mslxdff 0.1.203 → 0.1.208

package/src/bench/probe.js

@@ -1,7 +1,7 @@
 import { joinUrl } from "../providers/base.js";
 import { getCustomNormalizer } from "../providers/registry.js";
 import { createTransport } from "../transport/index.js";
-import { compatFetch } from "../compat.js";
+import { compatFetch } from "../compat.js";
 
 function normalizeModelsPayload(json, baseUrl = "") {
   if (!json) return [];

package/src/bench/runner.js

@@ -1,7 +1,7 @@
 import { joinUrl } from "../providers/base.js";
 import { computeMetrics, extractUsageFromJson } from "../metrics.js";
 import { createTransport } from "../transport/index.js";
-import { compatFetch } from "../compat.js";
+import { compatFetch } from "../compat.js";
 
 function extractInnerMessage(bodyText) {
   const t = String(bodyText || "");

package/src/bench/via-probe.js

@@ -1,7 +1,7 @@
 import { joinUrl } from "../providers/base.js";
 import { computeMetrics, extractUsageFromJson } from "../metrics.js";
 import { createTransport } from "../transport/index.js";
-import { compatFetch } from "../compat.js";
+import { compatFetch } from "../compat.js";
 
 function extractInnerMessage(bodyText) {
   const t = String(bodyText || "");

package/src/bench/via.js

@@ -1,144 +1,144 @@
-import { viaProbe } from "./via-probe.js";
-
-export async function resolveIncludeOpencode({ includeOpencode, isTTY, confirmFn, log = () => {} } = {}) {
-  if (!includeOpencode) return false;
-  if (!isTTY) {
-    try { log("via 已跳过 opencode（非 TTY，省额度；需对比请在 TTY 加 --include-opencode 并确认）"); } catch {}
-    return false;
-  }
-  if (typeof confirmFn === "function") {
-    try {
-      const ok = await confirmFn();
-      // confirmFn returns true for y, false otherwise
-      if (ok) return true;
-      try { log("已回落：via 跳过 opencode"); } catch {}
-      return false;
-    } catch { return false; }
-  }
-  return false;
-}
-
-export async function getOnlinePeers({ loadGroupsJoined, loadPeers, probeHealth } = {}) {
-  let candidates = [];
-  if (typeof loadPeers === "function") {
-    try { candidates = loadPeers() || []; } catch { candidates = []; }
-  } else {
-    try {
-      const m = await import("../state.js");
-      if (m.loadPeers) candidates = m.loadPeers() || [];
-    } catch {}
-  }
-  // also consider groupsJoined for validation - if no groups joined, via is meaningless, but we still allow static peers?
-  // For bench-via, if no groupsJoined at all, treat as no via capability -> return [] without probing?
-  // However static peers may exist without group; we probe them anyway.
-  // To satisfy "empty group -> skip probing" test, check loadGroupsJoined.
-  let joined = [];
-  if (typeof loadGroupsJoined === "function") {
-    try { joined = loadGroupsJoined() || []; } catch { joined = []; }
-  } else {
-    try {
-      const m = await import("../state.js");
-      if (m.loadGroupsJoined) joined = m.loadGroupsJoined() || [];
-    } catch {}
-  }
-  if (!joined.length) return [];
-  if (!candidates.length) return [];
-
-  // If probeHealth not provided, try import
-  let probe = probeHealth;
-  if (!probe) {
-    try {
-      const m = await import("../cli/group-helpers.js");
-      probe = m.probeHealth;
-    } catch { return candidates; }
-  }
-  if (!candidates.length) return [];
-  const online = [];
-  for (const p of candidates) {
-    try {
-      const r = await probe(p);
-      if (r && r.rank === 0 && !r.fail && !r.stale) online.push(p);
-    } catch {}
-  }
-  return online;
-}
-
-export async function orchestrateVia({
-  models = [],
-  peers = [],
-  directRunner = null,
-  viaProbeFn = viaProbe,
-  includeOpencode = false,
-  token,
-  timeoutMs = 30000,
-  clock = Date.now,
-  onProgress = null,
-  delayMs = 0,
-} = {}) {
-  const filtered = includeOpencode ? models : models.filter((m) => {
-    const s = typeof m === "string" ? m : (m.id || m.model || m.provider || "");
-    const first = String(s).split("/")[0].toLowerCase();
-    const prov = String(m?.provider || m?.providerId || first).toLowerCase();
-    return prov !== "opencode" && first !== "opencode";
-  });
-  // normalize model entries
-  const normModels = filtered.map((m) => {
-    if (typeof m === "string") return { provider: m.split("/")[0], model: m, id: m };
-    const id = m.id || m.model || "";
-    const provider = m.provider || m.providerId || (id.includes("/") ? id.split("/")[0] : "");
-    return { provider, model: id, id };
-  });
-  const results = [];
-  const total = normModels.length;
-  for (let idx = 0; idx < normModels.length; idx++) {
-    const entry = normModels[idx];
-    const provider = entry.provider;
-    const model = entry.model;
-    const seq = idx + 1;
-    // direct —— 串行，完成后立刻回调供实时进度
-    let direct = null;
-    if (directRunner) {
-      try { direct = await directRunner({ provider, model, timeoutMs, clock }); } catch (e) { direct = { ok: false, label: "网络错误", error: String(e), ttfbMs: null, totalMs: 0 }; }
-    } else {
-      direct = { ok: false, label: "未配置 directRunner", error: "missing directRunner", ttfbMs: null, totalMs: 0 };
-    }
-    if (typeof onProgress === "function") try { await onProgress({ phase: "direct", provider, model, seq, total, result: direct }); } catch {}
-    if (delayMs) await new Promise((r) => setTimeout(r, delayMs));
-    const via = {};
-    // 串行逐个 peer，避免 A/B/C 同时打同一上游并发限流
-    for (const peer of peers) {
-      const raw = String(peer.name || peer.id || peer.url || "peer");
-      let peerId = raw;
-      // 组员 name 可能存为完整 url（如 http://172.93...），统一短化为 host:port，防止 report 端 peerIds 与 via keys 不一致导致表格空白
-      if (raw.includes("://")) {
-        try { const u = new URL(raw); const host = u.hostname; const port = u.port ? `:${u.port}` : ""; if (host) peerId = `${host}${port}`; else peerId = raw.slice(-16); } catch { peerId = raw.slice(-16); }
-      }
-      const peerToken = peer.token || token || "";
-      let r;
-      try {
-        r = await viaProbeFn({ peerUrl: peer.url, token: peerToken, providerId: provider, model, prompt: "hi", maxTokens: 5, timeoutMs, clock });
-        via[peerId] = r;
-      } catch (e) {
-        r = { ok: false, label: "网络错误", error: String(e?.message || e), ttfbMs: null, totalMs: 0 };
-        via[peerId] = r;
-      }
-      if (typeof onProgress === "function") try { await onProgress({ phase: "via", provider, model, seq, total, peerId, result: r }); } catch {}
-      if (delayMs) await new Promise((rr) => setTimeout(rr, delayMs));
-    }
-    // best
-    let best = "direct";
-    let bestTtfb = direct?.ok ? (direct.ttfbMs ?? direct.totalMs) : Infinity;
-    let deltaMs = null;
-    for (const [pid, rv] of Object.entries(via)) {
-      if (!rv.ok) continue;
-      const t = rv.ttfbMs ?? rv.totalMs;
-      if (t < bestTtfb) { bestTtfb = t; best = `via:${pid}`; }
-    }
-    if (best.startsWith("via:")) {
-      const d = direct?.ttfbMs ?? direct?.totalMs ?? 0;
-      deltaMs = bestTtfb - d;
-    }
-    results.push({ provider, model, direct, via, best, deltaMs, opencodeSkipped: !includeOpencode });
-  }
-  return results;
-}
+import { viaProbe } from "./via-probe.js";
+
+export async function resolveIncludeOpencode({ includeOpencode, isTTY, confirmFn, log = () => {} } = {}) {
+  if (!includeOpencode) return false;
+  if (!isTTY) {
+    try { log("via 已跳过 opencode（非 TTY，省额度；需对比请在 TTY 加 --include-opencode 并确认）"); } catch {}
+    return false;
+  }
+  if (typeof confirmFn === "function") {
+    try {
+      const ok = await confirmFn();
+      // confirmFn returns true for y, false otherwise
+      if (ok) return true;
+      try { log("已回落：via 跳过 opencode"); } catch {}
+      return false;
+    } catch { return false; }
+  }
+  return false;
+}
+
+export async function getOnlinePeers({ loadGroupsJoined, loadPeers, probeHealth } = {}) {
+  let candidates = [];
+  if (typeof loadPeers === "function") {
+    try { candidates = loadPeers() || []; } catch { candidates = []; }
+  } else {
+    try {
+      const m = await import("../state.js");
+      if (m.loadPeers) candidates = m.loadPeers() || [];
+    } catch {}
+  }
+  // also consider groupsJoined for validation - if no groups joined, via is meaningless, but we still allow static peers?
+  // For bench-via, if no groupsJoined at all, treat as no via capability -> return [] without probing?
+  // However static peers may exist without group; we probe them anyway.
+  // To satisfy "empty group -> skip probing" test, check loadGroupsJoined.
+  let joined = [];
+  if (typeof loadGroupsJoined === "function") {
+    try { joined = loadGroupsJoined() || []; } catch { joined = []; }
+  } else {
+    try {
+      const m = await import("../state.js");
+      if (m.loadGroupsJoined) joined = m.loadGroupsJoined() || [];
+    } catch {}
+  }
+  if (!joined.length) return [];
+  if (!candidates.length) return [];
+
+  // If probeHealth not provided, try import
+  let probe = probeHealth;
+  if (!probe) {
+    try {
+      const m = await import("../cli/group-helpers.js");
+      probe = m.probeHealth;
+    } catch { return candidates; }
+  }
+  if (!candidates.length) return [];
+  const online = [];
+  for (const p of candidates) {
+    try {
+      const r = await probe(p);
+      if (r && r.rank === 0 && !r.fail && !r.stale) online.push(p);
+    } catch {}
+  }
+  return online;
+}
+
+export async function orchestrateVia({
+  models = [],
+  peers = [],
+  directRunner = null,
+  viaProbeFn = viaProbe,
+  includeOpencode = false,
+  token,
+  timeoutMs = 30000,
+  clock = Date.now,
+  onProgress = null,
+  delayMs = 0,
+} = {}) {
+  const filtered = includeOpencode ? models : models.filter((m) => {
+    const s = typeof m === "string" ? m : (m.id || m.model || m.provider || "");
+    const first = String(s).split("/")[0].toLowerCase();
+    const prov = String(m?.provider || m?.providerId || first).toLowerCase();
+    return prov !== "opencode" && first !== "opencode";
+  });
+  // normalize model entries
+  const normModels = filtered.map((m) => {
+    if (typeof m === "string") return { provider: m.split("/")[0], model: m, id: m };
+    const id = m.id || m.model || "";
+    const provider = m.provider || m.providerId || (id.includes("/") ? id.split("/")[0] : "");
+    return { provider, model: id, id };
+  });
+  const results = [];
+  const total = normModels.length;
+  for (let idx = 0; idx < normModels.length; idx++) {
+    const entry = normModels[idx];
+    const provider = entry.provider;
+    const model = entry.model;
+    const seq = idx + 1;
+    // direct —— 串行，完成后立刻回调供实时进度
+    let direct = null;
+    if (directRunner) {
+      try { direct = await directRunner({ provider, model, timeoutMs, clock }); } catch (e) { direct = { ok: false, label: "网络错误", error: String(e), ttfbMs: null, totalMs: 0 }; }
+    } else {
+      direct = { ok: false, label: "未配置 directRunner", error: "missing directRunner", ttfbMs: null, totalMs: 0 };
+    }
+    if (typeof onProgress === "function") try { await onProgress({ phase: "direct", provider, model, seq, total, result: direct }); } catch {}
+    if (delayMs) await new Promise((r) => setTimeout(r, delayMs));
+    const via = {};
+    // 串行逐个 peer，避免 A/B/C 同时打同一上游并发限流
+    for (const peer of peers) {
+      const raw = String(peer.name || peer.id || peer.url || "peer");
+      let peerId = raw;
+      // 组员 name 可能存为完整 url（如 http://172.93...），统一短化为 host:port，防止 report 端 peerIds 与 via keys 不一致导致表格空白
+      if (raw.includes("://")) {
+        try { const u = new URL(raw); const host = u.hostname; const port = u.port ? `:${u.port}` : ""; if (host) peerId = `${host}${port}`; else peerId = raw.slice(-16); } catch { peerId = raw.slice(-16); }
+      }
+      const peerToken = peer.token || token || "";
+      let r;
+      try {
+        r = await viaProbeFn({ peerUrl: peer.url, token: peerToken, providerId: provider, model, prompt: "hi", maxTokens: 5, timeoutMs, clock });
+        via[peerId] = r;
+      } catch (e) {
+        r = { ok: false, label: "网络错误", error: String(e?.message || e), ttfbMs: null, totalMs: 0 };
+        via[peerId] = r;
+      }
+      if (typeof onProgress === "function") try { await onProgress({ phase: "via", provider, model, seq, total, peerId, result: r }); } catch {}
+      if (delayMs) await new Promise((rr) => setTimeout(rr, delayMs));
+    }
+    // best
+    let best = "direct";
+    let bestTtfb = direct?.ok ? (direct.ttfbMs ?? direct.totalMs) : Infinity;
+    let deltaMs = null;
+    for (const [pid, rv] of Object.entries(via)) {
+      if (!rv.ok) continue;
+      const t = rv.ttfbMs ?? rv.totalMs;
+      if (t < bestTtfb) { bestTtfb = t; best = `via:${pid}`; }
+    }
+    if (best.startsWith("via:")) {
+      const d = direct?.ttfbMs ?? direct?.totalMs ?? 0;
+      deltaMs = bestTtfb - d;
+    }
+    results.push({ provider, model, direct, via, best, deltaMs, opencodeSkipped: !includeOpencode });
+  }
+  return results;
+}

package/src/bench/workbuddy-bench.js

@@ -1,6 +1,6 @@
 import { computeMetrics } from "../metrics.js";
 import { createTransport } from "../transport/index.js";
-import { compatFetch } from "../compat.js";
+import { compatFetch } from "../compat.js";
 
 function buildWorkbuddyHeaders(apiKey, auth) {
   const h = {

package/src/chat/config.js

@@ -1,11 +1,11 @@
-export const CHAT_PREFERRED = "mimo-v2.5-free";
-export const CHAT_FALLBACK = "big-pickle";
-// 128k 上下文 *95% ≈ 121.6k tokens，按 ~3.3 字符/token 估算 ≈ 400k 字符
-// 接近 95% 再压缩，避免频繁摘要
-export const CHAT_HISTORY_MAX_CHARS = 400000;
-export const CHAT_KEEP_RECENT = 40;
-export const CHAT_SUMMARY_TRIGGER = 400000;
-export const CHAT_MAX_TOOL_LOOPS = 6;
-export const CHAT_TIMEOUT_MS = 15000;
-export const CHAT_GATEWAY_TIMEOUT_MS = 25000;
-export const FORBIDDEN = ["-uninstall", "--uninstall"];
+export const CHAT_PREFERRED = "mimo-v2.5-free";
+export const CHAT_FALLBACK = "big-pickle";
+// 128k 上下文 *95% ≈ 121.6k tokens，按 ~3.3 字符/token 估算 ≈ 400k 字符
+// 接近 95% 再压缩，避免频繁摘要
+export const CHAT_HISTORY_MAX_CHARS = 400000;
+export const CHAT_KEEP_RECENT = 40;
+export const CHAT_SUMMARY_TRIGGER = 400000;
+export const CHAT_MAX_TOOL_LOOPS = 6;
+export const CHAT_TIMEOUT_MS = 15000;
+export const CHAT_GATEWAY_TIMEOUT_MS = 25000;
+export const FORBIDDEN = ["-uninstall", "--uninstall"];

package/src/chat/cooling.js

@@ -1,81 +1,81 @@
-/**
- * 冷却深模块：对 state 的 10min 冷却 + EMA 延迟做唯一封装
- * 注入化便于单测（内存 Map + now）
- */
-
-export function createCooling({
-  loadModelErrors,
-  saveModelErrors,
-  loadModelLatencies,
-  saveModelLatencies,
-  flush,
-  now = () => Date.now(),
-  cooldownMs = 10 * 60 * 1000,
-  slowCooldownMs = 10 * 60 * 1000,
-} = {}) {
-  const _loadErrors = loadModelErrors || (() => ({}));
-  const _saveErrors = saveModelErrors || (() => {});
-  const _loadLats = loadModelLatencies || (() => ({}));
-  const _saveLats = saveModelLatencies || (() => {});
-  const _flush = flush || (() => {});
-
-  async function isCooling(id) {
-    try {
-      const errors = _loadErrors() || {};
-      const e = errors[id];
-      if (!e || typeof e !== "object") return false;
-      const at = Number(e.at || 0);
-      if (!at) return false;
-      const isSlow = !!e.slow;
-      const cd = isSlow ? slowCooldownMs : cooldownMs;
-      return now() - at < cd && (e.status === "limit" || e.status === "error");
-    } catch {
-      return false;
-    }
-  }
-
-  async function recordError(id, status, { slow = false, latencyMs = 0 } = {}) {
-    try {
-      const errors = _loadErrors() || {};
-      const isLimit = Number(status) === 429 || String(status).includes("429");
-      const entryStatus = isLimit ? "limit" : "error";
-      errors[id] = { status: entryStatus, at: now(), code: Number.isInteger(Number(status)) ? Number(status) : null, slow: !!slow };
-      _saveErrors(errors);
-      try { _flush(); } catch {}
-      if (slow && Number.isFinite(latencyMs) && latencyMs > 0) {
-        try {
-          const lat = _loadLats() || {};
-          const prev = lat[id]?.emaMs;
-          const ema = prev ? Math.round(prev * 0.7 + latencyMs * 0.3) : Math.round(latencyMs);
-          lat[id] = { emaMs: ema, lastMs: Math.round(latencyMs), at: now(), count: (lat[id]?.count ?? 0) + 1 };
-          _saveLats(lat);
-          try { _flush(); } catch {}
-        } catch {}
-      }
-    } catch {}
-  }
-
-  async function recordOk(id, latencyMs) {
-    try {
-      const errors = _loadErrors() || {};
-      errors[id] = { status: "normal", at: now(), code: 200, slow: false };
-      _saveErrors(errors);
-      try { _flush(); } catch {}
-      if (Number.isFinite(latencyMs) && latencyMs > 0) {
-        const lat = _loadLats() || {};
-        const prev = lat[id]?.emaMs;
-        const ema = prev ? Math.round(prev * 0.7 + latencyMs * 0.3) : Math.round(latencyMs);
-        lat[id] = { emaMs: ema, lastMs: Math.round(latencyMs), at: now(), count: (lat[id]?.count ?? 0) + 1 };
-        _saveLats(lat);
-        try { _flush(); } catch {}
-      }
-    } catch {}
-  }
-
-  // 兼容旧命名
-  const recordChatError = recordError;
-  const recordChatOk = recordOk;
-  const isCoolingAsync = isCooling;
-
-  return { isCooling, isCoolingAsync, recordError, recordChatError, recordOk, recordChatOk };
-}
+/**
+ * 冷却深模块：对 state 的 10min 冷却 + EMA 延迟做唯一封装
+ * 注入化便于单测（内存 Map + now）
+ */
+
+export function createCooling({
+  loadModelErrors,
+  saveModelErrors,
+  loadModelLatencies,
+  saveModelLatencies,
+  flush,
+  now = () => Date.now(),
+  cooldownMs = 10 * 60 * 1000,
+  slowCooldownMs = 10 * 60 * 1000,
+} = {}) {
+  const _loadErrors = loadModelErrors || (() => ({}));
+  const _saveErrors = saveModelErrors || (() => {});
+  const _loadLats = loadModelLatencies || (() => ({}));
+  const _saveLats = saveModelLatencies || (() => {});
+  const _flush = flush || (() => {});
+
+  async function isCooling(id) {
+    try {
+      const errors = _loadErrors() || {};
+      const e = errors[id];
+      if (!e || typeof e !== "object") return false;
+      const at = Number(e.at || 0);
+      if (!at) return false;
+      const isSlow = !!e.slow;
+      const cd = isSlow ? slowCooldownMs : cooldownMs;
+      return now() - at < cd && (e.status === "limit" || e.status === "error");
+    } catch {
+      return false;
+    }
+  }
+
+  async function recordError(id, status, { slow = false, latencyMs = 0 } = {}) {
+    try {
+      const errors = _loadErrors() || {};
+      const isLimit = Number(status) === 429 || String(status).includes("429");
+      const entryStatus = isLimit ? "limit" : "error";
+      errors[id] = { status: entryStatus, at: now(), code: Number.isInteger(Number(status)) ? Number(status) : null, slow: !!slow };
+      _saveErrors(errors);
+      try { _flush(); } catch {}
+      if (slow && Number.isFinite(latencyMs) && latencyMs > 0) {
+        try {
+          const lat = _loadLats() || {};
+          const prev = lat[id]?.emaMs;
+          const ema = prev ? Math.round(prev * 0.7 + latencyMs * 0.3) : Math.round(latencyMs);
+          lat[id] = { emaMs: ema, lastMs: Math.round(latencyMs), at: now(), count: (lat[id]?.count ?? 0) + 1 };
+          _saveLats(lat);
+          try { _flush(); } catch {}
+        } catch {}
+      }
+    } catch {}
+  }
+
+  async function recordOk(id, latencyMs) {
+    try {
+      const errors = _loadErrors() || {};
+      errors[id] = { status: "normal", at: now(), code: 200, slow: false };
+      _saveErrors(errors);
+      try { _flush(); } catch {}
+      if (Number.isFinite(latencyMs) && latencyMs > 0) {
+        const lat = _loadLats() || {};
+        const prev = lat[id]?.emaMs;
+        const ema = prev ? Math.round(prev * 0.7 + latencyMs * 0.3) : Math.round(latencyMs);
+        lat[id] = { emaMs: ema, lastMs: Math.round(latencyMs), at: now(), count: (lat[id]?.count ?? 0) + 1 };
+        _saveLats(lat);
+        try { _flush(); } catch {}
+      }
+    } catch {}
+  }
+
+  // 兼容旧命名
+  const recordChatError = recordError;
+  const recordChatOk = recordOk;
+  const isCoolingAsync = isCooling;
+
+  return { isCooling, isCoolingAsync, recordError, recordChatError, recordOk, recordChatOk };
+}

package/src/chat/direct.js

@@ -1,58 +1,58 @@
-import { performance } from "node:perf_hooks";
-import { compatFetch } from "../compat.js";
-
-function isInput400(status, msg, hasTools) {
-  return status === 400 && /prompt|messages/i.test(String(msg || "")) && hasTools;
-}
-
-/**
- * 直连深模块：mimo/pickle 经 createUpstreamClient 的 stream:false 调用
- * 注入化：便于用 fake client 触发 400→去 tools 重试
- */
-export function createDirectClient({ createUpstreamClient, chatTimeoutMs = 15000, env = process.env, fetchImpl = compatFetch } = {}) {
-  const _create = createUpstreamClient || (() => { throw new Error("createUpstreamClient not injected"); });
-
-  async function doChat({ messages, tools, model }, withoutTools) {
-    const prevAnon = env.MSLXDFF_FREE_ANON;
-    const needDisable = model === "mimo-v2.5-free" || model === "big-pickle";
-    if (needDisable) env.MSLXDFF_FREE_ANON = "0";
-    const client = _create({ connectTimeoutMs: chatTimeoutMs, keepAlive: false, fetchImpl });
-    const body = { model: model || "mimo-v2.5-free", messages, stream: false };
-    if (!withoutTools && tools?.length) {
-      body.tools = tools;
-      body.tool_choice = "auto";
-    }
-    try {
-      const res = await client.chat(body);
-      const txt = await res.text();
-      let j;
-      try { j = JSON.parse(txt); } catch { return { ok: false, error: `non-json upstream: ${txt.slice(0, 800)}`, status: res.status }; }
-      if (!res.ok) {
-        const msg = j?.error?.message || txt.slice(0, 800);
-        if (!withoutTools && isInput400(res.status, msg, !!tools?.length)) {
-          try { await client.close(); } catch {}
-          // 重试去 tools
-          const retry = await doChat({ messages, model }, true);
-          if (retry.ok) return { ...retry, retriedWithoutTools: true };
-          return { ok: false, error: msg, status: res.status, retried: retry.error };
-        }
-        return { ok: false, error: msg, status: res.status };
-      }
-      const choice = j.choices?.[0];
-      if (!choice) return { ok: false, error: "no choice", status: res.status };
-      return { ok: true, message: choice.message, usage: j.usage, raw: j, status: res.status };
-    } finally {
-      try { await client.close(); } catch {}
-      if (needDisable) {
-        if (prevAnon === undefined) delete env.MSLXDFF_FREE_ANON;
-        else env.MSLXDFF_FREE_ANON = prevAnon;
-      }
-    }
-  }
-
-  async function chatOnce(opts) {
-    return doChat(opts, false);
-  }
-
-  return { chatOnce, _doChat: doChat };
-}
+import { performance } from "node:perf_hooks";
+import { compatFetch } from "../compat.js";
+
+function isInput400(status, msg, hasTools) {
+  return status === 400 && /prompt|messages/i.test(String(msg || "")) && hasTools;
+}
+
+/**
+ * 直连深模块：mimo/pickle 经 createUpstreamClient 的 stream:false 调用
+ * 注入化：便于用 fake client 触发 400→去 tools 重试
+ */
+export function createDirectClient({ createUpstreamClient, chatTimeoutMs = 15000, env = process.env, fetchImpl = compatFetch } = {}) {
+  const _create = createUpstreamClient || (() => { throw new Error("createUpstreamClient not injected"); });
+
+  async function doChat({ messages, tools, model }, withoutTools) {
+    const prevAnon = env.MSLXDFF_FREE_ANON;
+    const needDisable = model === "mimo-v2.5-free" || model === "big-pickle";
+    if (needDisable) env.MSLXDFF_FREE_ANON = "0";
+    const client = _create({ connectTimeoutMs: chatTimeoutMs, keepAlive: false, fetchImpl });
+    const body = { model: model || "mimo-v2.5-free", messages, stream: false };
+    if (!withoutTools && tools?.length) {
+      body.tools = tools;
+      body.tool_choice = "auto";
+    }
+    try {
+      const res = await client.chat(body);
+      const txt = await res.text();
+      let j;
+      try { j = JSON.parse(txt); } catch { return { ok: false, error: `non-json upstream: ${txt.slice(0, 800)}`, status: res.status }; }
+      if (!res.ok) {
+        const msg = j?.error?.message || txt.slice(0, 800);
+        if (!withoutTools && isInput400(res.status, msg, !!tools?.length)) {
+          try { await client.close(); } catch {}
+          // 重试去 tools
+          const retry = await doChat({ messages, model }, true);
+          if (retry.ok) return { ...retry, retriedWithoutTools: true };
+          return { ok: false, error: msg, status: res.status, retried: retry.error };
+        }
+        return { ok: false, error: msg, status: res.status };
+      }
+      const choice = j.choices?.[0];
+      if (!choice) return { ok: false, error: "no choice", status: res.status };
+      return { ok: true, message: choice.message, usage: j.usage, raw: j, status: res.status };
+    } finally {
+      try { await client.close(); } catch {}
+      if (needDisable) {
+        if (prevAnon === undefined) delete env.MSLXDFF_FREE_ANON;
+        else env.MSLXDFF_FREE_ANON = prevAnon;
+      }
+    }
+  }
+
+  async function chatOnce(opts) {
+    return doChat(opts, false);
+  }
+
+  return { chatOnce, _doChat: doChat };
+}

package/src/chat/index.js

@@ -1,6 +1,6 @@
-import { startRepl } from "./repl.js";
-
-export async function startChat({ singleShot } = {}) {
-  // 独立进程：本函数在用户终端前台运行，与 daemon 无父子关系，daemon 重启不影响
-  await startRepl({ singleShot });
-}
+import { startRepl } from "./repl.js";
+
+export async function startChat({ singleShot } = {}) {
+  // 独立进程：本函数在用户终端前台运行，与 daemon 无父子关系，daemon 重启不影响
+  await startRepl({ singleShot });
+}