dsh-llm-verifier 0.8.0 → 0.8.2

package/lib/index.js

@@ -1,6 +1,6 @@
 import { CRITERIA_PRESETS, CRITERIA_PRESET_IDS, DEFAULT_CRITERIA, DEFAULT_GROUND_TRUTH_NOTE, EMPTY_WORK_BASELINE, GRANULARITY, LETTERS, MAX_DIAGNOSTICS, MAX_DIAGNOSTIC_ACTION_CHARS, MAX_DIAGNOSTIC_CHARS, PROCESS_CRITERIA, PROPOSAL_CRITERIA, SCALE_DESCRIPTION, UNTRUSTED_EVIDENCE_NOTE, accumulatePairs, bradleyTerry, buildFindingContract, buildGenerationPrompt, buildPairwisePrompt, buildProgressPrompt, dedupeCriterionId, diagnosticKey, evidenceNonce, extractProgressScore, extractScore, normalizeScoreLetter, parseCriteriaMarkdown, parseDiagnostics, pivotRoundPairs, rankScores, renderDelimitedBlock, renderDiagnostics, renderReferenceContext, ringCycle, seededRandom, slugCriterionId, swapDiagnosticEvidence, topPivots } from "./core.js";
 import { a as attachUsage, c as emptyUsage, f as partialUsage, g as resolveCapabilityFile, h as TopLogprobCapabilityCache, i as addUsage, m as requestAttempts, o as callVerifier, p as predictScoringChannel, r as RequestLimiter, s as callVerifierText, u as generateCandidate } from "./caller-Bfp4HPK_.js";
-import { A as semanticDecision, B as resolveCacheFile, C as inspectDeliveryPhase, D as nextDiagnosticCycleId, E as latestDirectUserSeq, F as extractSession, I as sanitizeVerifierText, L as sessionEvents, M as semanticRouteHint, N as verificationFailed, O as parseSemanticRoute, P as verificationVerdict, R as ScoreCache, S as estimateRoutedCalls, T as itemBudget, V as stableHash, _ as analyzeStructuredRoute, a as compareRouteFeedbackDetail, b as buildSemanticRoutePrompt, c as inspectUserInteractionPause, d as selectRouteFeedbackDetail, f as sessionAccepted, g as RECOVERY_FAILURE_CONTEXT_CHARS, h as MAX_ROUTED_CHECKPOINTS, i as automaticFeedback, j as semanticReferencesVisible, k as routedRepeats, l as isAwaitingUserText, m as AutoVerifierRouter, n as PROPOSAL_FEEDBACK_NOTE, o as failedAcceptanceCriteria, p as topScoreIndices, r as analyzeAutoTask, s as hasPendingSubagents, t as MAX_ROUTE_FEEDBACK_CHARS, u as isSubagentSession, v as boundDecision, w as inspectRecoverySignal, x as buildSemanticRouteView, y as buildEvidenceIndex, z as SingleFlight } from "./auto-CGbckoaX.js";
+import { A as routedRepeats, B as SingleFlight, C as estimateRoutedCalls, D as latestDirectUserSeq, E as itemBudget, F as verificationVerdict, H as stableHash, I as extractSession, L as sanitizeVerifierText, M as semanticReferencesVisible, N as semanticRouteHint, O as nextDiagnosticCycleId, P as verificationFailed, R as sessionEvents, S as buildSemanticRouteView, T as inspectRecoverySignal, V as resolveCacheFile, _ as RECOVERY_FAILURE_CONTEXT_CHARS, a as compareRouteFeedbackDetail, b as buildEvidenceIndex, c as inspectUserInteractionPause, d as planModeActive, f as selectRouteFeedbackDetail, g as MAX_ROUTED_CHECKPOINTS, h as AutoVerifierRouter, i as automaticFeedback, j as semanticDecision, k as parseSemanticRoute, l as isAwaitingUserText, m as topScoreIndices, n as PROPOSAL_FEEDBACK_NOTE, o as failedAcceptanceCriteria, p as sessionAccepted, r as analyzeAutoTask, s as hasPendingSubagents, t as MAX_ROUTE_FEEDBACK_CHARS, u as isSubagentSession, v as analyzeStructuredRoute, w as inspectDeliveryPhase, x as buildSemanticRoutePrompt, y as boundDecision, z as ScoreCache } from "./auto-DpNooy0d.js";
 import { defineTool } from "@deepseek-ai/dsh-tools";
 import { createUserMessage, isAgentLoopRequest } from "@deepseek-ai/dsh-llm";
 import z from "schemastery";
@@ -6303,6 +6303,7 @@ function apply(ctx, config = {}) {
 		if (!selected.enabled || selected.autoVerifyMode !== "smart") return decision;
 		if (payload.signal.aborted) return decision;
 		if (!selected.autoVerifySubagents && isSubagentSession(payload.agent)) return decision;
+		if (planModeActive(sessionEvents(payload.agent.session))) return decision;
 		if (payload.messages.some((message) => {
 			const kind = message.source?.kind;
 			return kind === "user" || kind === "team-message";
@@ -6335,6 +6336,10 @@ function apply(ctx, config = {}) {
 		const subagentsPending = evidence.pendingSubagents || await hasLiveActiveSubagents(ctx, agent, signal);
 		const userPause = inspectUserInteractionPause(snapshot, evidence.taskStartSeq, turn);
 		const userInteractionPending = evidence.pendingUserInteraction || userPause !== void 0;
+		if (evidence.planMode) {
+			ctx.logger.info?.("llm-verifier automatic session verification skipped: plan mode is active");
+			return;
+		}
 		if (evidence.manualVerificationAccepted) {
 			autoRouter.acceptManual(agent);
 			return;
@@ -7408,6 +7413,6 @@ function apply(ctx, config = {}) {
 	}));
 }
 //#endregion
-export { AutoVerifierRouter, CRITERIA_PRESETS, CRITERIA_PRESET_IDS, Config, DEFAULT_CRITERIA, DEFAULT_GROUND_TRUTH_NOTE, DecisionStore, EMPTY_WORK_BASELINE, GRANULARITY, LETTERS, MAX_DIAGNOSTICS, MAX_DIAGNOSTIC_ACTION_CHARS, MAX_DIAGNOSTIC_CHARS, MAX_ROUTED_CHECKPOINTS, MAX_ROUTE_FEEDBACK_CHARS, MAX_TEAM_TASK_METADATA_CHARS, MAX_WORKSPACE_FILES, MODELS_DEV_URL, PRICE_TTL_MS, PROCESS_CRITERIA, PROPOSAL_CRITERIA, PROPOSAL_FEEDBACK_NOTE, PriceResolver, RECOVERY_FAILURE_CONTEXT_CHARS, RequestLimiter, SCALE_DESCRIPTION, ScoreCache, SingleFlight, StatisticsStore, UNPRICED, UNTRUSTED_EVIDENCE_NOTE, VERIFIER_TOOL_NAMES, VerifierEngine, accumulatePairs, analyzeAutoTask, analyzeStructuredRoute, apply, automaticFeedback, boundCaptureText, boundDecision, boundDecisionCalls, bradleyTerry, buildEvidenceIndex, buildFindingContract, buildGenerationPrompt, buildPairwisePrompt, buildPlanPreReviewPrompt, buildProgressPrompt, buildSemanticRoutePrompt, buildSemanticRouteView, buildTeamTaskVerificationPrompt, callVerifier, compareRouteFeedbackDetail, costUsd, dedupeCriterionId, diagnosticKey, emptyRunStats, errorDetails, estimateRoutedCalls, evidenceNonce, extractProgressScore, extractScore, failedAcceptanceCriteria, hasLiveActiveSubagents, hasPendingSubagents, inject, inspectDeliveryPhase, inspectRecoverySignal, inspectTeamTasks, inspectUserInteractionPause, isAwaitingUserText, isSubagentSession, itemBudget, latestDirectUserSeq, loadCatalogTable, mergeProviderDocument, mergeRunStats, mergeStatisticsOverviews, name, nextDiagnosticCycleId, normalizeCriteria, normalizeScoreLetter, orientRoundPairs, parseCriteriaMarkdown, parseDiagnostics, parseModelsDevDocument, parseSemanticRoute, parseStatisticsQuery, parseVerdictLetter, partialStats, pivotRoundPairs, planFromArguments, priceEntry, priceKey, probeWorkspaceChanges, rankScores, renderDelimitedBlock, renderDiagnostics, renderReferenceContext, renderWorkspaceChanges, resolveCacheFile, resolveCatalogDataDir, resolveDecisionsFile, resolveStatisticsFile, resolveTopicDataDir, ringCycle, routedRepeats, seededRandom, selectRouteFeedbackDetail, semanticDecision, semanticReferencesVisible, semanticRouteHint, sessionAccepted, slugCriterionId, stableHash, summarizeVerdict, swapDiagnosticEvidence, topPivots, topScoreIndices, verificationFailed, verificationVerdict };
+export { AutoVerifierRouter, CRITERIA_PRESETS, CRITERIA_PRESET_IDS, Config, DEFAULT_CRITERIA, DEFAULT_GROUND_TRUTH_NOTE, DecisionStore, EMPTY_WORK_BASELINE, GRANULARITY, LETTERS, MAX_DIAGNOSTICS, MAX_DIAGNOSTIC_ACTION_CHARS, MAX_DIAGNOSTIC_CHARS, MAX_ROUTED_CHECKPOINTS, MAX_ROUTE_FEEDBACK_CHARS, MAX_TEAM_TASK_METADATA_CHARS, MAX_WORKSPACE_FILES, MODELS_DEV_URL, PRICE_TTL_MS, PROCESS_CRITERIA, PROPOSAL_CRITERIA, PROPOSAL_FEEDBACK_NOTE, PriceResolver, RECOVERY_FAILURE_CONTEXT_CHARS, RequestLimiter, SCALE_DESCRIPTION, ScoreCache, SingleFlight, StatisticsStore, UNPRICED, UNTRUSTED_EVIDENCE_NOTE, VERIFIER_TOOL_NAMES, VerifierEngine, accumulatePairs, analyzeAutoTask, analyzeStructuredRoute, apply, automaticFeedback, boundCaptureText, boundDecision, boundDecisionCalls, bradleyTerry, buildEvidenceIndex, buildFindingContract, buildGenerationPrompt, buildPairwisePrompt, buildPlanPreReviewPrompt, buildProgressPrompt, buildSemanticRoutePrompt, buildSemanticRouteView, buildTeamTaskVerificationPrompt, callVerifier, compareRouteFeedbackDetail, costUsd, dedupeCriterionId, diagnosticKey, emptyRunStats, errorDetails, estimateRoutedCalls, evidenceNonce, extractProgressScore, extractScore, failedAcceptanceCriteria, hasLiveActiveSubagents, hasPendingSubagents, inject, inspectDeliveryPhase, inspectRecoverySignal, inspectTeamTasks, inspectUserInteractionPause, isAwaitingUserText, isSubagentSession, itemBudget, latestDirectUserSeq, loadCatalogTable, mergeProviderDocument, mergeRunStats, mergeStatisticsOverviews, name, nextDiagnosticCycleId, normalizeCriteria, normalizeScoreLetter, orientRoundPairs, parseCriteriaMarkdown, parseDiagnostics, parseModelsDevDocument, parseSemanticRoute, parseStatisticsQuery, parseVerdictLetter, partialStats, pivotRoundPairs, planFromArguments, planModeActive, priceEntry, priceKey, probeWorkspaceChanges, rankScores, renderDelimitedBlock, renderDiagnostics, renderReferenceContext, renderWorkspaceChanges, resolveCacheFile, resolveCatalogDataDir, resolveDecisionsFile, resolveStatisticsFile, resolveTopicDataDir, ringCycle, routedRepeats, seededRandom, selectRouteFeedbackDetail, semanticDecision, semanticReferencesVisible, semanticRouteHint, sessionAccepted, slugCriterionId, stableHash, summarizeVerdict, swapDiagnosticEvidence, topPivots, topScoreIndices, verificationFailed, verificationVerdict };
 
 //# sourceMappingURL=index.js.map