metheus-governance-mcp-cli 0.2.282 → 0.2.284

package/cli.mjs

@@ -25,9 +25,10 @@ import {
   resolveRolePlannerAuditorModelDisplayName,
   resolveRolePlannerModelDisplayName,
   resolveRolePlannerRepairModelDisplayName,
-  resolveResponderAdjudicatorModelDisplayName,
-  resolveGeminiReasoningConfig,
-  suggestLocalAIModelDisplayName,
+  resolveResponderAdjudicatorModelDisplayName,
+  resolveGeminiHeadlessExecutionModel,
+  resolveGeminiReasoningConfig,
+  suggestLocalAIModelDisplayName,
   SUPPORTED_LOCAL_AI_CLIENTS,
   normalizeLocalAIClientName,
   normalizeLocalAIPermissionMode,
@@ -144,10 +145,10 @@ import {
   findEarlierProcessableArchiveDuplicate,
   findRecentTelegramMessageEnvelope,
   isTelegramLocalInboundEnvelopeForRoute,
-  isInboundArchiveKind,
-  normalizeTelegramMessageEnvelope as normalizeRunnerTelegramMessageEnvelope,
-  normalizeArchiveCommentRecord,
-  selectPendingArchiveComments,
+  isInboundArchiveKind,
+  normalizeTelegramMessageEnvelope as normalizeRunnerTelegramMessageEnvelope,
+  normalizeArchiveCommentRecord,
+  selectPendingArchiveComments,
   printRunnerResult,
 } from "./lib/runner-helpers.mjs";
 import {
@@ -4144,12 +4145,14 @@ function buildRunnerValidationAndDeliverySummary({
   responseContractValidationStatus = "",
   responseContractValidationReason = "",
   responseContractValidationTargets = [],
-  assignmentValidationStatus = "",
-  assignmentValidationReason = "",
-  assignmentValidationModes = [],
-  deliveryStatus = "",
-  archiveStatus = "",
-  transportError = "",
+  assignmentValidationStatus = "",
+  assignmentValidationReason = "",
+  assignmentValidationModes = [],
+  failureReplyClassification = "",
+  failureFacts = {},
+  deliveryStatus = "",
+  archiveStatus = "",
+  transportError = "",
   archiveError = "",
   sourceMessageEnvelope = {},
   lastReplyMessageEnvelope = {},
@@ -6381,12 +6384,14 @@ function markRunnerRequestLifecycle({
   responseContractValidationStatus = "",
   responseContractValidationReason = "",
   responseContractValidationTargets = [],
-  assignmentValidationStatus = "",
-  assignmentValidationReason = "",
-  assignmentValidationModes = [],
-  deliveryStatus = "",
-  archiveStatus = "",
-  transportError = "",
+  assignmentValidationStatus = "",
+  assignmentValidationReason = "",
+  assignmentValidationModes = [],
+  failureReplyClassification = "",
+  failureFacts = {},
+  deliveryStatus = "",
+  archiveStatus = "",
+  transportError = "",
   archiveError = "",
   lastReplyMessageID = 0,
   lastReplyMessageThreadID = 0,
@@ -6425,10 +6430,7 @@ function markRunnerRequestLifecycle({
   const authoritativeDecisionBundle = resolvedDecisionBundleValidation.ok === true
     ? safeObject(resolvedDecisionBundleValidation.bundle)
     : runnerRequestAuthoritativeDecisionBundle(existing);
-  const effectiveReplyToMessageID = intFromRawAllowZero(
-    replyToMessageID,
-    intFromRawAllowZero(existing.last_reply_to_message_id, 0),
-  );
+  const effectiveReplyToMessageID = intFromRawAllowZero(replyToMessageID, 0);
   const lastReplyMessageEnvelope = buildTelegramBotReplyEnvelope({
     sourceEnvelope: sourceMessageEnvelope,
     chatID: existing.chat_id,
@@ -6449,14 +6451,14 @@ function markRunnerRequestLifecycle({
     senderUsername: normalizedCurrentBotSelector,
     body: aiReplyPreview,
   });
-  const shouldRefreshAttemptedDeliveryEnvelope = (
-    aiReplyGenerated === true
-    || String(aiReplyPreview || "").trim().length > 0
-    || String(deliveryStatus || "").trim().length > 0
-    || String(transportError || "").trim().length > 0
-    || intFromRawAllowZero(replyToMessageID, 0) > 0
-    || intFromRawAllowZero(lastReplyMessageThreadID, 0) > 0
-  );
+  const shouldRefreshAttemptedDeliveryEnvelope = (
+    aiReplyGenerated === true
+    || String(aiReplyPreview || "").trim().length > 0
+    || String(deliveryStatus || "").trim().length > 0
+    || String(transportError || "").trim().length > 0
+    || intFromRawAllowZero(replyToMessageID, 0) > 0
+    || intFromRawAllowZero(lastReplyMessageThreadID, 0) > 0
+  );
   const rootEffectiveExecutionContractTargets = uniqueOrderedStrings(
     [
       ...ensureArray(authoritativeDecisionBundle.execution_contract_targets),
@@ -6497,6 +6499,15 @@ function markRunnerRequestLifecycle({
     || "",
   ).trim().toLowerCase();
   const normalizedOutcome = String(outcome || "").trim().toLowerCase();
+  const normalizedFailureReplyClassification = String(failureReplyClassification || "").trim().toLowerCase();
+  const normalizedFailureFacts = safeObject(failureFacts);
+  const shouldPersistReplyAnchor = (
+    aiReplyGenerated === true
+    || intFromRawAllowZero(lastReplyMessageID, 0) > 0
+    || ["delivered", "dry_run", "archive_error", "failed_transport"].includes(normalizedDeliveryStatus)
+    || String(transportError || "").trim().length > 0
+    || ["replied", "delivery_failed_after_generation"].includes(normalizedOutcome)
+  );
   const shouldRemainRunningAfterReply = authoritativeDecisionBundle.should_close_after_reply === true
     ? false
     : authoritativeDecisionBundle.should_close_after_reply === false
@@ -6511,6 +6522,18 @@ function markRunnerRequestLifecycle({
       || rootEffectiveNextExpectedResponders.length > 0
       || continuationSelectors.length > 0
     );
+  const shouldRemainRunningAfterError = ["error", "execution_failed"].includes(normalizedOutcome)
+    && (
+      normalizedFailureFacts.retryable === true
+      || normalizedFailureReplyClassification === "retryable_failure"
+    )
+    && authoritativeDecisionBundle.should_close_after_reply !== true
+    && (
+      nextExecutionContractType === "delegation"
+      || rootEffectiveExecutionContractTargets.length > 0
+      || rootEffectiveNextExpectedResponders.length > 0
+      || continuationSelectors.length > 0
+    );
   const nextConversationIntentMode = String(
     authoritativeDecisionBundle.conversation_intent_mode
     || conversationIntentMode
@@ -6547,7 +6570,7 @@ function markRunnerRequestLifecycle({
       || normalizedOutcome === "execution_failed"
       || normalizedOutcome === "policy_violation"
     ) {
-      return "closed";
+      return shouldRemainRunningAfterError ? "running" : "closed";
     }
     return normalizeRunnerRequestStatus(existing.status);
   })();
@@ -6888,7 +6911,9 @@ function markRunnerRequestLifecycle({
     last_source_message_thread_id: intFromRawAllowZero(parsed.messageThreadID, 0) || existing.last_source_message_thread_id,
     last_reply_message_id: intFromRawAllowZero(lastReplyMessageID, 0) || existing.last_reply_message_id,
     last_reply_message_thread_id: intFromRawAllowZero(lastReplyMessageThreadID, 0) || existing.last_reply_message_thread_id,
-    last_reply_to_message_id: intFromRawAllowZero(replyToMessageID, 0) || existing.last_reply_to_message_id,
+    last_reply_to_message_id: shouldPersistReplyAnchor
+      ? effectiveReplyToMessageID
+      : existing.last_reply_to_message_id,
     last_reply_message_envelope: persistSuccessfulReplyEnvelope
       ? lastReplyMessageEnvelope
       : safeObject(existing.last_reply_message_envelope),
@@ -19554,12 +19579,13 @@ TELEGRAM_BOT_REVIEW_TOKEN=review-token
     push("runner_tui_frame_renders_route_statuses", false, String(err?.message || err));
   }
 
-  await runSelftestBotCommands(push, {
-    cliPath: fileURLToPath(import.meta.url),
-    parseSimpleEnvText,
-    resolveLocalAIExecutionModel,
-    suggestLocalAIModelDisplayName,
-    resolveGeminiReasoningConfig,
+  await runSelftestBotCommands(push, {
+    cliPath: fileURLToPath(import.meta.url),
+    parseSimpleEnvText,
+    resolveLocalAIExecutionModel,
+    resolveGeminiHeadlessExecutionModel,
+    suggestLocalAIModelDisplayName,
+    resolveGeminiReasoningConfig,
     stripLocalOnlyToolArgs: (requestObj, toolName) =>
       stripLocalOnlyToolArgs(requestObj, toolName),
     applyProxyResponsePatches: (params, deps = buildProxyResponsePipelineDeps()) =>

package/lib/local-ai-adapters.mjs

@@ -19,6 +19,7 @@ const GEMINI_HOME_SYNC_FILES = [
 ];
 const GEMINI_STDIN_BRIDGE_PROMPT = "Use the full task provided on standard input as the authoritative prompt. Follow it exactly and output only the final answer.";
 const GEMINI_CLI_TIMEOUT_MS = 90 * 1000;
+const GEMINI_RUNNER_STABLE_EXECUTION_MODEL = "gemini-3-flash-preview";
 const LOCAL_AI_MODEL_MAPPINGS = {
   gpt: [
     {
@@ -880,7 +881,12 @@ function runLocalAIPromptRawText({
   const normalizedClient = normalizeLocalAIClientName(client);
   const normalizedPermissionMode = normalizeLocalAIPermissionMode(permissionMode);
   const normalizedReasoningEffort = normalizeLocalAIReasoningEffort(reasoningEffort, "low");
-  const resolvedExecutionModel = resolveLocalAIExecutionModel(normalizedClient, model);
+  const resolvedExecutionModel = normalizedClient === "gemini"
+    ? resolveGeminiHeadlessExecutionModel(model, {
+      permissionMode: normalizedPermissionMode,
+      reasoningEffort: normalizedReasoningEffort,
+    })
+    : resolveLocalAIExecutionModel(normalizedClient, model);
   const resolvedWorkspaceDir = ensureWorkspaceDir(workspaceDir);
   const nextEnv = {
     ...process.env,
@@ -1386,6 +1392,23 @@ export function resolveLocalAIExecutionModel(clientName, rawModelValue = "") {
   return match ? String(match.execution || "").trim() : modelValue;
 }
 
+export function resolveGeminiHeadlessExecutionModel(
+  rawModelValue = "",
+  { permissionMode = "read_only", reasoningEffort = "low" } = {},
+) {
+  const resolvedExecutionModel = resolveLocalAIExecutionModel("gemini", rawModelValue);
+  const normalizedExecutionModel = normalizeModelAliasText(resolvedExecutionModel);
+  void normalizeLocalAIPermissionMode(permissionMode);
+  void normalizeLocalAIReasoningEffort(reasoningEffort, "low");
+  if (normalizedExecutionModel !== "auto-gemini-3") {
+    return resolvedExecutionModel;
+  }
+  // Headless runner turns should not depend on Gemini CLI's internal auto-router.
+  // Under heavier prompts it can escalate to capacity-constrained preview models,
+  // which makes one bot path look flaky even though the routing logic is correct.
+  return GEMINI_RUNNER_STABLE_EXECUTION_MODEL;
+}
+
 function buildCodexArgs({ workspaceDir, model, permissionMode, reasoningEffort, outputPath }) {
   const args = ["exec"];
   if (model) {
@@ -1513,7 +1536,10 @@ function buildGeminiThinkingConfig(model, reasoningEffort) {
 }
 
 export function resolveGeminiReasoningConfig(rawModelValue = "", reasoningEffort = "medium") {
-  const executionModel = resolveLocalAIExecutionModel("gemini", rawModelValue);
+  const executionModel = resolveGeminiHeadlessExecutionModel(rawModelValue, {
+    permissionMode: "read_only",
+    reasoningEffort,
+  });
   if (!executionModel) {
     return null;
   }
@@ -3256,7 +3282,12 @@ export function runLocalAIClient({
   const normalizedClient = normalizeLocalAIClientName(client);
   const normalizedPermissionMode = normalizeLocalAIPermissionMode(permissionMode);
   const normalizedReasoningEffort = normalizeLocalAIReasoningEffort(reasoningEffort);
-  const resolvedExecutionModel = resolveLocalAIExecutionModel(normalizedClient, model);
+  const resolvedExecutionModel = normalizedClient === "gemini"
+    ? resolveGeminiHeadlessExecutionModel(model, {
+      permissionMode: normalizedPermissionMode,
+      reasoningEffort: normalizedReasoningEffort,
+    })
+    : resolveLocalAIExecutionModel(normalizedClient, model);
   const resolvedWorkspaceDir = ensureWorkspaceDir(workspaceDir);
   const promptText = buildLocalBotPrompt(inputPayload);
   if (normalizedClient === "sample") {

package/lib/runner-delivery-archive-handoff.mjs

@@ -13,6 +13,14 @@ function intFromRawAllowZero(value, fallback = 0) {
   return Number.isFinite(parsed) ? parsed : fallback;
 }
 
+function buildReplyAnchorMismatchError(expectedReplyToMessageID, observedReplyToMessageID) {
+  return `reply anchor mismatch: expected ${String(expectedReplyToMessageID || 0)}, observed ${String(observedReplyToMessageID || 0)}`;
+}
+
+function buildMessageThreadMismatchError(expectedMessageThreadID, observedMessageThreadID) {
+  return `message thread mismatch: expected ${String(expectedMessageThreadID || 0)}, observed ${String(observedMessageThreadID || 0)}`;
+}
+
 function requireArchiveDependency(deps, key) {
   const candidate = deps?.[key];
   if (typeof candidate !== "function") {
@@ -83,11 +91,41 @@ export async function finalizeLocalBotDeliveryArchive({
       deliveredResult.message_id ?? deliveredBody.message_id ?? deliveredBody.ts,
       0,
     );
-    const deliveredMessageThreadID = intFromRawAllowZero(
+    const observedMessageThreadID = intFromRawAllowZero(
       deliveredResult.message_thread_id ?? deliveredBody.message_thread_id ?? delivery.effectiveMessageThreadID,
       intFromRawAllowZero(messageThreadID, 0),
     );
-    const archiveReplyToMessageID = intFromRawAllowZero(delivery.effectiveReplyToMessageID, replyToMessageID);
+    const expectedMessageThreadID = intFromRawAllowZero(messageThreadID, 0);
+    const expectedReplyToMessageID = intFromRawAllowZero(replyToMessageID, 0);
+    const observedReplyToMessageID = intFromRawAllowZero(delivery.effectiveReplyToMessageID, 0);
+    if (
+      (expectedReplyToMessageID > 0 || observedReplyToMessageID > 0)
+      && observedReplyToMessageID !== expectedReplyToMessageID
+    ) {
+      return {
+        ok: false,
+        error: buildReplyAnchorMismatchError(expectedReplyToMessageID, observedReplyToMessageID),
+        reply_anchor_mismatch: true,
+        expected_reply_to_message_id: expectedReplyToMessageID,
+        observed_reply_to_message_id: observedReplyToMessageID,
+        thread_id: thread.threadID,
+        work_item_id: thread.workItemID,
+      };
+    }
+    if (
+      (expectedMessageThreadID > 0 || observedMessageThreadID > 0)
+      && observedMessageThreadID !== expectedMessageThreadID
+    ) {
+      return {
+        ok: false,
+        error: buildMessageThreadMismatchError(expectedMessageThreadID, observedMessageThreadID),
+        message_thread_mismatch: true,
+        expected_message_thread_id: expectedMessageThreadID,
+        observed_message_thread_id: observedMessageThreadID,
+        thread_id: thread.threadID,
+        work_item_id: thread.workItemID,
+      };
+    }
     if (archiveDedupeOutbound && deliveredMessageID > 0) {
       const existingComments = await listThreadCommentsTail({
         siteBaseURL,
@@ -119,8 +157,8 @@ export async function finalizeLocalBotDeliveryArchive({
       destination,
       replyText: text,
       messageID: deliveredMessageID,
-      messageThreadID: deliveredMessageThreadID,
-      replyToMessageID: archiveReplyToMessageID,
+      messageThreadID: expectedMessageThreadID,
+      replyToMessageID: expectedReplyToMessageID,
       conversation: archiveConversation,
     });
     const createdComment = await createThreadComment({

package/lib/runner-helpers.mjs

@@ -314,6 +314,27 @@ export function buildTelegramMessageEnvelopeFromParsedArchive(parsedArchiveRaw,
   });
 }
 
+export function resolveTelegramReplyAnchorMessageID({
+  replyToMessageID = 0,
+  sourceEnvelope: sourceEnvelopeRaw = {},
+  fallbackReplyToMessageID = 0,
+} = {}) {
+  const explicitReplyToMessageID = intFromRawAllowZero(replyToMessageID, 0);
+  if (explicitReplyToMessageID > 0) {
+    return explicitReplyToMessageID;
+  }
+  const sourceEnvelope = normalizeTelegramMessageEnvelope(sourceEnvelopeRaw);
+  const sourceMessageID = intFromRawAllowZero(sourceEnvelope.message_id, 0);
+  if (sourceMessageID > 0) {
+    return sourceMessageID;
+  }
+  const sourceReplyToMessageID = intFromRawAllowZero(sourceEnvelope.reply_to_message_id, 0);
+  if (sourceReplyToMessageID > 0) {
+    return sourceReplyToMessageID;
+  }
+  return intFromRawAllowZero(fallbackReplyToMessageID, 0);
+}
+
 export function buildTelegramBotReplyEnvelope({
   sourceEnvelope: sourceEnvelopeRaw = {},
   chatID = "",

package/lib/runner-orchestration-failure.mjs

@@ -26,11 +26,14 @@ export function classifyExecutionFailureFacts(detail) {
   const normalizedDetail = String(detail || "").trim();
   const networkReset = /ECONNRESET|socket hang up|read ECONNRESET/i.test(normalizedDetail);
   const networkTimeout = /ETIMEDOUT|http timeout|ECONNABORTED|aborted/i.test(normalizedDetail);
-  const retryable = networkReset || networkTimeout;
+  const providerCapacityExhausted = /MODEL_CAPACITY_EXHAUSTED|RESOURCE_EXHAUSTED|No capacity available for model|rateLimitExceeded/i.test(normalizedDetail);
+  const retryable = networkReset || networkTimeout || providerCapacityExhausted;
   const base = {
     stage: "execution",
     operation: "runner_execution",
-    errorType: retryable
+    errorType: providerCapacityExhausted
+      ? "provider_capacity_exhausted"
+      : retryable
       ? (networkTimeout ? "network_timeout" : "network_reset")
       : "execution_failed",
     retryable,
@@ -42,6 +45,15 @@ export function classifyExecutionFailureFacts(detail) {
   if (!normalizedDetail) {
     return base;
   }
+  if (providerCapacityExhausted) {
+    return {
+      ...base,
+      stage: "provider_call",
+      operation: "local_ai_model_request",
+      errorType: "provider_capacity_exhausted",
+      retryable: true,
+    };
+  }
   if (/permission_mode=read_only|read[_ -]?only/i.test(normalizedDetail)) {
     return {
       ...base,

package/lib/runner-orchestration-selected-record-preparation.mjs

@@ -1,3 +1,7 @@
+import {
+  resolveTelegramReplyAnchorMessageID,
+} from "./runner-helpers.mjs";
+
 function safeObject(value) {
   if (!value || typeof value !== "object" || Array.isArray(value)) {
     return {};
@@ -154,9 +158,20 @@ export function prepareRunnerSelectedRecordIngress({
           currentBotSelector,
         });
   const replyMessageThreadID = intFromRawAllowZero(sourceMessageEnvelope.message_thread_id, 0);
-  const replyToMessageID = intFromRawAllowZero(sourceMessageEnvelope.message_id, 0);
+  const sourceMessageID = intFromRawAllowZero(sourceMessageEnvelope.message_id, 0);
+  const sourceReplyToMessageID = intFromRawAllowZero(sourceMessageEnvelope.reply_to_message_id, 0);
+  const replyToMessageID = resolveTelegramReplyAnchorMessageID({
+    replyToMessageID: sourceMessageID,
+    sourceEnvelope: sourceMessageEnvelope,
+  });
   const replyAnchorSource = String(sourceMessageEnvelope.source_origin || "").trim()
-    || (replyToMessageID > 0 ? "source_message_envelope" : "");
+    || (replyToMessageID > 0
+      ? sourceMessageID > 0
+        ? "source_message_envelope"
+        : sourceReplyToMessageID > 0
+          ? "source_message_envelope_reply_to"
+          : ""
+      : "");
 
   return {
     handledResult: null,

package/lib/runner-orchestration-selected-record-reply-outcome.mjs

@@ -29,6 +29,10 @@ function buildConversationSummaryDetail({
   ].filter(Boolean).join(" | ");
 }
 
+function buildMessageThreadMismatchError(expectedMessageThreadID, observedMessageThreadID) {
+  return `message thread mismatch: expected ${String(expectedMessageThreadID || 0)}, observed ${String(observedMessageThreadID || 0)}`;
+}
+
 function buildSelectedRecordReplyOutcomeBase({
   routeKey,
   normalizedRoute,
@@ -354,11 +358,19 @@ export async function finalizeRunnerSelectedRecordReplyOutcome({
     deliveryBody.result?.message_id ?? deliveryBody.message_id,
     0,
   );
-  const effectiveReplyMessageThreadID = intFromRawAllowZero(
+  const expectedReplyMessageThreadID = intFromRawAllowZero(replyMessageThreadID, 0);
+  const observedReplyMessageThreadID = intFromRawAllowZero(
     deliveryResult?.delivery?.effectiveMessageThreadID,
-    replyMessageThreadID,
+    0,
+  );
+  const messageThreadMismatch = (
+    (expectedReplyMessageThreadID > 0 || observedReplyMessageThreadID > 0)
+    && observedReplyMessageThreadID !== expectedReplyMessageThreadID
   );
-  const effectiveReplyToMessageID = intFromRawAllowZero(deliveryResult?.delivery?.effectiveReplyToMessageID, replyToMessageID);
+  const messageThreadError = messageThreadMismatch
+    ? buildMessageThreadMismatchError(expectedReplyMessageThreadID, observedReplyMessageThreadID)
+    : "";
+  const effectiveReplyToMessageID = intFromRawAllowZero(replyToMessageID, 0);
   const conversationDetail = buildConversationSummaryDetail({
     effectiveConversationContext,
     executionContract,
@@ -371,7 +383,7 @@ export async function finalizeRunnerSelectedRecordReplyOutcome({
       ...buildRunnerRouteStateFromComment(selectedRecord, {
         last_action: "replied",
         last_reply_message_id: replyMessageID,
-        last_reply_message_thread_id: effectiveReplyMessageThreadID,
+        last_reply_message_thread_id: expectedReplyMessageThreadID,
         last_reply_anchor_source: replyAnchorSource,
         last_contract_validation_status: String(responseContractValidation?.status || "").trim(),
         last_contract_validation_reason: String(responseContractValidation?.reason || "").trim(),
@@ -445,17 +457,21 @@ export async function finalizeRunnerSelectedRecordReplyOutcome({
       evidence_ids: ensureArray(aiResult?.evidenceItems).map((item) => String(safeObject(item).id || "").trim()).filter(Boolean),
       evidence_paths: ensureArray(aiResult?.evidenceItems).map((item) => String(safeObject(item).path || "").trim()).filter(Boolean),
       last_reply_message_id: replyMessageID,
-      last_reply_message_thread_id: effectiveReplyMessageThreadID,
+      last_reply_message_thread_id: expectedReplyMessageThreadID,
       reply_to_message_id: effectiveReplyToMessageID,
       last_reply_message_envelope: buildTelegramBotReplyEnvelope({
         sourceEnvelope: sourceMessageEnvelope,
         messageID: replyMessageID,
-        messageThreadID: effectiveReplyMessageThreadID,
+        messageThreadID: expectedReplyMessageThreadID,
         replyToMessageID: effectiveReplyToMessageID,
         sender: bot?.username ? `@${String(bot.username || "").trim().replace(/^@+/, "")}` : String(bot?.name || "bot").trim(),
         senderUsername: normalizeMentionSelector(bot?.username || bot?.name),
         body: sanitizedReplyText,
       }),
+      message_thread_mismatch: messageThreadMismatch,
+      expected_message_thread_id: expectedReplyMessageThreadID || undefined,
+      observed_message_thread_id: observedReplyMessageThreadID || undefined,
+      message_thread_error: messageThreadError,
       reply_anchor_source: replyAnchorSource,
       reply_fallback_used: deliveryResult?.delivery?.replyFallbackUsed === true,
       delivery_status: deliveryResult?.delivery?.dryRun ? "dry_run" : "delivered",

package/lib/runner-recorder-failure-delivery-handoff.mjs

@@ -36,14 +36,8 @@ export function prepareRunnerFailureReplyDeliveryHandoff({
   const sourceMessageEnvelope = Object.keys(safeObject(authoritativeSourceMessageEnvelope)).length > 0
     ? safeObject(authoritativeSourceMessageEnvelope)
     : safeObject(result.source_message_envelope);
-  const replyToMessageID = intFromRawAllowZero(
-    result.reply_to_message_id,
-    intFromRawAllowZero(sourceMessageEnvelope.message_id, 0),
-  );
-  const messageThreadID = intFromRawAllowZero(
-    result.reply_message_thread_id,
-    intFromRawAllowZero(sourceMessageEnvelope.message_thread_id, 0),
-  );
+  const replyToMessageID = intFromRawAllowZero(result.reply_to_message_id, 0);
+  const messageThreadID = intFromRawAllowZero(result.reply_message_thread_id, 0);
   return {
     siteBaseURL: normalizedRuntime.baseURL,
     token: normalizedRuntime.token,

package/lib/runner-recorder-failure-delivery-outcome-handoff.mjs

@@ -11,6 +11,14 @@ function intFromRawAllowZero(value, fallback = 0) {
   return Number.isFinite(parsed) ? parsed : fallback;
 }
 
+function buildReplyAnchorMismatchError(expectedReplyToMessageID, observedReplyToMessageID) {
+  return `reply anchor mismatch: expected ${String(expectedReplyToMessageID || 0)}, observed ${String(observedReplyToMessageID || 0)}`;
+}
+
+function buildMessageThreadMismatchError(expectedMessageThreadID, observedMessageThreadID) {
+  return `message thread mismatch: expected ${String(expectedMessageThreadID || 0)}, observed ${String(observedMessageThreadID || 0)}`;
+}
+
 export function finalizeRunnerFailureReplyDeliveryOutcome({
   processed = {},
   routeKey = "",
@@ -50,6 +58,24 @@ export function finalizeRunnerFailureReplyDeliveryOutcome({
   const delivery = safeObject(normalizedDeliveryResult.delivery);
   const archive = safeObject(normalizedDeliveryResult.archive);
   const handoff = safeObject(deliveryHandoff);
+  const expectedReplyToMessageID = intFromRawAllowZero(handoff.replyToMessageID, 0);
+  const observedReplyToMessageID = intFromRawAllowZero(delivery.effectiveReplyToMessageID, 0);
+  const expectedMessageThreadID = intFromRawAllowZero(handoff.messageThreadID, 0);
+  const observedMessageThreadID = intFromRawAllowZero(delivery.effectiveMessageThreadID, 0);
+  const replyAnchorMismatch = (
+    (expectedReplyToMessageID > 0 || observedReplyToMessageID > 0)
+    && observedReplyToMessageID !== expectedReplyToMessageID
+  );
+  const replyAnchorError = replyAnchorMismatch
+    ? buildReplyAnchorMismatchError(expectedReplyToMessageID, observedReplyToMessageID)
+    : "";
+  const messageThreadMismatch = (
+    (expectedMessageThreadID > 0 || observedMessageThreadID > 0)
+    && observedMessageThreadID !== expectedMessageThreadID
+  );
+  const messageThreadError = messageThreadMismatch
+    ? buildMessageThreadMismatchError(expectedMessageThreadID, observedMessageThreadID)
+    : "";
   const mergedResult = {
     ...result,
     failure_reply_sent: true,
@@ -69,14 +95,16 @@ export function finalizeRunnerFailureReplyDeliveryOutcome({
       safeObject(delivery.body).result?.message_id ?? safeObject(delivery.body).message_id,
       intFromRawAllowZero(result.last_reply_message_id, 0),
     ),
-    last_reply_message_thread_id: intFromRawAllowZero(
-      delivery.effectiveMessageThreadID,
-      intFromRawAllowZero(handoff.messageThreadID, 0),
-    ),
-    reply_to_message_id: intFromRawAllowZero(
-      delivery.effectiveReplyToMessageID,
-      intFromRawAllowZero(handoff.replyToMessageID, 0),
-    ),
+    last_reply_message_thread_id: expectedMessageThreadID,
+    reply_to_message_id: expectedReplyToMessageID,
+    reply_anchor_mismatch: replyAnchorMismatch,
+    expected_reply_to_message_id: expectedReplyToMessageID || undefined,
+    observed_reply_to_message_id: observedReplyToMessageID || undefined,
+    reply_anchor_error: replyAnchorError,
+    message_thread_mismatch: messageThreadMismatch,
+    expected_message_thread_id: expectedMessageThreadID || undefined,
+    observed_message_thread_id: observedMessageThreadID || undefined,
+    message_thread_error: messageThreadError,
   };
   if (normalizedRouteKey && typeof saveRunnerRouteState === "function") {
     const currentRouteState = typeof loadRouteState === "function"

package/lib/runner-recorder-lifecycle-handoff.mjs

@@ -88,6 +88,7 @@ export function buildRunnerProcessedLifecycleInput({
   const processed = safeObject(processedRaw);
   const result = safeObject(processed.result);
   const normalizedOutcome = normalizeRunnerProcessedLifecycleOutcome(processed);
+  const normalizedFailureFacts = safeObject(result.failure_facts);
   return {
     requestKey,
     selectedRecord,
@@ -95,6 +96,8 @@ export function buildRunnerProcessedLifecycleInput({
     outcome: normalizedOutcome,
     closedReason: normalizedOutcome === "skipped"
       ? String(processed.skippedRecord?.reason || result.detail || "skipped").trim() || "skipped"
+      : ["error", "execution_failed", "policy_violation"].includes(normalizedOutcome)
+        ? String(result.detail || "execution_error").trim() || "execution_error"
       : "",
     conversationIDRaw: String(result.conversation_id || "").trim(),
     conversationParticipants: ensureArray(result.conversation_participants),
@@ -126,6 +129,8 @@ export function buildRunnerProcessedLifecycleInput({
     assignmentValidationStatus: String(result.assignment_validation_status || "").trim(),
     assignmentValidationReason: String(result.assignment_validation_reason || "").trim(),
     assignmentValidationModes: ensureArray(result.assignment_validation_modes),
+    failureReplyClassification: String(result.failure_reply_classification || "").trim(),
+    failureFacts: normalizedFailureFacts,
     deliveryStatus: String(result.delivery_status || "").trim(),
     archiveStatus: String(result.archive_status || "").trim(),
     transportError: String(result.transport_error || "").trim(),

package/lib/selftest-bot-commands.mjs

@@ -361,6 +361,7 @@ export async function runSelftestBotCommands(push, deps) {
   const cliPath = String(requireDependency(deps, "cliPath") || "").trim();
   const parseSimpleEnvText = requireDependency(deps, "parseSimpleEnvText");
   const resolveLocalAIExecutionModel = requireDependency(deps, "resolveLocalAIExecutionModel");
+  const resolveGeminiHeadlessExecutionModel = requireDependency(deps, "resolveGeminiHeadlessExecutionModel");
   const suggestLocalAIModelDisplayName = requireDependency(deps, "suggestLocalAIModelDisplayName");
   const resolveGeminiReasoningConfig = requireDependency(deps, "resolveGeminiReasoningConfig");
   const stripLocalOnlyToolArgs = requireDependency(deps, "stripLocalOnlyToolArgs");
@@ -392,6 +393,18 @@ export async function runSelftestBotCommands(push, deps) {
       ].join(" "),
     );
 
+    push(
+      "gemini_headless_runner_uses_explicit_stable_execution_model",
+      resolveGeminiHeadlessExecutionModel("gemini-3.1-pro", {
+        permissionMode: "read_only",
+        reasoningEffort: "low",
+      }) === "gemini-3-flash-preview",
+      `gemini_headless=${resolveGeminiHeadlessExecutionModel("gemini-3.1-pro", {
+        permissionMode: "read_only",
+        reasoningEffort: "low",
+      })}`,
+    );
+
     push(
       "blank_model_defaults_to_first_display_model_for_each_client",
       suggestLocalAIModelDisplayName("gpt", "") === "gpt-5.4"
@@ -409,7 +422,7 @@ export async function runSelftestBotCommands(push, deps) {
     const geminiHighReasoning = resolveGeminiReasoningConfig("gemini-3.1-pro", "high");
     push(
       "gemini_reasoning_effort_maps_to_runtime_settings_override",
-      String(geminiLowReasoning?.model || "") === "auto-gemini-3"
+      String(geminiLowReasoning?.model || "") === "gemini-3-flash-preview"
         && String(safeObject(geminiLowReasoning?.thinkingConfig).thinkingLevel || "") === "LOW"
         && String(safeObject(geminiMediumReasoning?.thinkingConfig).thinkingLevel || "") === "THINKING_LEVEL_UNSPECIFIED"
         && String(safeObject(geminiHighReasoning?.thinkingConfig).thinkingLevel || "") === "HIGH",