agent-inspect 5.2.0 → 5.4.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +12 -0
- package/docs/CLI.md +28 -1
- package/package.json +1 -1
- package/packages/cli/dist/{chunk-SJ5R2XBE.mjs → chunk-YS2I722C.mjs} +112 -3
- package/packages/cli/dist/chunk-YS2I722C.mjs.map +1 -0
- package/packages/cli/dist/index.cjs +2487 -2163
- package/packages/cli/dist/index.cjs.map +1 -1
- package/packages/cli/dist/index.mjs +2313 -2112
- package/packages/cli/dist/index.mjs.map +1 -1
- package/packages/cli/dist/{src-PK22BBMH.mjs → src-FVNE44UA.mjs} +3 -3
- package/packages/cli/dist/{src-PK22BBMH.mjs.map → src-FVNE44UA.mjs.map} +1 -1
- package/packages/core/dist/advanced.cjs +118 -0
- package/packages/core/dist/advanced.cjs.map +1 -1
- package/packages/core/dist/advanced.d.cts +14 -4
- package/packages/core/dist/advanced.d.ts +14 -4
- package/packages/core/dist/advanced.mjs +114 -1
- package/packages/core/dist/advanced.mjs.map +1 -1
- package/packages/core/dist/checks.d.cts +1 -1
- package/packages/core/dist/checks.d.ts +1 -1
- package/packages/core/dist/{context-Cp8NRbBX.d.ts → context-BJ5UQLDi.d.ts} +1 -1
- package/packages/core/dist/{context-BbfkMYjz.d.cts → context-DyKbkmd4.d.cts} +1 -1
- package/packages/core/dist/index.d.cts +2 -2
- package/packages/core/dist/index.d.ts +2 -2
- package/packages/core/dist/{types-CvsKGweN.d.cts → types-J3Xtuf9x.d.cts} +13 -1
- package/packages/core/dist/{types-CvsKGweN.d.ts → types-J3Xtuf9x.d.ts} +13 -1
- package/packages/cli/dist/chunk-SJ5R2XBE.mjs.map +0 -1
|
@@ -1,13 +1,14 @@
|
|
|
1
|
-
export { C as CreateInspectorOptions, D as DEFAULT_MAX_EVENT_BYTES, f as DEFAULT_MAX_METADATA_VALUE_LENGTH, h as DEFAULT_MAX_PREVIEW_LENGTH, I as Inspector, a as InspectorCaptureOptions, b as InspectorObserveOptions, c as InspectorRunOptions, i as InspectorRuntime, j as InspectorRuntimeContext, k as InspectorRuntimeDiagnostics, l as InspectorRuntimeOptions, d as InspectorStepOptions, T as TraceSafetyOptions, e as createInspector, n as createInspectorRuntime, o as getCurrentContext, g as getCurrentCorrelationMetadata, p as getCurrentDepth, q as getCurrentRunId, r as getCurrentRunName, s as getCurrentStepId, t as getParentStepId, u as getTraceDirFromContext, v as getTraceSafetyFromContext, w as hasActiveContext, x as isAgentInspectEnabled, y as isSilentContext, m as maybeInspectRun, z as prepareMetadataForDisk, A as prepareTraceEventForDisk, B as resolveTraceSafetyOptions, E as runWithContext, F as runWithStepContext } from './context-
|
|
2
|
-
import { R as RedactionProfile, E as ErrorInfo, b as RunStatus, e as StepStatus,
|
|
1
|
+
export { C as CreateInspectorOptions, D as DEFAULT_MAX_EVENT_BYTES, f as DEFAULT_MAX_METADATA_VALUE_LENGTH, h as DEFAULT_MAX_PREVIEW_LENGTH, I as Inspector, a as InspectorCaptureOptions, b as InspectorObserveOptions, c as InspectorRunOptions, i as InspectorRuntime, j as InspectorRuntimeContext, k as InspectorRuntimeDiagnostics, l as InspectorRuntimeOptions, d as InspectorStepOptions, T as TraceSafetyOptions, e as createInspector, n as createInspectorRuntime, o as getCurrentContext, g as getCurrentCorrelationMetadata, p as getCurrentDepth, q as getCurrentRunId, r as getCurrentRunName, s as getCurrentStepId, t as getParentStepId, u as getTraceDirFromContext, v as getTraceSafetyFromContext, w as hasActiveContext, x as isAgentInspectEnabled, y as isSilentContext, m as maybeInspectRun, z as prepareMetadataForDisk, A as prepareTraceEventForDisk, B as resolveTraceSafetyOptions, E as runWithContext, F as runWithStepContext } from './context-DyKbkmd4.cjs';
|
|
2
|
+
import { R as RedactionProfile, h as TraceEvent, E as ErrorInfo, b as RunStatus, e as StepStatus, j as RunSummary, k as TraceMetadata, l as TraceMetadataStatus, f as StepType, g as TraceCorrelationMetadata } from './types-B1X87dUW.cjs';
|
|
3
3
|
export { A as ActiveStepContext, i as ExecutionContext, I as InspectRunOptions, O as ObserveOptions, a as Run, m as RunCompletedEvent, n as RunStartedEvent, c as Step, o as StepCompletedEvent, d as StepMetadata, S as StepOptions, p as StepStartedEvent, T as TokenMetadata, q as TraceEventBase, r as TraceSchemaVersion, s as isStepStatus, t as isStepType, u as isTraceEvent } from './types-B1X87dUW.cjs';
|
|
4
4
|
import { b as InspectRunTree } from './inspect-event-IqwOHbGo.cjs';
|
|
5
5
|
export { A as AttributionConfidence, E as EventSource, a as InspectEvent, I as InspectKind, c as InspectNode } from './inspect-event-IqwOHbGo.cjs';
|
|
6
6
|
import { P as PersistedInspectEvent } from './persisted-inspect-event-Gd-W8vB7.cjs';
|
|
7
7
|
import { Stats } from 'node:fs';
|
|
8
|
+
import { a as ObservedOutcome } from './types-J3Xtuf9x.cjs';
|
|
9
|
+
export { b as ObservedOutcomeStatus } from './types-J3Xtuf9x.cjs';
|
|
8
10
|
import { TraceCheckResult, TraceCheckStatus } from './checks.cjs';
|
|
9
11
|
import './writers.cjs';
|
|
10
|
-
import './types-CvsKGweN.cjs';
|
|
11
12
|
import './log-config-CW8w5YQ9.cjs';
|
|
12
13
|
import './readers.cjs';
|
|
13
14
|
|
|
@@ -20,6 +21,9 @@ interface ResolvedRedactionProfile {
|
|
|
20
21
|
/** Resolves named profile behavior (keys + metadata string caps). */
|
|
21
22
|
declare function resolveRedactionProfile(profile?: RedactionProfile): ResolvedRedactionProfile;
|
|
22
23
|
|
|
24
|
+
declare function extractOutcomesFromTraceEvents(events: readonly TraceEvent[]): ObservedOutcome[];
|
|
25
|
+
declare function extractOutcomesFromPersistedEvents(events: readonly PersistedInspectEvent[]): ObservedOutcome[];
|
|
26
|
+
|
|
23
27
|
/** Two spaces per nesting level in terminal output. */
|
|
24
28
|
declare const TERMINAL_INDENT = " ";
|
|
25
29
|
/** Max display length for names in terminal output. */
|
|
@@ -873,6 +877,12 @@ declare function runSuite(options?: RunSuiteOptions): Promise<SuiteRunResult>;
|
|
|
873
877
|
declare function renderSuiteReportMarkdown(result: SuiteRunResult): string;
|
|
874
878
|
declare function renderSuiteReport(result: SuiteRunResult, options?: RenderSuiteReportOptions): string;
|
|
875
879
|
|
|
880
|
+
declare const SUITE_TEMPLATE_IDS: readonly ["customer-support-agent", "refund-agent", "sales-assistant", "browser-task-agent", "mcp-tool-agent", "workflow-agent", "rag-answer-agent", "human-approval-agent"];
|
|
881
|
+
type SuiteTemplateId = (typeof SUITE_TEMPLATE_IDS)[number];
|
|
882
|
+
declare function listSuiteTemplates(): SuiteTemplateId[];
|
|
883
|
+
declare function getSuiteTemplate(id: string): SuiteConfig | undefined;
|
|
884
|
+
declare function resolveSuiteTemplate(id: string): SuiteConfig;
|
|
885
|
+
|
|
876
886
|
type CohortMetricId = "errorRate" | "duration" | "toolChoice" | "toolOrdering" | "llmCallCount" | "tokenUsage" | "retryCount" | "observationFailure" | "guardrailFailure" | "circuitViolation" | "redactionWarning";
|
|
877
887
|
declare const COHORT_METRIC_IDS: readonly CohortMetricId[];
|
|
878
888
|
interface CohortRunMetrics {
|
|
@@ -1006,4 +1016,4 @@ declare function renderGateGithubStepSummary(result: GateResult): string;
|
|
|
1006
1016
|
declare function renderGateJUnit(result: GateResult): string;
|
|
1007
1017
|
declare function renderGateReport(result: GateResult, options?: RenderGateReportOptions): string;
|
|
1008
1018
|
|
|
1009
|
-
export { type ActivityEntry, type ActivitySummary, type AnalyzeCohortOptions, type BuildActivitySummaryOptions, type BuildSessionIndexOptions, type BundleCheckResults, type BundleCheckRunResult, type BundleMetadata, type BundlePlaceholderArtifact, type BundleRedactionProfile, type BundleRedactionReport, type BundleRedactionReportRun, type BundleResolveOptions, type BundleResolveResult, type BundleSafeStatus, type BundleSafeStatusMetadata, COHORT_METRIC_IDS, type CohortAggregateMetrics, type CohortAnalysisResult, type CohortMetricComparison, type CohortMetricId, type CohortRunMetrics, type CriticalPathStep, DEFAULT_SUITE_ARTIFACTS_DIR, DEFAULT_SUITE_CONFIG_NAMES, DEFAULT_TRACE_DIR_NAME, type DurationStats, type EnrichSessionSummaryOptions, ErrorInfo, type ExplainFact, type ExplainInference, type ExplainMode, type ExplainOptions, type ExplainResult, FALLBACK_TRACE_DIR, type GateCheckId, type GateCheckResult, type GateExitCode, type GateResult, type GroupSessionCohortsOptions, type HandoffEdge, InspectRunTree, type LoadSuiteConfigOptions, MAX_NAME_LENGTH, MAX_TERMINAL_DEPTH, MAX_TERMINAL_NAME_LENGTH, type ParseTraceJsonlOptions, type ParseTraceJsonlResult, type ParsedDurationFilter, RUNS_DIR_NAME, RedactionProfile, type RenderCohortReportOptions, type RenderGateReportOptions, type RenderSuiteReportOptions, type RenderTimelineOptions, type RenderWhatOptions, type ResolvedRedactionProfile, type ResolvedSuiteCase, type RetryLink, type RunGateOptions, RunStatus, type RunSuiteOptions, RunSummary, type RunTimeline, type RunWhatSummary, SESSION_WORKFLOW_KEYS, type SessionCheckSummary, type SessionCohort, type SessionCohortKind, type SessionConfidence, type SessionEdgeSource, type SessionGroup, type SessionIndex, type SessionLastError, type SessionRunRecord, type SessionScopeOptions, type SessionScopeResult, type SessionStatus, type SessionSummary, type SessionWarning, type SessionWorkflowKey, type SessionWorkflowMetadata, StepStatus, StepType, type SuiteArtifactsConfig, type SuiteCaseConfig, type SuiteCaseResult, type SuiteCaseStatus, type SuiteChecksConfig, type SuiteConfig, type SuiteDiagnostic, type SuiteDiagnosticCode, type SuiteEvalConfig, type SuiteRunResult, type SuiteRunSummary, TERMINAL_INDENT, type TimelineEntry, type TimelineFocus, type TimelineOptions, TraceCorrelationMetadata, TraceDirectory, type TraceDirectoryOptions, TraceEvent, type TraceFilterOptions, type TraceJsonlFormat, TraceMetadata, TraceMetadataStatus, type TraceSearchOptions, type TraceSearchResult, type TraceSessionCheckResult, type TraceStats, type TraceStatsOptions, type TraceStatsRankedRun, type TraceStatsRankedStep, type ValidateSuiteConfigResult, aggregateBundleSafeStatus, aggregateSessionCheckResults, analyzeCohort, buildActivitySummary, buildBundleMetadata, buildBundleSummaryMarkdown, buildLocalExplanation, buildPlaceholderArtifact, buildRunSummary, buildRunTimeline, buildRunWhatSummary, buildSessionIndex, buildTraceStats, bundleFailsOnSafety, compareCohortAggregates, createRunId, createStepId, defaultBundleOutputPath, defaultSuiteConfigTemplate, deriveSessionStatus, enrichSessionRunRecord, enrichSessionSummary, ensureTraceDir, extractMetadata, extractSessionWorkflowMetadata, filterMetasBySessionScope, filterTraces, formatDuration, formatError, formatTerminalName, formatTimestamp, gateHasThresholds, getDefaultTraceDir, getIndent, getRunIdFromTraceFileName, getTraceFilePath, groupSessionCohorts, initializeTraceFile, isAgentInspectTrace, listTraceFiles, loadSessionRunRecords, loadSuiteConfig, loadTraceMetadataList, normalizeBundleOutputPath, normalizeSuiteConfig, parseCohortMetricList, parseDuration, parseDurationFilter, parseGateList, parseGateNumber, parseGroupBySpec, parseTraceJsonl, printError, printFailedAt, printRunComplete, printRunStart, printStepComplete, printStepStart, readTraceEvents, readTraceFile, renderActivitySummaryHuman, renderCohortReport, renderCohortSummaryMarkdown, renderErrorLine, renderGateGithubStepSummary, renderGateJUnit, renderGateReport, renderGateSummaryMarkdown, renderRunSummary, renderRunWhat, renderStepLine, renderSuiteReport, renderSuiteReportMarkdown, renderTimeline, renderTraceStats, resolveBundleRunIds, resolveRedactionProfile, resolveSuiteCaseTrace, resolveSuiteConfigPath, resolveTraceDir, runGate, runSuite, searchTraces, serializeEvent, sessionKeyForRun, toMetadataSafeStatus, traceMetasToSessionRunRecords, truncateName, unknownTraceFormatMessage, validateEvent, validateSuiteConfig, warn, writeTraceEvent };
|
|
1019
|
+
export { type ActivityEntry, type ActivitySummary, type AnalyzeCohortOptions, type BuildActivitySummaryOptions, type BuildSessionIndexOptions, type BundleCheckResults, type BundleCheckRunResult, type BundleMetadata, type BundlePlaceholderArtifact, type BundleRedactionProfile, type BundleRedactionReport, type BundleRedactionReportRun, type BundleResolveOptions, type BundleResolveResult, type BundleSafeStatus, type BundleSafeStatusMetadata, COHORT_METRIC_IDS, type CohortAggregateMetrics, type CohortAnalysisResult, type CohortMetricComparison, type CohortMetricId, type CohortRunMetrics, type CriticalPathStep, DEFAULT_SUITE_ARTIFACTS_DIR, DEFAULT_SUITE_CONFIG_NAMES, DEFAULT_TRACE_DIR_NAME, type DurationStats, type EnrichSessionSummaryOptions, ErrorInfo, type ExplainFact, type ExplainInference, type ExplainMode, type ExplainOptions, type ExplainResult, FALLBACK_TRACE_DIR, type GateCheckId, type GateCheckResult, type GateExitCode, type GateResult, type GroupSessionCohortsOptions, type HandoffEdge, InspectRunTree, type LoadSuiteConfigOptions, MAX_NAME_LENGTH, MAX_TERMINAL_DEPTH, MAX_TERMINAL_NAME_LENGTH, ObservedOutcome, type ParseTraceJsonlOptions, type ParseTraceJsonlResult, type ParsedDurationFilter, RUNS_DIR_NAME, RedactionProfile, type RenderCohortReportOptions, type RenderGateReportOptions, type RenderSuiteReportOptions, type RenderTimelineOptions, type RenderWhatOptions, type ResolvedRedactionProfile, type ResolvedSuiteCase, type RetryLink, type RunGateOptions, RunStatus, type RunSuiteOptions, RunSummary, type RunTimeline, type RunWhatSummary, SESSION_WORKFLOW_KEYS, SUITE_TEMPLATE_IDS, type SessionCheckSummary, type SessionCohort, type SessionCohortKind, type SessionConfidence, type SessionEdgeSource, type SessionGroup, type SessionIndex, type SessionLastError, type SessionRunRecord, type SessionScopeOptions, type SessionScopeResult, type SessionStatus, type SessionSummary, type SessionWarning, type SessionWorkflowKey, type SessionWorkflowMetadata, StepStatus, StepType, type SuiteArtifactsConfig, type SuiteCaseConfig, type SuiteCaseResult, type SuiteCaseStatus, type SuiteChecksConfig, type SuiteConfig, type SuiteDiagnostic, type SuiteDiagnosticCode, type SuiteEvalConfig, type SuiteRunResult, type SuiteRunSummary, type SuiteTemplateId, TERMINAL_INDENT, type TimelineEntry, type TimelineFocus, type TimelineOptions, TraceCorrelationMetadata, TraceDirectory, type TraceDirectoryOptions, TraceEvent, type TraceFilterOptions, type TraceJsonlFormat, TraceMetadata, TraceMetadataStatus, type TraceSearchOptions, type TraceSearchResult, type TraceSessionCheckResult, type TraceStats, type TraceStatsOptions, type TraceStatsRankedRun, type TraceStatsRankedStep, type ValidateSuiteConfigResult, aggregateBundleSafeStatus, aggregateSessionCheckResults, analyzeCohort, buildActivitySummary, buildBundleMetadata, buildBundleSummaryMarkdown, buildLocalExplanation, buildPlaceholderArtifact, buildRunSummary, buildRunTimeline, buildRunWhatSummary, buildSessionIndex, buildTraceStats, bundleFailsOnSafety, compareCohortAggregates, createRunId, createStepId, defaultBundleOutputPath, defaultSuiteConfigTemplate, deriveSessionStatus, enrichSessionRunRecord, enrichSessionSummary, ensureTraceDir, extractMetadata, extractOutcomesFromPersistedEvents, extractOutcomesFromTraceEvents, extractSessionWorkflowMetadata, filterMetasBySessionScope, filterTraces, formatDuration, formatError, formatTerminalName, formatTimestamp, gateHasThresholds, getDefaultTraceDir, getIndent, getRunIdFromTraceFileName, getSuiteTemplate, getTraceFilePath, groupSessionCohorts, initializeTraceFile, isAgentInspectTrace, listSuiteTemplates, listTraceFiles, loadSessionRunRecords, loadSuiteConfig, loadTraceMetadataList, normalizeBundleOutputPath, normalizeSuiteConfig, parseCohortMetricList, parseDuration, parseDurationFilter, parseGateList, parseGateNumber, parseGroupBySpec, parseTraceJsonl, printError, printFailedAt, printRunComplete, printRunStart, printStepComplete, printStepStart, readTraceEvents, readTraceFile, renderActivitySummaryHuman, renderCohortReport, renderCohortSummaryMarkdown, renderErrorLine, renderGateGithubStepSummary, renderGateJUnit, renderGateReport, renderGateSummaryMarkdown, renderRunSummary, renderRunWhat, renderStepLine, renderSuiteReport, renderSuiteReportMarkdown, renderTimeline, renderTraceStats, resolveBundleRunIds, resolveRedactionProfile, resolveSuiteCaseTrace, resolveSuiteConfigPath, resolveSuiteTemplate, resolveTraceDir, runGate, runSuite, searchTraces, serializeEvent, sessionKeyForRun, toMetadataSafeStatus, traceMetasToSessionRunRecords, truncateName, unknownTraceFormatMessage, validateEvent, validateSuiteConfig, warn, writeTraceEvent };
|
|
@@ -1,13 +1,14 @@
|
|
|
1
|
-
export { C as CreateInspectorOptions, D as DEFAULT_MAX_EVENT_BYTES, f as DEFAULT_MAX_METADATA_VALUE_LENGTH, h as DEFAULT_MAX_PREVIEW_LENGTH, I as Inspector, a as InspectorCaptureOptions, b as InspectorObserveOptions, c as InspectorRunOptions, i as InspectorRuntime, j as InspectorRuntimeContext, k as InspectorRuntimeDiagnostics, l as InspectorRuntimeOptions, d as InspectorStepOptions, T as TraceSafetyOptions, e as createInspector, n as createInspectorRuntime, o as getCurrentContext, g as getCurrentCorrelationMetadata, p as getCurrentDepth, q as getCurrentRunId, r as getCurrentRunName, s as getCurrentStepId, t as getParentStepId, u as getTraceDirFromContext, v as getTraceSafetyFromContext, w as hasActiveContext, x as isAgentInspectEnabled, y as isSilentContext, m as maybeInspectRun, z as prepareMetadataForDisk, A as prepareTraceEventForDisk, B as resolveTraceSafetyOptions, E as runWithContext, F as runWithStepContext } from './context-
|
|
2
|
-
import { R as RedactionProfile, E as ErrorInfo, b as RunStatus, e as StepStatus,
|
|
1
|
+
export { C as CreateInspectorOptions, D as DEFAULT_MAX_EVENT_BYTES, f as DEFAULT_MAX_METADATA_VALUE_LENGTH, h as DEFAULT_MAX_PREVIEW_LENGTH, I as Inspector, a as InspectorCaptureOptions, b as InspectorObserveOptions, c as InspectorRunOptions, i as InspectorRuntime, j as InspectorRuntimeContext, k as InspectorRuntimeDiagnostics, l as InspectorRuntimeOptions, d as InspectorStepOptions, T as TraceSafetyOptions, e as createInspector, n as createInspectorRuntime, o as getCurrentContext, g as getCurrentCorrelationMetadata, p as getCurrentDepth, q as getCurrentRunId, r as getCurrentRunName, s as getCurrentStepId, t as getParentStepId, u as getTraceDirFromContext, v as getTraceSafetyFromContext, w as hasActiveContext, x as isAgentInspectEnabled, y as isSilentContext, m as maybeInspectRun, z as prepareMetadataForDisk, A as prepareTraceEventForDisk, B as resolveTraceSafetyOptions, E as runWithContext, F as runWithStepContext } from './context-BJ5UQLDi.js';
|
|
2
|
+
import { R as RedactionProfile, h as TraceEvent, E as ErrorInfo, b as RunStatus, e as StepStatus, j as RunSummary, k as TraceMetadata, l as TraceMetadataStatus, f as StepType, g as TraceCorrelationMetadata } from './types-BVkjSd0q.js';
|
|
3
3
|
export { A as ActiveStepContext, i as ExecutionContext, I as InspectRunOptions, O as ObserveOptions, a as Run, m as RunCompletedEvent, n as RunStartedEvent, c as Step, o as StepCompletedEvent, d as StepMetadata, S as StepOptions, p as StepStartedEvent, T as TokenMetadata, q as TraceEventBase, r as TraceSchemaVersion, s as isStepStatus, t as isStepType, u as isTraceEvent } from './types-BVkjSd0q.js';
|
|
4
4
|
import { b as InspectRunTree } from './inspect-event-IqwOHbGo.js';
|
|
5
5
|
export { A as AttributionConfidence, E as EventSource, a as InspectEvent, I as InspectKind, c as InspectNode } from './inspect-event-IqwOHbGo.js';
|
|
6
6
|
import { P as PersistedInspectEvent } from './persisted-inspect-event-DruwqARZ.js';
|
|
7
7
|
import { Stats } from 'node:fs';
|
|
8
|
+
import { a as ObservedOutcome } from './types-J3Xtuf9x.js';
|
|
9
|
+
export { b as ObservedOutcomeStatus } from './types-J3Xtuf9x.js';
|
|
8
10
|
import { TraceCheckResult, TraceCheckStatus } from './checks.js';
|
|
9
11
|
import './writers.js';
|
|
10
|
-
import './types-CvsKGweN.js';
|
|
11
12
|
import './log-config-SRA4xTp8.js';
|
|
12
13
|
import './readers.js';
|
|
13
14
|
|
|
@@ -20,6 +21,9 @@ interface ResolvedRedactionProfile {
|
|
|
20
21
|
/** Resolves named profile behavior (keys + metadata string caps). */
|
|
21
22
|
declare function resolveRedactionProfile(profile?: RedactionProfile): ResolvedRedactionProfile;
|
|
22
23
|
|
|
24
|
+
declare function extractOutcomesFromTraceEvents(events: readonly TraceEvent[]): ObservedOutcome[];
|
|
25
|
+
declare function extractOutcomesFromPersistedEvents(events: readonly PersistedInspectEvent[]): ObservedOutcome[];
|
|
26
|
+
|
|
23
27
|
/** Two spaces per nesting level in terminal output. */
|
|
24
28
|
declare const TERMINAL_INDENT = " ";
|
|
25
29
|
/** Max display length for names in terminal output. */
|
|
@@ -873,6 +877,12 @@ declare function runSuite(options?: RunSuiteOptions): Promise<SuiteRunResult>;
|
|
|
873
877
|
declare function renderSuiteReportMarkdown(result: SuiteRunResult): string;
|
|
874
878
|
declare function renderSuiteReport(result: SuiteRunResult, options?: RenderSuiteReportOptions): string;
|
|
875
879
|
|
|
880
|
+
declare const SUITE_TEMPLATE_IDS: readonly ["customer-support-agent", "refund-agent", "sales-assistant", "browser-task-agent", "mcp-tool-agent", "workflow-agent", "rag-answer-agent", "human-approval-agent"];
|
|
881
|
+
type SuiteTemplateId = (typeof SUITE_TEMPLATE_IDS)[number];
|
|
882
|
+
declare function listSuiteTemplates(): SuiteTemplateId[];
|
|
883
|
+
declare function getSuiteTemplate(id: string): SuiteConfig | undefined;
|
|
884
|
+
declare function resolveSuiteTemplate(id: string): SuiteConfig;
|
|
885
|
+
|
|
876
886
|
type CohortMetricId = "errorRate" | "duration" | "toolChoice" | "toolOrdering" | "llmCallCount" | "tokenUsage" | "retryCount" | "observationFailure" | "guardrailFailure" | "circuitViolation" | "redactionWarning";
|
|
877
887
|
declare const COHORT_METRIC_IDS: readonly CohortMetricId[];
|
|
878
888
|
interface CohortRunMetrics {
|
|
@@ -1006,4 +1016,4 @@ declare function renderGateGithubStepSummary(result: GateResult): string;
|
|
|
1006
1016
|
declare function renderGateJUnit(result: GateResult): string;
|
|
1007
1017
|
declare function renderGateReport(result: GateResult, options?: RenderGateReportOptions): string;
|
|
1008
1018
|
|
|
1009
|
-
export { type ActivityEntry, type ActivitySummary, type AnalyzeCohortOptions, type BuildActivitySummaryOptions, type BuildSessionIndexOptions, type BundleCheckResults, type BundleCheckRunResult, type BundleMetadata, type BundlePlaceholderArtifact, type BundleRedactionProfile, type BundleRedactionReport, type BundleRedactionReportRun, type BundleResolveOptions, type BundleResolveResult, type BundleSafeStatus, type BundleSafeStatusMetadata, COHORT_METRIC_IDS, type CohortAggregateMetrics, type CohortAnalysisResult, type CohortMetricComparison, type CohortMetricId, type CohortRunMetrics, type CriticalPathStep, DEFAULT_SUITE_ARTIFACTS_DIR, DEFAULT_SUITE_CONFIG_NAMES, DEFAULT_TRACE_DIR_NAME, type DurationStats, type EnrichSessionSummaryOptions, ErrorInfo, type ExplainFact, type ExplainInference, type ExplainMode, type ExplainOptions, type ExplainResult, FALLBACK_TRACE_DIR, type GateCheckId, type GateCheckResult, type GateExitCode, type GateResult, type GroupSessionCohortsOptions, type HandoffEdge, InspectRunTree, type LoadSuiteConfigOptions, MAX_NAME_LENGTH, MAX_TERMINAL_DEPTH, MAX_TERMINAL_NAME_LENGTH, type ParseTraceJsonlOptions, type ParseTraceJsonlResult, type ParsedDurationFilter, RUNS_DIR_NAME, RedactionProfile, type RenderCohortReportOptions, type RenderGateReportOptions, type RenderSuiteReportOptions, type RenderTimelineOptions, type RenderWhatOptions, type ResolvedRedactionProfile, type ResolvedSuiteCase, type RetryLink, type RunGateOptions, RunStatus, type RunSuiteOptions, RunSummary, type RunTimeline, type RunWhatSummary, SESSION_WORKFLOW_KEYS, type SessionCheckSummary, type SessionCohort, type SessionCohortKind, type SessionConfidence, type SessionEdgeSource, type SessionGroup, type SessionIndex, type SessionLastError, type SessionRunRecord, type SessionScopeOptions, type SessionScopeResult, type SessionStatus, type SessionSummary, type SessionWarning, type SessionWorkflowKey, type SessionWorkflowMetadata, StepStatus, StepType, type SuiteArtifactsConfig, type SuiteCaseConfig, type SuiteCaseResult, type SuiteCaseStatus, type SuiteChecksConfig, type SuiteConfig, type SuiteDiagnostic, type SuiteDiagnosticCode, type SuiteEvalConfig, type SuiteRunResult, type SuiteRunSummary, TERMINAL_INDENT, type TimelineEntry, type TimelineFocus, type TimelineOptions, TraceCorrelationMetadata, TraceDirectory, type TraceDirectoryOptions, TraceEvent, type TraceFilterOptions, type TraceJsonlFormat, TraceMetadata, TraceMetadataStatus, type TraceSearchOptions, type TraceSearchResult, type TraceSessionCheckResult, type TraceStats, type TraceStatsOptions, type TraceStatsRankedRun, type TraceStatsRankedStep, type ValidateSuiteConfigResult, aggregateBundleSafeStatus, aggregateSessionCheckResults, analyzeCohort, buildActivitySummary, buildBundleMetadata, buildBundleSummaryMarkdown, buildLocalExplanation, buildPlaceholderArtifact, buildRunSummary, buildRunTimeline, buildRunWhatSummary, buildSessionIndex, buildTraceStats, bundleFailsOnSafety, compareCohortAggregates, createRunId, createStepId, defaultBundleOutputPath, defaultSuiteConfigTemplate, deriveSessionStatus, enrichSessionRunRecord, enrichSessionSummary, ensureTraceDir, extractMetadata, extractSessionWorkflowMetadata, filterMetasBySessionScope, filterTraces, formatDuration, formatError, formatTerminalName, formatTimestamp, gateHasThresholds, getDefaultTraceDir, getIndent, getRunIdFromTraceFileName, getTraceFilePath, groupSessionCohorts, initializeTraceFile, isAgentInspectTrace, listTraceFiles, loadSessionRunRecords, loadSuiteConfig, loadTraceMetadataList, normalizeBundleOutputPath, normalizeSuiteConfig, parseCohortMetricList, parseDuration, parseDurationFilter, parseGateList, parseGateNumber, parseGroupBySpec, parseTraceJsonl, printError, printFailedAt, printRunComplete, printRunStart, printStepComplete, printStepStart, readTraceEvents, readTraceFile, renderActivitySummaryHuman, renderCohortReport, renderCohortSummaryMarkdown, renderErrorLine, renderGateGithubStepSummary, renderGateJUnit, renderGateReport, renderGateSummaryMarkdown, renderRunSummary, renderRunWhat, renderStepLine, renderSuiteReport, renderSuiteReportMarkdown, renderTimeline, renderTraceStats, resolveBundleRunIds, resolveRedactionProfile, resolveSuiteCaseTrace, resolveSuiteConfigPath, resolveTraceDir, runGate, runSuite, searchTraces, serializeEvent, sessionKeyForRun, toMetadataSafeStatus, traceMetasToSessionRunRecords, truncateName, unknownTraceFormatMessage, validateEvent, validateSuiteConfig, warn, writeTraceEvent };
|
|
1019
|
+
export { type ActivityEntry, type ActivitySummary, type AnalyzeCohortOptions, type BuildActivitySummaryOptions, type BuildSessionIndexOptions, type BundleCheckResults, type BundleCheckRunResult, type BundleMetadata, type BundlePlaceholderArtifact, type BundleRedactionProfile, type BundleRedactionReport, type BundleRedactionReportRun, type BundleResolveOptions, type BundleResolveResult, type BundleSafeStatus, type BundleSafeStatusMetadata, COHORT_METRIC_IDS, type CohortAggregateMetrics, type CohortAnalysisResult, type CohortMetricComparison, type CohortMetricId, type CohortRunMetrics, type CriticalPathStep, DEFAULT_SUITE_ARTIFACTS_DIR, DEFAULT_SUITE_CONFIG_NAMES, DEFAULT_TRACE_DIR_NAME, type DurationStats, type EnrichSessionSummaryOptions, ErrorInfo, type ExplainFact, type ExplainInference, type ExplainMode, type ExplainOptions, type ExplainResult, FALLBACK_TRACE_DIR, type GateCheckId, type GateCheckResult, type GateExitCode, type GateResult, type GroupSessionCohortsOptions, type HandoffEdge, InspectRunTree, type LoadSuiteConfigOptions, MAX_NAME_LENGTH, MAX_TERMINAL_DEPTH, MAX_TERMINAL_NAME_LENGTH, ObservedOutcome, type ParseTraceJsonlOptions, type ParseTraceJsonlResult, type ParsedDurationFilter, RUNS_DIR_NAME, RedactionProfile, type RenderCohortReportOptions, type RenderGateReportOptions, type RenderSuiteReportOptions, type RenderTimelineOptions, type RenderWhatOptions, type ResolvedRedactionProfile, type ResolvedSuiteCase, type RetryLink, type RunGateOptions, RunStatus, type RunSuiteOptions, RunSummary, type RunTimeline, type RunWhatSummary, SESSION_WORKFLOW_KEYS, SUITE_TEMPLATE_IDS, type SessionCheckSummary, type SessionCohort, type SessionCohortKind, type SessionConfidence, type SessionEdgeSource, type SessionGroup, type SessionIndex, type SessionLastError, type SessionRunRecord, type SessionScopeOptions, type SessionScopeResult, type SessionStatus, type SessionSummary, type SessionWarning, type SessionWorkflowKey, type SessionWorkflowMetadata, StepStatus, StepType, type SuiteArtifactsConfig, type SuiteCaseConfig, type SuiteCaseResult, type SuiteCaseStatus, type SuiteChecksConfig, type SuiteConfig, type SuiteDiagnostic, type SuiteDiagnosticCode, type SuiteEvalConfig, type SuiteRunResult, type SuiteRunSummary, type SuiteTemplateId, TERMINAL_INDENT, type TimelineEntry, type TimelineFocus, type TimelineOptions, TraceCorrelationMetadata, TraceDirectory, type TraceDirectoryOptions, TraceEvent, type TraceFilterOptions, type TraceJsonlFormat, TraceMetadata, TraceMetadataStatus, type TraceSearchOptions, type TraceSearchResult, type TraceSessionCheckResult, type TraceStats, type TraceStatsOptions, type TraceStatsRankedRun, type TraceStatsRankedStep, type ValidateSuiteConfigResult, aggregateBundleSafeStatus, aggregateSessionCheckResults, analyzeCohort, buildActivitySummary, buildBundleMetadata, buildBundleSummaryMarkdown, buildLocalExplanation, buildPlaceholderArtifact, buildRunSummary, buildRunTimeline, buildRunWhatSummary, buildSessionIndex, buildTraceStats, bundleFailsOnSafety, compareCohortAggregates, createRunId, createStepId, defaultBundleOutputPath, defaultSuiteConfigTemplate, deriveSessionStatus, enrichSessionRunRecord, enrichSessionSummary, ensureTraceDir, extractMetadata, extractOutcomesFromPersistedEvents, extractOutcomesFromTraceEvents, extractSessionWorkflowMetadata, filterMetasBySessionScope, filterTraces, formatDuration, formatError, formatTerminalName, formatTimestamp, gateHasThresholds, getDefaultTraceDir, getIndent, getRunIdFromTraceFileName, getSuiteTemplate, getTraceFilePath, groupSessionCohorts, initializeTraceFile, isAgentInspectTrace, listSuiteTemplates, listTraceFiles, loadSessionRunRecords, loadSuiteConfig, loadTraceMetadataList, normalizeBundleOutputPath, normalizeSuiteConfig, parseCohortMetricList, parseDuration, parseDurationFilter, parseGateList, parseGateNumber, parseGroupBySpec, parseTraceJsonl, printError, printFailedAt, printRunComplete, printRunStart, printStepComplete, printStepStart, readTraceEvents, readTraceFile, renderActivitySummaryHuman, renderCohortReport, renderCohortSummaryMarkdown, renderErrorLine, renderGateGithubStepSummary, renderGateJUnit, renderGateReport, renderGateSummaryMarkdown, renderRunSummary, renderRunWhat, renderStepLine, renderSuiteReport, renderSuiteReportMarkdown, renderTimeline, renderTraceStats, resolveBundleRunIds, resolveRedactionProfile, resolveSuiteCaseTrace, resolveSuiteConfigPath, resolveSuiteTemplate, resolveTraceDir, runGate, runSuite, searchTraces, serializeEvent, sessionKeyForRun, toMetadataSafeStatus, traceMetasToSessionRunRecords, truncateName, unknownTraceFormatMessage, validateEvent, validateSuiteConfig, warn, writeTraceEvent };
|
|
@@ -3,6 +3,7 @@ export { MAX_TERMINAL_DEPTH, MAX_TERMINAL_NAME_LENGTH, TERMINAL_INDENT, TraceDir
|
|
|
3
3
|
import { buildRunSummary, extractMetadata } from './chunk-3USJVBLA.mjs';
|
|
4
4
|
export { buildRunSummary, buildRunTimeline, buildRunWhatSummary, extractMetadata, renderRunWhat, renderTimeline } from './chunk-3USJVBLA.mjs';
|
|
5
5
|
import { parseObservationFilter, extractOutcomesFromTraceEvents, extractOutcomesFromPersistedEvents } from './chunk-VFO76UH3.mjs';
|
|
6
|
+
export { extractOutcomesFromPersistedEvents, extractOutcomesFromTraceEvents } from './chunk-VFO76UH3.mjs';
|
|
6
7
|
import './chunk-BT7CATSD.mjs';
|
|
7
8
|
import { escapeHtml } from './chunk-CAWPF22J.mjs';
|
|
8
9
|
import { resolveRedactionProfile, readTraceEventsFromFile } from './chunk-ULCGIRTC.mjs';
|
|
@@ -2219,6 +2220,118 @@ function renderSuiteReport(result, options = {}) {
|
|
|
2219
2220
|
return renderSuiteReportMarkdown(result);
|
|
2220
2221
|
}
|
|
2221
2222
|
|
|
2223
|
+
// packages/core/src/suite/templates.ts
|
|
2224
|
+
var SUITE_TEMPLATE_IDS = [
|
|
2225
|
+
"customer-support-agent",
|
|
2226
|
+
"refund-agent",
|
|
2227
|
+
"sales-assistant",
|
|
2228
|
+
"browser-task-agent",
|
|
2229
|
+
"mcp-tool-agent",
|
|
2230
|
+
"workflow-agent",
|
|
2231
|
+
"rag-answer-agent",
|
|
2232
|
+
"human-approval-agent"
|
|
2233
|
+
];
|
|
2234
|
+
function baseTemplate(name, cases) {
|
|
2235
|
+
return {
|
|
2236
|
+
name,
|
|
2237
|
+
traces: "./.agent-inspect",
|
|
2238
|
+
cases,
|
|
2239
|
+
checks: {
|
|
2240
|
+
select: ["run.status", "outcome.status"],
|
|
2241
|
+
run: { maxDurationMs: 6e4 }
|
|
2242
|
+
},
|
|
2243
|
+
redactionProfile: "local",
|
|
2244
|
+
artifacts: { outputDir: ".agent-inspect/suite-runs" }
|
|
2245
|
+
};
|
|
2246
|
+
}
|
|
2247
|
+
var TEMPLATES = {
|
|
2248
|
+
"customer-support-agent": baseTemplate("customer-support-agent", [
|
|
2249
|
+
{
|
|
2250
|
+
id: "support-ticket",
|
|
2251
|
+
runId: "support-ticket",
|
|
2252
|
+
requireTools: ["searchDocs"],
|
|
2253
|
+
forbidTools: ["deleteAccount"],
|
|
2254
|
+
expectedObservations: ["answerReady"],
|
|
2255
|
+
maxDurationMs: 45e3
|
|
2256
|
+
}
|
|
2257
|
+
]),
|
|
2258
|
+
"refund-agent": baseTemplate("refund-agent", [
|
|
2259
|
+
{
|
|
2260
|
+
id: "refund-flow",
|
|
2261
|
+
runId: "refund-flow",
|
|
2262
|
+
requireTools: ["searchDocs"],
|
|
2263
|
+
forbidTools: ["deleteAccount"],
|
|
2264
|
+
expectedObservations: ["policyShown"],
|
|
2265
|
+
maxDurationMs: 3e4
|
|
2266
|
+
}
|
|
2267
|
+
]),
|
|
2268
|
+
"sales-assistant": baseTemplate("sales-assistant", [
|
|
2269
|
+
{
|
|
2270
|
+
id: "quote-request",
|
|
2271
|
+
runId: "quote-request",
|
|
2272
|
+
requireTools: ["lookupPricing"],
|
|
2273
|
+
maxDurationMs: 3e4
|
|
2274
|
+
}
|
|
2275
|
+
]),
|
|
2276
|
+
"browser-task-agent": baseTemplate("browser-task-agent", [
|
|
2277
|
+
{
|
|
2278
|
+
id: "browser-nav",
|
|
2279
|
+
runId: "browser-nav",
|
|
2280
|
+
requireTools: ["browserNavigate"],
|
|
2281
|
+
maxDurationMs: 12e4
|
|
2282
|
+
}
|
|
2283
|
+
]),
|
|
2284
|
+
"mcp-tool-agent": baseTemplate("mcp-tool-agent", [
|
|
2285
|
+
{
|
|
2286
|
+
id: "mcp-call",
|
|
2287
|
+
runId: "mcp-call",
|
|
2288
|
+
requireTools: ["mcp:tools/list"],
|
|
2289
|
+
maxDurationMs: 3e4
|
|
2290
|
+
}
|
|
2291
|
+
]),
|
|
2292
|
+
"workflow-agent": baseTemplate("workflow-agent", [
|
|
2293
|
+
{
|
|
2294
|
+
id: "workflow-handoff",
|
|
2295
|
+
runId: "workflow-handoff",
|
|
2296
|
+
requireTools: ["handoff"],
|
|
2297
|
+
maxDurationMs: 6e4
|
|
2298
|
+
}
|
|
2299
|
+
]),
|
|
2300
|
+
"rag-answer-agent": baseTemplate("rag-answer-agent", [
|
|
2301
|
+
{
|
|
2302
|
+
id: "rag-answer",
|
|
2303
|
+
runId: "rag-answer",
|
|
2304
|
+
requireTools: ["retrieve", "answer"],
|
|
2305
|
+
expectedObservations: ["citationShown"],
|
|
2306
|
+
maxDurationMs: 45e3
|
|
2307
|
+
}
|
|
2308
|
+
]),
|
|
2309
|
+
"human-approval-agent": baseTemplate("human-approval-agent", [
|
|
2310
|
+
{
|
|
2311
|
+
id: "approval-gate",
|
|
2312
|
+
runId: "approval-gate",
|
|
2313
|
+
expectedObservations: ["approvalRequested", "approvalGranted"],
|
|
2314
|
+
maxDurationMs: 6e4
|
|
2315
|
+
}
|
|
2316
|
+
])
|
|
2317
|
+
};
|
|
2318
|
+
function listSuiteTemplates() {
|
|
2319
|
+
return [...SUITE_TEMPLATE_IDS];
|
|
2320
|
+
}
|
|
2321
|
+
function getSuiteTemplate(id) {
|
|
2322
|
+
if (!SUITE_TEMPLATE_IDS.includes(id)) return void 0;
|
|
2323
|
+
return structuredClone(TEMPLATES[id]);
|
|
2324
|
+
}
|
|
2325
|
+
function resolveSuiteTemplate(id) {
|
|
2326
|
+
const template = getSuiteTemplate(id);
|
|
2327
|
+
if (template === void 0) {
|
|
2328
|
+
throw new Error(
|
|
2329
|
+
`Unknown suite template "${id}". Available: ${SUITE_TEMPLATE_IDS.join(", ")}`
|
|
2330
|
+
);
|
|
2331
|
+
}
|
|
2332
|
+
return template;
|
|
2333
|
+
}
|
|
2334
|
+
|
|
2222
2335
|
// packages/core/src/cohort/types.ts
|
|
2223
2336
|
var COHORT_METRIC_IDS = [
|
|
2224
2337
|
"errorRate",
|
|
@@ -3131,6 +3244,6 @@ function renderGateReport(result, options = {}) {
|
|
|
3131
3244
|
return renderGateSummaryMarkdown(result);
|
|
3132
3245
|
}
|
|
3133
3246
|
|
|
3134
|
-
export { COHORT_METRIC_IDS, DEFAULT_SUITE_ARTIFACTS_DIR, DEFAULT_SUITE_CONFIG_NAMES, SESSION_WORKFLOW_KEYS, aggregateBundleSafeStatus, aggregateSessionCheckResults, analyzeCohort, buildActivitySummary, buildBundleMetadata, buildBundleSummaryMarkdown, buildLocalExplanation, buildPlaceholderArtifact, buildSessionIndex, buildTraceStats, bundleFailsOnSafety, compareCohortAggregates, defaultBundleOutputPath, defaultSuiteConfigTemplate, deriveSessionStatus, enrichSessionRunRecord, enrichSessionSummary, extractSessionWorkflowMetadata, filterMetasBySessionScope, filterTraces, gateHasThresholds, groupSessionCohorts, isAgentInspectTrace, loadSessionRunRecords, loadSuiteConfig, loadTraceMetadataList, normalizeBundleOutputPath, normalizeSuiteConfig, parseCohortMetricList, parseDurationFilter, parseGateList, parseGateNumber, parseGroupBySpec, renderActivitySummaryHuman, renderCohortReport, renderCohortSummaryMarkdown, renderGateGithubStepSummary, renderGateJUnit, renderGateReport, renderGateSummaryMarkdown, renderSuiteReport, renderSuiteReportMarkdown, renderTraceStats, resolveBundleRunIds, resolveSuiteCaseTrace, resolveSuiteConfigPath, runGate, runSuite, searchTraces, sessionKeyForRun, toMetadataSafeStatus, traceMetasToSessionRunRecords, validateSuiteConfig };
|
|
3247
|
+
export { COHORT_METRIC_IDS, DEFAULT_SUITE_ARTIFACTS_DIR, DEFAULT_SUITE_CONFIG_NAMES, SESSION_WORKFLOW_KEYS, SUITE_TEMPLATE_IDS, aggregateBundleSafeStatus, aggregateSessionCheckResults, analyzeCohort, buildActivitySummary, buildBundleMetadata, buildBundleSummaryMarkdown, buildLocalExplanation, buildPlaceholderArtifact, buildSessionIndex, buildTraceStats, bundleFailsOnSafety, compareCohortAggregates, defaultBundleOutputPath, defaultSuiteConfigTemplate, deriveSessionStatus, enrichSessionRunRecord, enrichSessionSummary, extractSessionWorkflowMetadata, filterMetasBySessionScope, filterTraces, gateHasThresholds, getSuiteTemplate, groupSessionCohorts, isAgentInspectTrace, listSuiteTemplates, loadSessionRunRecords, loadSuiteConfig, loadTraceMetadataList, normalizeBundleOutputPath, normalizeSuiteConfig, parseCohortMetricList, parseDurationFilter, parseGateList, parseGateNumber, parseGroupBySpec, renderActivitySummaryHuman, renderCohortReport, renderCohortSummaryMarkdown, renderGateGithubStepSummary, renderGateJUnit, renderGateReport, renderGateSummaryMarkdown, renderSuiteReport, renderSuiteReportMarkdown, renderTraceStats, resolveBundleRunIds, resolveSuiteCaseTrace, resolveSuiteConfigPath, resolveSuiteTemplate, runGate, runSuite, searchTraces, sessionKeyForRun, toMetadataSafeStatus, traceMetasToSessionRunRecords, validateSuiteConfig };
|
|
3135
3248
|
//# sourceMappingURL=advanced.mjs.map
|
|
3136
3249
|
//# sourceMappingURL=advanced.mjs.map
|