metheus-governance-mcp-cli 0.2.282 → 0.2.283

package/cli.mjs

@@ -25,9 +25,10 @@ import {
   resolveRolePlannerAuditorModelDisplayName,
   resolveRolePlannerModelDisplayName,
   resolveRolePlannerRepairModelDisplayName,
-  resolveResponderAdjudicatorModelDisplayName,
-  resolveGeminiReasoningConfig,
-  suggestLocalAIModelDisplayName,
+  resolveResponderAdjudicatorModelDisplayName,
+  resolveGeminiHeadlessExecutionModel,
+  resolveGeminiReasoningConfig,
+  suggestLocalAIModelDisplayName,
   SUPPORTED_LOCAL_AI_CLIENTS,
   normalizeLocalAIClientName,
   normalizeLocalAIPermissionMode,
@@ -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,
@@ -6497,6 +6502,8 @@ function markRunnerRequestLifecycle({
     || "",
   ).trim().toLowerCase();
   const normalizedOutcome = String(outcome || "").trim().toLowerCase();
+  const normalizedFailureReplyClassification = String(failureReplyClassification || "").trim().toLowerCase();
+  const normalizedFailureFacts = safeObject(failureFacts);
   const shouldRemainRunningAfterReply = authoritativeDecisionBundle.should_close_after_reply === true
     ? false
     : authoritativeDecisionBundle.should_close_after_reply === false
@@ -6511,6 +6518,18 @@ 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 +6566,7 @@ function markRunnerRequestLifecycle({
       || normalizedOutcome === "execution_failed"
       || normalizedOutcome === "policy_violation"
     ) {
-      return "closed";
+      return shouldRemainRunningAfterError ? "running" : "closed";
     }
     return normalizeRunnerRequestStatus(existing.status);
   })();
@@ -19554,12 +19573,13 @@ 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-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-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",

package/lib/selftest-runner-scenarios.mjs

@@ -3245,6 +3245,73 @@ export async function runSelftestRunnerScenarios(push, deps) {
       `status=${String(failedRequest?.status || "(none)")} reason=${String(failedRequest?.closed_reason || "(none)")} closed_at=${String(failedRequest?.closed_at || "(none)")}`,
     );
 
+    saveBotRunnerState({
+      routes: {
+        [requestRouteKey]: {},
+      },
+      sharedInboxes: {},
+      excludedComments: {},
+      requests: {
+        "request-key-2f": {
+          request_key: "request-key-2f",
+          project_id: selftestProjectID,
+          provider: "telegram",
+          chat_id: "-100123",
+          source_message_id: 751,
+          conversation_id: "conv-request-2f",
+          execution_contract_type: "delegation",
+          execution_contract_targets: ["ryoai3_bot"],
+          next_expected_responders: ["ryoai3_bot"],
+          authoritative_decision_bundle: {
+            schema_version: "runner_conversation_decision.v1",
+            decision_type: "reply_outcome",
+            conversation_intent_mode: "delegated_single_lead",
+            allowed_responders: ["ryoai_bot", "ryoai2_bot", "ryoai3_bot"],
+            initial_responders: ["ryoai_bot"],
+            selected_bot_usernames: ["ryoai2_bot"],
+            allow_bot_to_bot: true,
+            execution_contract_type: "delegation",
+            execution_contract_targets: ["ryoai3_bot"],
+            next_expected_responders: ["ryoai3_bot"],
+            should_close_after_reply: false,
+          },
+          decision_bundle_validation_status: "valid",
+          status: "running",
+          claimed_by_route: requestRouteKey,
+        },
+      },
+      consumedComments: {},
+    });
+    const retryableFailedDelegation = markRunnerRequestLifecycle({
+      normalizedRoute: requestRoute,
+      requestKey: "request-key-2f",
+      selectedRecord: {
+        id: "comment-request-finish-2f",
+        parsedArchive: {
+          kind: "bot_reply",
+          chatID: "-100123",
+          messageID: 752,
+          conversationID: "conv-request-2f",
+        },
+      },
+      routeKey: requestRouteKey,
+      outcome: "error",
+      closedReason: "Gemini CLI timed out after 90s while waiting for a model response (No capacity available for model gemini-3.1-pro-preview on the server)",
+      currentBotSelector: "@RyoAI2_bot",
+      failureReplyClassification: "retryable_failure",
+      failureFacts: {
+        retryable: true,
+        error_type: "provider_capacity_exhausted",
+      },
+    });
+    push(
+      "runner_request_lifecycle_retryable_delegated_error_stays_running",
+      String(retryableFailedDelegation?.status || "") === "running"
+        && ensureArray(retryableFailedDelegation?.next_expected_responders).includes("ryoai3_bot")
+        && String(retryableFailedDelegation?.closed_reason || "").trim() === "",
+      `status=${String(retryableFailedDelegation?.status || "(none)")} next=${ensureArray(retryableFailedDelegation?.next_expected_responders).join(",")} closed_reason=${String(retryableFailedDelegation?.closed_reason || "(none)")}`,
+    );
+
     saveBotRunnerState({
       routes: {
         [requestRouteKey]: {},

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "metheus-governance-mcp-cli",
-  "version": "0.2.282",
+  "version": "0.2.283",
   "description": "Metheus Governance MCP CLI (setup + stdio proxy)",
   "type": "module",
   "files": [