSES, cD as Task, cE as TaskCheckTranscript, cF as TaskLinkReason, cG as TaskLinkedBy, cH as TaskNote, cI as TaskPackageFiles, cJ as TaskPackageFilesystemStatus, cK as TaskProviderVerdict, cL as TimingInfo, cM as TraceEventPage, cN as TraceOptions, cO as TrialArtifactStream, cP as TrialCounts, cQ as TrialFile, cR as TrialFilePage, cS as TrialFileRange, cT as TrialGpuCost, cU as TrialList, cV as TrialPage, cW as TrialStatus, cX as TrialStatusTally, cY as TrialTaskLink, cZ as TrialUploadProvenance, c_ as UploadJobOptions, c$ as UploadProvenance, d0 as UploadSkillOptions, d1 as UpstreamStatus, d2 as VerifierEnvironmentMode, d3 as VerifierResult, d4 as WatchAnalysisOptions, d5 as WatchImportOptions, d6 as WatchJobImportOptions, d7 as WatchJobOptions, d8 as gatewayUsageOf, d9 as isHostedErrorCode, da as passAtK } from './types-Dj0iMl5V.js';
9
9
 
10
10
  /**
11
11
  * The one skills resolver.
@@ -339,7 +339,28 @@ interface PlanEntry {
339
339
  * All possible session update types.
340
340
  * Discriminated union on `sessionUpdate` field.
341
341
  */
342
- type SessionUpdate = AgentMessageChunk | AgentThoughtChunk | UserMessageChunk | ToolCall | ToolCallUpdate | Plan | AgentError | AgentUsage;
342
+ type SessionUpdate = AgentMessageChunk | AgentThoughtChunk | UserMessageChunk | ToolCall | ToolCallUpdate | Plan | AgentError | AgentUsage | HarnessEvent;
343
+ /**
344
+ * A wire line the harness printed that has no ACP slot — kept, never dropped.
345
+ *
346
+ * DELIBERATE EXTENSION BEYOND ACP, like AgentError and AgentUsage. Harness
347
+ * streams are unversioned and grow between releases (the owner's ruling
348
+ * 2026-09-25: any event type outside the captured vocabulary is passed
349
+ * through as a generic event and logged, never a failure), and some captured
350
+ * types are real facts of the run with no ACP shape — a sub-agent's
351
+ * progress, an automatic retry, a compaction. `type` is the harness's own
352
+ * type word and `payload` the line's other fields, verbatim.
353
+ *
354
+ * Excluded from isAgentWorkUpdate: an unknown line is evidence the harness
355
+ * PRINTED something, never that the agent did work.
356
+ */
357
+ interface HarnessEvent {
358
+ sessionUpdate: "harness_event";
359
+ /** The harness's own type word for the line, verbatim. */
360
+ type: string;
361
+ /** Every other field of the line, verbatim. */
362
+ payload: Record<string, unknown>;
363
+ }
343
364
  /**
344
365
  * Token accounting as the harness reported it on one wire line.
345
366
  *
@@ -664,7 +685,7 @@ declare const DEFAULT_REASONING_EFFORT: ReasoningEffort;
664
685
  * takes there — pure data derivation from its effortSupport. The
665
686
  * harness-capabilities artifact generator and picker UIs share this.
666
687
  */
667
- declare function harnessEffortVocabulary(support: EffortSupport): {
688
+ declare function harnessEffortVocabulary(support: EffortSupport, efforts?: readonly ReasoningEffort[]): {
668
689
  efforts: readonly ReasoningEffort[];
669
690
  defaultEffort: ReasoningEffort | null;
670
691
  };
@@ -688,7 +709,7 @@ interface McpConfigInfo {
688
709
  /** Config filename (e.g., "settings.json" or "config.toml") */
689
710
  filename: string;
690
711
  /** Config format */
691
- format: "json" | "toml";
712
+ format: "json" | "toml" | "yaml";
692
713
  /** Whether to use workingDir for project-level config (Claude only) */
693
714
  projectConfig?: boolean;
694
715
  }
@@ -722,6 +743,8 @@ interface BuildCommandOptions {
722
743
  * names belong to the caller's gateway, never to Evolve's alias maps).
723
744
  */
724
745
  isExternalGateway?: boolean;
746
+ /** Whether the run configured MCP servers (pi loads its adapter extension only then). */
747
+ mcpConfigured?: boolean;
725
748
  /** Skills enabled for this run */
726
749
  skills?: string[];
727
750
  /** Sandbox home directory (default: "/home/user") */
@@ -737,6 +760,8 @@ interface AgentRegistryEntry {
737
760
  * advertises exactly the vocabulary the local SDK drives.
738
761
  */
739
762
  effortSupport: EffortSupport;
763
+ /** The subset of the level vocabulary this CLI can honor; absent = the whole vocabulary for its effortSupport. */
764
+ efforts?: readonly ReasoningEffort[];
740
765
  /** Environment variable name for API key */
741
766
  apiKeyEnv: string;
742
767
  /** Environment variable name for OAuth (file path or token depending on agent) */
@@ -856,6 +881,40 @@ interface AgentRegistryEntry {
856
881
  provider: "generic-chat-completion-api" | "openai" | "anthropic";
857
882
  maxOutputTokens?: number;
858
883
  };
884
+ /**
885
+ * The pi family's only route: a models.json provider entry written per run
886
+ * with the LITERAL base URL (pi never expands $VAR there), selected by --provider.
887
+ */
888
+ modelsJsonRoute?: {
889
+ /** The agent dir (~ expanded) that holds models.json. */
890
+ agentDir: string;
891
+ /** The provider name the command's --provider selects. */
892
+ providerName: string;
893
+ /** How the file names the key env: pi wants "$VAR" (docs/models.md), Prime Agent the bare "VAR". */
894
+ apiKeyRef: "dollar" | "bare";
895
+ };
896
+ /** Vendor defaults wrong for a metered run, deep-merged into the settings file at setup. */
897
+ settingsStamp?: {
898
+ path: string;
899
+ document: Record<string, unknown>;
900
+ };
901
+ /**
902
+ * Z Code's per-run provider file: the CLI's only source of model, level, base
903
+ * URL and key (no `--model` flag, no credential env; ZCode cli/src at v3.14.3).
904
+ */
905
+ zcodeProviderConfig?: {
906
+ path: string;
907
+ providerId: string;
908
+ providerName: string;
909
+ /** The window Z Code compacts at — the platform's one number, below every roster route's real ceiling. */
910
+ contextWindow: number;
911
+ /** Cap on `max_tokens` per request. */
912
+ maxOutputTokens: number;
913
+ };
914
+ /** Antigravity-only: the CLI's settings file, rewritten before every run (mcp/json.ts writeAntigravitySettings). */
915
+ antigravitySettings?: {
916
+ settingsPath: string;
917
+ };
859
918
  /** Environment variable that CLI reads for custom outbound HTTP headers */
860
919
  customHeadersEnv?: string;
861
920
  /** Format for custom headers env var: "newline" (Claude) or "comma" (Gemini). Default: "newline" */
@@ -904,6 +963,19 @@ interface AgentRegistryEntry {
904
963
  checkpointDirs?: string[];
905
964
  /** Additional relative paths to exclude from checkpoint tar. */
906
965
  checkpointExcludes?: string[];
966
+ /** The CLI exits 0 whatever happened: at exit 0 the last assistant message_end is the verdict (agent.ts). */
967
+ verdictFromStream?: true;
968
+ /** Where the SDK keeps the stream-captured session id for CLIs that resume by `--session-id` (droid, dsh). */
969
+ sessionIdStateFile?: string;
970
+ /** dsh only: the Evolve-owned `--patch` that routes the CLI (mcp/yaml.ts); it names env variables, never values. */
971
+ dshRoutePatch?: {
972
+ path: string;
973
+ /** The pi-ai provider route name the patch declares and the default model selects. */
974
+ providerName: string;
975
+ contextWindow: number;
976
+ /** The request's `max_tokens`. */
977
+ maxTokens: number;
978
+ };
907
979
  }
908
980
  /**
909
981
  * Registry of all supported agents.
@@ -1491,7 +1563,7 @@ interface SandboxProvider {
1491
1563
  prepareImage?(image?: string): Promise<void>;
1492
1564
  }
1493
1565
  /** Supported agent types (headless CLI agents only, no ACP) */
1494
- type AgentType = "claude" | "codex" | "gemini" | "qwen" | "kimi" | "opencode" | "droid";
1566
+ type AgentType = "claude" | "codex" | "gemini" | "qwen" | "kimi" | "opencode" | "droid" | "pi" | "prime-agent" | "dsh" | "zcode" | "antigravity";
1495
1567
  /** Agent type constants for use in code */
1496
1568
  declare const AGENT_TYPES: {
1497
1569
  readonly CLAUDE: "claude";
@@ -1501,6 +1573,11 @@ declare const AGENT_TYPES: {
1501
1573
  readonly KIMI: "kimi";
1502
1574
  readonly OPENCODE: "opencode";
1503
1575
  readonly DROID: "droid";
1576
+ readonly PI: "pi";
1577
+ readonly PRIME_AGENT: "prime-agent";
1578
+ readonly DSH: "dsh";
1579
+ readonly ZCODE: "zcode";
1580
+ readonly ANTIGRAVITY: "antigravity";
1504
1581
  };
1505
1582
  /** Workspace mode determines folder structure and system prompt */
1506
1583
  type WorkspaceMode = "knowledge" | "swe" | "task";
@@ -2289,6 +2366,7 @@ declare class Agent {
2289
2366
  private options;
2290
2367
  private sandbox?;
2291
2368
  private hasRun;
2369
+ private mcpConfigured;
2292
2370
  private readonly workingDir;
2293
2371
  private readonly homeDir;
2294
2372
  private lastRunTimestamp?;
@@ -2307,7 +2385,7 @@ declare class Agent {
2307
2385
  private interruptedOperations;
2308
2386
  private sandboxState;
2309
2387
  private agentState;
2310
- private droidSessionId?;
2388
+ private capturedSessionId?;
2311
2389
  private managedBrowserSession?;
2312
2390
  private providerRuntimeToken?;
2313
2391
  private managedSecretRuntimeToken?;
@@ -2416,6 +2494,8 @@ declare class Agent {
2416
2494
  private requiresPreRunDashboardIngest;
2417
2495
  private providerRuntimeHeaderUpdates;
2418
2496
  private shouldExposeProviderRuntimeTokenEnv;
2497
+ /** Z Code: the literal key lives in the per-run provider file, so no mode puts a key env in the process. */
2498
+ private credentialRidesProviderFile;
2419
2499
  private buildProviderRuntimeProcessEnvs;
2420
2500
  /**
2421
2501
  * Build per-run env overrides for spend tracking.
@@ -2425,13 +2505,42 @@ declare class Agent {
2425
2505
  */
2426
2506
  private buildRunEnvs;
2427
2507
  private writeCodexGatewayProviderConfig;
2428
- private captureDroidSession;
2429
- private extractDroidSessionId;
2430
- private findDroidSessionId;
2431
- private droidSessionStatePath;
2432
- private loadDroidSessionState;
2433
- private writeDroidSessionState;
2508
+ /**
2509
+ * The session id the CLI's own stream announced, for harnesses that resume
2510
+ * by id (registry sessionIdStateFile: droid `--session-id`, dsh
2511
+ * `--session-id`). Read off the parsed events first (every parser stamps
2512
+ * `sessionId`), else off the raw line for lines the parser drops (dsh's
2513
+ * opening `session` line, droid's JSON-RPC envelopes).
2514
+ */
2515
+ private captureHarnessSessionId;
2516
+ private extractSessionIdFromLine;
2517
+ private findSessionIdInLine;
2518
+ private sessionIdStatePath;
2519
+ private loadCapturedSessionId;
2520
+ private writeCapturedSessionId;
2521
+ /**
2522
+ * The model dsh's route patch names — the one string the provider receives
2523
+ * (dsh resolves nothing itself on the pi-ai route), per route, the droid
2524
+ * settings-file rule verbatim (droidSettingsModel): external gateway sends
2525
+ * the roster wire id (alias == id on this roster, so verbatim); gateway
2526
+ * mode the gatewayModelAliases spelling (none: verbatim); plain direct mode
2527
+ * the directModelAliases spelling — OpenRouter's own id without the
2528
+ * `openrouter/` route prefix.
2529
+ */
2530
+ private dshPatchModel;
2531
+ /**
2532
+ * dsh's route patch, rewritten before every run (registry dshRoutePatch;
2533
+ * mcp/yaml.ts). The key and the base URL are named as the env vars every
2534
+ * mode already sets (apiKeyEnv, baseUrlEnv), so the same file serves the
2535
+ * managed gateway, an external gateway and direct mode. Spend headers ride
2536
+ * env reads too, and only where the SDK sets those envs — the managed
2537
+ * gateway (spendTrackingEnvs + the runtime-binding env); an unset env would
2538
+ * turn into an undefined header value.
2539
+ */
2540
+ private writeDshPerRunPatch;
2434
2541
  private resolveCommandModel;
2542
+ /** The pi family's per-run models.json: the mode's LITERAL base URL (pi never expands $VAR there), the key by env name. */
2543
+ private writePiFamilyModelsJson;
2435
2544
  /**
2436
2545
  * The request model the Evolve-owned Droid settings file names. Droid
2437
2546
  * resolves nothing on this route — the custom model's `model` field is the
@@ -2466,6 +2575,11 @@ declare class Agent {
2466
2575
  */
2467
2576
  private setupAgentAuth;
2468
2577
  private writeGeminiGatewayAuthSettings;
2578
+ /**
2579
+ * Antigravity's settings for this run: API-key auth, telemetry off, and the slug buildCommand will name registered
2580
+ * under customModelsConfig (else `--model` refuses it); rewritten before each run, like droid's settings. No-op elsewhere.
2581
+ */
2582
+ private writeAntigravityRunSettings;
2469
2583
  private setupAgentPlugins;
2470
2584
  private assertProviderRuntimeDoesNotExposeGatewayKey;
2471
2585
  /**
@@ -2540,6 +2654,13 @@ declare class Agent {
2540
2654
  * custom_headers before spawning. CLI reads config at startup, so each run
2541
2655
  * gets a fresh write. Extracted so the effort pin on this path is testable. */
2542
2656
  private writeKimiPerRunConfig;
2657
+ /**
2658
+ * Z Code reads model, level, URL and key from its provider file only, so every
2659
+ * mode writes it before each spawn (gateway mode adds the spend headers).
2660
+ */
2661
+ private writeZcodePerRunConfig;
2662
+ /** The provider file holds the run's credential literally, so it lives only while the run does. */
2663
+ private removeZcodePerRunConfig;
2543
2664
  run(options: RunOptions, callbacks?: StreamCallbacks): Promise<AgentResponse>;
2544
2665
  /**
2545
2666
  * Execute arbitrary command in sandbox
@@ -4072,6 +4193,16 @@ declare function managedSandbox(provider?: ManagedSandboxProviderName, options?:
4072
4193
  */
4073
4194
  declare function resolveDefaultSandbox(): Promise<SandboxProvider>;
4074
4195
 
4196
+ /**
4197
+ * Antigravity CLI (`antigravity`, Google's `agy`) `--output-format stream-json` → ACP-style events. The CLI is closed
4198
+ * source: every shape here is from the live capture of agy 1.2.11 (the fixtures of tests/unit/antigravity-parser.test.ts),
4199
+ * then the vendor's headless docs page; the capture wins. Not on the wire: timestamps, text on user_input /
4200
+ * system_message / error_message steps, tool call ids (a step's updates pair by step_index), a failing command's exit
4201
+ * code. Usage rides per call on a DONE agent_response and whole on result; promptTokens = input + cache_read (Harbor).
4202
+ */
4203
+
4204
+ declare function createAntigravityParser(): (jsonLine: string) => OutputEvent[] | null;
4205
+
4075
4206
  /**
4076
4207
  * Claude JSONL → ACP-style events parser.
4077
4208
  *
@@ -4139,6 +4270,27 @@ declare function createCodexParser(): (jsonLine: string) => OutputEvent[] | null
4139
4270
 
4140
4271
  declare function createDroidParser(): (jsonLine: string) => OutputEvent[] | null;
4141
4272
 
4273
+ /**
4274
+ * DeepSeek Harness (dsh) `--profile headless --json` → ACP-style events.
4275
+ *
4276
+ * The vocabulary is the emitting source, not a published schema: deepseek-harness
4277
+ * packages/bundle/headless/src/json-stream.ts at dsh-v0.1.7-rc.2 (recon
4278
+ * 01-deepseek.md §G1; every shape observed live in 06-live-tests/dsh). Ten
4279
+ * line types: session, status (turn_start | step_start | step_end | turn_end),
4280
+ * thinking, text, tool_call, tool_result, final, error. No clock, model id,
4281
+ * message id or cost on any line; strings are capped at 8 KiB (`truncated`).
4282
+ *
4283
+ * Failure is `turn_end` with a `reason.kind` other than `completed` (round-2
4284
+ * E1: retries are invisible, then one turn_end, `final ""`, exit 1); a
4285
+ * `tool_result.status: "error"` is a tool failing, a shell's exit code is
4286
+ * not (E2/E2b); SIGINT ends the stream at `final` with no turn_end (E3).
4287
+ *
4288
+ * Any type or phase outside the vocabulary rides `harness_event` and is
4289
+ * logged once — the stream is unversioned and pre-stable by the vendor's word.
4290
+ */
4291
+
4292
+ declare function createDshParser(): (jsonLine: string) => OutputEvent[] | null;
4293
+
4142
4294
  /**
4143
4295
  * Gemini JSONL → ACP-style events parser.
4144
4296
  *
@@ -4161,6 +4313,22 @@ declare function createDroidParser(): (jsonLine: string) => OutputEvent[] | null
4161
4313
  */
4162
4314
  declare function createGeminiParser(): (jsonLine: string) => OutputEvent[] | null;
4163
4315
 
4316
+ /**
4317
+ * pi profile over the pi-family core: pi's built-in tools (read, bash, edit,
4318
+ * write; grep/find/ls when enabled) plus the pi-mcp-adapter's one proxy tool
4319
+ * `mcp`, whose own failures ride `details.error` with isError false.
4320
+ */
4321
+
4322
+ declare function createPiParser(): (jsonLine: string) => OutputEvent[] | null;
4323
+
4324
+ /**
4325
+ * Prime Agent profile over the pi-family core: one tool, `ipython {code}`
4326
+ * (shell, MCP, skills and sub-agents all run inside the cell), whose result
4327
+ * `details.status` — not the wire's isError — says whether the cell failed.
4328
+ */
4329
+
4330
+ declare function createPrimeAgentParser(): (jsonLine: string) => OutputEvent[] | null;
4331
+
4164
4332
  /**
4165
4333
  * Qwen NDJSON → ACP-style events parser.
4166
4334
  *
@@ -4193,6 +4361,23 @@ declare function createGeminiParser(): (jsonLine: string) => OutputEvent[] | nul
4193
4361
  */
4194
4362
  declare function parseQwenOutput(line: string): OutputEvent[] | null;
4195
4363
 
4364
+ /**
4365
+ * Z Code parser: `zcode -p … --output-format stream-json` (ZCode v3.14.3, CLI 0.16.9).
4366
+ *
4367
+ * Shapes come from the vendor's zod protocol (packages/shared/src/zcode-protocol)
4368
+ * and the CLI's NDJSON writer (headless-workflow.ts, prompt-command.ts); which
4369
+ * types appear and how the catch-all `session.updated` payloads are told apart
4370
+ * is live capture (2026-09-25). The stream is CLI-private and unversioned, so
4371
+ * an unknown type is logged once and passed through as `harness_event`.
4372
+ *
4373
+ * Sub-agents write into the same stream under their own sessionId; child lines
4374
+ * after the parent's announcement carry `parentToolCallId` and keep the run's
4375
+ * session id on the envelope. The exit code is no verdict: a cancelled or
4376
+ * failed turn is the `error` variant (fatal) with the turn's usage beside it.
4377
+ */
4378
+
4379
+ declare function createZcodeParser(): (jsonLine: string) => OutputEvent[] | null;
4380
+
4196
4381
  /**
4197
4382
  * Unified Parser Entry Point
4198
4383
  *
@@ -4230,14 +4415,16 @@ declare function parseNdjsonOutput(agentType: AgentType, output: string): Output
4230
4415
  /**
4231
4416
  * MCP JSON Configuration Writer
4232
4417
  *
4233
- * Handles MCP config for Claude, Gemini, Qwen, Kimi, Droid, and OpenCode agents.
4234
- * Uses registry for paths - no hardcoded values.
4418
+ * Handles MCP config for Claude, Gemini, Qwen, Kimi, Droid, OpenCode, and
4419
+ * Antigravity agents. Uses registry for paths - no hardcoded values.
4235
4420
  *
4236
4421
  * Transport formats by agent:
4237
4422
  * - Claude: { type: "http"|"sse"|"stdio", url: "..." }
4238
4423
  * - Gemini: { url: "...", type: "http"|"sse" } | { command: "..." }
4239
4424
  * - Qwen: { httpUrl: "..." } | { url: "..." } | { command: "..." }
4240
4425
  * - Kimi Code: { url: "...", transport?: "http"|"sse" } | { command: "...", transport: "stdio" }
4426
+ * - Z Code: { type: "stdio", command, args, env } | { type: "http"|"sse", url, headers } under `mcp.servers`
4427
+ * - Antigravity: { serverUrl: "...", headers? } | { command: "...", args?, env?, cwd? }
4241
4428
  */
4242
4429
 
4243
4430
  /**
@@ -4252,6 +4439,15 @@ declare function writeClaudeMcpConfig(sandbox: SandboxInstance, workingDir: stri
4252
4439
  declare function writeGeminiMcpConfig(sandbox: SandboxInstance, servers: Record<string, McpServerConfig>, homeDir?: string): Promise<void>;
4253
4440
  /** Write MCP config for Qwen agent */
4254
4441
  declare function writeQwenMcpConfig(sandbox: SandboxInstance, servers: Record<string, McpServerConfig>, homeDir?: string): Promise<void>;
4442
+ /** Write MCP config for Kimi agent (FastMCP-compatible transport field) */
4443
+ declare function writeKimiMcpConfig(sandbox: SandboxInstance, servers: Record<string, McpServerConfig>, homeDir?: string): Promise<void>;
4444
+ /** Write MCP config for the Antigravity agent (~/.gemini/config/mcp_config.json) */
4445
+ declare function writeAntigravityMcpConfig(sandbox: SandboxInstance, servers: Record<string, McpServerConfig>, homeDir?: string): Promise<void>;
4446
+ /**
4447
+ * The CLI's settings for a run, merged over the file it rewrites at every start: API-key auth (`modelProvider`), telemetry
4448
+ * off (`telemetryEnabled`, the key the binary keeps), non-workspace paths allowed, the run's slug registered (else `--model` exits 1).
4449
+ */
4450
+ declare function writeAntigravitySettings(sandbox: SandboxInstance, settingsPath: string, modelSlug: string, homeDir?: string): Promise<void>;
4255
4451
  /**
4256
4452
  * Write MCP config for Droid agent
4257
4453
  *
@@ -4275,6 +4471,52 @@ interface DroidGatewaySettingsConfig {
4275
4471
  * user's normal ~/.factory/settings.json inside the sandbox.
4276
4472
  */
4277
4473
  declare function writeDroidGatewaySettings(sandbox: SandboxInstance, config: DroidGatewaySettingsConfig, headers: Record<string, string>, homeDir?: string): Promise<void>;
4474
+ /**
4475
+ * Write MCP config for Z Code
4476
+ *
4477
+ * Z Code reads user-level servers from `mcp.servers` in ~/.zcode/cli/config.json
4478
+ * (the registry's mcpConfig); every other key of that file is preserved.
4479
+ */
4480
+ declare function writeZcodeMcpConfig(sandbox: SandboxInstance, servers: Record<string, McpServerConfig>, homeDir?: string): Promise<void>;
4481
+ interface ZcodeProviderConfigInput {
4482
+ /** The provider file, `~` = the sandbox home (registry zcodeProviderConfig.path). */
4483
+ path: string;
4484
+ providerId: string;
4485
+ providerName: string;
4486
+ /** OpenAI-compatible base URL INCLUDING `/v1` — Z Code appends `/chat/completions`. */
4487
+ baseUrl: string;
4488
+ /** The literal credential: Z Code expands no `${VAR}` in this file. */
4489
+ apiKey: string;
4490
+ /** The wire model id the request names. */
4491
+ model: string;
4492
+ /** One of ZCODE_REASONING_LEVELS. */
4493
+ reasoningLevel: string;
4494
+ contextWindow: number;
4495
+ maxOutputTokens: number;
4496
+ /** Extra request headers (the LiteLLM spend tags); {} when none. */
4497
+ headers: Record<string, string>;
4498
+ }
4499
+ /**
4500
+ * Z Code's provider file, built from scratch every run at mode 0600: the CLI reads its model,
4501
+ * level, base URL and key here and nowhere else (ZCode packages/provider/src/config at v3.14.3).
4502
+ */
4503
+ declare function writeZcodeProviderConfig(sandbox: SandboxInstance, config: ZcodeProviderConfigInput, homeDir?: string): Promise<void>;
4504
+ /**
4505
+ * Write MCP config for OpenCode agent
4506
+ *
4507
+ * OpenCode uses opencode.json in the working directory with an `mcp` key.
4508
+ * Format: { "mcp": { "name": { "type": "local"|"remote", "command": [...], "url": "..." } } }
4509
+ *
4510
+ * Key differences from other agents:
4511
+ * - Uses `mcp` key (not `mcpServers`)
4512
+ * - Local servers use `command` as array (not string)
4513
+ * - Remote servers use `type: "remote"` with `url`
4514
+ */
4515
+ declare function writeOpenCodeMcpConfig(sandbox: SandboxInstance, workingDir: string, servers: Record<string, McpServerConfig>): Promise<void>;
4516
+ /** pi's MCP config: `<agent-dir>/mcp.json`, read by the adapter extension the command loads. */
4517
+ declare function writePiMcpConfig(sandbox: SandboxInstance, servers: Record<string, McpServerConfig>, homeDir?: string): Promise<void>;
4518
+ /** Prime's MCP servers: the `mcpServers` map of the GLOBAL settings.json, the rest of the file kept. */
4519
+ declare function writePrimeAgentMcpConfig(sandbox: SandboxInstance, servers: Record<string, McpServerConfig>, homeDir?: string): Promise<void>;
4278
4520
 
4279
4521
  /**
4280
4522
  * MCP TOML Configuration Writer
@@ -4296,6 +4538,38 @@ declare function writeDroidGatewaySettings(sandbox: SandboxInstance, config: Dro
4296
4538
  */
4297
4539
  declare function writeCodexMcpConfig(sandbox: SandboxInstance, servers: Record<string, McpServerConfig>, homeDir?: string): Promise<void>;
4298
4540
 
4541
+ /**
4542
+ * dsh Configuration Writers (YAML patches)
4543
+ *
4544
+ * dsh has no model, effort or base-URL flag and reads no .mcp.json: Cordis
4545
+ * patch files (a YAML list of rows; `- insert:` adds rows) are its one config
4546
+ * surface, so two Evolve-owned files under ~/.dsh route it — the pi-ai route
4547
+ * (model, effort, spend headers) and one dsh-mcp-client row per MCP server.
4548
+ * The route patch names the key and URL as env variables (`!!js process.env.X`,
4549
+ * read by dsh at boot), so no secret lands on disk and one shape serves every mode.
4550
+ */
4551
+
4552
+ /** Rows must sit under `- insert:`; a bare row in a --patch file is "entry not found" (live M1). */
4553
+ declare function writeDshMcpConfig(sandbox: SandboxInstance, servers: Record<string, McpServerConfig>, homeDir?: string): Promise<void>;
4554
+ interface DshRoutePatchConfig {
4555
+ path: string;
4556
+ providerName: string;
4557
+ /** Env var NAME (pi-ai `apiKeyEnv` reads it per request). */
4558
+ apiKeyEnv: string;
4559
+ /** Env var NAME, read at boot via `!!js process.env`. */
4560
+ baseUrlEnv: string;
4561
+ /** The wire model id, sent verbatim. */
4562
+ model: string;
4563
+ /** pi-ai thinking level; absent = the route's default. */
4564
+ reasoningEffort?: string;
4565
+ /** Header name → env var NAME; only where the SDK sets every named env (an unset env would be an undefined header). */
4566
+ headerEnvs?: Record<string, string>;
4567
+ contextWindow: number;
4568
+ maxTokens: number;
4569
+ }
4570
+ /** Rewritten before every run; the run tag rides an env var, so the text is static per session. */
4571
+ declare function writeDshRoutePatch(sandbox: SandboxInstance, config: DshRoutePatchConfig, homeDir?: string): Promise<void>;
4572
+
4299
4573
  /**
4300
4574
  * MCP Configuration Module
4301
4575
  *
@@ -4313,6 +4587,11 @@ declare function writeCodexMcpConfig(sandbox: SandboxInstance, servers: Record<s
4313
4587
  * - Qwen: JSON to ~/.qwen/settings.json
4314
4588
  * - Droid: JSON to ${workingDir}/.factory/mcp.json
4315
4589
  * - OpenCode: JSON to ${workingDir}/opencode.json (mcp key)
4590
+ * - pi: JSON to ~/.pi/agent/mcp.json (the pi-mcp-adapter extension's file)
4591
+ * - Prime Agent: JSON to ~/.prime/agent/settings.json (mcpServers key)
4592
+ * - dsh: YAML patch rows to ~/.dsh/evolve-mcp.patch.yml
4593
+ * - Z Code: JSON to ~/.zcode/cli/config.json (mcp.servers key)
4594
+ * - Antigravity: JSON to ~/.gemini/config/mcp_config.json
4316
4595
  */
4317
4596
  declare function writeMcpConfig(agentType: AgentType, sandbox: SandboxInstance, workingDir: string, servers: Record<string, McpServerConfig>, homeDir?: string): Promise<void>;
4318
4597
 
@@ -4804,7 +5083,7 @@ declare function assembleTrialTree(parts: TrialTreeParts): Record<string, string
4804
5083
  /**
4805
5084
  * The analysis run's evolve.json: which run this analysis read (the
4806
5085
  * analyzed trial, its job, its task), the analyzer's model and status,
4807
- * which user downloaded it, and the analyzer's own meter — the verdict's
5086
+ * which user downloaded it, and the analyzer's own meter — the result's
4808
5087
  * one-home `usage` reading restated (never a second meter).
4809
5088
  */
4810
5089
  declare function analysisEvolveRecord(analysis: TrialAnalysis, userId: string | null): Record<string, unknown>;
@@ -5129,7 +5408,7 @@ interface HostedEvolve {
5129
5408
  readonly skills: SkillsClient;
5130
5409
  /** Globally addressable trials: get, trace, artifact, regrade, stop. */
5131
5410
  readonly trials: TrialsClient;
5132
- /** Analysis runs: verdict, the analyzer's own transcript, stored artifacts. */
5411
+ /** Analysis runs: the result, the analyzer's own transcript, stored artifacts. */
5133
5412
  readonly analyses: AnalysesClient;
5134
5413
  /** Task quality checks (Harbor's `harbor check`): create, get, list, watch. */
5135
5414
  readonly checks: ChecksClient;
@@ -5174,4 +5453,4 @@ declare function hosted(config?: HostedClientConfig): HostedEvolve;
5174
5453
  */
5175
5454
  declare function meta(config?: HostedClientConfig): Promise<CapabilityDocument>;
5176
5455
 
5177
- export { AGENT_REGISTRY, AGENT_TYPES, type ActionbookBrowserConfig, Agent, type AgentBrowserConfig, type AgentConfig, type AgentError, type AgentOptions, type AgentOverride, type AgentParser, type AgentPluginConfig, type AgentRegistryEntry, type AgentResponse, type AgentRuntimeState, type AgentType, type AgentUsage, AgentsClient, AnalysesClient, AuthClient, BINARY_EFFORT_VALUES, BROWSER_ACTIONBOOK_PROMPT, BROWSER_LOGIN_MCP_SERVER_NAME, type BaseMeta, type BestOfConfig, type BestOfParams, type BestOfResult, type BrowserConfig, type BrowserCredentialCreateInput, type BrowserCredentialDeleteInput, type BrowserCredentialListOptions, type BrowserCredentialMetadata, type BrowserCredentialScopeEntry, BrowserCredentialsClient, type BrowserCredentialsClientConfig, type BrowserCredentialsConfig, type BrowserProfileDeleteInput, type BrowserProfileMetadata, BrowserProfilesClient, type BrowserProfilesClientConfig, type BrowserProvider, type BrowserReplay, type BrowserReplayOptions, type BrowserRuntimeInfo, type CandidateCompleteEvent, CapabilityDocument, type CheckpointInfo, type CodexAgentPluginConfig, DEFAULT_HARNESS_TRIAL_LAYOUT, DEFAULT_REASONING_EFFORT, DatasetsClient, type DefaultBrowserConfig, type DownloadCheckpointOptions, type DownloadFilesOptions, type DownloadSessionOptions, type EffortSupport, type EmitOption, type EventHandler, type EventName, Evolve, EvolveApiError, type EvolveConfig, EvolveConfigError, EvolveDigestMismatchError, type EvolveEvents, EvolveIncompleteDownloadError, EvolveUploadTimeoutError, type ExecuteCommandOptions, type ExternalGatewayConfig, type FileInfo, type FileMap, type FileRange, type FilesystemEvent, type FilterConfig, type FilterParams, GatewayUsageEvent, type GeminiAgentPluginConfig, type GetEventsOptions, HARNESS_TRIAL_LAYOUTS, type HarnessTrialLayout, HostedClientConfig, HostedErrorCode, type HostedEvolve, ImportSettleError, type ImportSettleErrorCode, type IndexedMeta, type IntegrationAccount, type IntegrationAccountDeleteParams, type IntegrationAccountDeleteResult, type IntegrationAccountListParams, type IntegrationAccountUpdateParams, type IntegrationAccountUpdateResult, type IntegrationAuthParams, type IntegrationAuthResult, type IntegrationToolsFilter, type IntegrationsConfig, type IntegrationsSetup, type ItemInput, type ItemRetryEvent, JUDGE_PROMPT, Job, JobShareRequest, JobShares, JobVisibility, JobsClient, type JsonSchema, type JudgeCompleteEvent, type JudgeDecision, type JudgeMeta, type LifecycleEvent, type LifecycleReason, type ListSessionsOptions, MANAGED_SANDBOX_PROVIDERS, MAX_SKILL_BYTES, MAX_SKILL_FILES, type ManagedBrowserProvider, type ManagedSandboxCreateDefaults, type ManagedSandboxOptions, type ManagedSandboxProviderName, type ManagedSecretDeleteResult, type ManagedSecretDelivery, type ManagedSecretMetadata, type ManagedSecretRef, type ManagedSecretSetInput, type ManagedSecretWriteResult, type ManagedSecretsClient, type ManagedSecretsClientConfig, type MapConfig, type MapParams, type MarketplaceAgentPluginConfig, type McpConfigInfo, type McpServerConfig, type ModelInfo, NoActiveVersionError, type OnCandidateCompleteCallback, type OnItemRetryCallback, type OnJudgeCompleteCallback, type OnVerifierCompleteCallback, type OnWorkerCompleteCallback, type OperationType, OrgsClient, type OutputEvent, type OutputResult, type ParsedSkillRef, type PinnedSkillRef, Pipeline, type PipelineContext, type PipelineEventMap, type PipelineEvents, type PipelineResult, type ProcessInfo, type Prompt, type PromptFn, REASONING_EFFORTS, RETRY_FEEDBACK_PROMPT, type ReasoningEffort, type ReduceConfig, type ReduceMeta, type ReduceParams, type ReduceResult, type ResolvedSkill, type ResolvedStorageConfig, type RetryConfig, type RunCost, type RunOptions, SCHEMA_PROMPT, SKILL_FILE_NAME, SUPPORTED_RUN_OPTIONS, SWARM_RESULT_BRAND, SYSTEM_PROMPT, type SandboxCommandHandle, type SandboxCommandResult, type SandboxCommands, type SandboxCreateOptions, SandboxFeatureUnsupportedError, type SandboxFiles, type SandboxInspectOptions, type SandboxInstance, type SandboxLifecycleState, type SandboxMetrics, type SandboxNetworkPolicy, SandboxNotRunningError, SandboxPathNotFoundError, type SandboxProvider, type SandboxRunOptions, type SandboxSpawnOptions, type SchemaValidationOptions, Semaphore, type SessionCost, type SessionEvent, type SessionInfo, type SessionListScope, type SessionPage, type SessionStatus, type SessionTranscript, type SessionUpdate, type SessionsClient, type SessionsConfig, type SkillMountTarget, type SkillName, type SkillRef, SkillRefError, SkillResolveError, type SkillResolveOptions, SkillsClient, type SkillsConfig, type StepCompleteEvent, type StepErrorEvent, type StepEvent, type StepResult, type StepStartEvent, type StorageClient, type StorageConfig, type StreamCallbacks, type SupportedRunOption, Swarm, type SwarmConfig, type SwarmResult, SwarmResultList, TerminalPipeline, type TokenUsage, TraceEvent, Trial, TrialAnalysis, type TrialTreeParts, TrialsClient, UsageReading, VALIDATION_PRESETS, VERIFY_PROMPT, type ValidationMode, type VerifierCompleteEvent, type VerifyConfig, type VerifyDecision, type VerifyInfo, type VerifyMeta, WORKSPACE_PROMPT, WORKSPACE_SWE_PROMPT, type WatchHandle, type WatchOptions, type WorkerCompleteEvent, type WorkspaceMode, agents, analyses, analysisEvolveRecord, applyTemplate, assembleTrialTree, auth, browserCredentials, browserProfiles, buildWorkerSystemPrompt, checkEvolveRecord, checks, computeSkillDigest, createAgentParser, createClaudeParser, createCodexParser, createDroidParser, createGeminiParser, datasets, executeWithRetry, expandPath, getAgentConfig, getMcpSettingsDir, getMcpSettingsPath, harborCopyPath, harnessEffortVocabulary, harnessTrialLayout, homeRelativePath, hosted, isAgentWorkUpdate, isSandboxFeatureUnsupportedError, isSandboxNotRunningError, isSandboxPathNotFoundError, isValidAgentType, isZodSchema, jobEvolveRecord, jobs, jsonSchemaToString, managedSandbox, managedSecrets, meta, mountSkills, orgs, parseNdjsonLine, parseNdjsonOutput, parseQwenOutput, parseSkillRef, pinSkillRef, placeHomeObject, readLocalDir, resolveDefaultSandbox, resolveSkillSha, resolveSkills, resolveStorageConfig, saveLocalDir, sessions, skillCloneUrl, skillDisplayUrl, skills, storage, taskCheckEvolveRecord, trialEvolveRecord, trials, writeClaudeMcpConfig, writeCodexMcpConfig, writeDroidGatewaySettings, writeDroidMcpConfig, writeGeminiMcpConfig, writeMcpConfig, writeQwenMcpConfig, zodSchemaToJson };
5456
+ export { AGENT_REGISTRY, AGENT_TYPES, type ActionbookBrowserConfig, Agent, type AgentBrowserConfig, type AgentConfig, type AgentError, type AgentOptions, type AgentOverride, type AgentParser, type AgentPluginConfig, type AgentRegistryEntry, type AgentResponse, type AgentRuntimeState, type AgentType, type AgentUsage, AgentsClient, AnalysesClient, AuthClient, BINARY_EFFORT_VALUES, BROWSER_ACTIONBOOK_PROMPT, BROWSER_LOGIN_MCP_SERVER_NAME, type BaseMeta, type BestOfConfig, type BestOfParams, type BestOfResult, type BrowserConfig, type BrowserCredentialCreateInput, type BrowserCredentialDeleteInput, type BrowserCredentialListOptions, type BrowserCredentialMetadata, type BrowserCredentialScopeEntry, BrowserCredentialsClient, type BrowserCredentialsClientConfig, type BrowserCredentialsConfig, type BrowserProfileDeleteInput, type BrowserProfileMetadata, BrowserProfilesClient, type BrowserProfilesClientConfig, type BrowserProvider, type BrowserReplay, type BrowserReplayOptions, type BrowserRuntimeInfo, type CandidateCompleteEvent, CapabilityDocument, type CheckpointInfo, type CodexAgentPluginConfig, DEFAULT_HARNESS_TRIAL_LAYOUT, DEFAULT_REASONING_EFFORT, DatasetsClient, type DefaultBrowserConfig, type DownloadCheckpointOptions, type DownloadFilesOptions, type DownloadSessionOptions, type EffortSupport, type EmitOption, type EventHandler, type EventName, Evolve, EvolveApiError, type EvolveConfig, EvolveConfigError, EvolveDigestMismatchError, type EvolveEvents, EvolveIncompleteDownloadError, EvolveUploadTimeoutError, type ExecuteCommandOptions, type ExternalGatewayConfig, type FileInfo, type FileMap, type FileRange, type FilesystemEvent, type FilterConfig, type FilterParams, GatewayUsageEvent, type GeminiAgentPluginConfig, type GetEventsOptions, HARNESS_TRIAL_LAYOUTS, type HarnessEvent, type HarnessTrialLayout, HostedClientConfig, HostedErrorCode, type HostedEvolve, ImportSettleError, type ImportSettleErrorCode, type IndexedMeta, type IntegrationAccount, type IntegrationAccountDeleteParams, type IntegrationAccountDeleteResult, type IntegrationAccountListParams, type IntegrationAccountUpdateParams, type IntegrationAccountUpdateResult, type IntegrationAuthParams, type IntegrationAuthResult, type IntegrationToolsFilter, type IntegrationsConfig, type IntegrationsSetup, type ItemInput, type ItemRetryEvent, JUDGE_PROMPT, Job, JobShareRequest, JobShares, JobVisibility, JobsClient, type JsonSchema, type JudgeCompleteEvent, type JudgeDecision, type JudgeMeta, type LifecycleEvent, type LifecycleReason, type ListSessionsOptions, MANAGED_SANDBOX_PROVIDERS, MAX_SKILL_BYTES, MAX_SKILL_FILES, type ManagedBrowserProvider, type ManagedSandboxCreateDefaults, type ManagedSandboxOptions, type ManagedSandboxProviderName, type ManagedSecretDeleteResult, type ManagedSecretDelivery, type ManagedSecretMetadata, type ManagedSecretRef, type ManagedSecretSetInput, type ManagedSecretWriteResult, type ManagedSecretsClient, type ManagedSecretsClientConfig, type MapConfig, type MapParams, type MarketplaceAgentPluginConfig, type McpConfigInfo, type McpServerConfig, type ModelInfo, NoActiveVersionError, type OnCandidateCompleteCallback, type OnItemRetryCallback, type OnJudgeCompleteCallback, type OnVerifierCompleteCallback, type OnWorkerCompleteCallback, type OperationType, OrgsClient, type OutputEvent, type OutputResult, type ParsedSkillRef, type PinnedSkillRef, Pipeline, type PipelineContext, type PipelineEventMap, type PipelineEvents, type PipelineResult, type ProcessInfo, type Prompt, type PromptFn, REASONING_EFFORTS, RETRY_FEEDBACK_PROMPT, type ReasoningEffort, type ReduceConfig, type ReduceMeta, type ReduceParams, type ReduceResult, type ResolvedSkill, type ResolvedStorageConfig, type RetryConfig, type RunCost, type RunOptions, SCHEMA_PROMPT, SKILL_FILE_NAME, SUPPORTED_RUN_OPTIONS, SWARM_RESULT_BRAND, SYSTEM_PROMPT, type SandboxCommandHandle, type SandboxCommandResult, type SandboxCommands, type SandboxCreateOptions, SandboxFeatureUnsupportedError, type SandboxFiles, type SandboxInspectOptions, type SandboxInstance, type SandboxLifecycleState, type SandboxMetrics, type SandboxNetworkPolicy, SandboxNotRunningError, SandboxPathNotFoundError, type SandboxProvider, type SandboxRunOptions, type SandboxSpawnOptions, type SchemaValidationOptions, Semaphore, type SessionCost, type SessionEvent, type SessionInfo, type SessionListScope, type SessionPage, type SessionStatus, type SessionTranscript, type SessionUpdate, type SessionsClient, type SessionsConfig, type SkillMountTarget, type SkillName, type SkillRef, SkillRefError, SkillResolveError, type SkillResolveOptions, SkillsClient, type SkillsConfig, type StepCompleteEvent, type StepErrorEvent, type StepEvent, type StepResult, type StepStartEvent, type StorageClient, type StorageConfig, type StreamCallbacks, type SupportedRunOption, Swarm, type SwarmConfig, type SwarmResult, SwarmResultList, TerminalPipeline, type TokenUsage, TraceEvent, Trial, TrialAnalysis, type TrialTreeParts, TrialsClient, UsageReading, VALIDATION_PRESETS, VERIFY_PROMPT, type ValidationMode, type VerifierCompleteEvent, type VerifyConfig, type VerifyDecision, type VerifyInfo, type VerifyMeta, WORKSPACE_PROMPT, WORKSPACE_SWE_PROMPT, type WatchHandle, type WatchOptions, type WorkerCompleteEvent, type WorkspaceMode, agents, analyses, analysisEvolveRecord, applyTemplate, assembleTrialTree, auth, browserCredentials, browserProfiles, buildWorkerSystemPrompt, checkEvolveRecord, checks, computeSkillDigest, createAgentParser, createAntigravityParser, createClaudeParser, createCodexParser, createDroidParser, createDshParser, createGeminiParser, createPiParser, createPrimeAgentParser, createZcodeParser, datasets, executeWithRetry, expandPath, getAgentConfig, getMcpSettingsDir, getMcpSettingsPath, harborCopyPath, harnessEffortVocabulary, harnessTrialLayout, homeRelativePath, hosted, isAgentWorkUpdate, isSandboxFeatureUnsupportedError, isSandboxNotRunningError, isSandboxPathNotFoundError, isValidAgentType, isZodSchema, jobEvolveRecord, jobs, jsonSchemaToString, managedSandbox, managedSecrets, meta, mountSkills, orgs, parseNdjsonLine, parseNdjsonOutput, parseQwenOutput, parseSkillRef, pinSkillRef, placeHomeObject, readLocalDir, resolveDefaultSandbox, resolveSkillSha, resolveSkills, resolveStorageConfig, saveLocalDir, sessions, skillCloneUrl, skillDisplayUrl, skills, storage, taskCheckEvolveRecord, trialEvolveRecord, trials, writeAntigravityMcpConfig, writeAntigravitySettings, writeClaudeMcpConfig, writeCodexMcpConfig, writeDroidGatewaySettings, writeDroidMcpConfig, writeDshMcpConfig, writeDshRoutePatch, writeGeminiMcpConfig, writeKimiMcpConfig, writeMcpConfig, writeOpenCodeMcpConfig, writePiMcpConfig, writePrimeAgentMcpConfig, writeQwenMcpConfig, writeZcodeMcpConfig, writeZcodeProviderConfig, zodSchemaToJson };