@agentskit/harness 0.12.0 → 0.13.0

package/dist/index.d.ts

@@ -2240,9 +2240,10 @@ declare const modelFor: (policy: ModelPolicy, role: ModelRole) => ModelBinding;
  * PR/issue comment — a segment lifted from an issue description or a code-review finding can carry a secret that
  * was never meant to leave the private context it came from. Pure and kernel-safe: no adapters, no network, no
  * state. Pattern-based, not a claim of completeness — it catches the shapes that show up in practice (emails,
- * common provider API-key prefixes, phone numbers, card-number-shaped digit runs), not every possible secret.
+ * common provider API-key prefixes, PEM private-key blocks, phone numbers, card-number-shaped digit runs), not
+ * every possible secret.
  */
-type PiiKind = 'email' | 'api-key' | 'phone' | 'credit-card';
+type PiiKind = 'email' | 'api-key' | 'phone' | 'credit-card' | 'private-key';
 interface PiiMatch {
     readonly kind: PiiKind;
     readonly index: number;
@@ -2370,8 +2371,16 @@ declare const exportEvidenceBundle: ({ configPath, runId, outputPath, privateKey
     readonly privateKeyPath: string;
     readonly keyId: string;
 }) => Promise<EvidenceBundle>;
-declare const verifyEvidenceBundle: (path: string, { trustedKeys }?: {
+/** Per-file and total caps on decoded evidence content: without them, a corrupted or hostile bundle could carry
+ * arbitrarily large (or arbitrarily many) `contentBase64` blobs and exhaust memory during verification, before
+ * any hash or signature check ever runs. The base64-length pre-check happens before `Buffer.from` decodes
+ * anything, so an oversized single file is rejected without allocating its decoded buffer at all. */
+declare const EVIDENCE_MAX_FILE_BYTES: number;
+declare const EVIDENCE_MAX_TOTAL_BYTES: number;
+declare const verifyEvidenceBundle: (path: string, { trustedKeys, maxFileBytes, maxTotalBytes }?: {
     readonly trustedKeys?: readonly TrustedEvidenceKey[];
+    readonly maxFileBytes?: number;
+    readonly maxTotalBytes?: number;
 }) => EvidenceBundleVerification;
 declare const readEvidenceTrustStore: (path: string) => readonly TrustedEvidenceKey[];
 
@@ -4595,4 +4604,4 @@ declare const discoverIntake: (runner: CommandRunner, input: {
     readonly now: () => Date;
 }, options?: GitHubCliOptions) => Promise<readonly IntakeRecord[]>;
 
-export { AGENT_REGISTRY_SCHEMA_VERSION, ARTIFACT_SCHEMA_VERSION, ARTIFACT_TYPES, ASSURANCE_LEVELS, type AdapterMetadata, type AdapterTelemetry, type AdversarialReviewResult, type AgentAdapter, type AgentEvalCase, type AgentEvalReport, type AgentEvalSuite, type AgentMemoryAdapter, type AgentMemoryHit, type AgentMemoryKvStore, type AgentMemoryRecord, type AgentRegistry, type AgentRegistryEntry, AgentRegistryEntrySchema, AgentRegistrySchema, type AgentSessionOptions, type AgentUsage, type ApprovedAssumption, type ArgvRagContextProviderOptions, type ArtifactBinding, type ArtifactEnvelope, type ArtifactEnvelopeInput, type ArtifactType, type ArtificialAnalysisModel, type AssuranceLevel, type AutomationStatus, type AutonomyMode, BENCHMARK_SCHEMA_VERSION, BLOCK_STATUSES, type BenchmarkBinding, type BenchmarkComparison, type BenchmarkImprovementDirection, type BenchmarkManifest, type BenchmarkObservation, type BenchmarkObservationEvidence, type BenchmarkObservationInput, type BenchmarkObservationStatus, type BenchmarkReport, type BenchmarkRun, type BenchmarkSummary, type BenchmarkTask, type BlockAssessment, type BlockManifest, type BlockStatus, CAPABILITY_KINDS, CAPABILITY_MANIFEST_SCHEMA_VERSION, CHECK_CATEGORIES, COMPATIBILITY_COMPONENTS, COMPATIBILITY_SCHEMA_VERSION, CONTEXT_PROVIDER_SLOT, CONTRACT_CLOSE, CONTRACT_OPEN, CONTRACT_SCHEMA_VERSION, type CacheUsage, type CapabilityDescriptor, type CapabilityKind, type CapabilityManifest, type CapabilityManifestInput, type CatalogModel, type ChangedFile, type CheckCategory, type CheckOutcome, type CheckResult, type ChecksAssessment, type ClaimResult, type CodeReviewInput, type CodeReviewOutcome, type CodingAgentAdapter, type CodingAgentHandlerResult, type CodingAgentRequest, type CodingAgentResult, type CommandResult, type CommandRunOptions, type CommandRunner, type CompatibilityComponent, type CompatibilityComponentId, type CompatibilityManifest, type CompatibilityObservation, type CompatibilityReport, type CompatibilityStatus, type ContextProvider, type ContextQuery, type ContextReference, type ContextSnapshot, type ContractAssessment, type ContractOutcome, ContractOutcomeSchema, type ContractScope, type CooldownEntry, type CooldownState, type CoordinationIdentity, type CriterionStatus, type CycleIterationMetrics, type CycleMatrixRow, type CycleStepResult, type CycleStepStatus, type DebriefInput, type DebriefIssueRow, type DebriefReport, type DecisionPacket, type DeliverInput, type DeliverOutcome, type DeliverReport, type DeliverResult, type DeliveryState, type DetectProvidersInput, type DiscoveryAmbiguity, type DiscoveryCurrentInput, type DiscoveryCurrentResult, type DiscoveryDecisionLogEntry, type DiscoveryInput, type DiscoveryOption, type DiscoveryResult, type DispatchLease, type DispatchLedger, type DispatchRecord, type DispatchRecordFile, type Disposer, type DocBridgeIndexInspection, type DockerMount, type DockerRuntimeEvidence, type DockerToolDefinition, type DoctorCheck, type DoctorCheckStatus, EVAL_COMPONENTS, EVAL_LAYERS, EVAL_MANIFEST_SCHEMA_VERSION, EVENT_LOG_GENESIS, EVIDENCE_BUNDLE_SCHEMA_VERSION, type EffortLevel, type EvalBatteryReport, type EvalCaseDefinition, type EvalCaseReport, type EvalComponent, type EvalExpectation, type EvalLayer, type EvalManifest, type EvalObservation, type EvalObservationStatus, type EventLogLock, type EventLogLockRecovery, type EventLogLockStatus, type EventLogVerification, type EventStore, type EvidenceArtifact, type EvidenceBundle, type EvidenceBundleFile, type EvidenceBundleSignature, type EvidenceBundleVerification, type EvidenceReference, type ExecutePhaseProfileOptions, type FailureClass, type FailureClassification, type FetchQueueInput, FileArtifactStore, FileEventStore, type FilePreflightPlan, type GateAssessment, type GateBinding, type GateCriterion, type GenerateContractInput, type GitHubCliOptions, type GuidedInstallIO, type GuidedInstallInput, type GuidedInstallReport, HARNESS_ERROR_CATALOG, HARNESS_ERROR_CODES, HARNESS_EVENT_ENVELOPE_SCHEMA_VERSION, HARNESS_EVENT_SCHEMA_VERSION, HARNESS_EVENT_TYPES, HARNESS_PLUGIN_API_VERSION, HARNESS_REPO_URL, type HandoffBriefInput, HarnessError, type HarnessErrorClassification, type HarnessErrorDisposition, type HarnessEvent, type HarnessEventContext, type HarnessEventEnvelope, type HarnessEventEnvelopeInput, type HarnessEventInput, type HarnessEventListener, type HarnessEventPayloads, type HarnessEventProvenance, type HarnessEventType, type HarnessPlugin, type HarnessPluginContext, IMPROVEMENT_CYCLE_STEPS, type ImprovementCycleAssessment, type ImprovementCycleInput, type ImprovementCycleIteration, type ImprovementCycleStep, type InstallAction, type InstallInput, type InstallReport, type IntakeRecord, type IssueFailureRecord, type IssueFailureState, LEARNING_STATUSES, LEGAL_TRANSITIONS, LOOP_CONFIG_FILE, LOOP_CONFIG_SCHEMA_VERSION, LOOP_LOCAL_CONFIG_FILE, LOOP_STAGES, type LearningRecord, type LearningStatus, type LearningsLedger, type LinearIssueDetail, type LinearListInput, type LinearQueueFilter, type LinearWriteOptions, type LlmCache, type LlmCacheKeyInput, type LlmCacheStats, type LoadedConfig, type LoadedLoopConfig, type LocalConfigAnswers, type LocalConfigPrompter, type LoopConfig, type LoopConfigInput, LoopConfigSchema, type LoopDoctorInput, type LoopDoctorReport, type LoopEvent, type LoopEventBus, type LoopEventListener, type LoopEventPayload, type LoopHookListener, type LoopHookName, type LoopHookPayload, type LoopHookResult, type LoopIssue, type LoopPluginModule, type LoopProviderConfig, type LoopStage, type LoopStageName, type LoopState, type LoopStatusReport, MEMORY_SCOPES, MODEL_ROLES, type MachineMetrics, type MachineSample, type MachineThresholds, type McpPolicy, type McpToolBridge, type McpToolBridgeOptions, type McpToolCallInput, type McpToolCallResult, type MemoryContextPlan, type MemoryPromptSelection, type MemoryScope, type MemoryUsage, type MetricStatus, type ModelBinding, type ModelPolicy, type ModelQuality, type ModelReference, type ModelRole, type NormalizedPhaseProfile, type ObservabilityAnomaly, type ObservabilityMetrics, type ObservabilityReport, type ObservabilitySeverity, type ObservabilitySnapshot, type ObservabilityTerminal, type OptimizationComparison, type OptimizationObservation, type OrcaAgentHookState, type OrcaAutomation, type OrcaAutomationSpec, type OrcaCliOptions, type OrcaCreatedWorktree, type OrcaDispatchInput, type OrcaDispatchPlan, type OrcaLeaseState, type OrcaLifecycleInput, type OrcaLifecycleProjection, type OrcaMemorySample, type OrcaStatus as OrcaRuntimeStatus, type OrcaSendReceipt, type OrcaTerminal, type OrcaWorktree, type OutcomeProgress, type OutcomeProgressStatus, PHASE_DECISIONS, PHASE_EFFECTS, PHASE_EFFECT_ACTIONS, PHASE_MODES, PR_FIELDS, type ParallelismUsage, type PhaseAmbiguity, type PhaseContext, type PhaseDecision, type PhaseDecisionPacket, type PhaseDefinition, type PhaseEffect, type PhaseEffectAction, type PhaseEffectPolicy, type PhaseExecution, type PhaseExecutionReport, type PhaseGateEvaluator, type PhaseGateResult, type PhaseHandler, type PhaseHandlerResult, type PhaseMachineMetrics, type PhaseMode, type PhasePreflight, type PhasePreflightResult, type PhaseProfile, type PhaseResumeState, type PhaseRetryPolicy, type PhaseRoutePlan, type PhaseTelemetry, type PhaseTokenMetrics, type PiiKind, type PiiMatch, type PiiScanResult, type PilotAssessment, type PilotEntry, type PilotManifest, type PinnedSkill, type PinnedSkillRef, type PluginContribution, type PluginRegistry, type PluginSlot, type PolicyDecision, type PolicyGate, type PolicyRequest, type PolicyRule, type ProcessToolDefinition, type ProductionEvidence, type ProviderAuthStatus, type ProviderAvailability, type ProviderCatalog, type ProviderFailure, type ProviderSpec, type ProviderUsage, type PullRequestApproval, type PullRequestCheck, type PullRequestDraft, type PullRequestSnapshot, QUALITY_DIMENSIONS, type QaTransitionAssessment, type QualityDimension, type QualityDimensionScore, type QualityMatrix, REVIEW_SEVERITIES, RUN_STATES, type RagContextProviderOptions, type RagQueryResult, type RankedModel, type RecoveryObservation, type RecoveryPolicy, type RecoveryResult, type RepositoryProfile, type ResolvedAgent, type RetroInput, type RetroIssueRow, type RetroReport, type RetroStageReport, type RetroSuggestion, type RetroTarget, type RetroWindow, type ReviewFinding, type ReviewLens, type ReviewSeverity, type ReviewVerdict, type RichIO, type RoutingDecision, type RoutingSkip, type RunOutcome, type RunReconciliation, type RunState, type RuntimeConfig, type RuntimeExperimentCandidate, type RuntimeExperimentResult, STATES, SURFACE_NAMES, type SessionRecorder, type SlotAssessment, type SlotInput, type SourceSnapshot, type StagePauseEntry, type StagePauseState, type StateTransition, type StatusBlock, type StatusSnapshot, type StoredContract, type StructuredEvidence, type SurfaceName, type SurfaceRequirement, type TaskContract, TaskContractSchema, type TeamMember, type TickCandidateResult, type TickInput, type TickOutcome, type TickReport, type TokenUsage, type ToolDefinition, type ToolExecutionRequest, type ToolExecutionResult, type ToolRuntime, type TrackingAdapter, type TrackingConfig, type TrackingTransition, type TrustedEvidenceKey, type UsageMetric, type UsageWindow, type VerificationCheck, type VerificationConfig, type VerificationRun, WIP_STATES, type WatchEvent, type WatchEventKind, type WatchInput, type WatchReport, type WatchTargetSnapshot, type WatchdogBlocker, type WatchdogBudget, type WatchdogResult, type WipAssessment, type WipAssessmentInput, type WipEntry, type WipState, type WorkerBriefInput, type WorkflowNode, type WorkflowResult, activeCooldowns, adaptiveConcurrency, advanceQueueOwner, appendLoopEvent, approveRun, approvedDecision, artifactDigest, artifactFilePath, artifactIsFresh, artifactMarkdownPath, assertHuman, assessAcceptance, assessAgentEval, assessBlock, assessChecks, assessCompatibility, assessContract, assessDiscovery, assessImprovementCycle, assessIntegration, assessObservability, assessPilot, assessPreflight, assessProduction, assessQaTransition, assessSlots, assessWip, assessWorktreeCleanup, atLeast, authStatusFor, authorizeRun, automationName, automationPrompt, automationSpecs, availableMemoryBytes, benchmarkRuns, branchFor, briefPath, buildDebriefReport, buildListIssuesArgv, buildRetroReport, buildReviewArgv, buildSuggestions, busyIssues, cancelRun, classifyFailure, classifyHarnessError, classifyProviderFailure, classifyWatchEvent, classifyWatchPhase, cleanTaskArtifacts, clearIssueFailures, clearProviderCooldown, compareOptimization, compareVersions, composePullRequest, contractIsFresh, contractPath, cooldownPath, cooldownUntil, countRotationBlockingLeases, countRunningWorkers, createArgvRagContextProvider, createArtifactEnvelope, createCapabilityManifest, createCodingAgentAdapter, createCompatibilityManifest, createConfiguredToolRuntime, createDispatchLedger, createDocBridgeContextProvider, createDockerToolRuntime, createEvalManifest, createFileMemoryAdapter, createFileMemoryKvStore, createHarnessEventEnvelope, createInMemoryMemoryAdapter, createKvMemoryAdapter, createLinearTrackingAdapter, createLlmCache, createLlmCacheKey, createLoopEventBus, createMachineMonitor, createMcpToolBridge, createModelPolicy, createOrcaDispatchPlan, createOrcaLifecycleProjection, createPhaseArtifact, createPhaseProfile, createPluginRegistry, createPluginSlot, createPolicyGate, createProcessRunner, createProcessToolRuntime, createPullRequestApproval, createQualityMatrix, createRagContextProvider, createRichIO, createSessionRecorder, createStatusSnapshot, createToolRuntime, createTrackingAdapter, createTrackingTransition, deliveryStatePath, detectProviders, discoverIntake, dispatchRecordPath, evaluateWatchdog, executePhaseProfile, exportEvidenceBundle, extractResetsAt, fetchArtificialAnalysisModels, fetchLinearIssue, fetchLinearQueue, fetchTeamMembers, filterAndOrderQueue, findExecutable, formatWatchEvent, gatherLoopState, generateContract, githubComment, githubCommentArgv, githubCommentExists, githubLabelRemove, githubMerge, githubMergeArgv, githubOpenPullRequests, githubPullRequest, githubPullRequestsForBranch, hasLocalConfig, hashContextSnapshot, hashContextSnapshots, hashMcpArgs, inspectDocBridgeIndex, inspectEventLogLock, installLoopAutomations, installPreflight, intakeIssueId, intakePath, isDiscoveryCurrent, isIssuePaused, isStagePaused, isWsl, issueFailurePath, launchWorkerTerminal, learningToMemoryRecord, learningsPath, linearAttach, linearAttachArgv, linearCommentAdd, linearCommentAddArgv, linearLabelAdd, linearLabelArgv, linearLabelRemove, linearStatusSet, linearStatusSetArgv, listCliModels, listCliModelsCached, listDispatched, listIntake, listPausedIssues, loadAgentRegistry, loadAliases, loadBenchmarkManifest, loadBuiltinCatalog, loadConfig, loadLatestRun, loadLoopConfig, loadLoopPlugins, loadPinnedSkills, localConfigPath, loopStatus, markProviderExhausted, memoryDigestOf, mergeLoopConfig, modelFor, normalizeReason, openLoopMemory, orcaAccountList, orcaAgentHooks, orcaAutomationCreateArgv, orcaAutomationEditArgv, orcaAutomationRemove, orcaAutomationRun, orcaAutomationRuns, orcaAutomationsList, orcaDiagnosticsMemory, orcaJson, orcaStatus, orcaTerminalCreate, orcaTerminalList, orcaTerminalScreen, orcaTerminalSend, orcaTerminalWait, orcaVersion, orcaWorktreeCreate, orcaWorktreeRemove, orcaWorktreeSet, orcaWorktreeSetArgv, orcaWorktrees, parseAgentRegistryText, parseArtificialAnalysisPayload, parseAutomationRuns, parseContractOutput, parseGrokModelsOutput, parseJsonEnvelope, parseLinearIssueDetail, parseLinearIssues, parseLoopConfigText, parseMemInfo, parseModelRef, parseOrcaAgentHooks, parseOrcaAutomations, parseOrcaSendReceipt, parseOrcaStatus, parseOrcaTerminals, parseOrcaVersion, parseOrcaWorktreeCreate, parseOrcaWorktrees, parseProviderUsage, parsePullRequest, parseRagQueryOutput, parseRetro, parseReviewResult, parseSince, parseTeamMembers, parseUsageWindows, parseVmStat, pauseIssue, planFilePreflight, planMemoryContext, planPhaseProfile, planRun, precheckCommand, precheckDeliver, precheckTick, preferMemoryOverDocBridge, promoteLearnings, promoteLearningsToMemory, promptLocalConfig, providerIdentity, providerSpecs, queueOwner, rankModels, readAaCache, readArtifactFile, readCliModelsCache, readContextSnapshots, readCooldowns, readDeliveryState, readDispatchRecord, readEvidenceTrustStore, readIntake, readIssueFailures, readLearningsLedger, readLoopEvents, readOutcomeProgress, readStagePause, readStoredContract, reconcileRun, recordBenchmarkObservation, recordIssueFailure, recordStageRunResult, recoverEventLogLock, recoveryDelayMs, remainingUsagePercent, renderArtifactMarkdown, renderContractPrompt, renderDebriefMarkdown, renderFindingsForWorker, renderHandoffBrief, renderHeadlessArgv, renderLocalConfig, renderObservabilityMarkdown, renderPinnedSkills, renderRetroMarkdown, renderTuiCommand, renderWorkerBrief, resolveAgentForRole, resolveAlias, resolveCatalogCandidates, resolveDocContext, resumeIssue, resumeStage, resumeStateFromArtifacts, retroLearnings, retryRun, rotationStatePath, routeAllRoles, runAdversarialReview, runAgentEval, runCodeReview, runDeliver, runEvalBattery, runGuidedInstall, runLoopDoctor, runObservability, runRetroStage, runTick, runWithRecovery, runWorkflow, sampleMachine, scanForPii, selectMemoryForPrompt, selectModel, selectRuntime, severityRank, shellQuote, skillDigest, skillRefs, snapshotWatchTargets, stageEntry, stagePausePath, startRun, summarizeMachine, tiersFor, touchesProtectedPaths, transition, undeclaredOrcaProviders, uninstallLoopAutomations, unknownTelemetry, untrusted, upsertProposedLearnings, usageRankTuple, validateAdapterMetadata, validateArtifactEnvelope, validateBenchmarkManifest, validateBlockManifest, validateCacheableOperation, validateCapabilityManifest, validateCompatibilityManifest, validateConfig, validateContextSnapshot, validateContextSnapshots, validateEvalManifest, validateHarnessErrorClassification, validateHarnessEventEnvelope, validateLoopConfig, validateMemoryRecord, validateOptimizationObservation, validatePhaseTelemetry, validateSafeCommand, validateStatusSnapshot, verifyEvidenceBundle, verifyPullRequestApproval, verifyRun, watchDeliveries, worktreeNameFor, writeAaCache, writeCliModelsCache, writeDispatchRecord, writeIdFor, writeLearningsLedger, writeLocalConfig, writeStoredContract };
+export { AGENT_REGISTRY_SCHEMA_VERSION, ARTIFACT_SCHEMA_VERSION, ARTIFACT_TYPES, ASSURANCE_LEVELS, type AdapterMetadata, type AdapterTelemetry, type AdversarialReviewResult, type AgentAdapter, type AgentEvalCase, type AgentEvalReport, type AgentEvalSuite, type AgentMemoryAdapter, type AgentMemoryHit, type AgentMemoryKvStore, type AgentMemoryRecord, type AgentRegistry, type AgentRegistryEntry, AgentRegistryEntrySchema, AgentRegistrySchema, type AgentSessionOptions, type AgentUsage, type ApprovedAssumption, type ArgvRagContextProviderOptions, type ArtifactBinding, type ArtifactEnvelope, type ArtifactEnvelopeInput, type ArtifactType, type ArtificialAnalysisModel, type AssuranceLevel, type AutomationStatus, type AutonomyMode, BENCHMARK_SCHEMA_VERSION, BLOCK_STATUSES, type BenchmarkBinding, type BenchmarkComparison, type BenchmarkImprovementDirection, type BenchmarkManifest, type BenchmarkObservation, type BenchmarkObservationEvidence, type BenchmarkObservationInput, type BenchmarkObservationStatus, type BenchmarkReport, type BenchmarkRun, type BenchmarkSummary, type BenchmarkTask, type BlockAssessment, type BlockManifest, type BlockStatus, CAPABILITY_KINDS, CAPABILITY_MANIFEST_SCHEMA_VERSION, CHECK_CATEGORIES, COMPATIBILITY_COMPONENTS, COMPATIBILITY_SCHEMA_VERSION, CONTEXT_PROVIDER_SLOT, CONTRACT_CLOSE, CONTRACT_OPEN, CONTRACT_SCHEMA_VERSION, type CacheUsage, type CapabilityDescriptor, type CapabilityKind, type CapabilityManifest, type CapabilityManifestInput, type CatalogModel, type ChangedFile, type CheckCategory, type CheckOutcome, type CheckResult, type ChecksAssessment, type ClaimResult, type CodeReviewInput, type CodeReviewOutcome, type CodingAgentAdapter, type CodingAgentHandlerResult, type CodingAgentRequest, type CodingAgentResult, type CommandResult, type CommandRunOptions, type CommandRunner, type CompatibilityComponent, type CompatibilityComponentId, type CompatibilityManifest, type CompatibilityObservation, type CompatibilityReport, type CompatibilityStatus, type ContextProvider, type ContextQuery, type ContextReference, type ContextSnapshot, type ContractAssessment, type ContractOutcome, ContractOutcomeSchema, type ContractScope, type CooldownEntry, type CooldownState, type CoordinationIdentity, type CriterionStatus, type CycleIterationMetrics, type CycleMatrixRow, type CycleStepResult, type CycleStepStatus, type DebriefInput, type DebriefIssueRow, type DebriefReport, type DecisionPacket, type DeliverInput, type DeliverOutcome, type DeliverReport, type DeliverResult, type DeliveryState, type DetectProvidersInput, type DiscoveryAmbiguity, type DiscoveryCurrentInput, type DiscoveryCurrentResult, type DiscoveryDecisionLogEntry, type DiscoveryInput, type DiscoveryOption, type DiscoveryResult, type DispatchLease, type DispatchLedger, type DispatchRecord, type DispatchRecordFile, type Disposer, type DocBridgeIndexInspection, type DockerMount, type DockerRuntimeEvidence, type DockerToolDefinition, type DoctorCheck, type DoctorCheckStatus, EVAL_COMPONENTS, EVAL_LAYERS, EVAL_MANIFEST_SCHEMA_VERSION, EVENT_LOG_GENESIS, EVIDENCE_BUNDLE_SCHEMA_VERSION, EVIDENCE_MAX_FILE_BYTES, EVIDENCE_MAX_TOTAL_BYTES, type EffortLevel, type EvalBatteryReport, type EvalCaseDefinition, type EvalCaseReport, type EvalComponent, type EvalExpectation, type EvalLayer, type EvalManifest, type EvalObservation, type EvalObservationStatus, type EventLogLock, type EventLogLockRecovery, type EventLogLockStatus, type EventLogVerification, type EventStore, type EvidenceArtifact, type EvidenceBundle, type EvidenceBundleFile, type EvidenceBundleSignature, type EvidenceBundleVerification, type EvidenceReference, type ExecutePhaseProfileOptions, type FailureClass, type FailureClassification, type FetchQueueInput, FileArtifactStore, FileEventStore, type FilePreflightPlan, type GateAssessment, type GateBinding, type GateCriterion, type GenerateContractInput, type GitHubCliOptions, type GuidedInstallIO, type GuidedInstallInput, type GuidedInstallReport, HARNESS_ERROR_CATALOG, HARNESS_ERROR_CODES, HARNESS_EVENT_ENVELOPE_SCHEMA_VERSION, HARNESS_EVENT_SCHEMA_VERSION, HARNESS_EVENT_TYPES, HARNESS_PLUGIN_API_VERSION, HARNESS_REPO_URL, type HandoffBriefInput, HarnessError, type HarnessErrorClassification, type HarnessErrorDisposition, type HarnessEvent, type HarnessEventContext, type HarnessEventEnvelope, type HarnessEventEnvelopeInput, type HarnessEventInput, type HarnessEventListener, type HarnessEventPayloads, type HarnessEventProvenance, type HarnessEventType, type HarnessPlugin, type HarnessPluginContext, IMPROVEMENT_CYCLE_STEPS, type ImprovementCycleAssessment, type ImprovementCycleInput, type ImprovementCycleIteration, type ImprovementCycleStep, type InstallAction, type InstallInput, type InstallReport, type IntakeRecord, type IssueFailureRecord, type IssueFailureState, LEARNING_STATUSES, LEGAL_TRANSITIONS, LOOP_CONFIG_FILE, LOOP_CONFIG_SCHEMA_VERSION, LOOP_LOCAL_CONFIG_FILE, LOOP_STAGES, type LearningRecord, type LearningStatus, type LearningsLedger, type LinearIssueDetail, type LinearListInput, type LinearQueueFilter, type LinearWriteOptions, type LlmCache, type LlmCacheKeyInput, type LlmCacheStats, type LoadedConfig, type LoadedLoopConfig, type LocalConfigAnswers, type LocalConfigPrompter, type LoopConfig, type LoopConfigInput, LoopConfigSchema, type LoopDoctorInput, type LoopDoctorReport, type LoopEvent, type LoopEventBus, type LoopEventListener, type LoopEventPayload, type LoopHookListener, type LoopHookName, type LoopHookPayload, type LoopHookResult, type LoopIssue, type LoopPluginModule, type LoopProviderConfig, type LoopStage, type LoopStageName, type LoopState, type LoopStatusReport, MEMORY_SCOPES, MODEL_ROLES, type MachineMetrics, type MachineSample, type MachineThresholds, type McpPolicy, type McpToolBridge, type McpToolBridgeOptions, type McpToolCallInput, type McpToolCallResult, type MemoryContextPlan, type MemoryPromptSelection, type MemoryScope, type MemoryUsage, type MetricStatus, type ModelBinding, type ModelPolicy, type ModelQuality, type ModelReference, type ModelRole, type NormalizedPhaseProfile, type ObservabilityAnomaly, type ObservabilityMetrics, type ObservabilityReport, type ObservabilitySeverity, type ObservabilitySnapshot, type ObservabilityTerminal, type OptimizationComparison, type OptimizationObservation, type OrcaAgentHookState, type OrcaAutomation, type OrcaAutomationSpec, type OrcaCliOptions, type OrcaCreatedWorktree, type OrcaDispatchInput, type OrcaDispatchPlan, type OrcaLeaseState, type OrcaLifecycleInput, type OrcaLifecycleProjection, type OrcaMemorySample, type OrcaStatus as OrcaRuntimeStatus, type OrcaSendReceipt, type OrcaTerminal, type OrcaWorktree, type OutcomeProgress, type OutcomeProgressStatus, PHASE_DECISIONS, PHASE_EFFECTS, PHASE_EFFECT_ACTIONS, PHASE_MODES, PR_FIELDS, type ParallelismUsage, type PhaseAmbiguity, type PhaseContext, type PhaseDecision, type PhaseDecisionPacket, type PhaseDefinition, type PhaseEffect, type PhaseEffectAction, type PhaseEffectPolicy, type PhaseExecution, type PhaseExecutionReport, type PhaseGateEvaluator, type PhaseGateResult, type PhaseHandler, type PhaseHandlerResult, type PhaseMachineMetrics, type PhaseMode, type PhasePreflight, type PhasePreflightResult, type PhaseProfile, type PhaseResumeState, type PhaseRetryPolicy, type PhaseRoutePlan, type PhaseTelemetry, type PhaseTokenMetrics, type PiiKind, type PiiMatch, type PiiScanResult, type PilotAssessment, type PilotEntry, type PilotManifest, type PinnedSkill, type PinnedSkillRef, type PluginContribution, type PluginRegistry, type PluginSlot, type PolicyDecision, type PolicyGate, type PolicyRequest, type PolicyRule, type ProcessToolDefinition, type ProductionEvidence, type ProviderAuthStatus, type ProviderAvailability, type ProviderCatalog, type ProviderFailure, type ProviderSpec, type ProviderUsage, type PullRequestApproval, type PullRequestCheck, type PullRequestDraft, type PullRequestSnapshot, QUALITY_DIMENSIONS, type QaTransitionAssessment, type QualityDimension, type QualityDimensionScore, type QualityMatrix, REVIEW_SEVERITIES, RUN_STATES, type RagContextProviderOptions, type RagQueryResult, type RankedModel, type RecoveryObservation, type RecoveryPolicy, type RecoveryResult, type RepositoryProfile, type ResolvedAgent, type RetroInput, type RetroIssueRow, type RetroReport, type RetroStageReport, type RetroSuggestion, type RetroTarget, type RetroWindow, type ReviewFinding, type ReviewLens, type ReviewSeverity, type ReviewVerdict, type RichIO, type RoutingDecision, type RoutingSkip, type RunOutcome, type RunReconciliation, type RunState, type RuntimeConfig, type RuntimeExperimentCandidate, type RuntimeExperimentResult, STATES, SURFACE_NAMES, type SessionRecorder, type SlotAssessment, type SlotInput, type SourceSnapshot, type StagePauseEntry, type StagePauseState, type StateTransition, type StatusBlock, type StatusSnapshot, type StoredContract, type StructuredEvidence, type SurfaceName, type SurfaceRequirement, type TaskContract, TaskContractSchema, type TeamMember, type TickCandidateResult, type TickInput, type TickOutcome, type TickReport, type TokenUsage, type ToolDefinition, type ToolExecutionRequest, type ToolExecutionResult, type ToolRuntime, type TrackingAdapter, type TrackingConfig, type TrackingTransition, type TrustedEvidenceKey, type UsageMetric, type UsageWindow, type VerificationCheck, type VerificationConfig, type VerificationRun, WIP_STATES, type WatchEvent, type WatchEventKind, type WatchInput, type WatchReport, type WatchTargetSnapshot, type WatchdogBlocker, type WatchdogBudget, type WatchdogResult, type WipAssessment, type WipAssessmentInput, type WipEntry, type WipState, type WorkerBriefInput, type WorkflowNode, type WorkflowResult, activeCooldowns, adaptiveConcurrency, advanceQueueOwner, appendLoopEvent, approveRun, approvedDecision, artifactDigest, artifactFilePath, artifactIsFresh, artifactMarkdownPath, assertHuman, assessAcceptance, assessAgentEval, assessBlock, assessChecks, assessCompatibility, assessContract, assessDiscovery, assessImprovementCycle, assessIntegration, assessObservability, assessPilot, assessPreflight, assessProduction, assessQaTransition, assessSlots, assessWip, assessWorktreeCleanup, atLeast, authStatusFor, authorizeRun, automationName, automationPrompt, automationSpecs, availableMemoryBytes, benchmarkRuns, branchFor, briefPath, buildDebriefReport, buildListIssuesArgv, buildRetroReport, buildReviewArgv, buildSuggestions, busyIssues, cancelRun, classifyFailure, classifyHarnessError, classifyProviderFailure, classifyWatchEvent, classifyWatchPhase, cleanTaskArtifacts, clearIssueFailures, clearProviderCooldown, compareOptimization, compareVersions, composePullRequest, contractIsFresh, contractPath, cooldownPath, cooldownUntil, countRotationBlockingLeases, countRunningWorkers, createArgvRagContextProvider, createArtifactEnvelope, createCapabilityManifest, createCodingAgentAdapter, createCompatibilityManifest, createConfiguredToolRuntime, createDispatchLedger, createDocBridgeContextProvider, createDockerToolRuntime, createEvalManifest, createFileMemoryAdapter, createFileMemoryKvStore, createHarnessEventEnvelope, createInMemoryMemoryAdapter, createKvMemoryAdapter, createLinearTrackingAdapter, createLlmCache, createLlmCacheKey, createLoopEventBus, createMachineMonitor, createMcpToolBridge, createModelPolicy, createOrcaDispatchPlan, createOrcaLifecycleProjection, createPhaseArtifact, createPhaseProfile, createPluginRegistry, createPluginSlot, createPolicyGate, createProcessRunner, createProcessToolRuntime, createPullRequestApproval, createQualityMatrix, createRagContextProvider, createRichIO, createSessionRecorder, createStatusSnapshot, createToolRuntime, createTrackingAdapter, createTrackingTransition, deliveryStatePath, detectProviders, discoverIntake, dispatchRecordPath, evaluateWatchdog, executePhaseProfile, exportEvidenceBundle, extractResetsAt, fetchArtificialAnalysisModels, fetchLinearIssue, fetchLinearQueue, fetchTeamMembers, filterAndOrderQueue, findExecutable, formatWatchEvent, gatherLoopState, generateContract, githubComment, githubCommentArgv, githubCommentExists, githubLabelRemove, githubMerge, githubMergeArgv, githubOpenPullRequests, githubPullRequest, githubPullRequestsForBranch, hasLocalConfig, hashContextSnapshot, hashContextSnapshots, hashMcpArgs, inspectDocBridgeIndex, inspectEventLogLock, installLoopAutomations, installPreflight, intakeIssueId, intakePath, isDiscoveryCurrent, isIssuePaused, isStagePaused, isWsl, issueFailurePath, launchWorkerTerminal, learningToMemoryRecord, learningsPath, linearAttach, linearAttachArgv, linearCommentAdd, linearCommentAddArgv, linearLabelAdd, linearLabelArgv, linearLabelRemove, linearStatusSet, linearStatusSetArgv, listCliModels, listCliModelsCached, listDispatched, listIntake, listPausedIssues, loadAgentRegistry, loadAliases, loadBenchmarkManifest, loadBuiltinCatalog, loadConfig, loadLatestRun, loadLoopConfig, loadLoopPlugins, loadPinnedSkills, localConfigPath, loopStatus, markProviderExhausted, memoryDigestOf, mergeLoopConfig, modelFor, normalizeReason, openLoopMemory, orcaAccountList, orcaAgentHooks, orcaAutomationCreateArgv, orcaAutomationEditArgv, orcaAutomationRemove, orcaAutomationRun, orcaAutomationRuns, orcaAutomationsList, orcaDiagnosticsMemory, orcaJson, orcaStatus, orcaTerminalCreate, orcaTerminalList, orcaTerminalScreen, orcaTerminalSend, orcaTerminalWait, orcaVersion, orcaWorktreeCreate, orcaWorktreeRemove, orcaWorktreeSet, orcaWorktreeSetArgv, orcaWorktrees, parseAgentRegistryText, parseArtificialAnalysisPayload, parseAutomationRuns, parseContractOutput, parseGrokModelsOutput, parseJsonEnvelope, parseLinearIssueDetail, parseLinearIssues, parseLoopConfigText, parseMemInfo, parseModelRef, parseOrcaAgentHooks, parseOrcaAutomations, parseOrcaSendReceipt, parseOrcaStatus, parseOrcaTerminals, parseOrcaVersion, parseOrcaWorktreeCreate, parseOrcaWorktrees, parseProviderUsage, parsePullRequest, parseRagQueryOutput, parseRetro, parseReviewResult, parseSince, parseTeamMembers, parseUsageWindows, parseVmStat, pauseIssue, planFilePreflight, planMemoryContext, planPhaseProfile, planRun, precheckCommand, precheckDeliver, precheckTick, preferMemoryOverDocBridge, promoteLearnings, promoteLearningsToMemory, promptLocalConfig, providerIdentity, providerSpecs, queueOwner, rankModels, readAaCache, readArtifactFile, readCliModelsCache, readContextSnapshots, readCooldowns, readDeliveryState, readDispatchRecord, readEvidenceTrustStore, readIntake, readIssueFailures, readLearningsLedger, readLoopEvents, readOutcomeProgress, readStagePause, readStoredContract, reconcileRun, recordBenchmarkObservation, recordIssueFailure, recordStageRunResult, recoverEventLogLock, recoveryDelayMs, remainingUsagePercent, renderArtifactMarkdown, renderContractPrompt, renderDebriefMarkdown, renderFindingsForWorker, renderHandoffBrief, renderHeadlessArgv, renderLocalConfig, renderObservabilityMarkdown, renderPinnedSkills, renderRetroMarkdown, renderTuiCommand, renderWorkerBrief, resolveAgentForRole, resolveAlias, resolveCatalogCandidates, resolveDocContext, resumeIssue, resumeStage, resumeStateFromArtifacts, retroLearnings, retryRun, rotationStatePath, routeAllRoles, runAdversarialReview, runAgentEval, runCodeReview, runDeliver, runEvalBattery, runGuidedInstall, runLoopDoctor, runObservability, runRetroStage, runTick, runWithRecovery, runWorkflow, sampleMachine, scanForPii, selectMemoryForPrompt, selectModel, selectRuntime, severityRank, shellQuote, skillDigest, skillRefs, snapshotWatchTargets, stageEntry, stagePausePath, startRun, summarizeMachine, tiersFor, touchesProtectedPaths, transition, undeclaredOrcaProviders, uninstallLoopAutomations, unknownTelemetry, untrusted, upsertProposedLearnings, usageRankTuple, validateAdapterMetadata, validateArtifactEnvelope, validateBenchmarkManifest, validateBlockManifest, validateCacheableOperation, validateCapabilityManifest, validateCompatibilityManifest, validateConfig, validateContextSnapshot, validateContextSnapshots, validateEvalManifest, validateHarnessErrorClassification, validateHarnessEventEnvelope, validateLoopConfig, validateMemoryRecord, validateOptimizationObservation, validatePhaseTelemetry, validateSafeCommand, validateStatusSnapshot, verifyEvidenceBundle, verifyPullRequestApproval, verifyRun, watchDeliveries, worktreeNameFor, writeAaCache, writeCliModelsCache, writeDispatchRecord, writeIdFor, writeLearningsLedger, writeLocalConfig, writeStoredContract };

package/dist/index.js

@@ -1,6 +1,6 @@
 import { resolve, dirname, join, relative, isAbsolute, delimiter, basename, extname, sep } from 'path';
 import { createHash, randomUUID, createPrivateKey, createPublicKey, sign, verify } from 'crypto';
-import { existsSync, mkdirSync, openSync, writeSync, appendFileSync, closeSync, unlinkSync, readFileSync, writeFileSync, statSync, readdirSync, mkdtempSync, renameSync, rmSync } from 'fs';
+import { existsSync, mkdirSync, openSync, writeSync, appendFileSync, closeSync, unlinkSync, readFileSync, writeFileSync, statSync, readdirSync, mkdtempSync, renameSync, rmSync, accessSync, constants } from 'fs';
 import { execFile, spawn, execFileSync } from 'child_process';
 import { promisify } from 'util';
 import { cpus, loadavg, freemem, totalmem, tmpdir, release } from 'os';
@@ -1283,9 +1283,9 @@ var createDocBridgeContextProvider = ({ root, indexPath = ".doc-bridge/index.jso
     const started = Date.now();
     const ageBudget = maxAgeHours ?? 0;
     const inspection = ageBudget > 0 ? inspectDocBridgeIndex(root, indexPath, now4()) : null;
-    if (inspection?.error) throw new Error(`Doc Bridge index is unreadable: ${inspection.error}`);
+    if (inspection?.error) fail(`Doc Bridge index is unreadable: ${inspection.error}`, "INVALID_STATE");
     if (inspection?.ageHours !== null && inspection?.ageHours !== void 0 && inspection.ageHours > ageBudget) {
-      throw new Error(`Doc Bridge index is ${inspection.ageHours.toFixed(1)}h old; refresh it before resolving context.`);
+      fail(`Doc Bridge index is ${inspection.ageHours.toFixed(1)}h old; refresh it before resolving context.`, "STALE");
     }
     const document = index(root, indexPath);
     const contentHash = sourceHash(document);
@@ -1306,7 +1306,9 @@ var createDocBridgeContextProvider = ({ root, indexPath = ".doc-bridge/index.jso
 });
 var executable = (path) => {
   try {
-    return statSync(path).isFile();
+    if (!statSync(path).isFile()) return false;
+    accessSync(path, constants.X_OK);
+    return true;
   } catch {
     return false;
   }
@@ -1963,7 +1965,7 @@ var validateIteration = (iteration, index2) => {
     if (typeof result !== "object" || result === null || Array.isArray(result)) return fail(`iterations[${index2}].steps[${stepIndex}] must be an object.`, "INVALID_INPUT");
     if (result.step !== IMPROVEMENT_CYCLE_STEPS[stepIndex]) return fail(`iterations[${index2}].steps[${stepIndex}] must be ${IMPROVEMENT_CYCLE_STEPS[stepIndex]}.`, "INVALID_INPUT");
     if (!["passed", "failed", "blocked", "pending"].includes(result.status)) return fail(`iterations[${index2}].steps[${stepIndex}].status is invalid.`, "INVALID_INPUT");
-    if (result.status !== "passed" && !nonEmpty3(result.reason, `iterations[${index2}].steps[${stepIndex}].reason`)) return fail(`iterations[${index2}].steps[${stepIndex}].reason is required when the step does not pass.`, "INVALID_INPUT");
+    if (result.status !== "passed" && (typeof result.reason !== "string" || !result.reason.trim())) return fail(`iterations[${index2}].steps[${stepIndex}].reason is required when the step does not pass.`, "INVALID_INPUT");
   });
   if (iteration.adjustment !== void 0) nonEmpty3(iteration.adjustment, `iterations[${index2}].adjustment`);
   return { ...iteration, metrics: validateMetrics(iteration.metrics, index2) };
@@ -1977,8 +1979,10 @@ var assessImprovementCycle = (input) => {
   const iterations = input.iterations.map(validateIteration);
   iterations.forEach((iteration, index2) => {
     if (iteration.iteration !== index2 + 1) return fail("iterations must be sequential and start at 1.", "INVALID_INPUT");
-    if (index2 > 0 && iterations[index2 - 1]?.steps.every((step) => step.status === "passed")) return fail("a completed cycle cannot have later iterations.", "INVALID_INPUT");
-    if (index2 < iterations.length - 1 && !iteration.adjustment) return fail(`iterations[${index2}].adjustment is required before repeating.`, "INVALID_INPUT");
+    const isLast = index2 === iterations.length - 1;
+    const iterationComplete = iteration.steps.every((step) => step.status === "passed");
+    if (iterationComplete && !isLast) return fail("a completed cycle cannot have later iterations.", "INVALID_INPUT");
+    if (!isLast && !iteration.adjustment) return fail(`iterations[${index2}].adjustment is required before repeating.`, "INVALID_INPUT");
   });
   const matrix = iterations.map((iteration) => {
     const statuses = Object.fromEntries(iteration.steps.map((step) => [step.step, step.status]));
@@ -3230,7 +3234,6 @@ var createSessionRecorder = ({ stateDir, run, adapter, policy, runtime, sessionI
   const approvals = /* @__PURE__ */ new Map();
   const released = /* @__PURE__ */ new Map();
   const attempts = /* @__PURE__ */ new Map();
-  const executing = /* @__PURE__ */ new Set();
   let ended = false;
   if (resume) {
     const prior = store.read(run.runId).filter((event2) => event2.sessionId === id2);
@@ -3358,30 +3361,27 @@ var createSessionRecorder = ({ stateDir, run, adapter, policy, runtime, sessionI
       const actionId = required9(input.actionId, "actionId");
       const action = pending.get(actionId) ?? fail(`Tool action is not pending: ${actionId}.`, "INVALID_STATE");
       if (action.executionStarted) fail(`Tool action requires human recovery decision: ${actionId}.`, "HUMAN_APPROVAL_REQUIRED");
-      if (executing.has(actionId)) fail(`Tool action is already executing: ${actionId}.`, "INVALID_STATE");
-      executing.add(actionId);
       action.executionStarted = true;
       const attempt = (attempts.get(actionId) ?? 0) + 1;
       attempts.set(actionId, attempt);
       append("tool.execution.started", { actionId, turnId: action.turnId, toolId: action.toolId, attempt });
+      let result;
       try {
-        const result = await runtime.execute({ actionId, turnId: action.turnId, toolId: action.toolId, argumentsHash: action.argumentsHash, arguments: input.arguments });
-        if (result.status === "completed") {
-          complete2({ actionId, resultHash: result.resultHash, durationMs: result.durationMs, runtimeEvidence: result.runtimeEvidence });
-          return result;
-        }
-        if (result.status === "failed") {
-          failAction({ actionId, errorCode: result.errorCode, retryable: result.retryable, durationMs: result.durationMs, runtimeEvidence: result.runtimeEvidence });
-          return result;
-        }
-        return fail("Runtime returned an invalid execution result.", "HARNESS_ERROR");
+        result = await runtime.execute({ actionId, turnId: action.turnId, toolId: action.toolId, argumentsHash: action.argumentsHash, arguments: input.arguments });
       } catch {
-        const result = { status: "failed", errorCode: "RUNTIME_ERROR", retryable: true, durationMs: 0 };
-        if (pending.has(actionId)) failAction({ actionId, ...result });
+        const failure = { status: "failed", errorCode: "RUNTIME_ERROR", retryable: true, durationMs: 0 };
+        if (pending.has(actionId)) failAction({ actionId, ...failure });
+        return failure;
+      }
+      if (result.status === "completed") {
+        complete2({ actionId, resultHash: result.resultHash, durationMs: result.durationMs, runtimeEvidence: result.runtimeEvidence });
         return result;
-      } finally {
-        executing.delete(actionId);
       }
+      if (result.status === "failed") {
+        failAction({ actionId, errorCode: result.errorCode, retryable: result.retryable, durationMs: result.durationMs, runtimeEvidence: result.runtimeEvidence });
+        return result;
+      }
+      return fail("Runtime returned an invalid execution result.", "HARNESS_ERROR");
     },
     end: (status) => {
       open();
@@ -3946,7 +3946,17 @@ var modelFor = (policy, role) => policy.bindings.find((binding2) => binding2.rol
 
 // src/kernel/pii.ts
 var PATTERNS = [
-  { kind: "api-key", regex: /\b(?:sk-[A-Za-z0-9]{16,}|gh[opsu]_[A-Za-z0-9]{20,}|AKIA[0-9A-Z]{16}|xox[baprs]-[A-Za-z0-9-]{10,})\b/g },
+  // PEM key blocks first: large, unambiguous, and must claim their content before any narrower pattern below
+  // could otherwise match a substring inside the base64 body (unlikely, but claimed-range order matters).
+  { kind: "private-key", regex: /-----BEGIN (?:RSA |EC |DSA |OPENSSH |ENCRYPTED |)PRIVATE KEY-----[\s\S]*?-----END (?:RSA |EC |DSA |OPENSSH |ENCRYPTED |)PRIVATE KEY-----/g },
+  // `sk-` body allows `-`/`_` (not just alnum) so a project/scoped key like `sk-proj-...`/`sk-live-...` matches
+  // as one token instead of the hyphen splitting it into a too-short fragment. `github_pat_` (fine-grained PAT)
+  // and `AIza…` (Google API key) are current real-world formats missing from the original list entirely.
+  { kind: "api-key", regex: /\b(?:sk-[A-Za-z0-9_-]{16,}|sk_(?:live|test)_[A-Za-z0-9]{16,}|pk_(?:live|test)_[A-Za-z0-9]{16,}|gh[opsu]_[A-Za-z0-9]{20,}|github_pat_[A-Za-z0-9_]{20,}|AKIA[0-9A-Z]{16}|AIza[A-Za-z0-9_-]{30,}|xox[baprs]-[A-Za-z0-9-]{10,})\b/g },
+  // The AWS *secret* half (as opposed to the `AKIA…` access-key id above) has no recognizable prefix — a bare
+  // 40-char base64-shaped run is too generic to scan for on its own (matches hashes, tokens, arbitrary base64).
+  // Anchoring on the conventional key name it's almost always assigned to/from keeps this pattern high-signal.
+  { kind: "api-key", regex: /\b(?:aws_secret_access_key|AWS_SECRET_ACCESS_KEY|SecretAccessKey)\s*[:=]\s*["']?([A-Za-z0-9/+=]{40})["']?/g },
   { kind: "email", regex: /\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,}\b/g },
   { kind: "credit-card", regex: /\b(?:\d[ -]?){13,16}\b/g },
   { kind: "phone", regex: /\b\+?\d{1,3}?[\s().-]?\(?\d{2,4}\)?[\s.-]?\d{3,4}[\s.-]?\d{4}\b/g }
@@ -4111,7 +4121,9 @@ var exportEvidenceBundle = async ({ configPath, runId, outputPath, privateKeyPat
 `, "utf8");
   return bundle;
 };
-var verifyEvidenceBundle = (path, { trustedKeys = [] } = {}) => {
+var EVIDENCE_MAX_FILE_BYTES = 25 * 1048576;
+var EVIDENCE_MAX_TOTAL_BYTES = 200 * 1048576;
+var verifyEvidenceBundle = (path, { trustedKeys = [], maxFileBytes = EVIDENCE_MAX_FILE_BYTES, maxTotalBytes = EVIDENCE_MAX_TOTAL_BYTES } = {}) => {
   const bundle = parseBundle(path);
   if (bundle.type !== "agentskit-harness-evidence-bundle" || bundle.schemaVersion !== EVIDENCE_BUNDLE_SCHEMA_VERSION || !bundle.runId || !validKeyId(bundle.signerKeyId) || !validDigest(bundle.payloadHash) || bundle.signature?.algorithm !== "ed25519" || bundle.signature.keyId !== bundle.signerKeyId || typeof bundle.signature.publicKeyPem !== "string" || typeof bundle.signature.signatureBase64 !== "string" || !Array.isArray(bundle.files)) fail("Evidence bundle metadata is invalid.", "HARNESS_ERROR");
   if (trustedKeys.length) {
@@ -4121,10 +4133,14 @@ var verifyEvidenceBundle = (path, { trustedKeys = [] } = {}) => {
     if (trusted.publicKeyPem !== bundle.signature.publicKeyPem) fail(`Evidence bundle key does not match trust store: ${bundle.signerKeyId}`, "HARNESS_ERROR");
   }
   const paths = /* @__PURE__ */ new Set();
+  let totalBytes = 0;
   for (const file of bundle.files) {
     if (!file || typeof file.path !== "string" || paths.has(file.path) || !validDigest(file.sha256) || typeof file.contentBase64 !== "string") fail("Evidence bundle file metadata is invalid.", "HARNESS_ERROR");
     paths.add(file.path);
+    if (file.contentBase64.length > Math.ceil(maxFileBytes / 3) * 4) fail(`Evidence bundle file exceeds the maximum allowed size: ${file.path}`, "HARNESS_ERROR");
     const content = Buffer.from(file.contentBase64, "base64");
+    totalBytes += content.length;
+    if (totalBytes > maxTotalBytes) fail("Evidence bundle exceeds the maximum total allowed size.", "HARNESS_ERROR");
     if (sha256(content) !== file.sha256) fail(`Evidence bundle file hash mismatch: ${file.path}`, "HARNESS_ERROR");
   }
   if (!paths.has(`runs/${bundle.runId}/run.json`) || !paths.has(`runs/${bundle.runId}/events.ndjson`)) fail("Evidence bundle is missing the run projection or event log.", "HARNESS_ERROR");
@@ -5877,6 +5893,14 @@ var githubCommentExists = async (runner, input, options = {}) => {
   const list2 = await ghJson(runner, ["api", "--paginate", `repos/${input.repo}/issues/${input.number}/comments`, "--jq", "[.[].body]"], options);
   return Array.isArray(list2) && list2.some((body3) => typeof body3 === "string" && body3.includes(input.marker));
 };
+var writeJsonAtomic = (path, value) => {
+  const dir = dirname(path);
+  mkdirSync(dir, { recursive: true });
+  const tmp = join(dir, `.${basename(path)}.${process.pid}.${Date.now()}.tmp`);
+  writeFileSync(tmp, `${JSON.stringify(value, null, 2)}
+`, "utf8");
+  renameSync(tmp, path);
+};
 var clip = (text7, max) => text7.length <= max ? text7 : `${text7.slice(0, Math.max(0, max - 1))}\u2026`;
 var createFileMemoryKvStore = (dir) => {
   mkdirSync(dir, { recursive: true });
@@ -6086,9 +6110,7 @@ var readStoredContract = (stateDir, identifier) => {
 };
 var writeStoredContract = (stateDir, stored) => {
   const path = contractPath(stateDir, stored.issue);
-  mkdirSync(dirname(path), { recursive: true });
-  writeFileSync(path, `${JSON.stringify(stored, null, 2)}
-`, "utf8");
+  writeJsonAtomic(path, stored);
   return path;
 };
 var contractIsFresh = (stored, issue, reuseHours, now4, memoryDigest) => stored.issueUpdatedAt === issue.updatedAt && (reuseHours === 0 || now4.getTime() - Date.parse(stored.generatedAt) <= reuseHours * 36e5) && (memoryDigest === void 0 || (stored.memoryDigest ?? hashJson([])) === memoryDigest);
@@ -6523,26 +6545,56 @@ var readDispatchRecord = (stateDir, identifier) => {
     return null;
   }
 };
-var writeJson2 = (path, value) => {
-  mkdirSync(dirname(path), { recursive: true });
-  writeFileSync(path, `${JSON.stringify(value, null, 2)}
-`, "utf8");
-};
 var writeDispatchRecord = (stateDir, record3) => {
   const path = dispatchRecordPath(stateDir, record3.issue);
-  writeJson2(path, record3);
+  writeJsonAtomic(path, record3);
   return path;
 };
 var EVENTS_ROTATE_AT_BYTES = 10 * 1024 * 1024;
+var EVENTS_LOCK_STALE_MS = 5e3;
+var EVENTS_LOCK_MAX_ATTEMPTS = 100;
+var EVENTS_LOCK_RETRY_MS = 10;
+var acquireEventsLock = (lockFilePath) => {
+  for (let attempt = 0; attempt < EVENTS_LOCK_MAX_ATTEMPTS; attempt += 1) {
+    try {
+      return openSync(lockFilePath, "wx");
+    } catch (error) {
+      if (error.code !== "EEXIST") throw error;
+      try {
+        if (Date.now() - statSync(lockFilePath).mtimeMs > EVENTS_LOCK_STALE_MS) unlinkSync(lockFilePath);
+      } catch {
+      }
+      Atomics.wait(new Int32Array(new SharedArrayBuffer(4)), 0, 0, EVENTS_LOCK_RETRY_MS);
+    }
+  }
+  return null;
+};
 var appendLoopEvent = (stateDir, event2, bus, now4 = () => /* @__PURE__ */ new Date()) => {
   const path = join(stateDir, "events.ndjson");
   mkdirSync(dirname(path), { recursive: true });
+  const lockFilePath = `${path}.lock`;
+  const lockFd = acquireEventsLock(lockFilePath);
   try {
-    if (statSync(path).size > EVENTS_ROTATE_AT_BYTES) renameSync(path, join(stateDir, `events-archive-${now4().getTime()}.ndjson`));
-  } catch {
-  }
-  appendFileSync(path, `${JSON.stringify(event2)}
+    if (lockFd !== null) {
+      try {
+        if (statSync(path).size > EVENTS_ROTATE_AT_BYTES) renameSync(path, join(stateDir, `events-archive-${now4().getTime()}.ndjson`));
+      } catch {
+      }
+    }
+    appendFileSync(path, `${JSON.stringify(event2)}
 `, "utf8");
+  } finally {
+    if (lockFd !== null) {
+      try {
+        closeSync(lockFd);
+      } catch {
+      }
+      try {
+        unlinkSync(lockFilePath);
+      } catch {
+      }
+    }
+  }
   if (bus && typeof event2["type"] === "string") bus.emit(event2);
 };
 var gatherLoopState = async (input) => {
@@ -6838,7 +6890,7 @@ The loop will not retry this issue until you remove the \`${config.resilience.pa
       if (!launched.accepted) notes.push(`${detail.identifier}: terminal ${launched.terminal} did not confirm the brief; deliver will nudge it if it stays idle`);
       ledger.recordDispatch({ lease: claim.lease, idempotencyKey: plan.idempotencyKey, commandDigest: plan.commandDigest });
       const record3 = { issue: detail.identifier, worktreeId: created.id, worktree, branch: actualBranch, terminal: launched.terminal, provider: builder.provider, model: builder.model, contractDigest: stored.digest, leaseKey: claim.lease.key, leaseId: claim.lease.leaseId, dispatchedAt: now4().toISOString(), url: detail.url, briefDigest, skills: skillRefs(pinnedSkills), setup: setupResult, effort: builder.effort, initialRemainingPercent: builder.remainingPercent, worktreePath: created.path };
-      writeJson2(dispatchRecordPath(loaded.stateDir, detail.identifier), record3);
+      writeJsonAtomic(dispatchRecordPath(loaded.stateDir, detail.identifier), record3);
       appendLoopEvent(loaded.stateDir, { at: record3.dispatchedAt, type: "worker.dispatched", ...record3, command: builder.tui, briefAccepted: launched.accepted, tuiIdle: launched.idle }, bus);
       await bus.runHook("afterDispatch", { issue: detail.identifier, provider: record3.provider, model: record3.model, branch: record3.branch, worktreeId: record3.worktreeId });
       clearIssueFailures(loaded.stateDir, detail.identifier);
@@ -6955,11 +7007,6 @@ var discoverIntake = async (runner, input, options = {}) => {
 // src/loop/deliver.ts
 var message3 = (error) => error instanceof HarnessError ? `${error.code}: ${error.message}` : error instanceof Error ? error.message : String(error);
 var isMissingOrcaWorktree = (error) => message3(error).includes("selector_not_found");
-var writeJson3 = (path, value) => {
-  mkdirSync(dirname(path), { recursive: true });
-  writeFileSync(path, `${JSON.stringify(value, null, 2)}
-`, "utf8");
-};
 var deliveryStatePath = (stateDir, identifier) => join(stateDir, "issues", identifier, "delivery.json");
 var readDeliveryState = (stateDir, identifier) => {
   const path = deliveryStatePath(stateDir, identifier);
@@ -6986,7 +7033,7 @@ var minutesBetween = (later, earlier) => earlier === null ? Number.POSITIVE_INFI
 var orcaOptions = (config) => ({ bin: config.orca.bin, timeoutMs: config.orca.timeoutMs });
 var linearOptions = (config) => ({ bin: config.orca.bin, workspaceId: config.linear.workspaceId, orca: { timeoutMs: config.orca.timeoutMs } });
 var saveState = (ctx, state) => {
-  if (!ctx.dryRun) writeJson3(deliveryStatePath(ctx.loaded.stateDir, state.issue), state);
+  if (!ctx.dryRun) writeJsonAtomic(deliveryStatePath(ctx.loaded.stateDir, state.issue), state);
 };
 var event = (ctx, payload) => {
   if (!ctx.dryRun) appendLoopEvent(ctx.loaded.stateDir, { at: ctx.now().toISOString(), ...payload }, ctx.bus);
@@ -7658,7 +7705,7 @@ var runDeliver = async (input) => {
         const candidates = (await githubOpenPullRequests(input.runner, { repo: config.project.repo, limit: 100 })).filter((item) => item.headRef === record3.branch || item.headRef.endsWith(`/${record3.worktree}`) || item.headRef === record3.worktree);
         if (candidates.length) {
           open = candidates;
-          if (!dryRun) writeJson3(dispatchRecordPath(loaded.stateDir, record3.issue), { ...record3, branch: candidates[0].headRef });
+          if (!dryRun) writeJsonAtomic(dispatchRecordPath(loaded.stateDir, record3.issue), { ...record3, branch: candidates[0].headRef });
           notes.push(`${record3.issue}: PR found on branch ${candidates[0].headRef}; dispatch record updated`);
         }
       }
@@ -8267,7 +8314,7 @@ var parseSince = (value, now4) => {
     return new Date(now4.getTime() - amount * unit);
   }
   const parsed = Date.parse(value);
-  if (Number.isNaN(parsed)) throw new Error(`Unrecognised --since value: ${value} (use 7d, 12h, 30m or an ISO date)`);
+  if (Number.isNaN(parsed)) fail(`Unrecognised --since value: ${value} (use 7d, 12h, 30m or an ISO date)`, "INVALID_INPUT");
   return new Date(parsed);
 };
 var median3 = (values) => {
@@ -8516,6 +8563,7 @@ var phaseOf = (dispatch, delivery) => {
   return "in-flight";
 };
 var summarize2 = (phase2, delivery, dispatch) => {
+  if (phase2 === "idle") return "Not yet dispatched";
   if (phase2 === "merged") return `Merged PR #${delivery.prNumber ?? "?"}`;
   if (phase2 === "held" || phase2 === "held-incomplete-review") {
     if (delivery.heldFor) return `Held for a human (self-edit or protected path at ${delivery.heldFor.slice(0, 7)})`;
@@ -8602,7 +8650,7 @@ var buildDebriefReport = (input) => {
         });
         continue;
       }
-      continue;
+      if (!input.issue) continue;
     }
     rows.push(rowFor({ issue, dispatch, delivery, intent, repo: config.project.repo, now: now4 }));
   }
@@ -8770,7 +8818,7 @@ var runObservability = async (input) => {
   const events = readLoopEvents(loaded.stateDir).filter((event2) => Date.parse(event2.at) >= since.getTime() && Date.parse(event2.at) <= at.getTime());
   const ledger = createDispatchLedger(loaded.stateDir);
   const active = ledger.active();
-  const missingDeliveryIssues = active.filter((lease) => !existsSync(deliveryStatePath(loaded.stateDir, lease.issue))).map((lease) => lease.issue);
+  const missingDeliveryIssues = active.filter((lease) => !existsSync(deliveryStatePath(loaded.stateDir, lease.issue)) && !existsSync(dispatchRecordPath(loaded.stateDir, lease.issue))).map((lease) => lease.issue);
   const records = listDispatched(loaded.stateDir);
   const completed = records.map((record3) => ({ record: record3, state: readDeliveryState(loaded.stateDir, record3.issue) })).filter(({ state }) => state.finishedAt && Date.parse(state.finishedAt) >= since.getTime());
   const leadTimes = completed.map(({ record: record3, state }) => state.finishedAt ? (Date.parse(state.finishedAt) - Date.parse(record3.dispatchedAt)) / 6e4 : null).filter((value) => value !== null && Number.isFinite(value)).sort((a, b) => a - b);