@evolvingmachines/evolve 0.0.64 → 0.0.65

package/dist/{types-LYvFBfCo.d.cts → types-B_qmzlMk.d.cts}

@@ -112,6 +112,13 @@ type EvalSandboxProvider = (typeof EVAL_SANDBOX_PROVIDERS)[number];
 declare const JOB_LIST_SCOPES: readonly ["my", "shared", "org"];
 /** One list scope — see JOB_LIST_SCOPES. */
 type JobListScope = (typeof JOB_LIST_SCOPES)[number];
+/**
+ * A check is a job, so the list takes a `kind` (the ruling and the Harbor lines live on the spec's
+ * JobListKind parameter). A runtime value so the CLI validates `--kind` against it.
+ */
+declare const JOB_LIST_KINDS: readonly ["job", "check", "all"];
+/** One list kind — see JOB_LIST_KINDS. */
+type JobListKind = (typeof JOB_LIST_KINDS)[number];
 /**
  * An analysis's own lifecycle ladder — lowercase, the object's Harbor
  * dialect (spec TrialAnalysis.status). A runtime value so the CLI validates
@@ -568,6 +575,22 @@ interface AnalyzeConfigInput {
      * anything else is refused `invalid_input` naming `analyze.n_trials`.
      */
     n_trials?: number;
+    /**
+     * Analyze only these trials of the job — Harbor's `harbor analyze <trial
+     * directory>` for one trial (their cli/analyze.py:242-245), given as ids
+     * here; combinable with `passing`/`failing` (a listed trial on the other
+     * side of the filter is skipped) and applied before `n_trials` (Harbor's
+     * one-trial path skips its filter and cap, analyzer.py:180-181 before
+     * :189-191; hosted, both apply to the named trials too). An id that is not
+     * a trial of this job is refused `invalid_input` naming `analyze.trial_ids`
+     * with the unknown ids named in the message and in
+     * `details.unknown_trial_ids`; a list that leaves nothing analyzable is the
+     * 409 `no_analyzable_trials`; an empty list, a duplicate, an empty string
+     * or a non-string is refused `invalid_input`. Only `jobs().analyze()` takes
+     * it: on `start({ analyze })` (the embedded trigger) it is refused — the
+     * trials do not exist yet.
+     */
+    trial_ids?: string[];
 }
 /**
  * The RESOLVED trace-analysis policy — the caller's values or the defaults of
@@ -611,6 +634,11 @@ interface AnalyzeConfig {
     failing: boolean;
     /** The trial cap as stored (`AnalyzeConfigInput.n_trials`); null = no cap. */
     n_trials: number | null;
+    /**
+     * Always null on the job body today: the stored policy is the create-time
+     * one, which refuses `trial_ids`; a manual wave's named trials are the rows it enqueues.
+     */
+    trial_ids: string[] | null;
 }
 /** The job-creation body — POST /api/jobs. */
 interface JobCreate {
@@ -1235,11 +1263,20 @@ interface JobBuildExclusion {
     failed_task_names: string[];
     note: string;
 }
+/**
+ * The caller's relation to a job on a read (Job.viewer): `creator` made it,
+ * `member` belongs to its organization, `shared` reads it through an email
+ * share, `link` through its unlisted link. Acting verbs (analyze, cancel,
+ * resume, retry, regrade) are open to `creator` and `member` only.
+ */
+type JobViewer = "creator" | "member" | "shared" | "link";
 /**
  * THE job body — the same shape from create, get, list items, cancel, resume,
  * and regrade responses; no field appears on some responses and not others.
  */
 interface Job {
+    /** The row's kind on the jobs list (JobListItem): always `job` on a Job body; a CheckRow says `check`. */
+    kind: "job";
     id: string;
     /** User-facing label. */
     job_name: string;
@@ -1328,6 +1365,14 @@ interface Job {
     source_jobs: SourceJob[];
     /** Derived: any source_jobs entry with action "regrade". */
     is_regrade: boolean;
+    /**
+     * The caller's relation on a read (JobViewer). Null on the responses that
+     * echo a job the caller just acted on (create, analyze, cancel, resume,
+     * retry, regrade), where the caller is the creator or a member by
+     * construction, and on list rows (batched; read one job to learn it);
+     * null too from a server older than the field.
+     */
+    viewer: JobViewer | null;
     /**
      * The upload provenance echo — null for every job this platform executed,
      * non-null only on a job ingested by jobs().upload(). See UploadProvenance.
@@ -2225,6 +2270,54 @@ type CheckPage = Page<Check>;
 /** The handle returned by checks().list() — one page on await, every page on for-await. */
 interface CheckList extends Awaitable<CheckPage>, AsyncIterable<Check> {
 }
+/**
+ * Spec CheckTaskTally. A stopped task check is stored `failed` with the stop phase; the tally counts it
+ * under `stopped`, never `failed`, so a stop never reads as a failure.
+ */
+interface CheckTaskTally {
+    total: number;
+    byStatus: Record<AnalysisStatus | "stopped", number>;
+}
+/**
+ * Spec CheckRow, one row of `jobs().list({ kind: "check" | "all" })`: a check is a job, so it lists beside them.
+ * Not a Job (no arms, attempts, caps, retry policy, trials or upload provenance), so those fields are absent, never faked.
+ */
+interface CheckRow {
+    kind: "check";
+    id: string;
+    /** Check.name — Harbor's `--job-name`: the caller's, or the accept stamp. */
+    name: string;
+    status: CheckStatus;
+    source: CheckSource;
+    /** The checker's model. */
+    model_name: string;
+    /** The effort every task's checker ran at. */
+    reasoning_effort: string;
+    sandbox_provider: EvalSandboxProvider;
+    /** The owning organization's slug (every check has one). */
+    org: string;
+    visibility: JobVisibility;
+    /** How many task checks, and how they break down; done = completed + failed + stopped. */
+    tasks: CheckTaskTally;
+    /** The sum of the measured task costs; null when none was measured. */
+    cost_usd: number | null;
+    /** The accept instant; the list orders it with the jobs' `started_at`. */
+    created_at: string;
+    /** When the last task settled; null until every task has. */
+    finished_at: string | null;
+}
+/** One row of the jobs list, told apart by `kind` (spec JobListItem). */
+type JobListItem = Job | CheckRow;
+/** Cursor page of check rows (newest first) — `jobs().list({ kind: "check" })`. */
+type CheckRowPage = Page<CheckRow>;
+/** The handle returned by `jobs().list({ kind: "check" })`. */
+interface CheckRowList extends Awaitable<CheckRowPage>, AsyncIterable<CheckRow> {
+}
+/** Cursor page of jobs and check rows (newest first) — `jobs().list({ kind: "all" })`. */
+type JobListItemPage = Page<JobListItem>;
+/** The handle returned by `jobs().list({ kind: "all" })`. */
+interface JobListItemList extends Awaitable<JobListItemPage>, AsyncIterable<JobListItem> {
+}
 /**
  * One task's rollup within a job: its trial tally, mean reward over SCORED
  * trials, and measured cost. Sits between the job body and the trial list so
@@ -3140,6 +3233,8 @@ interface ListJobsOptions extends PageOptions {
      * See JOB_LIST_SCOPES.
      */
     scope?: JobListScope;
+    /** `job` (the server's default), `check` or `all`; see JOB_LIST_KINDS. */
+    kind?: JobListKind;
 }
 /** Options for analyses().list() (default page 50, max 200) */
 interface ListAnalysesOptions extends PageOptions {
@@ -3755,11 +3850,19 @@ interface JobsClient {
     /** Get one job */
     get(id: string): Promise<Job>;
     /**
-     * List the caller's jobs, newest first (cursor-paged). Await the
-     * result for one page, or `for await` it to walk every job across
-     * cursor pages transparently.
+     * List the caller's jobs, newest first (cursor-paged): await one page, or `for await` every page.
+     * One overload per `kind`, so the rows are typed as what was asked for.
      */
-    list(options?: ListJobsOptions): JobList;
+    list(options?: ListJobsOptions & {
+        kind?: "job";
+    }): JobList;
+    list(options: ListJobsOptions & {
+        kind: "check";
+    }): CheckRowList;
+    list(options: ListJobsOptions & {
+        kind: "all";
+    }): JobListItemList;
+    list(options: ListJobsOptions): JobListItemList;
     /**
      * List a job's trials (cursor-paged; { status } filters, e.g. to
      * the failed trials). Await the result for one page, or `for await` it to
@@ -4466,6 +4569,8 @@ interface AnalysesClient {
      * feed's species-blind events door, which answers `trial_not_found`.
      */
     list(options?: ListAnalysesOptions): AnalysisList;
+    /** The defaults an analysis runs under when its config names nothing (GET /api/analyses/defaults): model, effort, provider, rubric and the unrendered prompt template. */
+    defaults(): Promise<AnalyzeDefaults>;
     /**
      * The verdict document — the wire's TrialAnalysis, statuses and typed
      * failure included, for EVERY analysis (not only completed ones). The same
@@ -4697,6 +4802,21 @@ interface Check {
     /** When the last task settled; null until every task has. */
     finished_at: string | null;
 }
+/**
+ * The policy an empty analyze config resolves to (GET /api/analyses/defaults):
+ * each key the value `AnalyzeConfig` echoes for a job created with
+ * `analyze: {}`, except `prompt`, which `AnalyzeConfig` serves as null and
+ * this serves as the template text.
+ */
+interface AnalyzeDefaults {
+    model_name: string;
+    rubric: Rubric;
+    /** The built-in analyze prompt template, unrendered — pass it as `prompt` to run the default body explicitly, or edit it from here. */
+    prompt: string;
+    /** The effort the default model runs at when the config names none. */
+    reasoning_effort: string;
+    sandbox_provider: EvalSandboxProvider;
+}
 /**
  * The policy an empty check config resolves to (GET /api/checks/defaults):
  * each key the value `Check` echoes for a check created with no config,
@@ -5288,4 +5408,4 @@ interface CapabilityDocument {
     error_codes: string[];
 }
 
-export { type AnalysisEvidence as $, type AgentInput as A, type AgentArm as B, type Check as C, type DatasetVersion as D, type AgentArmInput as E, type AgentCapability as F, type GatewayUsageEvent as G, type HostedErrorCode as H, type AgentDatasetStats as I, type JobCreate as J, type AgentEffortSupport as K, type AgentInfo as L, type AgentList as M, type AgentModelOption as N, type OrgsClient as O, type PublishDatasetInput as P, type AgentPage as Q, type Rubric as R, type SkillsClient as S, type Trial as T, type UsageReading as U, type AgentResult as V, type AgentSource as W, type AgentSourceInput as X, type AgentUpsertInput as Y, type AnalysisArtifactStream as Z, type AnalysisCheck as _, type TrialAnalysis as a, type FilesystemStreamEvent as a$, type AnalysisFailure as a0, type AnalysisLabel as a1, type AnalysisList as a2, type AnalysisPage as a3, type AnalysisStatus as a4, type AnalysisTranscript as a5, type AnalysisTranscriptOptions as a6, type AnalyzeConfig as a7, type AnalyzeConfigInput as a8, type ApiKey as a9, type DatasetVersionHubSource as aA, type DatasetVersionSource as aB, type DatasetVersionState as aC, type DownloadDatasetOptions as aD, type DownloadJobOptions as aE, EVAL_SANDBOX_PROVIDERS as aF, type Agent as aG, type ModelInfo as aH, type EvalSandboxProvider as aI, type StepResult as aJ, type ExceptionInfo as aK, type FilesystemArchiveOptions as aL, type FilesystemBox as aM, type FilesystemCapture as aN, type FilesystemChange as aO, type FilesystemChanges as aP, type FilesystemChangesOptions as aQ, type FilesystemEntry as aR, type FilesystemListOptions as aS, type FilesystemListing as aT, type FilesystemReadOptions as aU, type FilesystemSearchHit as aV, type FilesystemSearchOptions as aW, type FilesystemSearchResult as aX, type FilesystemSource as aY, type FilesystemState as aZ, type FilesystemStatus as a_, type AttemptPhase as aa, type AuthStatus as ab, type Awaitable as ac, CHECK_STATUSES as ad, type CheckDefaults as ae, type CheckLabel as af, type CompareCell as ag, type CompareCoverage as ah, type CompareJobAggregate as ai, type CompareResponse as aj, type CompareTaskRow as ak, type Dataset as al, type DatasetImportFailure as am, type DatasetImportList as an, type DatasetImportPage as ao, type DatasetImportStatus as ap, type DatasetList as aq, type DatasetPage as ar, type DatasetPatch as as, type DatasetPreflight as at, type DatasetRef as au, type DatasetSelector as av, type DatasetSource as aw, type DatasetVersionArchiveSource as ax, type DatasetVersionArchiveUrlSource as ay, type DatasetVersionGitSource as az, type DatasetImport as b, type PreflightDatasetInput as b$, type FilesystemStreamOptions as b0, type FilesystemWatchResult as b1, GATEWAY_TRACE_SEQ_BASE as b2, type GatewayUsage as b3, type GetDatasetOptions as b4, type GrepJobOptions as b5, HOSTED_ERROR_CODES as b6, type ImportPhase as b7, type ImportPhaseProgress as b8, type ImportWarning as b9, type JobTaskRollupPage as bA, type JobWatch as bB, type JudgeResult as bC, type ListAgentsOptions as bD, type ListAnalysesOptions as bE, type ListDatasetsOptions as bF, type ListImportsOptions as bG, type ListJobImportsOptions as bH, type ListJobTasksOptions as bI, type ListJobsOptions as bJ, type ListSkillsOptions as bK, type ListTrialFilesOptions as bL, type ListTrialsOptions as bM, type ManagedProviderCapability as bN, type OrgInvite as bO, type OrgInviteCreated as bP, type OrgJoined as bQ, type OrgMember as bR, type OrgQuota as bS, type OrgRole as bT, type OrgUsage as bU, type Organization as bV, type OrganizationDetail as bW, type Page as bX, type PageOptions as bY, type PassAtKGroup as bZ, type PassAtKPoint as b_, type InfraFailureSignature as ba, JOB_LIST_SCOPES as bb, type JobAnalysisStats as bc, type JobDeleteResult as bd, type JobFailure as be, type JobGrepGroup as bf, type JobGrepPage as bg, type JobImport as bh, type JobImportFailure as bi, type JobImportList as bj, type JobImportPage as bk, type JobImportPhaseName as bl, type JobImportPhaseProgress as bm, type JobImportProgress as bn, type JobImportSkippedTrial as bo, type JobImportSource as bp, type JobList as bq, type JobListScope as br, type JobPage as bs, type JobShareEmail as bt, type JobShareLink as bu, type JobStats as bv, type JobStatus as bw, type JobTaskLink as bx, type JobTaskRollup as by, type JobTaskRollupList as bz, type JobEvent as c, gatewayUsageOf as c$, type PreflightDeferredCheck as c0, type PreflightManifestVerdict as c1, type PreflightTaskVerdict as c2, type ProviderCapability as c3, type PublishDatasetOptions as c4, type RegradeRequest as c5, type ResumeRequest as c6, type RetryRequest as c7, type RubricCriterion as c8, type RunFilesystem as c9, type TaskPackageFilesystemStatus as cA, type TaskProviderVerdict as cB, type TimingInfo as cC, type TraceEventPage as cD, type TraceOptions as cE, type TrialArtifactStream as cF, type TrialCounts as cG, type TrialFile as cH, type TrialFilePage as cI, type TrialFileRange as cJ, type TrialGpuCost as cK, type TrialList as cL, type TrialPage as cM, type TrialStatus as cN, type TrialStatusTally as cO, type TrialTaskLink as cP, type TrialUploadProvenance as cQ, type UploadJobOptions as cR, type UploadProvenance as cS, type UploadSkillOptions as cT, type UpstreamStatus as cU, type VerifierEnvironmentMode as cV, type VerifierResult as cW, type WatchAnalysisOptions as cX, type WatchImportOptions as cY, type WatchJobImportOptions as cZ, type WatchJobOptions as c_, SANDBOX_LOG_STREAMS as ca, type SandboxLogEvent as cb, type SandboxLogLine as cc, type SandboxLogLines as cd, type SandboxLogOptions as ce, type SandboxLogStream as cf, type SandboxProcs as cg, type SkillLock as ch, type SkillUpload as ci, type SkillUploadList as cj, type SkillUploadPage as ck, type SourceJob as cl, type SpendSource as cm, type StartJobOptions as cn, type StatusVocabulary as co, type StopResponse as cp, TASK_LINKED_BY as cq, TASK_LINK_REASONS as cr, TRIAL_ARTIFACT_STREAMS as cs, TRIAL_STATUSES as ct, type Task as cu, type TaskCheckTranscript as cv, type TaskLinkReason as cw, type TaskLinkedBy as cx, type TaskNote as cy, type TaskPackageFiles as cz, type DatasetFailedTask as d, isHostedErrorCode as d0, passAtK as d1, type DatasetImportProgress as e, type JobSecretInline as f, type JobSecretRef as g, type TraceEvent as h, type JobVisibility as i, type JobShareRequest as j, type JobShares as k, type Job as l, type TaskCheck as m, type DatasetsClient as n, type AgentsClient as o, type JobsClient as p, type TrialsClient as q, type AnalysesClient as r, type ChecksClient as s, type CapabilityDocument as t, type HostedClientConfig as u, type AuthClient as v, AGENT_EFFORT_SUPPORT_VALUES as w, ANALYSIS_ARTIFACT_STREAMS as x, ANALYSIS_STATUSES as y, type ActiveDataset as z };
+export { type AnalysisEvidence as $, type AgentInput as A, type AgentArm as B, type Check as C, type DatasetVersion as D, type AgentArmInput as E, type AgentCapability as F, type GatewayUsageEvent as G, type HostedErrorCode as H, type AgentDatasetStats as I, type JobCreate as J, type AgentEffortSupport as K, type AgentInfo as L, type AgentList as M, type AgentModelOption as N, type OrgsClient as O, type PublishDatasetInput as P, type AgentPage as Q, type Rubric as R, type SkillsClient as S, type Trial as T, type UsageReading as U, type AgentResult as V, type AgentSource as W, type AgentSourceInput as X, type AgentUpsertInput as Y, type AnalysisArtifactStream as Z, type AnalysisCheck as _, type TrialAnalysis as a, type FilesystemSearchOptions as a$, type AnalysisFailure as a0, type AnalysisLabel as a1, type AnalysisList as a2, type AnalysisPage as a3, type AnalysisStatus as a4, type AnalysisTranscript as a5, type AnalysisTranscriptOptions as a6, type AnalyzeConfig as a7, type AnalyzeConfigInput as a8, type AnalyzeDefaults as a9, type DatasetSelector as aA, type DatasetSource as aB, type DatasetVersionArchiveSource as aC, type DatasetVersionArchiveUrlSource as aD, type DatasetVersionGitSource as aE, type DatasetVersionHubSource as aF, type DatasetVersionSource as aG, type DatasetVersionState as aH, type DownloadDatasetOptions as aI, type DownloadJobOptions as aJ, EVAL_SANDBOX_PROVIDERS as aK, type Agent as aL, type ModelInfo as aM, type EvalSandboxProvider as aN, type StepResult as aO, type ExceptionInfo as aP, type FilesystemArchiveOptions as aQ, type FilesystemBox as aR, type FilesystemCapture as aS, type FilesystemChange as aT, type FilesystemChanges as aU, type FilesystemChangesOptions as aV, type FilesystemEntry as aW, type FilesystemListOptions as aX, type FilesystemListing as aY, type FilesystemReadOptions as aZ, type FilesystemSearchHit as a_, type ApiKey as aa, type AttemptPhase as ab, type AuthStatus as ac, type Awaitable as ad, CHECK_STATUSES as ae, type CheckDefaults as af, type CheckLabel as ag, type CheckRow as ah, type CheckRowList as ai, type CheckRowPage as aj, type CheckTaskTally as ak, type CompareCell as al, type CompareCoverage as am, type CompareJobAggregate as an, type CompareResponse as ao, type CompareTaskRow as ap, type Dataset as aq, type DatasetImportFailure as ar, type DatasetImportList as as, type DatasetImportPage as at, type DatasetImportStatus as au, type DatasetList as av, type DatasetPage as aw, type DatasetPatch as ax, type DatasetPreflight as ay, type DatasetRef as az, type DatasetImport as b, type OrgJoined as b$, type FilesystemSearchResult as b0, type FilesystemSource as b1, type FilesystemState as b2, type FilesystemStatus as b3, type FilesystemStreamEvent as b4, type FilesystemStreamOptions as b5, type FilesystemWatchResult as b6, GATEWAY_TRACE_SEQ_BASE as b7, type GatewayUsage as b8, type GetDatasetOptions as b9, type JobListKind as bA, type JobListScope as bB, type JobPage as bC, type JobShareEmail as bD, type JobShareLink as bE, type JobStats as bF, type JobStatus as bG, type JobTaskLink as bH, type JobTaskRollup as bI, type JobTaskRollupList as bJ, type JobTaskRollupPage as bK, type JobViewer as bL, type JobWatch as bM, type JudgeResult as bN, type ListAgentsOptions as bO, type ListAnalysesOptions as bP, type ListDatasetsOptions as bQ, type ListImportsOptions as bR, type ListJobImportsOptions as bS, type ListJobTasksOptions as bT, type ListJobsOptions as bU, type ListSkillsOptions as bV, type ListTrialFilesOptions as bW, type ListTrialsOptions as bX, type ManagedProviderCapability as bY, type OrgInvite as bZ, type OrgInviteCreated as b_, type GrepJobOptions as ba, HOSTED_ERROR_CODES as bb, type ImportPhase as bc, type ImportPhaseProgress as bd, type ImportWarning as be, type InfraFailureSignature as bf, JOB_LIST_KINDS as bg, JOB_LIST_SCOPES as bh, type JobAnalysisStats as bi, type JobDeleteResult as bj, type JobFailure as bk, type JobGrepGroup as bl, type JobGrepPage as bm, type JobImport as bn, type JobImportFailure as bo, type JobImportList as bp, type JobImportPage as bq, type JobImportPhaseName as br, type JobImportPhaseProgress as bs, type JobImportProgress as bt, type JobImportSkippedTrial as bu, type JobImportSource as bv, type JobList as bw, type JobListItem as bx, type JobListItemList as by, type JobListItemPage as bz, type JobEvent as c, type TrialUploadProvenance as c$, type OrgMember as c0, type OrgQuota as c1, type OrgRole as c2, type OrgUsage as c3, type Organization as c4, type OrganizationDetail as c5, type Page as c6, type PageOptions as c7, type PassAtKGroup as c8, type PassAtKPoint as c9, type StopResponse as cA, TASK_LINKED_BY as cB, TASK_LINK_REASONS as cC, TRIAL_ARTIFACT_STREAMS as cD, TRIAL_STATUSES as cE, type Task as cF, type TaskCheckTranscript as cG, type TaskLinkReason as cH, type TaskLinkedBy as cI, type TaskNote as cJ, type TaskPackageFiles as cK, type TaskPackageFilesystemStatus as cL, type TaskProviderVerdict as cM, type TimingInfo as cN, type TraceEventPage as cO, type TraceOptions as cP, type TrialArtifactStream as cQ, type TrialCounts as cR, type TrialFile as cS, type TrialFilePage as cT, type TrialFileRange as cU, type TrialGpuCost as cV, type TrialList as cW, type TrialPage as cX, type TrialStatus as cY, type TrialStatusTally as cZ, type TrialTaskLink as c_, type PreflightDatasetInput as ca, type PreflightDeferredCheck as cb, type PreflightManifestVerdict as cc, type PreflightTaskVerdict as cd, type ProviderCapability as ce, type PublishDatasetOptions as cf, type RegradeRequest as cg, type ResumeRequest as ch, type RetryRequest as ci, type RubricCriterion as cj, type RunFilesystem as ck, SANDBOX_LOG_STREAMS as cl, type SandboxLogEvent as cm, type SandboxLogLine as cn, type SandboxLogLines as co, type SandboxLogOptions as cp, type SandboxLogStream as cq, type SandboxProcs as cr, type SkillLock as cs, type SkillUpload as ct, type SkillUploadList as cu, type SkillUploadPage as cv, type SourceJob as cw, type SpendSource as cx, type StartJobOptions as cy, type StatusVocabulary as cz, type DatasetFailedTask as d, type UploadJobOptions as d0, type UploadProvenance as d1, type UploadSkillOptions as d2, type UpstreamStatus as d3, type VerifierEnvironmentMode as d4, type VerifierResult as d5, type WatchAnalysisOptions as d6, type WatchImportOptions as d7, type WatchJobImportOptions as d8, type WatchJobOptions as d9, gatewayUsageOf as da, isHostedErrorCode as db, passAtK as dc, type DatasetImportProgress as e, type JobSecretInline as f, type JobSecretRef as g, type TraceEvent as h, type JobVisibility as i, type JobShareRequest as j, type JobShares as k, type Job as l, type TaskCheck as m, type DatasetsClient as n, type AgentsClient as o, type JobsClient as p, type TrialsClient as q, type AnalysesClient as r, type ChecksClient as s, type CapabilityDocument as t, type HostedClientConfig as u, type AuthClient as v, AGENT_EFFORT_SUPPORT_VALUES as w, ANALYSIS_ARTIFACT_STREAMS as x, ANALYSIS_STATUSES as y, type ActiveDataset as z };

package/dist/{types-LYvFBfCo.d.ts → types-B_qmzlMk.d.ts}

@@ -112,6 +112,13 @@ type EvalSandboxProvider = (typeof EVAL_SANDBOX_PROVIDERS)[number];
 declare const JOB_LIST_SCOPES: readonly ["my", "shared", "org"];
 /** One list scope — see JOB_LIST_SCOPES. */
 type JobListScope = (typeof JOB_LIST_SCOPES)[number];
+/**
+ * A check is a job, so the list takes a `kind` (the ruling and the Harbor lines live on the spec's
+ * JobListKind parameter). A runtime value so the CLI validates `--kind` against it.
+ */
+declare const JOB_LIST_KINDS: readonly ["job", "check", "all"];
+/** One list kind — see JOB_LIST_KINDS. */
+type JobListKind = (typeof JOB_LIST_KINDS)[number];
 /**
  * An analysis's own lifecycle ladder — lowercase, the object's Harbor
  * dialect (spec TrialAnalysis.status). A runtime value so the CLI validates
@@ -568,6 +575,22 @@ interface AnalyzeConfigInput {
      * anything else is refused `invalid_input` naming `analyze.n_trials`.
      */
     n_trials?: number;
+    /**
+     * Analyze only these trials of the job — Harbor's `harbor analyze <trial
+     * directory>` for one trial (their cli/analyze.py:242-245), given as ids
+     * here; combinable with `passing`/`failing` (a listed trial on the other
+     * side of the filter is skipped) and applied before `n_trials` (Harbor's
+     * one-trial path skips its filter and cap, analyzer.py:180-181 before
+     * :189-191; hosted, both apply to the named trials too). An id that is not
+     * a trial of this job is refused `invalid_input` naming `analyze.trial_ids`
+     * with the unknown ids named in the message and in
+     * `details.unknown_trial_ids`; a list that leaves nothing analyzable is the
+     * 409 `no_analyzable_trials`; an empty list, a duplicate, an empty string
+     * or a non-string is refused `invalid_input`. Only `jobs().analyze()` takes
+     * it: on `start({ analyze })` (the embedded trigger) it is refused — the
+     * trials do not exist yet.
+     */
+    trial_ids?: string[];
 }
 /**
  * The RESOLVED trace-analysis policy — the caller's values or the defaults of
@@ -611,6 +634,11 @@ interface AnalyzeConfig {
     failing: boolean;
     /** The trial cap as stored (`AnalyzeConfigInput.n_trials`); null = no cap. */
     n_trials: number | null;
+    /**
+     * Always null on the job body today: the stored policy is the create-time
+     * one, which refuses `trial_ids`; a manual wave's named trials are the rows it enqueues.
+     */
+    trial_ids: string[] | null;
 }
 /** The job-creation body — POST /api/jobs. */
 interface JobCreate {
@@ -1235,11 +1263,20 @@ interface JobBuildExclusion {
     failed_task_names: string[];
     note: string;
 }
+/**
+ * The caller's relation to a job on a read (Job.viewer): `creator` made it,
+ * `member` belongs to its organization, `shared` reads it through an email
+ * share, `link` through its unlisted link. Acting verbs (analyze, cancel,
+ * resume, retry, regrade) are open to `creator` and `member` only.
+ */
+type JobViewer = "creator" | "member" | "shared" | "link";
 /**
  * THE job body — the same shape from create, get, list items, cancel, resume,
  * and regrade responses; no field appears on some responses and not others.
  */
 interface Job {
+    /** The row's kind on the jobs list (JobListItem): always `job` on a Job body; a CheckRow says `check`. */
+    kind: "job";
     id: string;
     /** User-facing label. */
     job_name: string;
@@ -1328,6 +1365,14 @@ interface Job {
     source_jobs: SourceJob[];
     /** Derived: any source_jobs entry with action "regrade". */
     is_regrade: boolean;
+    /**
+     * The caller's relation on a read (JobViewer). Null on the responses that
+     * echo a job the caller just acted on (create, analyze, cancel, resume,
+     * retry, regrade), where the caller is the creator or a member by
+     * construction, and on list rows (batched; read one job to learn it);
+     * null too from a server older than the field.
+     */
+    viewer: JobViewer | null;
     /**
      * The upload provenance echo — null for every job this platform executed,
      * non-null only on a job ingested by jobs().upload(). See UploadProvenance.
@@ -2225,6 +2270,54 @@ type CheckPage = Page<Check>;
 /** The handle returned by checks().list() — one page on await, every page on for-await. */
 interface CheckList extends Awaitable<CheckPage>, AsyncIterable<Check> {
 }
+/**
+ * Spec CheckTaskTally. A stopped task check is stored `failed` with the stop phase; the tally counts it
+ * under `stopped`, never `failed`, so a stop never reads as a failure.
+ */
+interface CheckTaskTally {
+    total: number;
+    byStatus: Record<AnalysisStatus | "stopped", number>;
+}
+/**
+ * Spec CheckRow, one row of `jobs().list({ kind: "check" | "all" })`: a check is a job, so it lists beside them.
+ * Not a Job (no arms, attempts, caps, retry policy, trials or upload provenance), so those fields are absent, never faked.
+ */
+interface CheckRow {
+    kind: "check";
+    id: string;
+    /** Check.name — Harbor's `--job-name`: the caller's, or the accept stamp. */
+    name: string;
+    status: CheckStatus;
+    source: CheckSource;
+    /** The checker's model. */
+    model_name: string;
+    /** The effort every task's checker ran at. */
+    reasoning_effort: string;
+    sandbox_provider: EvalSandboxProvider;
+    /** The owning organization's slug (every check has one). */
+    org: string;
+    visibility: JobVisibility;
+    /** How many task checks, and how they break down; done = completed + failed + stopped. */
+    tasks: CheckTaskTally;
+    /** The sum of the measured task costs; null when none was measured. */
+    cost_usd: number | null;
+    /** The accept instant; the list orders it with the jobs' `started_at`. */
+    created_at: string;
+    /** When the last task settled; null until every task has. */
+    finished_at: string | null;
+}
+/** One row of the jobs list, told apart by `kind` (spec JobListItem). */
+type JobListItem = Job | CheckRow;
+/** Cursor page of check rows (newest first) — `jobs().list({ kind: "check" })`. */
+type CheckRowPage = Page<CheckRow>;
+/** The handle returned by `jobs().list({ kind: "check" })`. */
+interface CheckRowList extends Awaitable<CheckRowPage>, AsyncIterable<CheckRow> {
+}
+/** Cursor page of jobs and check rows (newest first) — `jobs().list({ kind: "all" })`. */
+type JobListItemPage = Page<JobListItem>;
+/** The handle returned by `jobs().list({ kind: "all" })`. */
+interface JobListItemList extends Awaitable<JobListItemPage>, AsyncIterable<JobListItem> {
+}
 /**
  * One task's rollup within a job: its trial tally, mean reward over SCORED
  * trials, and measured cost. Sits between the job body and the trial list so
@@ -3140,6 +3233,8 @@ interface ListJobsOptions extends PageOptions {
      * See JOB_LIST_SCOPES.
      */
     scope?: JobListScope;
+    /** `job` (the server's default), `check` or `all`; see JOB_LIST_KINDS. */
+    kind?: JobListKind;
 }
 /** Options for analyses().list() (default page 50, max 200) */
 interface ListAnalysesOptions extends PageOptions {
@@ -3755,11 +3850,19 @@ interface JobsClient {
     /** Get one job */
     get(id: string): Promise<Job>;
     /**
-     * List the caller's jobs, newest first (cursor-paged). Await the
-     * result for one page, or `for await` it to walk every job across
-     * cursor pages transparently.
+     * List the caller's jobs, newest first (cursor-paged): await one page, or `for await` every page.
+     * One overload per `kind`, so the rows are typed as what was asked for.
      */
-    list(options?: ListJobsOptions): JobList;
+    list(options?: ListJobsOptions & {
+        kind?: "job";
+    }): JobList;
+    list(options: ListJobsOptions & {
+        kind: "check";
+    }): CheckRowList;
+    list(options: ListJobsOptions & {
+        kind: "all";
+    }): JobListItemList;
+    list(options: ListJobsOptions): JobListItemList;
     /**
      * List a job's trials (cursor-paged; { status } filters, e.g. to
      * the failed trials). Await the result for one page, or `for await` it to
@@ -4466,6 +4569,8 @@ interface AnalysesClient {
      * feed's species-blind events door, which answers `trial_not_found`.
      */
     list(options?: ListAnalysesOptions): AnalysisList;
+    /** The defaults an analysis runs under when its config names nothing (GET /api/analyses/defaults): model, effort, provider, rubric and the unrendered prompt template. */
+    defaults(): Promise<AnalyzeDefaults>;
     /**
      * The verdict document — the wire's TrialAnalysis, statuses and typed
      * failure included, for EVERY analysis (not only completed ones). The same
@@ -4697,6 +4802,21 @@ interface Check {
     /** When the last task settled; null until every task has. */
     finished_at: string | null;
 }
+/**
+ * The policy an empty analyze config resolves to (GET /api/analyses/defaults):
+ * each key the value `AnalyzeConfig` echoes for a job created with
+ * `analyze: {}`, except `prompt`, which `AnalyzeConfig` serves as null and
+ * this serves as the template text.
+ */
+interface AnalyzeDefaults {
+    model_name: string;
+    rubric: Rubric;
+    /** The built-in analyze prompt template, unrendered — pass it as `prompt` to run the default body explicitly, or edit it from here. */
+    prompt: string;
+    /** The effort the default model runs at when the config names none. */
+    reasoning_effort: string;
+    sandbox_provider: EvalSandboxProvider;
+}
 /**
  * The policy an empty check config resolves to (GET /api/checks/defaults):
  * each key the value `Check` echoes for a check created with no config,
@@ -5288,4 +5408,4 @@ interface CapabilityDocument {
     error_codes: string[];
 }
 
-export { type AnalysisEvidence as $, type AgentInput as A, type AgentArm as B, type Check as C, type DatasetVersion as D, type AgentArmInput as E, type AgentCapability as F, type GatewayUsageEvent as G, type HostedErrorCode as H, type AgentDatasetStats as I, type JobCreate as J, type AgentEffortSupport as K, type AgentInfo as L, type AgentList as M, type AgentModelOption as N, type OrgsClient as O, type PublishDatasetInput as P, type AgentPage as Q, type Rubric as R, type SkillsClient as S, type Trial as T, type UsageReading as U, type AgentResult as V, type AgentSource as W, type AgentSourceInput as X, type AgentUpsertInput as Y, type AnalysisArtifactStream as Z, type AnalysisCheck as _, type TrialAnalysis as a, type FilesystemStreamEvent as a$, type AnalysisFailure as a0, type AnalysisLabel as a1, type AnalysisList as a2, type AnalysisPage as a3, type AnalysisStatus as a4, type AnalysisTranscript as a5, type AnalysisTranscriptOptions as a6, type AnalyzeConfig as a7, type AnalyzeConfigInput as a8, type ApiKey as a9, type DatasetVersionHubSource as aA, type DatasetVersionSource as aB, type DatasetVersionState as aC, type DownloadDatasetOptions as aD, type DownloadJobOptions as aE, EVAL_SANDBOX_PROVIDERS as aF, type Agent as aG, type ModelInfo as aH, type EvalSandboxProvider as aI, type StepResult as aJ, type ExceptionInfo as aK, type FilesystemArchiveOptions as aL, type FilesystemBox as aM, type FilesystemCapture as aN, type FilesystemChange as aO, type FilesystemChanges as aP, type FilesystemChangesOptions as aQ, type FilesystemEntry as aR, type FilesystemListOptions as aS, type FilesystemListing as aT, type FilesystemReadOptions as aU, type FilesystemSearchHit as aV, type FilesystemSearchOptions as aW, type FilesystemSearchResult as aX, type FilesystemSource as aY, type FilesystemState as aZ, type FilesystemStatus as a_, type AttemptPhase as aa, type AuthStatus as ab, type Awaitable as ac, CHECK_STATUSES as ad, type CheckDefaults as ae, type CheckLabel as af, type CompareCell as ag, type CompareCoverage as ah, type CompareJobAggregate as ai, type CompareResponse as aj, type CompareTaskRow as ak, type Dataset as al, type DatasetImportFailure as am, type DatasetImportList as an, type DatasetImportPage as ao, type DatasetImportStatus as ap, type DatasetList as aq, type DatasetPage as ar, type DatasetPatch as as, type DatasetPreflight as at, type DatasetRef as au, type DatasetSelector as av, type DatasetSource as aw, type DatasetVersionArchiveSource as ax, type DatasetVersionArchiveUrlSource as ay, type DatasetVersionGitSource as az, type DatasetImport as b, type PreflightDatasetInput as b$, type FilesystemStreamOptions as b0, type FilesystemWatchResult as b1, GATEWAY_TRACE_SEQ_BASE as b2, type GatewayUsage as b3, type GetDatasetOptions as b4, type GrepJobOptions as b5, HOSTED_ERROR_CODES as b6, type ImportPhase as b7, type ImportPhaseProgress as b8, type ImportWarning as b9, type JobTaskRollupPage as bA, type JobWatch as bB, type JudgeResult as bC, type ListAgentsOptions as bD, type ListAnalysesOptions as bE, type ListDatasetsOptions as bF, type ListImportsOptions as bG, type ListJobImportsOptions as bH, type ListJobTasksOptions as bI, type ListJobsOptions as bJ, type ListSkillsOptions as bK, type ListTrialFilesOptions as bL, type ListTrialsOptions as bM, type ManagedProviderCapability as bN, type OrgInvite as bO, type OrgInviteCreated as bP, type OrgJoined as bQ, type OrgMember as bR, type OrgQuota as bS, type OrgRole as bT, type OrgUsage as bU, type Organization as bV, type OrganizationDetail as bW, type Page as bX, type PageOptions as bY, type PassAtKGroup as bZ, type PassAtKPoint as b_, type InfraFailureSignature as ba, JOB_LIST_SCOPES as bb, type JobAnalysisStats as bc, type JobDeleteResult as bd, type JobFailure as be, type JobGrepGroup as bf, type JobGrepPage as bg, type JobImport as bh, type JobImportFailure as bi, type JobImportList as bj, type JobImportPage as bk, type JobImportPhaseName as bl, type JobImportPhaseProgress as bm, type JobImportProgress as bn, type JobImportSkippedTrial as bo, type JobImportSource as bp, type JobList as bq, type JobListScope as br, type JobPage as bs, type JobShareEmail as bt, type JobShareLink as bu, type JobStats as bv, type JobStatus as bw, type JobTaskLink as bx, type JobTaskRollup as by, type JobTaskRollupList as bz, type JobEvent as c, gatewayUsageOf as c$, type PreflightDeferredCheck as c0, type PreflightManifestVerdict as c1, type PreflightTaskVerdict as c2, type ProviderCapability as c3, type PublishDatasetOptions as c4, type RegradeRequest as c5, type ResumeRequest as c6, type RetryRequest as c7, type RubricCriterion as c8, type RunFilesystem as c9, type TaskPackageFilesystemStatus as cA, type TaskProviderVerdict as cB, type TimingInfo as cC, type TraceEventPage as cD, type TraceOptions as cE, type TrialArtifactStream as cF, type TrialCounts as cG, type TrialFile as cH, type TrialFilePage as cI, type TrialFileRange as cJ, type TrialGpuCost as cK, type TrialList as cL, type TrialPage as cM, type TrialStatus as cN, type TrialStatusTally as cO, type TrialTaskLink as cP, type TrialUploadProvenance as cQ, type UploadJobOptions as cR, type UploadProvenance as cS, type UploadSkillOptions as cT, type UpstreamStatus as cU, type VerifierEnvironmentMode as cV, type VerifierResult as cW, type WatchAnalysisOptions as cX, type WatchImportOptions as cY, type WatchJobImportOptions as cZ, type WatchJobOptions as c_, SANDBOX_LOG_STREAMS as ca, type SandboxLogEvent as cb, type SandboxLogLine as cc, type SandboxLogLines as cd, type SandboxLogOptions as ce, type SandboxLogStream as cf, type SandboxProcs as cg, type SkillLock as ch, type SkillUpload as ci, type SkillUploadList as cj, type SkillUploadPage as ck, type SourceJob as cl, type SpendSource as cm, type StartJobOptions as cn, type StatusVocabulary as co, type StopResponse as cp, TASK_LINKED_BY as cq, TASK_LINK_REASONS as cr, TRIAL_ARTIFACT_STREAMS as cs, TRIAL_STATUSES as ct, type Task as cu, type TaskCheckTranscript as cv, type TaskLinkReason as cw, type TaskLinkedBy as cx, type TaskNote as cy, type TaskPackageFiles as cz, type DatasetFailedTask as d, isHostedErrorCode as d0, passAtK as d1, type DatasetImportProgress as e, type JobSecretInline as f, type JobSecretRef as g, type TraceEvent as h, type JobVisibility as i, type JobShareRequest as j, type JobShares as k, type Job as l, type TaskCheck as m, type DatasetsClient as n, type AgentsClient as o, type JobsClient as p, type TrialsClient as q, type AnalysesClient as r, type ChecksClient as s, type CapabilityDocument as t, type HostedClientConfig as u, type AuthClient as v, AGENT_EFFORT_SUPPORT_VALUES as w, ANALYSIS_ARTIFACT_STREAMS as x, ANALYSIS_STATUSES as y, type ActiveDataset as z };
+export { type AnalysisEvidence as $, type AgentInput as A, type AgentArm as B, type Check as C, type DatasetVersion as D, type AgentArmInput as E, type AgentCapability as F, type GatewayUsageEvent as G, type HostedErrorCode as H, type AgentDatasetStats as I, type JobCreate as J, type AgentEffortSupport as K, type AgentInfo as L, type AgentList as M, type AgentModelOption as N, type OrgsClient as O, type PublishDatasetInput as P, type AgentPage as Q, type Rubric as R, type SkillsClient as S, type Trial as T, type UsageReading as U, type AgentResult as V, type AgentSource as W, type AgentSourceInput as X, type AgentUpsertInput as Y, type AnalysisArtifactStream as Z, type AnalysisCheck as _, type TrialAnalysis as a, type FilesystemSearchOptions as a$, type AnalysisFailure as a0, type AnalysisLabel as a1, type AnalysisList as a2, type AnalysisPage as a3, type AnalysisStatus as a4, type AnalysisTranscript as a5, type AnalysisTranscriptOptions as a6, type AnalyzeConfig as a7, type AnalyzeConfigInput as a8, type AnalyzeDefaults as a9, type DatasetSelector as aA, type DatasetSource as aB, type DatasetVersionArchiveSource as aC, type DatasetVersionArchiveUrlSource as aD, type DatasetVersionGitSource as aE, type DatasetVersionHubSource as aF, type DatasetVersionSource as aG, type DatasetVersionState as aH, type DownloadDatasetOptions as aI, type DownloadJobOptions as aJ, EVAL_SANDBOX_PROVIDERS as aK, type Agent as aL, type ModelInfo as aM, type EvalSandboxProvider as aN, type StepResult as aO, type ExceptionInfo as aP, type FilesystemArchiveOptions as aQ, type FilesystemBox as aR, type FilesystemCapture as aS, type FilesystemChange as aT, type FilesystemChanges as aU, type FilesystemChangesOptions as aV, type FilesystemEntry as aW, type FilesystemListOptions as aX, type FilesystemListing as aY, type FilesystemReadOptions as aZ, type FilesystemSearchHit as a_, type ApiKey as aa, type AttemptPhase as ab, type AuthStatus as ac, type Awaitable as ad, CHECK_STATUSES as ae, type CheckDefaults as af, type CheckLabel as ag, type CheckRow as ah, type CheckRowList as ai, type CheckRowPage as aj, type CheckTaskTally as ak, type CompareCell as al, type CompareCoverage as am, type CompareJobAggregate as an, type CompareResponse as ao, type CompareTaskRow as ap, type Dataset as aq, type DatasetImportFailure as ar, type DatasetImportList as as, type DatasetImportPage as at, type DatasetImportStatus as au, type DatasetList as av, type DatasetPage as aw, type DatasetPatch as ax, type DatasetPreflight as ay, type DatasetRef as az, type DatasetImport as b, type OrgJoined as b$, type FilesystemSearchResult as b0, type FilesystemSource as b1, type FilesystemState as b2, type FilesystemStatus as b3, type FilesystemStreamEvent as b4, type FilesystemStreamOptions as b5, type FilesystemWatchResult as b6, GATEWAY_TRACE_SEQ_BASE as b7, type GatewayUsage as b8, type GetDatasetOptions as b9, type JobListKind as bA, type JobListScope as bB, type JobPage as bC, type JobShareEmail as bD, type JobShareLink as bE, type JobStats as bF, type JobStatus as bG, type JobTaskLink as bH, type JobTaskRollup as bI, type JobTaskRollupList as bJ, type JobTaskRollupPage as bK, type JobViewer as bL, type JobWatch as bM, type JudgeResult as bN, type ListAgentsOptions as bO, type ListAnalysesOptions as bP, type ListDatasetsOptions as bQ, type ListImportsOptions as bR, type ListJobImportsOptions as bS, type ListJobTasksOptions as bT, type ListJobsOptions as bU, type ListSkillsOptions as bV, type ListTrialFilesOptions as bW, type ListTrialsOptions as bX, type ManagedProviderCapability as bY, type OrgInvite as bZ, type OrgInviteCreated as b_, type GrepJobOptions as ba, HOSTED_ERROR_CODES as bb, type ImportPhase as bc, type ImportPhaseProgress as bd, type ImportWarning as be, type InfraFailureSignature as bf, JOB_LIST_KINDS as bg, JOB_LIST_SCOPES as bh, type JobAnalysisStats as bi, type JobDeleteResult as bj, type JobFailure as bk, type JobGrepGroup as bl, type JobGrepPage as bm, type JobImport as bn, type JobImportFailure as bo, type JobImportList as bp, type JobImportPage as bq, type JobImportPhaseName as br, type JobImportPhaseProgress as bs, type JobImportProgress as bt, type JobImportSkippedTrial as bu, type JobImportSource as bv, type JobList as bw, type JobListItem as bx, type JobListItemList as by, type JobListItemPage as bz, type JobEvent as c, type TrialUploadProvenance as c$, type OrgMember as c0, type OrgQuota as c1, type OrgRole as c2, type OrgUsage as c3, type Organization as c4, type OrganizationDetail as c5, type Page as c6, type PageOptions as c7, type PassAtKGroup as c8, type PassAtKPoint as c9, type StopResponse as cA, TASK_LINKED_BY as cB, TASK_LINK_REASONS as cC, TRIAL_ARTIFACT_STREAMS as cD, TRIAL_STATUSES as cE, type Task as cF, type TaskCheckTranscript as cG, type TaskLinkReason as cH, type TaskLinkedBy as cI, type TaskNote as cJ, type TaskPackageFiles as cK, type TaskPackageFilesystemStatus as cL, type TaskProviderVerdict as cM, type TimingInfo as cN, type TraceEventPage as cO, type TraceOptions as cP, type TrialArtifactStream as cQ, type TrialCounts as cR, type TrialFile as cS, type TrialFilePage as cT, type TrialFileRange as cU, type TrialGpuCost as cV, type TrialList as cW, type TrialPage as cX, type TrialStatus as cY, type TrialStatusTally as cZ, type TrialTaskLink as c_, type PreflightDatasetInput as ca, type PreflightDeferredCheck as cb, type PreflightManifestVerdict as cc, type PreflightTaskVerdict as cd, type ProviderCapability as ce, type PublishDatasetOptions as cf, type RegradeRequest as cg, type ResumeRequest as ch, type RetryRequest as ci, type RubricCriterion as cj, type RunFilesystem as ck, SANDBOX_LOG_STREAMS as cl, type SandboxLogEvent as cm, type SandboxLogLine as cn, type SandboxLogLines as co, type SandboxLogOptions as cp, type SandboxLogStream as cq, type SandboxProcs as cr, type SkillLock as cs, type SkillUpload as ct, type SkillUploadList as cu, type SkillUploadPage as cv, type SourceJob as cw, type SpendSource as cx, type StartJobOptions as cy, type StatusVocabulary as cz, type DatasetFailedTask as d, type UploadJobOptions as d0, type UploadProvenance as d1, type UploadSkillOptions as d2, type UpstreamStatus as d3, type VerifierEnvironmentMode as d4, type VerifierResult as d5, type WatchAnalysisOptions as d6, type WatchImportOptions as d7, type WatchJobImportOptions as d8, type WatchJobOptions as d9, gatewayUsageOf as da, isHostedErrorCode as db, passAtK as dc, type DatasetImportProgress as e, type JobSecretInline as f, type JobSecretRef as g, type TraceEvent as h, type JobVisibility as i, type JobShareRequest as j, type JobShares as k, type Job as l, type TaskCheck as m, type DatasetsClient as n, type AgentsClient as o, type JobsClient as p, type TrialsClient as q, type AnalysesClient as r, type ChecksClient as s, type CapabilityDocument as t, type HostedClientConfig as u, type AuthClient as v, AGENT_EFFORT_SUPPORT_VALUES as w, ANALYSIS_ARTIFACT_STREAMS as x, ANALYSIS_STATUSES as y, type ActiveDataset as z };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@evolvingmachines/evolve",
-  "version": "0.0.64",
+  "version": "0.0.65",
   "keywords": [
     "ai",
     "agents",
@@ -129,9 +129,9 @@
   },
   "dependencies": {
     "@agentclientprotocol/sdk": "^0.5.1",
-    "@evolvingmachines/daytona": "0.0.64",
-    "@evolvingmachines/e2b": "0.0.64",
-    "@evolvingmachines/modal": "0.0.64",
+    "@evolvingmachines/daytona": "0.0.65",
+    "@evolvingmachines/e2b": "0.0.65",
+    "@evolvingmachines/modal": "0.0.65",
     "ajv": "^8.17.1",
     "p-map": "^7.0.2",
     "smol-toml": "^1.7.0",

package/skills/evolve-evals/SKILL.md

@@ -57,7 +57,7 @@ The pages under `references/` are the documentation site's pages, byte for byte,
 | [Datasets](references/dashboard/datasets.mdx) | Browse the catalog, and each dataset's tasks and versions, in the dashboard. |
 | [Jobs](references/dashboard/jobs.mdx) | Start a job from the browser, follow it, and read its results. |
 | [Trial viewer](references/dashboard/trial-viewer.mdx) | Read one trial's trace event by event, its tool calls, its spend, and download its artifacts. |
-| [Checks](references/dashboard/checks.mdx) | Read a job's task quality checks in the dashboard. |
+| [Checks](references/dashboard/checks.mdx) | Read task quality checks in the dashboard, on a job's CHECK tab or on the check's own page. |
 | [Traces](references/dashboard/traces.mdx) | Every recorded run in one list. |
 
 ### SDK

package/skills/evolve-evals/references/cli-reference/analyze.mdx

@@ -7,6 +7,7 @@ import GlobalOptions from '/snippets/global-options.mdx';
 
 ```bash
 evolve analyze <job-id> [options]
+evolve analyze --show-defaults
 ```
 
 Analyze a terminal job's trial traces against a rubric, server-side. Without `--watch` the command returns at once with the job block and the hint `Follow it with: evolve job show <id>`; `--json` prints the accepted job body. Exit 0 on an accepted wave; with `--watch`, exit 1 when any analysis failed.
@@ -19,6 +20,7 @@ evolve analyze 3e1f9a2c-… \
   -l 20 \
   -n 2 \
   --watch
+evolve analyze 3e1f9a2c-… -t d1a10c4e-… -t d1a10f72-…
 ```
 
 ## Options
@@ -43,6 +45,10 @@ evolve analyze 3e1f9a2c-… \
   Sandbox provider the analyzer runs on. Defaults to `daytona`.
 </ParamField>
 
+<ParamField path="--show-defaults">
+  Print the defaults an analysis runs under when you name none: model, effort, provider, the rubric's criteria and the built-in prompt template, then exit. With `--json`, one object. Takes no job id and no other analyze flag.
+</ParamField>
+
 <ParamField path="-n, --n-concurrent" type="n">
   Max concurrent trial analyses, 1 to 150, beneath the organization's `max_concurrent_analyses` (fleet default 60). Defaults to the ceiling.
 </ParamField>
@@ -55,8 +61,12 @@ evolve analyze 3e1f9a2c-… \
   Only analyze failing trials (reward below 1.0, or an exception).
 </ParamField>
 
+<ParamField path="-t, --trial" type="trial-id">
+  Analyze only this trial; repeatable. Combines with `--passing` or `--failing` (a named trial on the other side is skipped) and comes before `-l`. An id that is not a trial of this job is refused, with the unknown ids named.
+</ParamField>
+
 <ParamField path="-l, --n-trials" type="n">
-  Max trials to analyze, after `--passing` or `--failing`, in the job's trial order.
+  Max trials to analyze, after `--passing`, `--failing` or `--trial`, in the job's trial order.
 </ParamField>
 
 <ParamField path="--watch">

package/skills/evolve-evals/references/cli-reference/job.mdx

@@ -26,7 +26,7 @@ Start a job; add `--watch` to follow it. Exactly the options of [`evolve run`](/
 evolve job list [options]
 ```
 
-List jobs, newest first: yours, `--scope shared` for what teammates created, `--scope org` for everything in your organizations.
+List jobs, newest first: yours, `--scope shared` for what teammates created, `--scope org` for everything in your organizations. A task quality check is a job too (Harbor runs `harbor check` as one job): `--kind all` lists jobs and checks in one list, each row marked by its kind, and `--kind check` lists checks alone.
 
 <ListFlags />
 
@@ -38,8 +38,13 @@ List jobs, newest first: yours, `--scope shared` for what teammates created, `--
   Visibility scope: `my`, what you created (the default); `shared`, your organizations' rows that teammates created; or `org`, every row in your organizations, your own included.
 </ParamField>
 
+<ParamField path="--kind" type="job | check | all">
+  Which rows to list: `job` (the default), `check` for task quality checks, or `all` for both merged newest first. With `check` or `all` a `KIND` column leads each row; a check row reads its id, status, dataset or archive, task check count, cost and accept time, and under `--columns` its `name` and its checker's model and effort (`agents`).
+</ParamField>
+
 ```bash
 evolve job list --limit 20 -q
+evolve job list --kind all
 ```
 
 ## job show

package/skills/evolve-evals/references/core-concepts/analyze.mdx

@@ -42,9 +42,9 @@ evolve analyze <job-id> \
   --watch
 ```
 
-`-r` is the rubric file, `-p` the prompt file. `--passing` and `--failing` restrict the batch of analyses to trials with reward 1.0, or below it; both together is refused. `-l` caps the number of trials, `-n` how many analyses run at once, beneath the organization's `max_concurrent_analyses` (fleet default 60). `-m` picks the analyzer's model, `--effort` its reasoning effort, `-e` its sandbox provider.
+`-r` is the rubric file, `-p` the prompt file. `--passing` and `--failing` restrict the batch of analyses to trials with reward 1.0, or below it; both together is refused. `-t` names one trial to analyze, repeatable, applied with the filter and before `-l`. `-l` caps the number of trials, `-n` how many analyses run at once, beneath the organization's `max_concurrent_analyses` (fleet default 60). `-m` picks the analyzer's model, `--effort` its reasoning effort, `-e` its sandbox provider. `--show-defaults` prints the defaults a bare `evolve analyze` runs under and exits.
 
-A result that misses or invents a criterion is a failed analysis, never a partial pass. A rubric with unknown keys, empty or duplicate criteria, or a length bound exceeded is refused with `invalid_rubric`; an off-roster model or an unknown effort or provider with `invalid_input`; a job with no analyzable trial, or a `--passing` or `--failing` filter that selects none, with `no_analyzable_trials`; a job still running with `job_not_terminal`.
+A result that misses or invents a criterion is a failed analysis, never a partial pass. A rubric with unknown keys, empty or duplicate criteria, or a length bound exceeded is refused with `invalid_rubric`; an off-roster model, an unknown effort or provider, or a `-t` id that is not a trial of the job with `invalid_input`; a job with no analyzable trial, or a `--passing`, `--failing` or `-t` selection that leaves none, with `no_analyzable_trials`; a job still running with `job_not_terminal`.
 
 On `evolve run` the same knobs are `--analyze-model`, `--analyze-rubric`, `--analyze-prompt`, `--analyze-effort` and `--analyze-provider`. Each of them implies `--analyze`.
 

package/skills/evolve-evals/references/core-concepts/sharing.mdx

@@ -28,7 +28,7 @@ In the browser a shared job opens on the run, its trials and each trial's trace,
 
 ## In the dashboard
 
-SHARE sits on a job's page, on a task check's trace opened from the job's CHECK tab, and on a session's trace under Traces. It opens the dialog: type an address and INVITE, or under General access turn on **Anyone with the link** and copy it. Turning it off stops the link at once. Only the run's creator sees SHARE. Runs shared with you by email are listed under SHARED WITH ME on Traces.
+SHARE sits on a job's page, on a check's page, on a task check's trace opened from either, and on a session's trace under Traces. It opens the dialog: type an address and INVITE, or under General access turn on **Anyone with the link** and copy it. Turning it off stops the link at once. Only the run's creator sees SHARE. Runs shared with you by email are listed under SHARED WITH ME on your personal workspace's Traces.
 
 <Card title="job reference" icon="terminal" href="/cli-reference/job">
   `job share`, `job unshare`, `job shares` and their flags.

package/skills/evolve-evals/references/core-concepts/teams.mdx

@@ -25,11 +25,11 @@ In the SDK the same choice is `org: "acme"` on `jobs().start()` and `datasets().
 
 ## In the dashboard
 
-The box under your account in the sidebar is the team switcher. On Jobs, Datasets and Settings it names the team you are in; on Traces and Skills it names none. Its menu lists your teams and Personal, **Manage members** while a team is open, and **Create team**. Pick a team and the pages under it show that team's work.
+The box under your account in the sidebar is the workspace switcher. It always names one workspace: the one in the page's URL, otherwise the one you last opened in this tab, otherwise Personal. Its menu lists Personal and every team you belong to, each with your role, and **Create team**. Pick a workspace and Datasets, Jobs, Traces, Members and Settings show that workspace's work; on a page that is yours alone (Skills, Integrations, Secrets, API Keys, Usage, Billing) the pick is remembered for the next one.
 
-Team settings has three tabs. GENERAL holds the name, and DELETE TEAM for an owner. MEMBERS lists people with their role: an owner has MAKE OWNER, DEMOTE and REMOVE on every other row, and everyone has LEAVE on their own. INVITES has NEW INVITE LINK, shown once, and REVOKE. A person who opens an invite link signs in, or signs up, and joins.
+TEAM in the sidebar has two pages. Members lists people with their role: an owner has MAKE OWNER, DEMOTE and REMOVE on every other row, and everyone has LEAVE on their own; below the list, an owner has NEW INVITE LINK, shown once, and REVOKE. A person who opens an invite link signs in, or signs up, and joins. Settings holds the team's name, and DELETE TEAM for an owner. In Personal, Members lists only you and Settings shows your workspace's name.
 
-Traces and Skills are not under a team. Their tabs are MINE (yours, in any team), TEAM (everything in every team you belong to, yours included) and SHARED WITH ME (what teammates made, plus runs shared with you by email).
+Traces is under the workspace too. In a team it is one list of every run in that team, with a CREATED BY column naming who ran each. In Personal it has two tabs: MINE (yours, in any team) and SHARED WITH ME (what teammates made, plus runs shared with you by email). Skills lists your own uploads.
 
 <Card title="auth reference" icon="terminal" href="/cli-reference/auth">
   Every `evolve auth org` command.