nolo-cli 0.19.0 → 0.24.0-alpha.14

package/{chunk-BV4VEE3W.js → chunk-S7AV3NCK.js}

@@ -8,24 +8,26 @@ import {
   writeUsage
 } from "./chunk-DZ3XJVET.js";
 import {
-  getReadableCliDb
-} from "./chunk-WFHXMLHZ.js";
-import {
-  normalizeAgentHandle,
-  readDbRecord,
-  resolveAgentRecordFromHybridStore
-} from "./chunk-DFCFMCY6.js";
+  resolveMachineId
+} from "./chunk-6HYVJV7B.js";
 import {
   agentRunCardLabels,
-  createCliCallAgentToolExecutor,
   createCliLocalRuntimeAdapter,
   getCliLocale,
   isBuiltinNoloAgentRef,
-  isQuotaExhaustedError,
+  isSubtaskRun,
   parseAgentRunEvent,
   t,
   toolLabel
-} from "./chunk-23FZYPLC.js";
+} from "./chunk-PTKI7TIN.js";
+import {
+  getReadableCliDb
+} from "./chunk-WFHXMLHZ.js";
+import {
+  normalizeAgentHandle,
+  readDbRecord,
+  resolveAgentRecordFromHybridStore
+} from "./chunk-J2WT5CEO.js";
 import {
   NOLO_PROJECT_MANAGER_AGENT_KEY,
   isLocalCliAgentKey
@@ -34,12 +36,9 @@ import {
   applyModelLayerOverride,
   buildModelLayerOverride
 } from "./chunk-LTRXGFMS.js";
-import {
-  resolveMachineId
-} from "./chunk-6HYVJV7B.js";
 import {
   formatAgentListCard
-} from "./chunk-EDV7EBOI.js";
+} from "./chunk-GGB6MPHD.js";
 import {
   isGatewayHttpStatus
 } from "./chunk-RQSOAMPJ.js";
@@ -51,23 +50,19 @@ import {
   chooseMemoryOwners,
   loadMemoryCandidatesFromDb,
   resolveAgentMemoryPolicy
-} from "./chunk-IYEXFJPC.js";
+} from "./chunk-GNGOZMU2.js";
 import {
   API_REPORTED_COST_MULTIPLIER,
   normalizeUsage,
   parseUiAskChoiceContent
-} from "./chunk-SCTYAHYK.js";
+} from "./chunk-ADE6577S.js";
 import {
   buildSkillDiscoveryContextBlock
 } from "./chunk-7OBMZCRD.js";
 import {
   parseExternalSkillMarkdown,
   parseSkillDocProtocol
-} from "./chunk-VC6OZXO5.js";
-import {
-  buildPlatformChatCompletionRequest,
-  parsePlatformChatCompletionData
-} from "./chunk-N54HIJVJ.js";
+} from "./chunk-ZWJOGMKH.js";
 import {
   resolveAgentRuntimeConfigFromRecord
 } from "./chunk-OIQPORYI.js";
@@ -76,29 +71,25 @@ import {
   LOCAL_TURN_ABORTED_CODE,
   readActionGate,
   readCommandActionGatePayload
-} from "./chunk-YGW4G5W5.js";
+} from "./chunk-P3RZ6DHN.js";
 import {
   createThinkParserState,
   flushThinkParser,
   processThinkChunk
-} from "./chunk-LSFAWXKC.js";
+} from "./chunk-N7J63Q6C.js";
 import {
-  resolvePlatformAuthToken,
-  resolvePlatformChatCompletionsEndpoint
-} from "./chunk-MHSGJW4Z.js";
+  resolvePlatformAuthToken
+} from "./chunk-PXPISYEH.js";
 import {
   DEFAULT_CONTEXT_WINDOW,
   getModelContextWindow
-} from "./chunk-I2ZJKTG5.js";
+} from "./chunk-PY7VGC5M.js";
 import {
   clipCompactText
 } from "./chunk-5MMAK75Q.js";
 import {
   findModelConfig
-} from "./chunk-UAIVG4J6.js";
-import {
-  toErrorMessage
-} from "./chunk-HNMXYYJR.js";
+} from "./chunk-LEHTC2FI.js";
 import {
   buildMemoryOverlayLayer,
   buildMemoryUseGuidanceLayer
@@ -110,7 +101,7 @@ import {
   createRunActivityTracker,
   finalizeRunRecord,
   spawnLocalBackgroundRun
-} from "./chunk-VNAEBGDK.js";
+} from "./chunk-RZNYL4LW.js";
 import {
   formatFinishedRunCard,
   formatListRunsCard,
@@ -120,10 +111,13 @@ import {
   formatStopRunCard,
   isAgentRunTerminalStatus,
   runShowsLogTail
-} from "./chunk-FQZZLA3C.js";
+} from "./chunk-KXKQO4Q6.js";
 import {
   CliProviderQuotaError
-} from "./chunk-X7W4TNB4.js";
+} from "./chunk-CMKUBIGW.js";
+import {
+  toErrorMessage
+} from "./chunk-HNMXYYJR.js";
 import {
   readPipeText,
   spawnProcess
@@ -136,7 +130,7 @@ import {
 } from "./chunk-Y5GCAMGP.js";
 import {
   createMemoryKey
-} from "./chunk-6Y7A3OQQ.js";
+} from "./chunk-NRY4GCKJ.js";
 import {
   asTrimmedLowercaseString
 } from "./chunk-QSJ3HFEF.js";
@@ -144,7 +138,7 @@ import {
   BUILTIN_NOLO_AGENT_KEY,
   PUBLIC_DEEPSEEK_V4_FLASH_AGENT_KEY,
   PUBLIC_KIMI_K26_IMAGE_AGENT_KEY
-} from "./chunk-3ND52WIL.js";
+} from "./chunk-ATWUOTJP.js";
 import {
   asTrimmedString
 } from "./chunk-3EUW3WEO.js";
@@ -155,97 +149,6 @@ import {
   asOptionalTrimmedString
 } from "./chunk-G3SUCT4W.js";
 
-// packages/agent-runtime/quickChatIntentCore.ts
-var INTENT_MODEL = "deepseek-v4-flash";
-var INTENT_PROVIDER = "nolo";
-var QUICK_CHAT_INTENT_TIMEOUT_MS = 4e3;
-var SHORT_GREETING_MAX_LENGTH = 20;
-var SHORT_GREETING_PATTERN = /^(?:hi|hello|hey|hiya|yo|sup|howdy|嗨|哈喽|哈囉|你好|您好|在吗|在嘛|喂|早上好|中午好|下午好|晚上好|拜拜|再见|谢谢|thanks|thx|ok|好的|收到|了解)[!.。，、~～]?$/iu;
-var VALID_QUICK_CHAT_SKILLS = /* @__PURE__ */ new Set([
-  "table",
-  "doc",
-  "code",
-  "pagebuilder"
-]);
-function isShortGreeting(text) {
-  const trimmed = text.trim();
-  if (!trimmed) return false;
-  if (trimmed.length > SHORT_GREETING_MAX_LENGTH) return false;
-  return SHORT_GREETING_PATTERN.test(trimmed);
-}
-var COMPLEXITY_SIMPLE_MAX_LENGTH = 80;
-var COMPLEXITY_MEDIUM_MAX_LENGTH = 500;
-var COMPLEXITY_KEYWORDS = /为什么|如何|分析|对比|比较|设计|架构|实现|原理|推导|评价|区别|关系|影响|优化|重构/;
-var COMPLEXITY_CODE_BLOCK = /```[\s\S]*```/;
-function estimateComplexity(text) {
-  const trimmed = text.trim();
-  const len = trimmed.length;
-  if (len > COMPLEXITY_MEDIUM_MAX_LENGTH) return "complex";
-  if (COMPLEXITY_CODE_BLOCK.test(trimmed)) return "complex";
-  if (len > COMPLEXITY_SIMPLE_MAX_LENGTH) return "medium";
-  if (COMPLEXITY_KEYWORDS.test(trimmed)) return "medium";
-  return "simple";
-}
-var TIER_DESCRIPTIONS = {
-  flash: "\u5FEB\u901F\u7B80\u5355\uFF1A\u95EE\u5019\u3001\u95F2\u804A\u3001\u5FEB\u901F\u95EE\u7B54\u3001\u7FFB\u8BD1\u3001\u77ED\u6587",
-  balanced: "\u5E73\u8861\u63A8\u7406\uFF1A\u5206\u6790\u3001\u4E2D\u7B49\u957F\u5EA6\u5199\u4F5C\u3001\u4EE3\u7801\u3001\u63A8\u7406\uFF08\u5F53\u524D\u8DEF\u7531\u81F3 DeepSeek V4 Flash\uFF09",
-  quality: "\u9AD8\u8D28\u91CF\u6DF1\u5EA6\uFF1A\u590D\u6742\u63A8\u7406\u3001\u957F\u6587\u3001\u67B6\u6784\u8BBE\u8BA1\u3001\u6DF1\u5EA6\u5206\u6790\uFF08\u5F53\u524D\u8DEF\u7531\u81F3 DeepSeek V4 Flash\uFF09"
-};
-function buildQuickChatIntentSystemPrompt(tierAgents) {
-  const agentList = tierAgents.map((t2) => `  - "${t2.agentKey}"\uFF1A${t2.description}`).join("\n");
-  return `\u4F60\u662F\u4E00\u4E2A\u610F\u56FE\u8DEF\u7531\u5668\u3002\u6839\u636E\u7528\u6237\u6D88\u606F\uFF0C\u5224\u65AD\u5E94\u8BE5\u4EA4\u7ED9\u54EA\u4E2A agent \u5904\u7406\u3002
-
-\u53EF\u9009 agent \u5217\u8868\uFF08agentKey \u4E0E\u63CF\u8FF0\uFF09\uFF1A
-${agentList}
-
-\u5224\u65AD\u539F\u5219\uFF08\u9ED8\u8BA4\u504F\u5411\u901A\u7528\u6863\u4F4D\uFF1B\u4E13\u804C agent \u5728\u660E\u786E\u5339\u914D\u65F6\u4F7F\u7528\uFF1B**\u7F16\u7801/\u672C\u5730\u4ED3\u5E93\u4EFB\u52A1\u8D70\u901A\u7528\u6863 balanced/quality**\uFF09\uFF1A
-- \u9ED8\u8BA4\u60C5\u51B5\u4E0B\uFF0C**\u9664\u975E\u7528\u6237\u660E\u786E\u8868\u8FBE\u8981\u5EFA\u5E94\u7528/\u5EFA Agent/\u63D0\u4EA4\u53CD\u9988/\u591A Agent \u7F16\u6392**\uFF0C\u5426\u5219\u9009\u901A\u7528\u6863\u4F4D\uFF08flash/balanced/quality\uFF09\uFF1A
-  - \u7B80\u5355\u95EE\u5019\u3001\u95F2\u804A\u3001\u5FEB\u901F\u95EE\u7B54\u3001\u7FFB\u8BD1\u3001\u77ED\u6587 \u2192 \u9009 flash \u6863
-  - \u9700\u8981\u63A8\u7406\u3001\u5206\u6790\u3001\u4E2D\u7B49\u957F\u5EA6\u5199\u4F5C\u3001\u666E\u901A\u4EE3\u7801\u89E3\u91CA/\u67B6\u6784\u95EE\u7B54 \u2192 \u9009 balanced \u6863
-  - \u590D\u6742\u63A8\u7406\u3001\u957F\u6587\u3001\u67B6\u6784\u8BBE\u8BA1\u3001\u6DF1\u5EA6\u5206\u6790 \u2192 \u9009 quality \u6863
-  - **\u660E\u786E\u7F16\u7801\u4EFB\u52A1\uFF08\u5B9E\u73B0\u529F\u80FD\u3001\u4FEE bug\u3001refactor\u3001\u5199/\u8865\u6D4B\u8BD5\u3001\u8DD1 build/tests/ci\uFF09\u6216\u67E5\u770B\u672C\u673A\u4ED3\u5E93/\u4EE3\u7801\u72B6\u6001\uFF08git/\u5206\u652F/commit/\u672C\u5730\u4E0E\u8FDC\u7A0B\u5DEE\u8DDD/\u672A\u63D0\u4EA4\u6539\u52A8\u7B49\uFF09** \u2192 \u9009 balanced \u6216 quality \u6863\uFF08\u6309\u590D\u6742\u5EA6\uFF1B**\u4E0D\u8981\u9009 flash**\uFF09
-- \u4E13\u804C agent \u8DEF\u7531\u89C4\u5219\uFF08\u6709\u660E\u786E\u5339\u914D\u65F6\u4F7F\u7528\uFF09\uFF1A
-  - \u8981\u63D0\u4EA4/\u8BB0\u5F55 bug\u3001\u4F53\u9A8C\u95EE\u9898\u3001\u6570\u636E\u5F02\u5E38\u3001\u4EA7\u54C1\u5EFA\u8BAE\uFF0C\u6216\u53E3\u8BED\u5982\u300C\u60F3\u53CD\u9988\u4E00\u4E9B\u95EE\u9898\u300D\u300C\u53CD\u9988\u4E00\u4E0B\u300D\u2192 \u9009\u610F\u89C1\u53CD\u9988\u4E13\u804C agent
-  - \u8981\u65B0\u5EFA\u6216\u5B9A\u5236 AI/Agent/\u667A\u80FD\u4F53\u52A9\u624B\uFF08\u542B\u914D\u7F6E\u5DE5\u5177\u3001\u77E5\u8BC6\u3001\u786E\u8BA4\u521B\u5EFA\uFF09\u2192 \u9009\u521B\u5EFA Agent \u4E13\u804C agent
-  - \u8981\u505A/\u6539/\u53D1\u5E03\u7F51\u7AD9\u3001\u7F51\u9875\u3001\u5C0F\u5E94\u7528\u3001\u9884\u7EA6\u9875\u3001\u535A\u5BA2\u3001\u770B\u677F \u2192 \u9009\u5E94\u7528\u6784\u5EFA\u4E13\u804C agent
-  - \u8981\u6D3E\u591A\u4E2A agent\u3001\u591A agent \u5E76\u884C\u3001\u62C6\u6210\u4EFB\u52A1\u540E\u53F0\u6267\u884C\u3001\u591A\u4E2A agent review\u3001\u8BA9\u4E0D\u540C agent \u8BA8\u8BBA/\u4F1A\u5546 \u2192 \u9009\u591A Agent \u7F16\u6392\u4E13\u804C agent\uFF08\u663E\u5F0F multi-agent \u4F18\u5148\uFF09
-- **\u62FF\u4E0D\u51C6\u65F6**\uFF1A\u7EAF\u95F2\u804A\u9009 flash\uFF1B**\u82E5\u6CBE\u8FB9\u672C\u5730\u4ED3\u5E93/\u5F53\u524D\u4EE3\u7801\u72B6\u6001\uFF0C\u9009 balanced/quality**\u3002
-
-\u5728\u7ED9\u51FA agentKey \u4E4B\u524D\uFF0C\u5148\u5728\u5FC3\u91CC\u533A\u5206\u5019\u9009\u6863\u4F4D\u7684\u8FB9\u754C\u518D\u7ED9\u7F6E\u4FE1\u5EA6\uFF1Abalanced \u4E0E quality \u7684\u8FB9\u754C\u5728\u4E8E\u590D\u6742\u5EA6/\u7BC7\u5E45/\u662F\u5426\u9700\u8981\u6DF1\u5EA6\u5206\u6790\uFF0C\u62FF\u4E0D\u51C6\u5C31\u7ED9\u4F4E confidence\u3002
-
-\u540C\u65F6\u5224\u65AD\u672C\u8F6E\u6D88\u606F\u662F\u5426\u9700\u8981\u6302\u8F7D\u5BF9\u8C61\u64CD\u4F5C\u6280\u80FD\uFF08skills \u6570\u7EC4\uFF0C\u672A\u547D\u4E2D\u7ED9 []\uFF09\uFF1A
-- \u7528\u6237\u8981\u65B0\u5EFA/\u6574\u7406/\u5F55\u5165/\u6279\u91CF\u5904\u7406\u8868\u683C\u6570\u636E\uFF08\u5EFA\u8868\u3001\u52A0\u884C\u3001\u6539\u5217\u3001\u6309\u6761\u4EF6\u6574\u7406\u6570\u636E\uFF09\u2192 skills \u542B "table"
-- \u7528\u6237\u8981\u5199/\u6DA6\u8272/\u6539\u5199/\u7EED\u5199/\u6392\u7248\u4E00\u7BC7\u6587\u6863\u6216\u6587\u7AE0 \u2192 skills \u542B "doc"
-- \u7528\u6237\u8981\u5199/\u6539/\u5B9E\u73B0/\u4FEE bug/refactor/\u8865\u6D4B\u8BD5/\u5BA1\u67E5\u5E76\u4FEE\u6539\u4EE3\u7801 \u2192 skills \u542B "code"\uFF08\u4EC5\u95EE\u901A\u7528\u77E5\u8BC6\u3001\u7EAF\u95F2\u804A\u3001\u53EA\u67E5\u770B git/\u4ED3\u5E93\u72B6\u6001\u800C\u4E0D\u4FEE\u6539\u4EE3\u7801\u65F6\u4E0D\u6302 "code"\uFF09
-- \u7528\u6237\u60F3\u8981\u4E00\u4E2A\u53EF\u4EA4\u4E92\u7684\u53EF\u89C6\u5316\u9875\u9762/\u62A5\u544A/\u770B\u677F/\u4E3B\u9875/\u6559\u7A0B/\u5BF9\u6BD4\u9875/\u8BA1\u5212\u9875\uFF08\u5982"\u505A\u4E2A\u65E5\u62A5""\u505A\u4E00\u4E2A\u4E2A\u4EBA\u4E3B\u9875""\u505A\u4E2A\u9500\u552E\u770B\u677F""\u505A\u4E2A\u65B9\u6848\u5BF9\u6BD4\u9875""\u505A\u4E2A\u5B66\u4E60\u8BA1\u5212\u9875"\uFF09\u2192 skills \u542B "pagebuilder"
-- \u666E\u901A\u95EE\u7B54/\u7EAF\u95F2\u804A/\u5EFA\u5E94\u7528 \u2192 skills \u4E3A []
-
-\u53EA\u8F93\u51FA JSON\uFF0C\u4E0D\u8981\u8F93\u51FA\u5176\u4ED6\u5185\u5BB9\uFF1A
-{"confidence": <0\u52301\u4E4B\u95F4\u7684\u5C0F\u6570\uFF0C\u8868\u793A\u4F60\u5BF9\u8FD9\u4E2A\u5206\u7C7B\u5224\u65AD\u7684\u628A\u63E1\u7A0B\u5EA6>, "agentKey": "<\u4ECE\u5217\u8868\u4E2D\u9009\u62E9\u7684 agentKey>", "skills": <\u5B57\u7B26\u4E32\u6570\u7EC4\uFF0C\u5143\u7D20\u53EA\u80FD\u662F ${[...VALID_QUICK_CHAT_SKILLS].map((s) => `"${s}"`).join(" | ")}\uFF0C\u672A\u547D\u4E2D\u7ED9 []>}`;
-}
-function parseQuickChatIntentResult(content, tierAgents) {
-  try {
-    const parsed = JSON.parse(content.trim());
-    if (typeof parsed !== "object" || parsed === null || !("agentKey" in parsed)) return null;
-    const key = parsed.agentKey;
-    if (typeof key !== "string") return null;
-    if (!tierAgents.some((t2) => t2.agentKey === key)) return null;
-    const rawConfidence = parsed.confidence;
-    const confidence = typeof rawConfidence === "number" && Number.isFinite(rawConfidence) && rawConfidence >= 0 && rawConfidence <= 1 ? rawConfidence : void 0;
-    const rawSkills = parsed.skills;
-    let skills;
-    if (Array.isArray(rawSkills)) {
-      const filtered = [...new Set(rawSkills)].filter(
-        (s) => typeof s === "string" && VALID_QUICK_CHAT_SKILLS.has(s)
-      );
-      if (filtered.length > 0) skills = filtered;
-    }
-    return { agentKey: key, confidence, skills };
-  } catch {
-    return null;
-  }
-}
-
 // packages/cli/client/autoModelRouter.ts
 var CLI_AUTO_TIER_AGENT_KEYS = {
   flash: PUBLIC_DEEPSEEK_V4_FLASH_AGENT_KEY,
@@ -260,9 +163,7 @@ var CLI_AUTO_TIER_MODELS = {
 var CLI_IMAGE_AGENT_KEY = PUBLIC_KIMI_K26_IMAGE_AGENT_KEY;
 var CLI_IMAGE_AGENT_MODEL = "kimi-k2.6";
 var CLI_AUTO_TIER_AGENT_KEY_TABLE = {
-  [CLI_AUTO_TIER_AGENT_KEYS.flash]: true,
-  [CLI_AUTO_TIER_AGENT_KEYS.balanced]: true,
-  [CLI_AUTO_TIER_AGENT_KEYS.quality]: true
+  [CLI_AUTO_TIER_AGENT_KEYS.flash]: true
 };
 function resolveCliAutoAgentModel(agentKey) {
   if (agentKey === CLI_AUTO_TIER_AGENT_KEYS.flash) return CLI_AUTO_TIER_MODELS.flash;
@@ -275,90 +176,15 @@ function resolveCliAutoAgentModel(agentKey) {
   if (agentKey === CLI_IMAGE_AGENT_KEY) return CLI_IMAGE_AGENT_MODEL;
   return void 0;
 }
-var buildRouteOptions = () => ["flash", "balanced", "quality"].map((tier) => ({
-  tier,
-  agentKey: CLI_AUTO_TIER_AGENT_KEYS[tier],
-  description: TIER_DESCRIPTIONS[tier]
-}));
-var tierForAgentKey = (agentKey) => {
-  for (const tier of ["flash", "balanced", "quality"]) {
-    if (CLI_AUTO_TIER_AGENT_KEYS[tier] === agentKey) return tier;
-  }
-  return null;
-};
-function resolveCliAutoFallbackTier(text) {
-  switch (estimateComplexity(text)) {
-    case "complex":
-      return "quality";
-    case "medium":
-      return "balanced";
-    case "simple":
-    default:
-      return "flash";
-  }
-}
-async function classifyCliAutoRoute(text, options) {
-  const timeoutMs = options.timeoutMs ?? QUICK_CHAT_INTENT_TIMEOUT_MS;
-  const fallbackTier = resolveCliAutoFallbackTier(text);
-  const fallback = () => ({
-    agentKey: CLI_AUTO_TIER_AGENT_KEYS[fallbackTier],
-    tier: fallbackTier,
-    classified: false
-  });
-  if (!text.trim()) return fallback();
-  if (isShortGreeting(text)) {
-    return {
-      agentKey: CLI_AUTO_TIER_AGENT_KEYS.flash,
-      tier: "flash",
-      classified: true
-    };
-  }
-  if (!options.authToken) return fallback();
-  const fetchImpl = options.fetchImpl ?? globalThis.fetch;
-  const controller = new AbortController();
-  const timeoutHandle = setTimeout(() => controller.abort(), timeoutMs);
-  try {
-    const routeOptions = buildRouteOptions();
-    const messages = [
-      { role: "system", content: buildQuickChatIntentSystemPrompt(routeOptions) },
-      { role: "user", content: text }
-    ];
-    const { url, init } = buildPlatformChatCompletionRequest({
-      providerConfig: {
-        serverUrl: options.serverUrl,
-        authToken: options.authToken,
-        agentKey: CLI_AUTO_TIER_AGENT_KEYS.flash,
-        model: INTENT_MODEL,
-        provider: INTENT_PROVIDER,
-        endpoint: resolvePlatformChatCompletionsEndpoint(INTENT_PROVIDER) ?? "",
-        requestOptions: {}
-      },
-      messages,
-      stream: false
-    });
-    const response = await fetchImpl(url, {
-      ...init,
-      signal: controller.signal
-    });
-    if (!response.ok) return fallback();
-    const data = parsePlatformChatCompletionData(await response.text());
-    const content = data?.choices?.[0]?.message?.content;
-    if (typeof content !== "string" || !content.trim()) return fallback();
-    const parsed = parseQuickChatIntentResult(content, routeOptions);
-    if (!parsed) return fallback();
-    const tier = tierForAgentKey(parsed.agentKey) ?? fallbackTier;
-    return {
-      agentKey: parsed.agentKey,
-      tier,
-      classified: true,
-      confidence: parsed.confidence,
-      skills: parsed.skills
-    };
-  } catch {
-    return fallback();
-  } finally {
-    clearTimeout(timeoutHandle);
+function classifyCliAutoRoute(_text, options) {
+  if (options.hasImages) {
+    return { agentKey: CLI_IMAGE_AGENT_KEY, tier: "image", classified: true };
   }
+  return {
+    agentKey: CLI_AUTO_TIER_AGENT_KEYS.flash,
+    tier: "flash",
+    classified: true
+  };
 }
 
 // packages/cli/client/tokenUsage.ts
@@ -459,9 +285,20 @@ function formatUsage(usage, dialogId) {
   const parts = [];
   if (typeof dialogId === "string" && dialogId)
     parts.push(`dialog=${dialogId}`);
-  const input = usage?.input_tokens ?? usage?.prompt_tokens ?? 0;
-  const output = usage?.output_tokens ?? usage?.completion_tokens ?? 0;
-  if (input || output) parts.push(`tokens=${input}+${output}`);
+  if (usage && typeof usage === "object") {
+    const normalized = normalizeUsage(usage);
+    const input = Number(normalized.input_tokens || usage?.input_tokens || usage?.prompt_tokens || 0);
+    const output = Number(normalized.output_tokens || usage?.output_tokens || usage?.completion_tokens || 0);
+    const cacheHit = Number(normalized.cache_read_input_tokens || 0);
+    if (input || output) {
+      let tokenStr = `tokens=${input}+${output}`;
+      if (cacheHit > 0 && input > 0) {
+        const hitPercent = (cacheHit / (input + cacheHit) * 100).toFixed(1).replace(/\.0$/, "");
+        tokenStr += ` (cache: ${formatTokenCount(cacheHit)} / ${hitPercent}%)`;
+      }
+      parts.push(tokenStr);
+    }
+  }
   return parts.length ? `  (${parts.join("  ")})` : "";
 }
 
@@ -706,7 +543,7 @@ function wrapTranscriptLine(line, columns, hangingIndent = "") {
     }
     const openingStyles = [...activeStyles];
     const isContinuation = result.length > 0 && hangingIndent.length > 0;
-    const indentWidth = isContinuation ? displayWidth(hangingIndent) : 0;
+    const indentWidth = isContinuation ? visibleWidth(hangingIndent) : 0;
     const maxSegmentWidth = Math.max(1, columns - indentWidth);
     let width = 0;
     let end = start;
@@ -808,13 +645,17 @@ var THEME_PALETTES = {
       // denser slate — 687584 read too blue in dense prose (inline code/tool labels); 5E6A78 pulls blue down while staying darker than the old value
     },
     dark: {
-      accent: { hex: "89B4FA", ansiFallback: "\x1B[34m" },
-      // blue
+      accent: { hex: "5BA3D9", ansiFallback: "\x1B[34m" },
+      // trail dark primary (was wrongly copied from catppuccin 89B4FA)
       chrome: { hex: "6C7086", ansiFallback: "\x1B[90m" },
-      success: { hex: "A6E3A1", ansiFallback: "\x1B[32m" },
-      warning: { hex: "F9E2AF", ansiFallback: "\x1B[33m" },
-      info: { hex: "94E2D5", ansiFallback: "\x1B[36m" },
-      danger: { hex: "F38BA8", ansiFallback: "\x1B[31m" },
+      success: { hex: "5CB87A", ansiFallback: "\x1B[32m" },
+      // trail dark success (was A6E3A1)
+      warning: { hex: "E6B35C", ansiFallback: "\x1B[33m" },
+      // trail dark warning (was F9E2AF)
+      info: { hex: "6BB8E8", ansiFallback: "\x1B[36m" },
+      // trail dark info (was 94E2D5)
+      danger: { hex: "E07060", ansiFallback: "\x1B[31m" },
+      // trail dark error (was F38BA8)
       muted: { hex: "9AA3B8", ansiFallback: "\x1B[90m" }
       // dimmed from A6ADC8 so secondary text sits a step below body text
     }
@@ -1096,6 +937,13 @@ function diffLineSequences(env = process.env, brightness = resolveTuiBrightness(
     context: { fg: "", bg: "", bar: " " }
   };
 }
+function renderSurfaceLine(args) {
+  const { text, surface, padTo } = args;
+  if (!surface) return text;
+  const reopened = text.split("\x1B[0m").join(`\x1B[0m${surface}`);
+  const padding = " ".repeat(Math.max(0, padTo - visibleWidth(text)));
+  return `${surface}${reopened}${padding}\x1B[0m`;
+}
 function renderDiffLine(args) {
   const { kind, text } = args;
   const env = args.env ?? process.env;
@@ -1108,13 +956,23 @@ function renderDiffLine(args) {
     return `${themeColorSequence(token, env)}${text}\x1B[39m`;
   }
   const seq = seqs[kind];
-  const padding = args.padTo === void 0 ? "" : " ".repeat(Math.max(0, args.padTo - displayWidth(text)));
-  return `${seq.bar}${seq.bg}${seq.fg}${text}${padding}\x1B[0m`;
+  return renderSurfaceLine({
+    text: `${seq.fg}${text}`,
+    surface: seq.bg,
+    padTo: args.padTo ?? visibleWidth(text)
+  });
 }
 function surfaceBackgroundSequence(env = process.env, brightness = resolveTuiBrightness(env)) {
   if (!supportsTruecolor(env)) return "";
   return hexToBgSgr((SURFACE_HEX[activeThemeName] ?? SURFACE_HEX.trail)[brightness]);
 }
+function userSurfaceBackgroundSequence(env = process.env, brightness = resolveTuiBrightness(env)) {
+  if (!supportsTruecolor(env)) return "";
+  const palette = THEME_PALETTES[activeThemeName] ?? THEME_PALETTES.trail;
+  const accentHex = palette[brightness].accent.hex;
+  const weight = brightness === "dark" ? 0.24 : 0.16;
+  return hexToBgSgr(blendHex(accentHex, resolveTerminalBaseHex(brightness), weight));
+}
 function themeColorSequence(token, env = process.env, brightness = resolveTuiBrightness(env)) {
   const palette = THEME_PALETTES[activeThemeName] ?? THEME_PALETTES.trail;
   const entry = palette[brightness][token];
@@ -1558,9 +1416,28 @@ function createRenderAwareStreamWriter(args) {
   let inFence = false;
   let fenceLang = "unknown";
   let lastKind = null;
+  const PARTIAL_FLUSH_THRESHOLD = 48;
   const emitBreathIfNeeded = (nextKind) => {
     if (needsListProseBreath(lastKind, nextKind)) args.write("\n");
   };
+  const flushPartialProse = () => {
+    if (inFence || buffer.length < PARTIAL_FLUSH_THRESHOLD) return;
+    const firstLine = buffer.split("\n", 1)[0] ?? "";
+    if (isTableRow(firstLine) || classifyStreamLine(firstLine) !== "prose") return;
+    const Segmenter = Intl.Segmenter;
+    const graphemes = typeof Segmenter === "function" ? Array.from(new Segmenter(void 0, { granularity: "grapheme" }).segment(firstLine), (part) => part.segment) : Array.from(firstLine);
+    if (graphemes.length < PARTIAL_FLUSH_THRESHOLD) return;
+    let cut = PARTIAL_FLUSH_THRESHOLD;
+    const whitespaceIndex = graphemes.slice(0, cut).findLastIndex((part) => /\s/.test(part));
+    if (whitespaceIndex > 0) cut = whitespaceIndex + 1;
+    const prefix = graphemes.slice(0, cut).join("");
+    const hasUnsafeMarkdownPrefix = /[`*_~\[]/.test(prefix) || /^\s*[#>]/.test(prefix) || /!\[/.test(prefix);
+    if (hasUnsafeMarkdownPrefix) return;
+    emitBreathIfNeeded("prose");
+    emitFormattedAssistantBlock(args.write, prefix, false);
+    lastKind = "prose";
+    buffer = prefix.length === buffer.length ? "" : buffer.slice(prefix.length);
+  };
   const flushCompleteBlocks = () => {
     while (buffer.includes("\n")) {
       const lines = buffer.split("\n");
@@ -1620,6 +1497,7 @@ function createRenderAwareStreamWriter(args) {
       if (!chunk) return;
       buffer += chunk;
       flushCompleteBlocks();
+      flushPartialProse();
     },
     flush() {
       if (!buffer) return;
@@ -2426,7 +2304,7 @@ function isReadToolName(name) {
 }
 function isSearchToolName(name) {
   if (!name) return false;
-  return name === "grep" || name === "searchFiles" || name === "search_files" || name === "globFiles" || name === "glob_files" || name === "glob" || name === "searchWorkspace" || name === "search";
+  return name === "grep" || name === "searchFiles" || name === "search_files" || name === "globFiles" || name === "glob_files" || name === "glob" || name === "searchWorkspace" || name === "search_workspace" || name === "search_all_spaces" || name === "search";
 }
 function isRunToolName(name) {
   if (!name) return false;
@@ -2573,8 +2451,43 @@ function resolveLoadSkillName(event) {
   if (match?.[1]) return match[1];
   return argName || "skill";
 }
+function formatTodoListForCli(event, colorEnabled) {
+  if (event.type !== "tool-result" || event.toolName !== "setTodoList") {
+    return void 0;
+  }
+  let raw = event.metadata?.displayData;
+  let parsed = typeof raw === "object" && raw !== null;
+  if (typeof raw === "string") {
+    try {
+      raw = JSON.parse(raw);
+      parsed = true;
+    } catch {
+      raw = void 0;
+    }
+  }
+  if (!parsed && typeof event.content === "string") {
+    try {
+      raw = JSON.parse(event.content);
+      parsed = true;
+    } catch {
+      raw = void 0;
+    }
+  }
+  const todos = Array.isArray(raw?.todos) ? raw.todos : void 0;
+  if (!todos) return void 0;
+  if (todos.length === 0) return "\u2611 Todo\n  (empty)\n";
+  const lines = todos.map((todo) => {
+    const status = todo?.status === "done" ? "\u2713" : todo?.status === "in_progress" ? "\u25D0" : "\u25CB";
+    const title = typeof todo?.title === "string" ? todo.title : "Untitled task";
+    return `  ${status} ${title}`;
+  });
+  const text = [`\u2611 Todo (${todos.length})`, ...lines].join("\n") + "\n";
+  return colorEnabled ? themeText(text, "chrome") : text;
+}
 function formatCompactToolLine(event, pending, colorEnabled) {
   const toolName = event.toolName || pending?.toolName || "tool";
+  const todoBlock = formatTodoListForCli(event, colorEnabled);
+  if (todoBlock) return todoBlock;
   if (isReadToolName(toolName) && event.type === "tool-result") {
     const rawPath = (typeof event.metadata?.path === "string" ? event.metadata.path : void 0) || (typeof event.metadata?.filePath === "string" ? event.metadata.filePath : void 0) || event.argumentsPreview || pending?.argumentsPreview || "";
     return formatReadTreeBlockForCli([{ path: rawPath, metadata: event.metadata }], colorEnabled);
@@ -2656,7 +2569,8 @@ function formatEditDetailBlock(lines, colorEnabled) {
   return lines.map((line) => `  ${renderDiffLine({ kind: line.kind, text: line.text, padTo, colorEnabled: true })}
 `).join("");
 }
-function createToolEventFormatter(mode, colorEnabled = resolveCliColorEnabled()) {
+function createToolEventFormatter(mode, colorEnabled = resolveCliColorEnabled(), opts = {}) {
+  const suppressRunProgressCards = opts.suppressRunProgressCards === true;
   const pending = /* @__PURE__ */ new Map();
   let readBuffer = [];
   let searchBuffer = [];
@@ -2680,9 +2594,16 @@ function createToolEventFormatter(mode, colorEnabled = resolveCliColorEnabled())
     statusFold = null;
     const substance = cardSubstance(fold);
     if (lastPrintedCardByRunId.get(fold.runId) === substance) return "";
-    lastPrintedCardByRunId.set(fold.runId, substance);
     const prevLog = lastEmittedLogByRunId.get(fold.runId);
     const changed = Boolean(fold.logKey) && fold.logKey !== prevLog;
+    if (suppressRunProgressCards) {
+      const terminal = isAgentRunTerminalStatus(fold.status);
+      if (!terminal && !fold.errorMessage && !changed) {
+        lastPrintedCardByRunId.set(fold.runId, substance);
+        return "";
+      }
+    }
+    lastPrintedCardByRunId.set(fold.runId, substance);
     if (changed && runShowsLogTail(fold.status)) {
       lastEmittedLogByRunId.set(fold.runId, fold.logKey);
     }
@@ -3060,7 +2981,9 @@ function createCliTurnOutput(params) {
   const spinner = params.spinner ?? new Spinner(options.output, workingLabel, Boolean(options.activityReporter));
   const toolDisplayMode = resolveToolDisplayMode(options.env);
   const traceLocalTools = shouldEmitToolEvents(toolDisplayMode);
-  const formatToolEvent = createToolEventFormatter(toolDisplayMode);
+  const formatToolEvent = createToolEventFormatter(toolDisplayMode, void 0, {
+    suppressRunProgressCards: typeof options.onAgentRunStatus === "function"
+  });
   const eventMode = resolveAgentEventMode(options);
   let streamedAssistantText = false;
   let everStreamedAnyText = false;
@@ -3557,7 +3480,7 @@ function findCollapsedPasteSpanAt(buffer, pos, opts = {}) {
 // packages/cli/client/agentRun.ts
 import { ulid } from "ulid";
 async function loadRunLocalAgentTurn() {
-  const { runLocalAgentTurn } = await import("./agentRuntimeLocal-7WYWCGL4.js");
+  const { runLocalAgentTurn } = await import("./agentRuntimeLocal-GVNERRKQ.js");
   return runLocalAgentTurn;
 }
 async function resolveCurrentMachineId(options) {
@@ -3996,55 +3919,7 @@ async function runLocalAgentTurnForCli(options, settings) {
     ...options.parentDialogId ? { parentThreadId: options.parentDialogId } : {}
   } : void 0;
   const currentDialogId = options.continueDialogId ?? ulid();
-  const runChildTurn = async (input) => {
-    const { runLocalAgentTurn } = await import("./agentRuntimeLocal-7WYWCGL4.js");
-    return runLocalAgentTurn(input);
-  };
-  const createFreshChildBaseAdapter = () => {
-    const fresh = options.localRuntimeAdapterFactory?.(options.env, {
-      cwd: options.localRuntimeCwd
-    }) ?? buildDefaultLocalRuntimeAdapter(options);
-    return applyEphemeralIfRequested(options, fresh);
-  };
-  const withLocalDelegation = (args) => {
-    let adapter2;
-    const callAgentExecutor = createCliCallAgentToolExecutor(
-      {
-        env: options.env,
-        fetchImpl: options.fetchImpl,
-        cwd: options.localRuntimeCwd,
-        output: options.output
-      },
-      {
-        createChildAdapter: (child) => withLocalDelegation({
-          base: createFreshChildBaseAdapter(),
-          dialogId: child.dialogId,
-          spaceId: child.spaceId,
-          runtimeContext: child.runtimeContext
-        }),
-        runChildTurn,
-        dialogId: args.dialogId,
-        spaceId: args.spaceId,
-        runtimeContext: args.runtimeContext
-      }
-    );
-    adapter2 = {
-      ...args.base,
-      executeTool: async (call) => {
-        if (call.name === "callAgent") {
-          return callAgentExecutor(call);
-        }
-        return args.base.executeTool(call);
-      }
-    };
-    return adapter2;
-  };
-  const adapter = withLocalDelegation({
-    base: baseAdapter,
-    dialogId: currentDialogId,
-    spaceId: options.spaceId,
-    runtimeContext
-  });
+  const adapter = baseAdapter;
   const expandedMessage = options.pastedTextStore?.items.size ? expandCollapsedPastes(options.message, options.pastedTextStore) : options.message;
   const workingLabel = `${options.agentName} -> working locally`;
   const turnOutput = createCliTurnOutput({
@@ -4795,6 +4670,40 @@ var resolveMemoryRuntime = async (input) => {
   };
 };
 
+// packages/ai/tools/agent/quotaCircuitBreaker.ts
+var QUOTA_ERROR_PATTERNS = [
+  /429/,
+  /quota/i,
+  /rate\s*limit/i,
+  /too\s*many\s*requests/i,
+  /额度/,
+  /上限/,
+  /用尽/,
+  /CliProviderQuotaError/i
+];
+function isCliProviderQuotaErrorShape(error) {
+  if (!error || typeof error !== "object") return false;
+  const e = error;
+  return e.name === "CliProviderQuotaError" || typeof e.message === "string" && e.message.startsWith("[QUOTA_LIMITED:");
+}
+function isQuotaExhaustedError(error) {
+  if (!error) return false;
+  if (isCliProviderQuotaErrorShape(error)) return true;
+  if (error instanceof Error) {
+    return QUOTA_ERROR_PATTERNS.some((pattern) => pattern.test(error.message));
+  }
+  if (typeof error === "string") {
+    return QUOTA_ERROR_PATTERNS.some((pattern) => pattern.test(error));
+  }
+  const maybeStatus = error.status;
+  if (typeof maybeStatus === "number" && maybeStatus === 429) return true;
+  const maybeStatusCode = error.statusCode;
+  if (typeof maybeStatusCode === "number" && maybeStatusCode === 429) return true;
+  return false;
+}
+var DEFAULT_QUOTA_BREAKER_MS = 6 * 3600 * 1e3;
+var DEFAULT_AGENT_NOT_FOUND_BREAKER_MS = 15 * 60 * 1e3;
+
 // packages/cli/agentRunCommand.ts
 function resolvePositiveMs(value, fallback) {
   return parsePositiveFiniteNumberOrFallback(value, fallback);
@@ -4918,7 +4827,7 @@ async function runAgentRunCommand(args, deps) {
   if (wantsFallbackSuggestions) {
     effectiveMessage = `${parsed.message}
 
-[Quota fallback context for agent decision: If this execution for ${parsed.agentKey} hits subscription quota, YOU (the agent) must decide the next agent to use. Do not rely on automatic wrapper. Reason based on the specific task (complexity, length, domain), agent strengths, cost, previous attempt results. Suggested alternatives: ${parsed.fallbackAgentKeys.join(", ")}. Update the task row with attempt history and dispatch the chosen next (via new agent run or callAgent({ background: true })).]`;
+[Quota fallback context for agent decision: If this execution for ${parsed.agentKey} hits subscription quota, YOU (the agent) must decide the next agent to use. Do not rely on automatic wrapper. Reason based on the specific task (complexity, length, domain), agent strengths, cost, previous attempt results. Suggested alternatives: ${parsed.fallbackAgentKeys.join(", ")}. Update the task row with attempt history and dispatch the chosen next (via startAgentRun \u5F02\u6B65\u6D3E\u53D1)).]`;
   }
   const resolvedAgentKey = parsed.runtimeMode !== "local" ? await (deps.resolveAgentRunAgentKey ?? resolveAgentRunAgentKey)({
     agentInput: resolveRawAgentInput(args, deps.commandPath) ?? parsed.agentKey,
@@ -4936,7 +4845,9 @@ async function runAgentRunCommand(args, deps) {
     const hasExplicitAgent = Boolean(readFlagValue(args, "--agent"));
     const route = await classifyCliAutoRoute(effectiveMessage, {
       serverUrl,
-      authToken
+      authToken,
+      // 有图 → kimi vision 档；无图 → flash 档。漏传会永远走 flash。
+      hasImages: parsed.imageUrls.length > 0
     });
     if (hasExplicitAgent) {
       modelOverride = buildModelLayerOverride(
@@ -5054,92 +4965,99 @@ async function runAgentRunCommand(args, deps) {
     return 0;
   }
   const cliCwd = parsed.cwd ?? process.cwd();
+  const isSubtask = isSubtaskRun(env);
   const extraContextBlocks = [];
-  const AGENTS_MD_MAX = 8192;
-  for (const name of ["AGENTS.md", "CLAUDE.md"]) {
-    const agentsMdPath = join2(cliCwd, name);
-    if (existsSync2(agentsMdPath)) {
-      try {
-        let mdContent = readFileSync2(agentsMdPath, "utf8").trim();
-        if (!mdContent) continue;
-        if (Buffer.byteLength(mdContent, "utf8") > AGENTS_MD_MAX) {
-          mdContent = Buffer.from(mdContent, "utf8").subarray(0, AGENTS_MD_MAX).toString("utf8") + "\n\n<!-- AGENTS.md truncated -->";
-        }
-        extraContextBlocks.push(`--- \u9879\u76EE\u6307\u4EE4\uFF08${name}\uFF09---
+  const contextBlockScopes = [];
+  let memoryPromptBlock = null;
+  let memoryOverlayLayer = null;
+  let memoryUseGuidanceLayer = null;
+  if (!isSubtask) {
+    const AGENTS_MD_MAX = 8192;
+    for (const name of ["AGENTS.md", "CLAUDE.md"]) {
+      const agentsMdPath = join2(cliCwd, name);
+      if (existsSync2(agentsMdPath)) {
+        try {
+          let mdContent = readFileSync2(agentsMdPath, "utf8").trim();
+          if (!mdContent) continue;
+          if (Buffer.byteLength(mdContent, "utf8") > AGENTS_MD_MAX) {
+            mdContent = Buffer.from(mdContent, "utf8").subarray(0, AGENTS_MD_MAX).toString("utf8") + "\n\n<!-- AGENTS.md truncated -->";
+          }
+          extraContextBlocks.push(`--- \u9879\u76EE\u6307\u4EE4\uFF08${name}\uFF09---
 ${mdContent}`);
-        break;
-      } catch {
+          break;
+        } catch {
+        }
       }
     }
-  }
-  extraContextBlocks.push(...buildSkillContextBlocks(skillReferences));
-  const skillDiscoveryBlock = buildSkillDiscoveryContextBlock(cliCwd);
-  if (skillDiscoveryBlock) {
-    extraContextBlocks.push(skillDiscoveryBlock);
-  }
-  let memoryPromptBlock = null;
-  if (!deps.memoryRecallDisabled) {
-    const memoryServerUrl = resolveServerUrl(env);
-    const memoryAuthToken = resolveAuthToken(args, env);
-    const memoryAgentKey = effectiveAgentKey;
-    const memorySpaceId = parsed.spaceId ?? void 0;
-    try {
-      if (memoryAuthToken && memoryServerUrl) {
-        const memoryResponse = await fetch(`${memoryServerUrl}/api/memory/query`, {
-          method: "POST",
-          headers: {
-            "Content-Type": "application/json",
-            Authorization: `Bearer ${memoryAuthToken}`
-          },
-          body: JSON.stringify({
+    extraContextBlocks.push(...buildSkillContextBlocks(skillReferences));
+    const skillDiscoveryBlock = buildSkillDiscoveryContextBlock(cliCwd);
+    if (skillDiscoveryBlock) {
+      extraContextBlocks.push(skillDiscoveryBlock);
+    }
+    if (!deps.memoryRecallDisabled) {
+      const memoryServerUrl = resolveServerUrl(env);
+      const memoryAuthToken = resolveAuthToken(args, env);
+      const memoryAgentKey = effectiveAgentKey;
+      const memorySpaceId = parsed.spaceId ?? void 0;
+      try {
+        const resolvedRuntimeMode = parsed.runtimeMode ?? (env.NOLO_RUNTIME_MODE === "local" || env.NOLO_RUNTIME_MODE === "server" || env.NOLO_RUNTIME_MODE === "auto" ? env.NOLO_RUNTIME_MODE : "auto");
+        const usesExplicitServerRuntime = resolvedRuntimeMode === "server";
+        if (usesExplicitServerRuntime && memoryAuthToken && memoryServerUrl) {
+          const memoryResponse = await fetch(`${memoryServerUrl}/api/memory/query`, {
+            method: "POST",
+            headers: {
+              "Content-Type": "application/json",
+              Authorization: `Bearer ${memoryAuthToken}`
+            },
+            body: JSON.stringify({
+              agentKey: memoryAgentKey,
+              userInput: effectiveMessage,
+              ...memorySpaceId ? { spaceId: memorySpaceId } : {}
+            }),
+            signal: AbortSignal.timeout(5e3)
+          });
+          if (memoryResponse.ok) {
+            const memoryPayload = await memoryResponse.json().catch(() => null);
+            const block = typeof memoryPayload?.promptBlock === "string" ? memoryPayload.promptBlock.trim() : "";
+            memoryPromptBlock = block || null;
+          } else {
+            console.warn("[cli-memory] memory query returned non-ok status:", memoryResponse.status);
+          }
+        } else {
+          const localDb = await getDefaultCliLocalRuntimeDb({ env });
+          const machineId = resolveMachineId();
+          const resolution = await resolveMemoryRuntime({
+            db: localDb,
+            userId: machineId,
             agentKey: memoryAgentKey,
             userInput: effectiveMessage,
             ...memorySpaceId ? { spaceId: memorySpaceId } : {}
-          }),
-          signal: AbortSignal.timeout(5e3)
-        });
-        if (memoryResponse.ok) {
-          const memoryPayload = await memoryResponse.json().catch(() => null);
-          const block = typeof memoryPayload?.promptBlock === "string" ? memoryPayload.promptBlock.trim() : "";
-          memoryPromptBlock = block || null;
-        } else {
-          console.warn("[cli-memory] memory query returned non-ok status:", memoryResponse.status);
+          });
+          memoryPromptBlock = resolution.promptBlock;
         }
-      } else {
-        const localDb = await getDefaultCliLocalRuntimeDb({ env });
-        const machineId = resolveMachineId();
-        const resolution = await resolveMemoryRuntime({
-          db: localDb,
-          userId: machineId,
-          agentKey: memoryAgentKey,
-          userInput: effectiveMessage,
-          ...memorySpaceId ? { spaceId: memorySpaceId } : {}
-        });
-        memoryPromptBlock = resolution.promptBlock;
+      } catch (memoryError) {
+        console.warn("[cli-memory] memory recall failed, omitting memory layer:", memoryError);
       }
-    } catch (memoryError) {
-      console.warn("[cli-memory] memory recall failed, omitting memory layer:", memoryError);
     }
-  }
-  const memoryOverlayLayer = buildMemoryOverlayLayer({ promptBlock: memoryPromptBlock });
-  const memoryUseGuidanceLayer = buildMemoryUseGuidanceLayer({ promptBlock: memoryPromptBlock });
-  const contextBlockScopes = [];
-  const agentsMdMarker = "--- \u9879\u76EE\u6307\u4EE4\uFF08";
-  for (const block of extraContextBlocks) {
-    if (block.startsWith(agentsMdMarker)) {
-      contextBlockScopes.push({ content: block, cacheScope: "session" });
+    memoryOverlayLayer = buildMemoryOverlayLayer({ promptBlock: memoryPromptBlock });
+    memoryUseGuidanceLayer = buildMemoryUseGuidanceLayer({ promptBlock: memoryPromptBlock });
+    const agentsMdMarker = "--- \u9879\u76EE\u6307\u4EE4\uFF08";
+    for (const block of extraContextBlocks) {
+      if (block.startsWith(agentsMdMarker)) {
+        contextBlockScopes.push({ content: block, cacheScope: "session" });
+      }
     }
-  }
-  if (memoryUseGuidanceLayer?.content?.trim()) {
-    contextBlockScopes.push({ content: memoryUseGuidanceLayer.content, cacheScope: "session" });
-  }
-  for (const block of extraContextBlocks) {
-    if (!block.startsWith(agentsMdMarker)) {
-      contextBlockScopes.push({ content: block, cacheScope: "turn" });
+    if (memoryUseGuidanceLayer?.content?.trim()) {
+      contextBlockScopes.push({ content: memoryUseGuidanceLayer.content, cacheScope: "session" });
+    }
+    for (const block of extraContextBlocks) {
+      if (!block.startsWith(agentsMdMarker)) {
+        contextBlockScopes.push({ content: block, cacheScope: "turn" });
+      }
+    }
+    if (memoryOverlayLayer?.content?.trim()) {
+      contextBlockScopes.push({ content: memoryOverlayLayer.content, cacheScope: "turn" });
     }
-  }
-  if (memoryOverlayLayer?.content?.trim()) {
-    contextBlockScopes.push({ content: memoryOverlayLayer.content, cacheScope: "turn" });
   }
   const buildRunOptions = (targetAgentKey) => ({
     agentName: targetAgentKey,
@@ -5287,10 +5205,8 @@ function isBalanceExhaustedError(error) {
 }
 
 export {
-  CLI_AUTO_TIER_AGENT_KEYS,
   CLI_IMAGE_AGENT_KEY,
   CLI_AUTO_TIER_AGENT_KEY_TABLE,
-  resolveCliAutoFallbackTier,
   classifyCliAutoRoute,
   resolveAgentContextWindow,
   formatTokenCount,
@@ -5321,7 +5237,9 @@ export {
   resolveTuiBrightness,
   setActiveTerminalBaseHex,
   applyDetectedBackground,
+  renderSurfaceLine,
   surfaceBackgroundSequence,
+  userSurfaceBackgroundSequence,
   themeColorSequence,
   themeText,
   formatAssistantDisplay,
@@ -5343,6 +5261,8 @@ export {
   prependSubjectDialogMarker,
   normalizeCliImageInput,
   inspectLocalRunWorkspace,
+  QUOTA_ERROR_PATTERNS,
+  isQuotaExhaustedError,
   runAgentRunCommand,
   isBalanceExhaustedError
 };