nolo-cli 0.54.0-alpha.1 → 0.54.0

package/{authCommand-5HM57KPL.js → authCommand-OK6HLHGS.js}

@@ -1,12 +1,13 @@
 import {
   defaultOpenBrowser
-} from "./chunk-6FGXGFAH.js";
+} from "./chunk-75MSWWUE.js";
 import {
   generateCloudflareEmailRoutingToken,
   parseFlagWithOptionalValue,
   runCloudflareOAuthLogin,
   upsertEnvVariable
-} from "./chunk-ZMV5PCRS.js";
+} from "./chunk-SWJGJMTW.js";
+import "./chunk-ESHYE7WU.js";
 import {
   ANTHROPIC_OAUTH_AUTHORIZE_URL,
   ANTHROPIC_OAUTH_CALLBACK_PORT,
@@ -45,18 +46,17 @@ import {
   startCallbackServer,
   validateXAIEndpoint,
   xaiOAuthDiscovery
-} from "./chunk-45E65JER.js";
-import "./chunk-ESHYE7WU.js";
+} from "./chunk-NFLA6GSU.js";
 import "./chunk-JBTJAPUF.js";
-import "./chunk-CULR4XB4.js";
+import "./chunk-BG7J6Y5X.js";
 import "./chunk-IA3A5373.js";
 import "./chunk-GLKOYSRU.js";
-import "./chunk-5Y5G76BJ.js";
-import "./chunk-ET7BXWAW.js";
+import "./chunk-KLIBF3BY.js";
+import "./chunk-N4JKMCY3.js";
+import "./chunk-SVDMVX4G.js";
 import {
   createOAuthTokenStore
 } from "./chunk-FD36SCCD.js";
-import "./chunk-7UFH6THF.js";
 import "./chunk-ZHXJ6M7T.js";
 import {
   toErrorMessage
@@ -67,20 +67,20 @@ import {
 } from "./chunk-UNZDCI3O.js";
 import "./chunk-A7GB5UGH.js";
 import "./chunk-OLZBQWZU.js";
-import "./chunk-J7KVJAUQ.js";
-import "./chunk-SVEUG45B.js";
+import "./chunk-HPW5MNDQ.js";
+import "./chunk-NHRQCXIQ.js";
 import {
   asOptionalFiniteNumber
 } from "./chunk-WWC7I6TA.js";
 import "./chunk-I42VC5KR.js";
 import "./chunk-QSJ3HFEF.js";
-import {
-  isRecord
-} from "./chunk-KUG52IAP.js";
 import {
   asTrimmedString
 } from "./chunk-3EUW3WEO.js";
 import "./chunk-U6VC4MU4.js";
+import {
+  isRecord
+} from "./chunk-KUG52IAP.js";
 import "./chunk-LCBAMY2M.js";
 import "./chunk-4SN6J65J.js";
 import "./chunk-CF7PE763.js";

package/{authCommands-SBQVTD6V.js → authCommands-2DUIUEMT.js}

@@ -6,7 +6,7 @@ import {
   runLoginCommand,
   runLogoutCommand,
   runWhoamiCommand
-} from "./chunk-6FGXGFAH.js";
+} from "./chunk-75MSWWUE.js";
 import "./chunk-ESHYE7WU.js";
 import "./chunk-JBTJAPUF.js";
 import "./chunk-IA3A5373.js";
@@ -14,7 +14,7 @@ import "./chunk-GLKOYSRU.js";
 import "./chunk-UNZDCI3O.js";
 import "./chunk-A7GB5UGH.js";
 import "./chunk-OLZBQWZU.js";
-import "./chunk-SVEUG45B.js";
+import "./chunk-NHRQCXIQ.js";
 import "./chunk-WWC7I6TA.js";
 import "./chunk-I42VC5KR.js";
 import "./chunk-3EUW3WEO.js";

package/{chatQueueReduxAdapter-ZXXP6E4O.js → chatQueueReduxAdapter-TV2TUOQS.js}

@@ -4,54 +4,57 @@ import {
 } from "./chunk-YPREESF4.js";
 import {
   handleSendMessage
-} from "./chunk-BKPUPZ6W.js";
-import "./chunk-M2CGG6VJ.js";
-import "./chunk-H2G6JXTR.js";
+} from "./chunk-YBFJQC2M.js";
+import "./chunk-EVBFYPDC.js";
+import "./chunk-HQUORJTM.js";
 import "./chunk-YYLHWIW6.js";
-import "./chunk-K7CNV7GI.js";
 import "./chunk-AGGIDPCZ.js";
-import "./chunk-SG65C33A.js";
+import "./chunk-TC6P4NDR.js";
+import "./chunk-YLCUU3WM.js";
 import "./chunk-5RGU3J7P.js";
-import "./chunk-M76ECQRD.js";
 import {
   clearPendingUserInputQueue,
   dequeueUserInput,
   enqueueUserInput,
   getLoopStopReason,
   getPendingUserInputQueue
-} from "./chunk-MRWGHF56.js";
-import "./chunk-PTHLNXGY.js";
-import "./chunk-HR5XDXYM.js";
-import "./chunk-ET7BXWAW.js";
-import "./chunk-NQ4N4IGO.js";
-import "./chunk-FD36SCCD.js";
-import "./chunk-7UFH6THF.js";
+} from "./chunk-6YTOJY35.js";
+import "./chunk-M76ECQRD.js";
+import "./chunk-YDH3FDFQ.js";
+import "./chunk-K4NISKOR.js";
+import "./chunk-BBTN4NQD.js";
+import "./chunk-N4JKMCY3.js";
+import "./chunk-BYXBJQAS.js";
+import "./chunk-SVDMVX4G.js";
 import "./chunk-DWLEOOGR.js";
+import "./chunk-CS6TCJ7E.js";
+import "./chunk-FD36SCCD.js";
 import "./chunk-ZHXJ6M7T.js";
 import "./chunk-HNMXYYJR.js";
-import "./chunk-FA4SY4F5.js";
-import "./chunk-WJV3G7BM.js";
-import "./chunk-J7KVJAUQ.js";
-import "./chunk-X7S7OFUX.js";
+import "./chunk-BYEJ7K2Z.js";
+import "./chunk-NJUQBYFJ.js";
+import "./chunk-HPW5MNDQ.js";
+import "./chunk-SVC3GGUA.js";
 import "./chunk-Y5GCAMGP.js";
 import "./chunk-UZ6XXCVR.js";
 import "./chunk-KX6YBIHG.js";
-import "./chunk-4RXXP6WD.js";
-import "./chunk-76UA7376.js";
-import "./chunk-SVEUG45B.js";
+import "./chunk-MYMBBB6Q.js";
+import "./chunk-UAD2WT54.js";
+import "./chunk-7NYDJAQ7.js";
+import "./chunk-NHRQCXIQ.js";
 import "./chunk-WWC7I6TA.js";
 import "./chunk-I42VC5KR.js";
-import "./chunk-YZ2WE6CW.js";
+import "./chunk-SAJSBPI2.js";
 import "./chunk-QSJ3HFEF.js";
-import "./chunk-2JYSSXLY.js";
+import "./chunk-2WBFJHQA.js";
 import "./chunk-BJIRLHJR.js";
-import "./chunk-HHBXCRJS.js";
-import "./chunk-L3JWI7WX.js";
+import "./chunk-XOJRZUPF.js";
+import "./chunk-QOKAKDPE.js";
 import "./chunk-OW765XBU.js";
 import "./chunk-GBRG75BQ.js";
-import "./chunk-KUG52IAP.js";
 import "./chunk-3EUW3WEO.js";
 import "./chunk-U6VC4MU4.js";
+import "./chunk-KUG52IAP.js";
 import "./chunk-LCBAMY2M.js";
 import "./chunk-4SN6J65J.js";
 import "./chunk-CF7PE763.js";

package/{chunk-7CH6U3FL.js → chunk-2GRGFCF6.js}

@@ -78,10 +78,8 @@ var MEMORY_USE_GUIDANCE = `--- \u8BB0\u5FC6\u4F7F\u7528\u65B9\u5F0F ---
 - \u8BB0\u5FC6\u662F\u4E2A\u6027\u5316\u589E\u5F3A\u5C42\uFF1B\u5F53\u524D\u8F93\u5165/\u5BF9\u8BDD/\u7CFB\u7EDF\u89C4\u5219/Agent prompt/skill/\u7528\u6237\u5168\u5C40\u504F\u597D\u90FD\u4F18\u5148\u4E8E\u5B83\u3002\u5F53\u524D\u8F93\u5165\u7ED9\u51FA\u65B0\u8BED\u8A00\u3001\u6280\u672F\u6808\u3001\u6570\u503C\u3001\u7EA6\u675F\u6216\u660E\u786E\u8986\u76D6\u65E7\u504F\u597D\u65F6\uFF0C\u91C7\u7528\u5F53\u524D\u8F93\u5165\uFF0C\u4E0D\u8981\u628A\u65E7\u8BB0\u5FC6\u5F53\u66F4\u9AD8\u771F\u503C\u3002
 - \u8BB0\u5FC6\u542B\u7528\u6237\u8EAB\u4EFD/\u79F0\u547C/\u5173\u7CFB/\u957F\u671F\u504F\u597D/\u9879\u76EE\u80CC\u666F\u65F6\uFF0C\u76F8\u5173\u5904\u81EA\u7136\u4F53\u73B0\uFF08\u5F00\u573A\u79F0\u547C\u3001\u4E0A\u4E0B\u6587\u786E\u8BA4\u3001\u56DE\u7B54\u7ED3\u6784\u3001\u53D6\u820D\u6807\u51C6\uFF09\uFF0C\u4E0D\u8981\u6BCF\u53E5\u673A\u68B0\u79F0\u547C\u7528\u6237\u3002
 - \u201C\u4E0A\u6B21/\u7EE7\u7EED/\u8FD9\u4E2A\u9879\u76EE\u201D\u7B49\u6307\u4EE3\u4F18\u5148\u6309\u5F53\u524D dialog/space/project/agent/sourceDialog \u7B49 KV \u8DEF\u5F84\u548C\u65F6\u95F4\u7EBF\u5B9A\u4F4D\uFF0C\u4E0D\u8981\u53EA\u6309\u8BED\u4E49\u76F8\u4F3C\u5EA6\u635E\u4E00\u6761\u3002
-- \u63A8\u65AD\u578B\u8BB0\u5FC6\uFF08inferred\uFF09\u53EA\u7528\u4E8E\u628A\u63E1\u8BED\u6C14\u3001\u72B6\u6001\u548C\u672A\u5B8C\u6210\u4E8B\u9879\uFF0C\u4E0D\u8981\u8BF4\u6210\u201C\u4F60\u660E\u786E\u544A\u8BC9\u8FC7\u6211\u201D\u6216\u5F53\u6210\u7528\u6237\u5DF2\u6388\u6743\u4FDD\u5B58\u7684\u4E8B\u5B9E\u3002\u51B2\u7A81\u6216\u573A\u666F\u4E0D\u660E\u65F6\u8BF4\u660E\u4F9D\u636E\u6216\u7B80\u77ED\u786E\u8BA4\uFF0C\u4E0D\u8981\u786C\u5957\u3002
-- \u3010\u7F6E\u4FE1\u5EA6\u6765\u6E90\u3011\u6BCF\u6761\u8BB0\u5FC6\u6709\u6765\u6E90\uFF1Averified\uFF08\u5B9E\u6D4B\u9A8C\u8BC1\uFF0C\u9AD8\u7F6E\u4FE1\u5EA6\uFF09\u3001stated\uFF08\u7528\u6237\u9648\u8FF0\uFF0C\u4E2D\u9AD8\uFF09\u3001inferred\uFF08\u6A21\u578B\u63A8\u65AD\uFF0C\u4F4E\u2014\u2014\u5BB9\u6613\u7F16\u9020\uFF0C\u4F18\u5148\u5B58\u7591\uFF09\u3002\u53EC\u56DE\u65F6\u4F18\u5148\u4F7F\u7528 verified/stated\uFF0Cinferred \u9700\u9A8C\u8BC1\u540E\u518D\u91C7\u7EB3\u3002
-- \u3010\u53EC\u56DE\u89C4\u5219\u3011\u53EC\u56DE\u8BB0\u5FC6\u5FC5\u987B\u5E26\u5B8C\u6574\u5386\u53F2\u4E0A\u4E0B\u6587\uFF08\u6765\u6E90\u3001\u7F6E\u4FE1\u5EA6\u3001\u53D8\u66F4\u8BB0\u5F55\uFF09\uFF0C\u7981\u6B62\u81EA\u884C\u63A8\u7406\u586B\u8865\u3002\u4E0D\u8981\u51ED\u5370\u8C61\u62FC\u51FA\u672A\u9A8C\u8BC1\u7684\u4FE1\u606F\u3002
-- \u3010\u964D\u6743\u4E0E\u5220\u9664\u3011\u65E5\u5E38\u9519\u8BEF\u8BB0\u5FC6\u4F18\u5148\u964D\u6743/\u5F52\u6863\u4E0E\u8986\u76D6\uFF08\u964D\u4F4E\u7F6E\u4FE1\u5EA6\uFF09\uFF0C\u4FDD\u7559\u6863\u6848\u4F9B\u5BA1\u8BA1\uFF1B\u82E5\u7528\u6237\u660E\u786E/\u5F3A\u5236\u8981\u6C42\u5220\u9664\uFF0C\u4E14\u5728\u7528\u6237\u81EA\u8EAB\u6743\u9650\u8303\u56F4\u5185\uFF0C\u5141\u8BB8\u7269\u7406\u5220\u9664\uFF08\u4EC5\u9650\u4E8E\u5F53\u524D\u7528\u6237\u62E5\u6709\u7684\u8BB0\u5FC6\uFF0C\u4E25\u7981\u8D8A\u6743\uFF09\u3002`;
+- \u63A8\u65AD\u578B\u8BB0\u5FC6\uFF08inferred\uFF09\u53EA\u7528\u4E8E\u628A\u63E1\u8BED\u6C14\u3001\u72B6\u6001\u548C\u672A\u5B8C\u6210\u4E8B\u9879\uFF0C\u4E0D\u8981\u8BF4\u6210\u201C\u4F60\u660E\u786E\u544A\u8BC9\u8FC7\u6211\u201D\u6216\u5F53\u6210\u7528\u6237\u5DF2\u6388\u6743\u4FDD\u5B58\u7684\u4E8B\u5B9E\uFF1B\u51B2\u7A81\u6216\u573A\u666F\u4E0D\u660E\u65F6\u8BF4\u660E\u4F9D\u636E\u6216\u7B80\u77ED\u786E\u8BA4\uFF0C\u4E0D\u8981\u786C\u5957\u3002
+- \u7F6E\u4FE1\u5EA6\uFF08verified/stated/inferred\uFF09\u3001\u53EC\u56DE\u4E0A\u4E0B\u6587\u8981\u6C42\u4E0E\u964D\u6743/\u5220\u9664\u8FB9\u754C\u4EE5\u8BB0\u5FC6\u5DE5\u5177\uFF08rememberMemory / queryMemory / deleteMemory\uFF09schema \u4E3A\u51C6\u3002`;
 
 export {
   buildUserResponseLanguageContext,

package/{chunk-T27IEBNQ.js → chunk-2QDDKFG5.js}

@@ -70,6 +70,9 @@ function sanitizeOutboundHistory(messages, options) {
       const name = call?.function?.name;
       const declaredAllows = !declared || isString(name) && declared.has(name);
       if (!declaredAllows) return;
+      if (!hasParsableObjectArguments(call?.function?.arguments)) {
+        return;
+      }
       const id = isNonEmptyString(call?.id) ? call.id : stableToolCallId(i, callIndex);
       candidateCallIds.add(id);
     });
@@ -143,7 +146,77 @@ ${extra}` : extra;
   }
   return extra;
 }
+function hasParsableObjectArguments(raw) {
+  if (raw === void 0) return true;
+  if (raw === null || typeof raw === "number" || typeof raw === "boolean") {
+    return false;
+  }
+  if (typeof raw !== "string") return true;
+  if (raw.trim() === "") return true;
+  try {
+    const parsed = JSON.parse(raw);
+    return parsed !== null && typeof parsed === "object";
+  } catch {
+    return false;
+  }
+}
+function downgradeUnparsableToolCalls(messages) {
+  if (!Array.isArray(messages) || messages.length === 0) {
+    return { messages, downgraded: 0 };
+  }
+  const poisonIds = /* @__PURE__ */ new Set();
+  for (let i = 0; i < messages.length; i++) {
+    const m = messages[i];
+    if (m.role !== "assistant" || !Array.isArray(m.tool_calls)) continue;
+    m.tool_calls.forEach((call, callIndex) => {
+      if (!hasParsableObjectArguments(call?.function?.arguments)) {
+        const id = isNonEmptyString(call?.id) ? call.id : stableToolCallId(i, callIndex);
+        poisonIds.add(id);
+      }
+    });
+  }
+  if (poisonIds.size === 0) return { messages, downgraded: 0 };
+  let downgraded = 0;
+  const out = [];
+  for (const m of messages) {
+    if (m.role === "assistant" && Array.isArray(m.tool_calls) && m.tool_calls.length > 0) {
+      const kept = [];
+      const lines = [];
+      for (const call of m.tool_calls) {
+        if (!hasParsableObjectArguments(call?.function?.arguments)) {
+          lines.push(renderToolCallAsText(call));
+          downgraded++;
+        } else {
+          kept.push(call);
+        }
+      }
+      if (lines.length === 0) {
+        out.push(m);
+        continue;
+      }
+      const sanitized = {
+        ...m,
+        content: combineContentWithText(m.content, joinLines(...lines)),
+        tool_calls: kept.length > 0 ? kept : void 0
+      };
+      if (kept.length === 0) delete sanitized.tool_calls;
+      out.push(sanitized);
+      continue;
+    }
+    if (m.role === "tool") {
+      const id = isNonEmptyString(m.tool_call_id) ? m.tool_call_id : "";
+      if (id && poisonIds.has(id)) {
+        out.push({ role: "assistant", content: renderToolResultAsText(m) });
+        continue;
+      }
+    }
+    out.push(m);
+  }
+  return { messages: out, downgraded };
+}
 
 export {
-  sanitizeForOutbound
+  sanitizeForOutbound,
+  hasParsableObjectArguments,
+  downgradeUnparsableToolCalls
 };

package/{chunk-2JYSSXLY.js → chunk-2WBFJHQA.js}

@@ -10,14 +10,14 @@ import {
   scheduleExistingRecordReplication,
   shouldReplaceLocalWithRemoteRecord,
   shouldReplicateLocalRecord
-} from "./chunk-HHBXCRJS.js";
+} from "./chunk-XOJRZUPF.js";
 import {
   fetchFromClientDb,
   fetchFromServer,
   isAbortError,
   isReadTimeoutError,
   logger
-} from "./chunk-L3JWI7WX.js";
+} from "./chunk-QOKAKDPE.js";
 import {
   shouldReplaceWithNextRecord
 } from "./chunk-GBRG75BQ.js";
@@ -152,30 +152,9 @@ var BUILTIN_AGENT_CATALOG = [
     model: "gpt-5.5-pro",
     runtimeFallback: true
   },
-  {
-    id: "01CLSO50DIPB00000000A1DVVU",
-    group: "public",
-    name: "Claude Sonnet 5",
-    provider: "deepinfra",
-    model: "anthropic/claude-sonnet-5",
-    runtimeFallback: true
-  },
-  {
-    id: "01CLOP48DIPB0000000001OI1W",
-    group: "public",
-    name: "Claude Opus 5",
-    provider: "deepinfra",
-    model: "anthropic/claude-opus-5",
-    runtimeFallback: true
-  },
-  {
-    id: "01FABLE5DIPB00000001VJBKM4",
-    group: "public",
-    name: "Claude Fable 5",
-    provider: "deepinfra",
-    model: "anthropic/claude-fable-5",
-    runtimeFallback: true
-  },
+  // Claude Sonnet/Opus/Fable 5 已从广场下架（2026-09-01，平台停止维护 Claude 系）：
+  // 存量记录由 modelUpgradeTable 迁移到 nolo/glm-5-3-flash，兼容期请求由
+  // platformHostedRoutingTable 重映射；客户端下架提示走 publicAgents.removedFromPlaza。
   // Gemini 文本档只上架最新一代：旧版留在广场只会分散选择、拉长列表，而
   // 用户想要的是「最新的那个 Gemini」。gemini-3.6-flash 这个**模型**仍然保留
   // （modelAbility / 定价表照旧），下架的只是广场入口。
@@ -726,7 +705,6 @@ export {
   BUILTIN_NOLO_AGENT_KEY,
   BUILTIN_NOLO_AGENT_NAME,
   BUILTIN_NOLO_AGENT_MODEL,
-  PUBLIC_DEEPSEEK_V4_FLASH_AGENT_KEY,
   DEFAULT_CODE_PLANNER_EXECUTOR_CANDIDATE_KEYS,
   isSystemBuiltinTrustedAgentKey,
   isBuiltinPlatformAgentKey,

package/{chunk-FZMM2KZC.js → chunk-3LKZRP7E.js}

@@ -6,10 +6,10 @@ import {
 } from "./chunk-AGGIDPCZ.js";
 import {
   getModelConfig
-} from "./chunk-7UFH6THF.js";
+} from "./chunk-SVDMVX4G.js";
 import {
   isPlatformHostedImageModel
-} from "./chunk-J7KVJAUQ.js";
+} from "./chunk-HPW5MNDQ.js";
 import {
   asOptionalFiniteNumber
 } from "./chunk-WWC7I6TA.js";

package/{chunk-WUMYQ2FE.js → chunk-3Y2K5O2R.js}

@@ -1,6 +1,6 @@
 import {
   sanitizeForOutbound
-} from "./chunk-T27IEBNQ.js";
+} from "./chunk-2QDDKFG5.js";
 import {
   accumulateToolCallDelta,
   createThinkParserState,
@@ -23,7 +23,7 @@ import {
   shouldStripReasoningContentForOutbound,
   toOpenAiCompatibleMessages,
   toResponsesTools
-} from "./chunk-6BYAXNH5.js";
+} from "./chunk-VZMUOPYG.js";
 import {
   buildProviderExecutionPlan,
   canUsePlatformChatProvider,
@@ -33,13 +33,13 @@ import {
   resolvePlatformAuthToken,
   resolvePlatformServerUrl,
   resolveProviderTransportDecision
-} from "./chunk-5Y5G76BJ.js";
+} from "./chunk-KLIBF3BY.js";
 import {
   clipCompactText
-} from "./chunk-HR5XDXYM.js";
+} from "./chunk-BBTN4NQD.js";
 import {
   NOLO_CLIENT_VERSION_HEADER
-} from "./chunk-J7KVJAUQ.js";
+} from "./chunk-HPW5MNDQ.js";
 import {
   isRecord
 } from "./chunk-KUG52IAP.js";
@@ -78,6 +78,33 @@ function extractUsageFromSsePayload(parsed) {
   return void 0;
 }
 
+// packages/agent-runtime/promptCacheKey.ts
+function collectStablePromptPrefix(messages, fallbackPrompt) {
+  const systemTexts = [];
+  if (fallbackPrompt?.trim()) systemTexts.push(fallbackPrompt.trim());
+  for (const raw of messages) {
+    if (!raw || typeof raw !== "object") continue;
+    const message = raw;
+    const role = String(message.role ?? "");
+    if (role !== "system" && role !== "developer") continue;
+    if (typeof message.content === "string") {
+      const boundary = Number(message.stable_prefix_chars);
+      const text = Number.isFinite(boundary) && boundary > 0 ? message.content.slice(0, boundary) : message.content;
+      if (text.trim()) systemTexts.push(text.trim());
+    }
+  }
+  return systemTexts.join("\n\n");
+}
+function stablePromptCacheKey(parts, namespace) {
+  const value = JSON.stringify(parts);
+  let hash = 2166136261;
+  for (let index = 0; index < value.length; index += 1) {
+    hash ^= value.charCodeAt(index);
+    hash = Math.imul(hash, 16777619) >>> 0;
+  }
+  return `${namespace}-${hash.toString(16).padStart(8, "0")}`;
+}
+
 // packages/core/chat/providerFailureMessage.ts
 var RAW_BODY_CLIP = 400;
 function describeProviderFailure(args) {
@@ -477,7 +504,8 @@ function buildPlatformChatCompletionRequest(args) {
     },
     stream: args.stream ?? false,
     ...args.stream ? getUsageRequestOptions(args.providerConfig.usageProvider ?? args.providerConfig.provider, {
-      api: usesResponsesApi ? "responses" : "chat-completions"
+      api: usesResponsesApi ? "responses" : "chat-completions",
+      endpoint: usesResponsesApi ? null : args.providerConfig.endpoint
     }) : {},
     ...requestOptions,
     ...args.tools && args.tools.length > 0 ? {
@@ -485,6 +513,23 @@ function buildPlatformChatCompletionRequest(args) {
       ...usesResponsesApi ? {} : { tool_choice: "auto" }
     } : {},
     ...shouldDisableThinking(args.providerConfig) ? { thinking: { type: "disabled" } } : {},
+    // 缓存路由亲和提示：托管上游是多副本共享池，前缀缓存是节点本地的，
+    // 请求落到另一副本就整条 miss（实测占该线全部 miss token 的 52%，
+    // 依据见 promptCacheKey.ts）。键只取跨轮稳定的部分（model + system 稳定
+    // 前缀 + 工具定义），历史增长不参与，否则键每轮都变等于没有键。
+    // 已实测 RunInfra 接受该字段且不破坏既有命中；效果待上线后按
+    // promptCacheKey.ts 末尾的口径复测。
+    prompt_cache_key: stablePromptCacheKey(
+      [
+        args.providerConfig.model,
+        // 只从消息里取：本 builder 只拿得到 providerConfig，没有 agentConfig，
+        // agent prompt 早已由 localLoop 的 buildMessages 拼进 system 稳定前缀，
+        // 所以 sanitizedMessages 里的 system 就是完整取材，无需 fallback。
+        collectStablePromptPrefix(sanitizedMessages),
+        args.tools ?? []
+      ],
+      "nolo-chat"
+    ),
     url: args.providerConfig.endpoint,
     provider: args.providerConfig.provider,
     agentKey: args.providerConfig.agentKey,
@@ -500,6 +545,17 @@ function buildPlatformChatCompletionRequest(args) {
       headers: {
         "Content-Type": "application/json",
         Authorization: `Bearer ${args.providerConfig.authToken}`,
+        // 流式请求显式放弃压缩协商。实测记录（2026-09-01，glm-5-3-flash 经
+        // nolo.chat）：某个时间窗内 SSE 响应带 content-encoding: gzip 返回，
+        // 整条流被攒成 1 个 chunk，首字节 = 总生成时长（5.8/10.6/16.3s）；
+        // 同一分钟交替发起的 identity 请求为 123-138 chunks、首字节
+        // 1.2-3.6s。即压缩中间件一旦介入，流式的全部 TTFT 收益归零。
+        // 该窗口之后线上不再压缩 SSE（同一探针 6/6 未压缩），具体是哪一跳
+        // 压的没有定论；边缘侧已把流式路径排除出 encode
+        // （configureCaddyProxy.sh），这里是不依赖任何中间跳行为的客户端兜底
+        // ——用户侧本地代理同样可能压。SSE 正文只有几十 KB，放弃压缩的带宽
+        // 代价可忽略。
+        ...args.stream ? { "Accept-Encoding": "identity" } : {},
         // 客户端版本闸门第 2 层：与鉴权头同处一处统一注入，避免某条命令路径
         // 漏发。server 据此拒绝「该客户端认不了 wire 的托管模型」；不发这个头
         // 的旧客户端在 server 侧 fail-open（已知局限）。
@@ -785,6 +841,8 @@ export {
   createResponsesToolAccumulator,
   applyResponsesToolEvent,
   finalizeResponsesToolCalls,
+  collectStablePromptPrefix,
+  stablePromptCacheKey,
   resolvePlatformChatProviderConfig,
   buildPlatformChatCompletionRequest,
   parsePlatformChatCompletionData,

package/{chunk-5XCW6RQG.js → chunk-3ZTR7ZO4.js}

@@ -199,7 +199,7 @@ async function readDocFunc(args, thunkApi) {
     throw new Error(`\u65E0\u6548\u7684\u9875\u9762 ID: ${id}\u3002\u9875\u9762 ID \u901A\u5E38\u4EE5 "page-" \u5F00\u5934\u3002`);
   }
   try {
-    const { readAction } = await import("./read-J7RVVCHH.js");
+    const { readAction } = await import("./read-K4JU4MQC.js");
     const pageData = await readAction({ dbKey: id }, thunkApi);
     if (!pageData) {
       throw new Error(`\u672A\u627E\u5230 ID \u4E3A ${id} \u7684\u9875\u9762\u3002`);