mslxdff 0.1.83 → 0.1.85

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mslxdff",
-  "version": "0.1.83",
+  "version": "0.1.85",
   "description": "测试项目，请勿使用。",
   "type": "module",
   "bin": {

package/src/bench/cline-bench.js

@@ -0,0 +1,54 @@
+import { clineHeaders } from "../providers/cline/headers.js";
+import { computeMetrics } from "../metrics.js";
+
+export async function clineBenchOne({ baseUrl, model, accessToken, prompt, maxTokens, timeoutMs, fetchImpl }) {
+  const controller = new AbortController();
+  const timer = setTimeout(() => controller.abort(new Error(`timeout ${timeoutMs}ms`)), timeoutMs);
+  const t0 = performance.now();
+  let ttfbMs = null;
+  let content = "";
+  try {
+    const res = await fetchImpl(`${baseUrl}/api/v1/chat/completions`, {
+      method: "POST",
+      headers: { ...clineHeaders(`sess_bench_${Date.now()}`, accessToken), Accept: "text/event-stream" },
+      body: JSON.stringify({ model, messages: [{ role: "user", content: prompt }], stream: true, max_tokens: maxTokens, session_id: `sess_bench_${Date.now()}`, reasoning_effort: "high" }),
+      signal: controller.signal,
+    });
+    if (res instanceof Error) throw res;
+    if (!res.ok) {
+      let txt = "";
+      try { txt = await res.text(); } catch {}
+      const label = res.status === 401 ? "鉴权失败" : res.status === 429 ? "限流" : res.status >= 500 ? `上游错误 ${res.status}` : `HTTP ${res.status}`;
+      return { id: model, ok: false, status: res.status, label, error: txt.slice(0, 300), ttfbMs, totalMs: Math.round(performance.now() - t0), tps: null, charsPerSec: null, tokens: null };
+    }
+    const reader = res.body.getReader();
+    const decoder = new TextDecoder();
+    let buf = "";
+    for (;;) {
+      const { done, value } = await reader.read();
+      if (done) break;
+      if (ttfbMs === null) ttfbMs = Math.round(performance.now() - t0);
+      buf += decoder.decode(value, { stream: true });
+      let idx;
+      while ((idx = buf.indexOf("\n")) >= 0) {
+        const line = buf.slice(0, idx);
+        buf = buf.slice(idx + 1);
+        if (!line.startsWith("data:")) continue;
+        const payload = line.slice(5).trim();
+        if (!payload || payload === "[DONE]") continue;
+        try {
+          const j = JSON.parse(payload);
+          const c = j?.choices?.[0]?.delta?.content || j?.choices?.[0]?.message?.content || "";
+          if (typeof c === "string") content += c;
+        } catch {}
+      }
+    }
+    const totalMs = Math.round(performance.now() - t0);
+    const chars = content.length;
+    const { tps, charsPerSec } = computeMetrics({ ttfbMs, totalMs, promptTokens: null, completionTokens: null, chars });
+    return { id: model, ok: true, status: 200, label: "成功", ttfbMs, totalMs, tps, charsPerSec, tokens: null, chars };
+  } catch (e) {
+    const msg = e?.message || String(e);
+    return { id: model, ok: false, label: /timeout|abort/i.test(msg) ? "超时" : "网络错误", error: msg.slice(0, 300), ttfbMs, totalMs: Math.round(performance.now() - t0), tps: null, charsPerSec: null, tokens: null };
+  } finally { clearTimeout(timer); }
+}

package/src/bench/via-routes.js

@@ -0,0 +1,87 @@
+import { readFileSync, existsSync } from "node:fs";
+import { join, dirname } from "node:path";
+import os from "node:os";
+import { defaultStateFile } from "../state/store.js";
+import { atomicWriteSync } from "../state/persist.js";
+
+export function defaultViaRoutesFile() {
+  if (process.env.MSLXDFF_VIA_ROUTES_FILE) return String(process.env.MSLXDFF_VIA_ROUTES_FILE).trim();
+  const sf = defaultStateFile();
+  return join(dirname(sf), "via-routes.json");
+}
+
+function viaTtlMs() {
+  const raw = process.env.MSLXDFF_VIA_ROUTE_TTL_MS;
+  if (raw === undefined || raw === null || raw === "") return 0;
+  const s = String(raw).trim().toLowerCase();
+  if (s === "0" || s === "off" || s === "false" || s === "no" || s === "disable" || s === "disabled") return 0;
+  const n = Number(s);
+  if (Number.isInteger(n) && n >= 0) return n;
+  return 0;
+}
+
+export function loadViaRoutes(file) {
+  const f = file || defaultViaRoutesFile();
+  try {
+    if (!existsSync(f)) return { version: 1, at: null, routes: {}, meta: {} };
+    const j = JSON.parse(readFileSync(f, "utf8"));
+    if (j && typeof j === "object" && j.routes && typeof j.routes === "object") return j;
+    if (j && typeof j === "object" && !j.routes) return { version: 1, at: j.at || null, routes: j, meta: {} };
+    return { version: 1, at: null, routes: {}, meta: {} };
+  } catch {
+    return { version: 1, at: null, routes: {}, meta: {} };
+  }
+}
+
+export function getViaRoute(model, { file, ttlMs } = {}) {
+  const id = String(model || "").trim();
+  if (!id) return null;
+  const f = file || defaultViaRoutesFile();
+  const data = loadViaRoutes(f);
+  const entry = data.routes?.[id];
+  if (!entry) return null;
+  const t = ttlMs !== undefined ? ttlMs : viaTtlMs();
+  if (t > 0 && entry.at) {
+    const atMs = Date.parse(entry.at);
+    if (Number.isFinite(atMs) && Date.now() - atMs > t) return null;
+  }
+  return entry;
+}
+
+export function saveViaRoutes(results, { file, meta } = {}) {
+  const f = file || defaultViaRoutesFile();
+  const now = new Date().toISOString();
+  const prev = loadViaRoutes(f);
+  const nextRoutes = { ...(prev.routes || {}) };
+  for (const r of results || []) {
+    const id = String(r.model || r.id || "").trim();
+    if (!id) continue;
+    const best = String(r.best || "direct").trim() || "direct";
+    const direct = r.direct ? { ok: Boolean(r.direct.ok), ttfbMs: r.direct.ttfbMs ?? r.direct.totalMs ?? null, totalMs: r.direct.totalMs ?? null, label: r.direct.label || null, error: r.direct.error ? String(r.direct.error).slice(0, 300) : null } : null;
+    const via = {};
+    for (const [k, v] of Object.entries(r.via || {})) {
+      via[k] = v?.ok ? { ok: true, ttfbMs: v.ttfbMs ?? v.totalMs ?? null, totalMs: v.totalMs ?? null } : { ok: false, ttfbMs: v?.ttfbMs ?? null, totalMs: v?.totalMs ?? null, label: v?.label || v?.error || "offline" };
+    }
+    nextRoutes[id] = {
+      best,
+      direct,
+      via,
+      deltaMs: r.deltaMs ?? null,
+      provider: r.provider || id.split("/")[0] || "",
+      at: now,
+    };
+  }
+  const out = {
+    version: 1,
+    at: now,
+    routes: nextRoutes,
+    meta: meta || prev.meta || {},
+  };
+  atomicWriteSync(f, out);
+  return out;
+}
+
+export function clearViaRoutes(file) {
+  const f = file || defaultViaRoutesFile();
+  atomicWriteSync(f, { version: 1, at: new Date().toISOString(), routes: {}, meta: {} });
+}

package/src/bench/via.js

@@ -72,6 +72,8 @@ export async function orchestrateVia({
   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 || "");
@@ -87,17 +89,23 @@ export async function orchestrateVia({
     return { provider, model: id, id };
   });
   const results = [];
-  for (const entry of normModels) {
+  const total = normModels.length;
+  for (let idx = 0; idx < normModels.length; idx++) {
+    const entry = normModels[idx];
     const provider = entry.provider;
     const model = entry.model;
-    // direct
+    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;
@@ -106,12 +114,16 @@ export async function orchestrateVia({
         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 {
-        const r = await viaProbeFn({ peerUrl: peer.url, token: peerToken, providerId: provider, model, prompt: "hi", maxTokens: 5, timeoutMs, clock });
+        r = await viaProbeFn({ peerUrl: peer.url, token: peerToken, providerId: provider, model, prompt: "hi", maxTokens: 5, timeoutMs, clock });
         via[peerId] = r;
       } catch (e) {
-        via[peerId] = { ok: false, label: "网络错误", error: String(e?.message || e), ttfbMs: null, totalMs: 0 };
+        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";

package/src/bench/workbuddy-bench.js

@@ -0,0 +1,70 @@
+import { computeMetrics } from "../metrics.js";
+
+function buildWorkbuddyHeaders(apiKey, auth) {
+  const h = {
+    "Content-Type": "application/json",
+    Accept: "text/event-stream",
+    "User-Agent": "CLI/2.115.0 WorkBuddy/2.115.0",
+    Origin: "https://www.codebuddy.cn",
+    Referer: "https://www.codebuddy.cn/",
+    "X-Product": "SaaS",
+  };
+  if (apiKey) h["Authorization"] = `Bearer ${apiKey}`;
+  if (auth?.uid) h["X-User-Id"] = auth.uid;
+  h["X-Domain"] = auth?.domain || "www.codebuddy.cn";
+  if (auth?.enterpriseId) { h["X-Enterprise-Id"] = auth.enterpriseId; h["X-Tenant-Id"] = auth.enterpriseId; }
+  return h;
+}
+
+export async function workbuddyBenchOne({ baseUrl, chatPath = "/v2/chat/completions", model, apiKey, auth, prompt = "hi", maxTokens = 5, timeoutMs = 30000, fetchImpl = globalThis.fetch }) {
+  const controller = new AbortController();
+  const timer = setTimeout(() => controller.abort(new Error(`timeout ${timeoutMs}ms`)), timeoutMs);
+  const t0 = typeof performance !== "undefined" && performance.now ? performance.now() : Date.now();
+  let ttfbMs = null;
+  let content = "";
+  try {
+    const url = String(baseUrl).replace(/\/+$/, "") + String(chatPath || "/v2/chat/completions");
+    const headers = buildWorkbuddyHeaders(apiKey, auth);
+    const rawModel = String(model || "").trim();
+    const body = { model: rawModel, stream: true, messages: [{ role: "user", content: prompt }], max_tokens: maxTokens };
+    const res = await fetchImpl(url, { method: "POST", headers, body: JSON.stringify(body), signal: controller.signal });
+    if (res instanceof Error) throw res;
+    if (!res.ok) {
+      let txt = "";
+      try { txt = await res.text(); } catch {}
+      const label = res.status === 401 ? "鉴权失败" : res.status === 402 ? "余额不足" : res.status === 429 ? "限流" : res.status >= 500 ? `上游错误 ${res.status}` : `HTTP ${res.status}`;
+      return { id: model, ok: false, status: res.status, label, error: txt.slice(0, 300), ttfbMs, totalMs: Math.round((typeof performance !== "undefined" && performance.now ? performance.now() : Date.now()) - t0), tps: null, charsPerSec: null, tokens: null };
+    }
+    // SSE stream parsing — 复用 cline 逻辑
+    const reader = res.body.getReader();
+    const decoder = new TextDecoder();
+    let buf = "";
+    for (;;) {
+      const { done, value } = await reader.read();
+      if (done) break;
+      if (ttfbMs === null) ttfbMs = Math.round((typeof performance !== "undefined" && performance.now ? performance.now() : Date.now()) - t0);
+      buf += decoder.decode(value, { stream: true });
+      let idx;
+      while ((idx = buf.indexOf("\n")) >= 0) {
+        const line = buf.slice(0, idx);
+        buf = buf.slice(idx + 1);
+        if (!line.startsWith("data:")) continue;
+        const payload = line.slice(5).trim();
+        if (!payload || payload === "[DONE]") continue;
+        try {
+          const j = JSON.parse(payload);
+          const c = j?.choices?.[0]?.delta?.content || j?.choices?.[0]?.message?.content || "";
+          if (typeof c === "string") content += c;
+        } catch {}
+      }
+    }
+    const totalMs = Math.round((typeof performance !== "undefined" && performance.now ? performance.now() : Date.now()) - t0);
+    const chars = content.length;
+    const { tps, charsPerSec } = computeMetrics({ ttfbMs, totalMs, promptTokens: null, completionTokens: null, chars });
+    return { id: model, ok: true, status: 200, label: "成功", ttfbMs, totalMs, tps, charsPerSec, tokens: null, chars };
+  } catch (e) {
+    const msg = e?.message || String(e);
+    const totalMs = Math.round((typeof performance !== "undefined" && performance.now ? performance.now() : Date.now()) - t0);
+    return { id: model, ok: false, label: /timeout|abort/i.test(msg) ? "超时" : "网络错误", error: msg.slice(0, 300), ttfbMs, totalMs, tps: null, charsPerSec: null, tokens: null };
+  } finally { clearTimeout(timer); }
+}

package/src/cli/commands/provider/bench-via.js

@@ -0,0 +1,183 @@
+import { defaultChatPath } from "../../../state/provider-config.js";
+import { isRefreshToken, clineHeaders } from "../../../providers/cline/headers.js";
+import { refreshTokenForBase } from "../../../providers/cline/auth.js";
+import { runOne } from "../../../bench/runner.js";
+import { formatViaReport } from "../../../bench/report.js";
+import { clineBenchOne } from "../../../bench/cline-bench.js";
+import { workbuddyBenchOne } from "../../../bench/workbuddy-bench.js";
+
+export function buildHeadersForProvider(providerId, apiKey, auth) {
+  const h = {};
+  if (String(providerId).toLowerCase() === "workbuddy") {
+    h["Content-Type"] = "application/json";
+    h["Accept"] = "text/event-stream";
+    h["User-Agent"] = "CLI/2.115.0 WorkBuddy/2.115.0";
+    h["Origin"] = "https://www.codebuddy.cn";
+    h["Referer"] = "https://www.codebuddy.cn/";
+    h["X-Product"] = "SaaS";
+    if (apiKey) h["Authorization"] = `Bearer ${apiKey}`;
+    if (auth?.uid) h["X-User-Id"] = auth.uid;
+    h["X-Domain"] = auth?.domain || "www.codebuddy.cn";
+    if (auth?.enterpriseId) { h["X-Enterprise-Id"] = auth.enterpriseId; h["X-Tenant-Id"] = auth.enterpriseId; }
+    return h;
+  }
+  if (apiKey) h["Authorization"] = `Bearer ${apiKey}`;
+  return h;
+}
+
+export async function handleVia({ providerId, opts, fetchImpl, loadConfigs, loadKeys, loadAllowed, loadBaseUrl }) {
+  const { getOnlinePeers, orchestrateVia, resolveIncludeOpencode } = await import("../../../bench/via.js");
+  const peers = await getOnlinePeers();
+  if (!peers.length) {
+    const msg = "未加入组或无在线 peer，--via 无意义。先 mslxdff -group list / -addtogroup";
+    if (opts.json) console.log(JSON.stringify({ meta: { at: new Date().toISOString(), samples: opts.samples, timeout: opts.timeoutMs, includeOpencode: false, peers: [], opencodeSkipped: true }, results: [], advice: msg }, null, 2));
+    else console.log(msg);
+    process.exit(0);
+  }
+  const isTTY = Boolean(process.stdin.isTTY && process.stdout.isTTY);
+  let includeOpencode = opts.includeOpencode;
+  if (includeOpencode) {
+    const confirmFn = async () => {
+      const readline = await import("node:readline");
+      return new Promise((res) => {
+        const rl = readline.createInterface({ input: process.stdin, output: process.stdout });
+        rl.question("将消耗 B/C/D 的 opencode 额度，确认测 opencode via？y/N ", (ans) => { rl.close(); res(ans.trim().toLowerCase() === "y"); });
+      });
+    };
+    const log = (s) => (opts.json ? console.error(s) : console.log(s));
+    includeOpencode = await resolveIncludeOpencode({ includeOpencode, isTTY, confirmFn, log });
+  }
+  const { loadToken } = await import("../../../state.js");
+  let token = "";
+  try { token = (await loadToken()).token || ""; } catch {}
+  let targetIds = [];
+  const isAll = providerId === "bench" || providerId === "all";
+  if (isAll) {
+    const configs = loadConfigs();
+    const ids = new Set(Object.keys(configs));
+    ids.add("opencode"); ids.add("openrouter");
+    for (const pid of [...ids]) {
+      const allowed = loadAllowed(pid) || [];
+      if (allowed.length) targetIds.push(pid);
+    }
+    try {
+      const { readFileSync } = await import("node:fs");
+      const { defaultStateFile } = await import("../../../state.js");
+      const raw = JSON.parse(readFileSync(defaultStateFile(), "utf8"));
+      for (const k of Object.keys(raw.providerConfigs || {})) if (!targetIds.includes(k) && (loadAllowed(k) || []).length) targetIds.push(k);
+      for (const k of Object.keys(raw.providerKeys || {})) if (!targetIds.includes(k) && (loadAllowed(k) || []).length) targetIds.push(k);
+    } catch {}
+    if (!targetIds.length) targetIds = ["openrouter", "workbuddy", "clinebot"].filter((p) => (loadAllowed(p) || []).length);
+  } else {
+    targetIds = [providerId];
+  }
+  const allResults = [];
+  const viaLog = (s) => (opts.json ? console.error(s) : console.log(s));
+  viaLog(`bench-via: direct vs via peers (samples=${opts.samples}, timeout=${opts.timeoutMs}ms, ${includeOpencode ? "opencode=included" : "opencode=skipped"}) peers=${peers.map((p) => p.id).join(",")}  模式: 串行(直连→逐 peer)`);
+  const delayMs = Number(process.env.MSLXDFF_BENCH_DELAY_MS || 120) || 0;
+  for (const pid of targetIds) {
+    const cfg = (loadConfigs()[pid] || {});
+    const keys = loadKeys(pid) || [];
+    const allowed = loadAllowed(pid) || [];
+    const baseUrl = (loadBaseUrl(pid) || cfg.baseUrl || "").trim();
+    if (!allowed.length) { viaLog(`provider ${pid}: 无勾选模型，跳过`); continue; }
+    if (!baseUrl && pid !== "opencode") { viaLog(`provider ${pid}: missing baseUrl 跳过`); continue; }
+    if (!keys.length && pid !== "opencode") { viaLog(`provider ${pid}: 未配置 Key 跳过`); continue; }
+    const chatPath = cfg.chatPath || defaultChatPath(pid);
+    let auths = [];
+    try { const m = await import("../../../state.js"); auths = m.loadProviderAuths ? m.loadProviderAuths(pid) : []; } catch {}
+    const models = allowed.map((id) => ({ provider: pid, model: String(id), id: String(id) }));
+    const directRunner = async ({ provider, model }) => {
+      const p = provider || pid;
+      const idx = models.findIndex((x) => x.model === model);
+      const kIdx = idx >= 0 ? idx % (keys.length || 1) : 0;
+      const aIdx = Math.min(kIdx, Math.max(auths.length - 1, 0));
+      const key = keys[kIdx] || keys[0] || "";
+      const auth = auths[aIdx] || auths[0] || null;
+      if (String(p).toLowerCase() === "workbuddy") {
+        return workbuddyBenchOne({ baseUrl, chatPath, model, apiKey: key, auth, prompt: "hi", maxTokens: 5, timeoutMs: opts.timeoutMs, fetchImpl });
+      }
+      const cKeys = keys.filter((k) => isRefreshToken(k, p));
+      if (cKeys.length) {
+        const normBase = String(baseUrl).replace(/\/+$/, "");
+        const chatBase = normBase.endsWith("/api/v1") ? normBase.slice(0, -7) : normBase;
+        const rt = cKeys[kIdx % cKeys.length];
+        const at = await refreshTokenForBase({ refreshToken: rt, baseUrl: normBase, fetchImpl });
+        if (!at) return { ok: false, label: "鉴权失败", error: "refresh failed", ttfbMs: null, totalMs: 0 };
+        return clineBenchOne({ baseUrl: chatBase, model, accessToken: at, prompt: "hi", maxTokens: 5, timeoutMs: opts.timeoutMs, fetchImpl });
+      }
+      const headers = buildHeadersForProvider(p, key, auth);
+      return runOne({ baseUrl: p === "opencode" ? (process.env.UPSTREAM_BASE_URL || "https://opencode.ai") : baseUrl, chatPath, model, providerId: p, apiKey: key || "public", headers, prompt: "hi", maxTokens: 5, timeoutMs: opts.timeoutMs, fetchImpl });
+    };
+    const { viaProbe } = await import("../../../bench/via-probe.js");
+    const viaProbeFn = async (args) => {
+      const { peerUrl, providerId, model } = args;
+      const p = providerId || pid;
+      const idx = models.findIndex((x) => x.model === model);
+      const kIdx = idx >= 0 ? idx % (keys.length || 1) : 0;
+      const aIdx = Math.min(kIdx, Math.max(auths.length - 1, 0));
+      const key = keys[kIdx] || keys[0] || "";
+      const auth = auths[aIdx] || auths[0] || null;
+      // workbuddy 强制 stream:true SSE 中继（与直连一致）
+      if (String(p).toLowerCase() === "workbuddy") {
+        const rawModel = String(model).startsWith(`${p}/`) ? String(model).slice(p.length + 1) : String(model);
+        const targetUrl = `${String(baseUrl).replace(/\/+$/, "")}${chatPath}`;
+        const headers = buildHeadersForProvider(p, key, auth);
+        const body = { model: rawModel, stream: true, messages: [{ role: "user", content: "hi" }], max_tokens: 5 };
+        const peer = peers.find((pe) => pe.url === peerUrl || (pe.name || pe.id) === peerUrl);
+        const peerToken = peer?.token || token;
+        return viaProbe({ peerUrl, token: peerToken, relayTarget: targetUrl, relayHeaders: headers, relayBody: body, timeoutMs: opts.timeoutMs, fetchImpl });
+      }
+      const cKeys = keys.filter((k) => isRefreshToken(k, p));
+      if (cKeys.length) {
+        const normBase = String(baseUrl).replace(/\/+$/, "");
+        const chatBase = normBase.endsWith("/api/v1") ? normBase.slice(0, -7) : normBase;
+        const rt = cKeys[0];
+        const at = await refreshTokenForBase({ refreshToken: rt, baseUrl: normBase, fetchImpl });
+        if (!at) return { ok: false, label: "鉴权失败", error: "refresh failed", ttfbMs: null, totalMs: 0 };
+        const targetUrl = `${chatBase}/api/v1/chat/completions`;
+        const headers = clineHeaders(`sess_bench_via_${Date.now()}`, at);
+        const rawModel = String(model).startsWith(`${p}/`) ? String(model).slice(p.length + 1) : String(model);
+        const body = { model: rawModel, messages: [{ role: "user", content: "hi" }], stream: true, max_tokens: 5, session_id: `sess_bench_via_${Date.now()}`, reasoning_effort: "high" };
+        const peer = peers.find((pe) => pe.url === peerUrl || (pe.name || pe.id) === peerUrl);
+        const peerToken = peer?.token || token;
+        return viaProbe({ peerUrl, token: peerToken, relayTarget: targetUrl, relayHeaders: headers, relayBody: body, timeoutMs: opts.timeoutMs, fetchImpl });
+      }
+      const rawModel = String(model).startsWith(`${p}/`) ? String(model).slice(p.length + 1) : String(model);
+      const targetUrl = p === "opencode" ? `${process.env.UPSTREAM_BASE_URL || "https://opencode.ai"}/zen/v1/chat/completions` : `${String(baseUrl).replace(/\/+$/, "")}${chatPath}`;
+      const headers = buildHeadersForProvider(p, key, auth);
+      const body = { model: rawModel, stream: false, messages: [{ role: "user", content: "hi" }], max_tokens: 5 };
+      const peer = peers.find((pe) => pe.url === peerUrl || (pe.name || pe.id) === peerUrl);
+      const peerToken = peer?.token || token;
+      return viaProbe({ peerUrl, token: peerToken, relayTarget: targetUrl, relayHeaders: headers, relayBody: body, timeoutMs: opts.timeoutMs, fetchImpl });
+    };
+    if (!opts.json) viaLog(`\n[${pid}] 共 ${models.length} 个模型，串行测试中...`);
+    else viaLog(`[${pid}] ${models.length} models sequential`);
+    const onProgress = async ({ phase, provider, model, seq, total, peerId, result }) => {
+      const short = String(model).length > 28 ? String(model).slice(0, 28) : String(model);
+      const ms = result?.ttfbMs != null ? `${result.ttfbMs}ms` : result?.totalMs != null ? `${result.totalMs}ms` : "—";
+      const okTag = result?.ok ? "成功" : (result?.label || "失败");
+      const extra = result?.ok ? "" : result?.error ? ` (${String(result.error).slice(0, 40)})` : "";
+      if (phase === "direct") viaLog(`  [${seq}/${total}] ${provider}/${short}  直连 ${ms} ${okTag}${extra} 完成`);
+      else viaLog(`    ↳ via ${peerId}  ${ms} ${okTag}${extra}`);
+    };
+    const { orchestrateVia } = await import("../../../bench/via.js");
+    const part = await orchestrateVia({ models, peers, directRunner, viaProbeFn, includeOpencode, token, timeoutMs: opts.timeoutMs, onProgress, delayMs });
+    allResults.push(...part);
+    if (!opts.json) viaLog(`  ${pid}: ${part.length} 模型完成 ✓`);
+  }
+  const meta = { at: new Date().toISOString(), samples: opts.samples, timeout: opts.timeoutMs, includeOpencode, peers: peers.map((p) => p.id), opencodeSkipped: !includeOpencode };
+  const report = formatViaReport(allResults, { peers, meta, json: opts.json });
+  if (opts.apply) {
+    const { saveViaRoutes } = await import("../../../bench/via-routes.js");
+    const saved = saveViaRoutes(allResults, { meta });
+    const viaLog2 = (s) => (opts.json ? console.error(s) : console.log(s));
+    viaLog2(`\nvia-routes 已落盘: ${saved.at} 共 ${Object.keys(saved.routes).length} 条 → ${saved.routes[Object.keys(saved.routes)[0]] ? "" : ""}${(await import("../../../bench/via-routes.js")).defaultViaRoutesFile()}`);
+    for (const [m, e] of Object.entries(saved.routes)) {
+      if (allResults.some((r) => r.model === m)) viaLog2(`  ${m} → ${e.best}${e.deltaMs ? ` (${e.deltaMs}ms)` : ""}`);
+    }
+  }
+  if (opts.json) console.log(report.text);
+  else console.log("\n" + report.text);
+  process.exit(0);
+}

package/src/cli/commands/provider/bench.js

@@ -1,69 +1,20 @@
 import { probeModels } from "../../../bench/probe.js";
 import { runOne } from "../../../bench/runner.js";
-import { formatReport, formatViaReport } from "../../../bench/report.js";
+import { formatReport } from "../../../bench/report.js";
 import { defaultModelsPath, defaultChatPath } from "../../../state/provider-config.js";
-import { isRefreshToken, clineHeaders } from "../../../providers/cline/headers.js";
+import { isRefreshToken } from "../../../providers/cline/headers.js";
 import { refreshTokenForBase } from "../../../providers/cline/auth.js";
-import { computeMetrics } from "../../../metrics.js";
-
-async function clineBenchOne({ baseUrl, model, accessToken, prompt, maxTokens, timeoutMs, fetchImpl }) {
-  const controller = new AbortController();
-  const timer = setTimeout(() => controller.abort(new Error(`timeout ${timeoutMs}ms`)), timeoutMs);
-  const t0 = performance.now();
-  let ttfbMs = null;
-  let content = "";
-  try {
-    const res = await fetchImpl(`${baseUrl}/api/v1/chat/completions`, {
-      method: "POST",
-      headers: { ...clineHeaders(`sess_bench_${Date.now()}`, accessToken), Accept: "text/event-stream" },
-      body: JSON.stringify({ model, messages: [{ role: "user", content: prompt }], stream: true, max_tokens: maxTokens, session_id: `sess_bench_${Date.now()}`, reasoning_effort: "high" }),
-      signal: controller.signal,
-    });
-    if (res instanceof Error) throw res;
-    if (!res.ok) {
-      let txt = "";
-      try { txt = await res.text(); } catch {}
-      const label = res.status === 401 ? "鉴权失败" : res.status === 429 ? "限流" : res.status >= 500 ? `上游错误 ${res.status}` : `HTTP ${res.status}`;
-      return { id: model, ok: false, status: res.status, label, error: txt.slice(0, 300), ttfbMs, totalMs: Math.round(performance.now() - t0), tps: null, charsPerSec: null, tokens: null };
-    }
-    const reader = res.body.getReader();
-    const decoder = new TextDecoder();
-    let buf = "";
-    for (;;) {
-      const { done, value } = await reader.read();
-      if (done) break;
-      if (ttfbMs === null) ttfbMs = Math.round(performance.now() - t0);
-      buf += decoder.decode(value, { stream: true });
-      let idx;
-      while ((idx = buf.indexOf("\n")) >= 0) {
-        const line = buf.slice(0, idx);
-        buf = buf.slice(idx + 1);
-        if (!line.startsWith("data:")) continue;
-        const payload = line.slice(5).trim();
-        if (!payload || payload === "[DONE]") continue;
-        try {
-          const j = JSON.parse(payload);
-          const c = j?.choices?.[0]?.delta?.content || j?.choices?.[0]?.message?.content || "";
-          if (typeof c === "string") content += c;
-        } catch {}
-      }
-    }
-    const totalMs = Math.round(performance.now() - t0);
-    const chars = content.length;
-    const { tps, charsPerSec } = computeMetrics({ ttfbMs, totalMs, promptTokens: null, completionTokens: null, chars });
-    return { id: model, ok: true, status: 200, label: "成功", ttfbMs, totalMs, tps, charsPerSec, tokens: null, chars };
-  } catch (e) {
-    const msg = e?.message || String(e);
-    return { id: model, ok: false, label: /timeout|abort/i.test(msg) ? "超时" : "网络错误", error: msg.slice(0, 300), ttfbMs, totalMs: Math.round(performance.now() - t0), tps: null, charsPerSec: null, tokens: null };
-  } finally { clearTimeout(timer); }
-}
+import { clineBenchOne } from "../../../bench/cline-bench.js";
+import { workbuddyBenchOne } from "../../../bench/workbuddy-bench.js";
+import { buildHeadersForProvider, handleVia } from "./bench-via.js";
 
 function parseBenchArgs(rest) {
-  const opts = { json: false, prompt: "hi", maxTokens: 32, timeoutMs: 30000, via: false, includeOpencode: false, samples: 1 };
+  const opts = { json: false, prompt: "hi", maxTokens: 32, timeoutMs: 30000, via: false, includeOpencode: false, samples: 1, apply: false };
   for (let i = 0; i < rest.length; i++) {
     const a = rest[i];
     if (a === "--json" || a === "-json") opts.json = true;
     else if (a === "--via" || a === "--bench-via") opts.via = true;
+    else if (a === "--apply" || a === "--write" || a === "--save") opts.apply = true;
     else if (a === "--include-opencode") opts.includeOpencode = true;
     else if (a === "--samples" && rest[i + 1]) opts.samples = Number(rest[++i]) || 1;
     else if (a.startsWith("--samples=")) opts.samples = Number(a.split("=")[1]) || 1;
@@ -77,160 +28,6 @@ function parseBenchArgs(rest) {
   return opts;
 }
 
-function buildHeadersForProvider(providerId, apiKey, auth) {
-  const h = {};
-  if (String(providerId).toLowerCase() === "workbuddy") {
-    h["Content-Type"] = "application/json";
-    h["Accept"] = "text/event-stream";
-    h["User-Agent"] = "CLI/2.115.0 WorkBuddy/2.115.0";
-    h["Origin"] = "https://www.codebuddy.cn";
-    h["Referer"] = "https://www.codebuddy.cn/";
-    h["X-Product"] = "SaaS";
-    if (apiKey) h["Authorization"] = `Bearer ${apiKey}`;
-    if (auth?.uid) h["X-User-Id"] = auth.uid;
-    h["X-Domain"] = auth?.domain || "www.codebuddy.cn";
-    if (auth?.enterpriseId) { h["X-Enterprise-Id"] = auth.enterpriseId; h["X-Tenant-Id"] = auth.enterpriseId; }
-    return h;
-  }
-  if (apiKey) h["Authorization"] = `Bearer ${apiKey}`;
-  return h;
-}
-
-async function handleVia({ providerId, opts, fetchImpl, loadConfigs, loadKeys, loadAllowed, loadBaseUrl }) {
-  const { getOnlinePeers, orchestrateVia, resolveIncludeOpencode } = await import("../../../bench/via.js");
-  const peers = await getOnlinePeers();
-  if (!peers.length) {
-    const msg = "未加入组或无在线 peer，--via 无意义。先 mslxdff -group list / -addtogroup";
-    if (opts.json) console.log(JSON.stringify({ meta: { at: new Date().toISOString(), samples: opts.samples, timeout: opts.timeoutMs, includeOpencode: false, peers: [], opencodeSkipped: true }, results: [], advice: msg }, null, 2));
-    else console.log(msg);
-    process.exit(0);
-  }
-  const isTTY = Boolean(process.stdin.isTTY && process.stdout.isTTY);
-  let includeOpencode = opts.includeOpencode;
-  if (includeOpencode) {
-    const confirmFn = async () => {
-      const readline = await import("node:readline");
-      return new Promise((res) => {
-        const rl = readline.createInterface({ input: process.stdin, output: process.stdout });
-        rl.question("将消耗 B/C/D 的 opencode 额度，确认测 opencode via？y/N ", (ans) => { rl.close(); res(ans.trim().toLowerCase() === "y"); });
-      });
-    };
-    const log = (s) => (opts.json ? console.error(s) : console.log(s));
-    includeOpencode = await resolveIncludeOpencode({ includeOpencode, isTTY, confirmFn, log });
-  }
-  const { loadToken } = await import("../../../state.js");
-  let token = "";
-  try { token = (await loadToken()).token || ""; } catch {}
-  // target providers
-  let targetIds = [];
-  const isAll = providerId === "bench" || providerId === "all";
-  if (isAll) {
-    const configs = loadConfigs();
-    const ids = new Set(Object.keys(configs));
-    ids.add("opencode"); ids.add("openrouter");
-    try { const m = await import("../../../state.js"); const raw = m.loadProviderKeys ? null : null; } catch {}
-    // collect from providerKeys via loadKeys probing? simple: iterate ids and keep those with allowed
-    for (const pid of [...ids]) {
-      const allowed = loadAllowed(pid) || [];
-      if (allowed.length) targetIds.push(pid);
-      else if (pid === "opencode" && !includeOpencode) continue;
-    }
-    // also check generic ids from state raw
-    try {
-      const { readFileSync } = await import("node:fs");
-      const { defaultStateFile } = await import("../../../state.js");
-      const raw = JSON.parse(readFileSync(defaultStateFile(), "utf8"));
-      for (const k of Object.keys(raw.providerConfigs || {})) if (!targetIds.includes(k) && (loadAllowed(k) || []).length) targetIds.push(k);
-      for (const k of Object.keys(raw.providerKeys || {})) if (!targetIds.includes(k) && (loadAllowed(k) || []).length) targetIds.push(k);
-    } catch {}
-    if (!targetIds.length) targetIds = ["openrouter", "workbuddy", "clinebot"].filter((p) => (loadAllowed(p) || []).length);
-  } else {
-    targetIds = [providerId];
-  }
-  const allResults = [];
-  const viaLog = (s) => (opts.json ? console.error(s) : console.log(s));
-  viaLog(`bench-via: direct vs via peers (samples=${opts.samples}, timeout=${opts.timeoutMs}ms, ${includeOpencode ? "opencode=included" : "opencode=skipped"}) peers=${peers.map((p) => p.id).join(",")}`);
-  for (const pid of targetIds) {
-    const cfg = (loadConfigs()[pid] || {});
-    const keys = loadKeys(pid) || [];
-    const allowed = loadAllowed(pid) || [];
-    const baseUrl = (loadBaseUrl(pid) || cfg.baseUrl || "").trim();
-    if (!allowed.length) {
-      viaLog(`provider ${pid}: 无勾选模型，跳过`);
-      continue;
-    }
-    if (!baseUrl && pid !== "opencode") { viaLog(`provider ${pid}: missing baseUrl 跳过`); continue; }
-    if (!keys.length && pid !== "opencode") { viaLog(`provider ${pid}: 未配置 Key 跳过`); continue; }
-    const chatPath = cfg.chatPath || defaultChatPath(pid);
-    let auths = [];
-    try { const m = await import("../../../state.js"); auths = m.loadProviderAuths ? m.loadProviderAuths(pid) : []; } catch {}
-    const models = allowed.map((id) => ({ provider: pid, model: String(id), id: String(id) }));
-    // directRunner per provider
-    const directRunner = async ({ provider, model }) => {
-      const p = provider || pid;
-      const idx = models.findIndex((x) => x.model === model);
-      const kIdx = idx >= 0 ? idx % (keys.length || 1) : 0;
-      const aIdx = Math.min(kIdx, Math.max(auths.length - 1, 0));
-      const key = keys[kIdx] || keys[0] || "";
-      const auth = auths[aIdx] || auths[0] || null;
-      const cKeys = keys.filter((k) => isRefreshToken(k, p));
-      if (cKeys.length) {
-        const normBase = String(baseUrl).replace(/\/+$/, "");
-        const chatBase = normBase.endsWith("/api/v1") ? normBase.slice(0, -7) : normBase;
-        const rt = cKeys[kIdx % cKeys.length];
-        const at = await refreshTokenForBase({ refreshToken: rt, baseUrl: normBase, fetchImpl });
-        if (!at) return { ok: false, label: "鉴权失败", error: "refresh failed", ttfbMs: null, totalMs: 0 };
-        return clineBenchOne({ baseUrl: chatBase, model, accessToken: at, prompt: "hi", maxTokens: 5, timeoutMs: opts.timeoutMs, fetchImpl });
-      }
-      const headers = buildHeadersForProvider(p, key, auth);
-      return runOne({ baseUrl: p === "opencode" ? (process.env.UPSTREAM_BASE_URL || "https://opencode.ai") : baseUrl, chatPath, model, providerId: p, apiKey: key || "public", headers, prompt: "hi", maxTokens: 5, timeoutMs: opts.timeoutMs, fetchImpl });
-    };
-    const { viaProbe } = await import("../../../bench/via-probe.js");
-    const { joinUrl: _joinUrl } = await import("../../../providers/base.js");
-    const viaProbeFn = async (args) => {
-      // 纯中继：A 把 targetUrl+headers+body 发给 B，B 原样 fetch 到上游（不查 B 本地 providerConfigs）
-      const { peerUrl, providerId, model } = args;
-      const p = providerId || pid;
-      const idx = models.findIndex((x) => x.model === model);
-      const kIdx = idx >= 0 ? idx % (keys.length || 1) : 0;
-      const aIdx = Math.min(kIdx, Math.max(auths.length - 1, 0));
-      const key = keys[kIdx] || keys[0] || "";
-      const auth = auths[aIdx] || auths[0] || null;
-      const cKeys = keys.filter((k) => isRefreshToken(k, p));
-      if (cKeys.length) {
-        const normBase = String(baseUrl).replace(/\/+$/, "");
-        const chatBase = normBase.endsWith("/api/v1") ? normBase.slice(0, -7) : normBase;
-        const rt = cKeys[0];
-        const at = await refreshTokenForBase({ refreshToken: rt, baseUrl: normBase, fetchImpl });
-        if (!at) return { ok: false, label: "鉴权失败", error: "refresh failed", ttfbMs: null, totalMs: 0 };
-        const targetUrl = `${chatBase}/api/v1/chat/completions`;
-        const headers = clineHeaders(`sess_bench_via_${Date.now()}`, at);
-        const rawModel = String(model).startsWith(`${p}/`) ? String(model).slice(p.length + 1) : String(model);
-        const body = { model: rawModel, messages: [{ role: "user", content: "hi" }], stream: true, max_tokens: 5, session_id: `sess_bench_via_${Date.now()}`, reasoning_effort: "high" };
-        const peer = peers.find((pe) => pe.url === peerUrl || (pe.name || pe.id) === peerUrl);
-        const peerToken = peer?.token || token;
-        return viaProbe({ peerUrl, token: peerToken, relayTarget: targetUrl, relayHeaders: headers, relayBody: body, timeoutMs: opts.timeoutMs, fetchImpl });
-      }
-      const rawModel = String(model).startsWith(`${p}/`) ? String(model).slice(p.length + 1) : String(model);
-      const targetUrl = p === "opencode" ? `${process.env.UPSTREAM_BASE_URL || "https://opencode.ai"}/zen/v1/chat/completions` : `${String(baseUrl).replace(/\/+$/, "")}${chatPath}`;
-      const headers = buildHeadersForProvider(p, key, auth);
-      const body = { model: rawModel, stream: false, messages: [{ role: "user", content: "hi" }], max_tokens: 5 };
-      // 用对端 peer.token 做 relay 鉴权（B 只认自己的 Bearer）
-      const peer = peers.find((pe) => pe.url === peerUrl || (pe.name || pe.id) === peerUrl);
-      const peerToken = peer?.token || token;
-      return viaProbe({ peerUrl, token: peerToken, relayTarget: targetUrl, relayHeaders: headers, relayBody: body, timeoutMs: opts.timeoutMs, fetchImpl });
-    };
-    const part = await orchestrateVia({ models, peers, directRunner, viaProbeFn, includeOpencode, token, timeoutMs: opts.timeoutMs });
-    allResults.push(...part);
-    if (!opts.json) viaLog(`  ${pid}: ${part.length} 模型完成`);
-  }
-  const meta = { at: new Date().toISOString(), samples: opts.samples, timeout: opts.timeoutMs, includeOpencode, peers: peers.map((p) => p.id), opencodeSkipped: !includeOpencode };
-  const report = formatViaReport(allResults, { peers, meta, json: opts.json });
-  if (opts.json) console.log(report.text);
-  else console.log("\n" + report.text);
-  process.exit(0);
-}
-
 export async function handleProviderBench(id, sub, rest, args, deps = {}) {
   const _restArr = rest || [];
   const _hasVia = _restArr.includes("--via") || _restArr.includes("--bench-via");
@@ -238,7 +35,6 @@ export async function handleProviderBench(id, sub, rest, args, deps = {}) {
   const isBench = sub === "bench" || sub === "benchmark" || sub === "eval" || sub === "test" || _isBenchViaAll;
   if (!isBench) return false;
   const opts = parseBenchArgs(_restArr);
-  // via branch
   if (opts.via) {
     const loadConfigs = deps.loadProviderConfigs || (await import("../../../state.js")).loadProviderConfigs;
     const loadKeys = deps.loadProviderKeys || (await import("../../../state.js")).loadProviderKeys;
@@ -301,6 +97,12 @@ export async function handleProviderBench(id, sub, rest, args, deps = {}) {
     const aIdx = Math.min(kIdx, Math.max(auths.length - 1, 0));
     const key = keys[kIdx];
     const auth = auths[aIdx] || auths[0] || null;
+    if (String(providerId).toLowerCase() === "workbuddy") {
+      const r = await workbuddyBenchOne({ baseUrl, chatPath, model, apiKey: key, auth, prompt: opts.prompt, maxTokens: opts.maxTokens, timeoutMs: opts.timeoutMs, fetchImpl });
+      results.push(r);
+      if (!opts.json) { if (r.ok) console.log(`OK  TTFB ${r.ttfbMs}ms  总 ${r.totalMs}ms  ${r.tps != null ? `${r.tps} t/s` : r.charsPerSec != null ? `${r.charsPerSec} 字/秒` : "—"}`); else console.log(`FAIL ${r.label} ${r.error ? `(${r.error.slice(0, 60)})` : ""}`); }
+      continue;
+    }
     if (rtKeys.length) {
       const rt = rtKeys[kIdx % rtKeys.length];
       const at = await refreshTokenForBase({ refreshToken: rt, baseUrl: normBase, fetchImpl });

package/src/routes/chat/gateway.js

@@ -10,6 +10,7 @@ import { handleHedge } from "./hedge-handler.js";
 import { handleLocalRelay } from "./local-handler.js";
 import { handlePeerRelay } from "./peer-handler.js";
 import { handleBroadbandRelay } from "./broadband-handler.js";
+import { handleViaRoute } from "./via-route-handler.js";
 import { handleExhaustedLocal, handleExhaustedAll } from "./exhausted-handler.js";
 import { normalizeFullId, getModelAlias } from "../../providers/model-id.js";
 
@@ -111,7 +112,7 @@ export function createChatGateway({ upstream, auto, logs, peers, maxHops, groups
       for (const e of sel.errors) evt("plugin-hook-error", { reqId, hook: "model:select", plugin: e.plugin, error: e.error });
     }
 
-    const handlerCtx = { reqId, model: null, body, hops, peers, plugins, evt, logError, logCall, logs };
+    const handlerCtx = { reqId, model: null, body, hops, peers, plugins, evt, logError, logCall, logs, workbuddyUid };
 
     // ===== Selector: 首次 auto 并发择优 =====
     if (useAuto && order.length > 1 && auto && !lockModel) {
@@ -176,8 +177,38 @@ export function createChatGateway({ upstream, auto, logs, peers, maxHops, groups
       }
     }
 
+    // ===== VIA-ROUTE 单路径择路（显式锁模型，不并发） =====
+    let viaRouteLastErr = null;
+    if (!useAuto && requested && requested.includes("/") && canForwardPeers && !lockModel && peers) {
+      try {
+        const vr = await handleViaRoute({
+          model: requested,
+          body,
+          peers,
+          handlerCtx,
+          evt,
+          logCall,
+          logError,
+          mark,
+          perf0,
+          stages,
+          startedAt,
+          plugins,
+          res,
+          requested,
+          useAuto,
+          lockModel,
+          auto,
+        });
+        if (vr.handled) return;
+        if (vr.lastErr) viaRouteLastErr = vr.lastErr;
+      } catch (e) {
+        evt("via-route-exception", { reqId, model: requested, error: errMsg(e) });
+      }
+    }
+
     // ===== Executor: 串行 trial =====
-    let lastErr = null;
+    let lastErr = viaRouteLastErr;
     for (let idx = 0; idx < order.length; idx++) {
       const model = order[idx];
       handlerCtx.model = model;

package/src/routes/chat/via-route-handler.js

@@ -0,0 +1,144 @@
+import { createRelayPipeline } from "./relay-pipeline.js";
+import { relay, SLOW_TOTAL_MS, STREAM_TIMEOUT_MS, STALL_TIMEOUT_MS, SCORE_STALL_MS } from "../stream.js";
+import { buildFallbackInfo } from "../fallback.js";
+import { getViaRoute } from "../../bench/via-routes.js";
+import { loadProviderKeys } from "../../state.js";
+import { SHARE_KEYS_HEADER } from "../../providers/share-keys.js";
+import { errMsg } from "../helpers.js";
+
+function shortLabel(p) {
+  const raw = String(p?.name || p?.id || p?.url || "").trim();
+  if (!raw) return "";
+  if (raw.includes("://")) {
+    try { const u = new URL(raw); return `${u.hostname}${u.port ? `:${u.port}` : ""}`; } catch { return raw.slice(-16); }
+  }
+  return raw;
+}
+
+export async function handleViaRoute({
+  model,
+  body,
+  peers,
+  handlerCtx,
+  evt,
+  logCall,
+  logError,
+  mark,
+  perf0,
+  stages,
+  startedAt,
+  plugins,
+  res,
+  requested,
+  useAuto,
+  lockModel,
+  auto,
+}) {
+  const route = getViaRoute(model);
+  if (!route || !route.best || route.best === "direct" || !String(route.best).startsWith("via:")) return { handled: false };
+  const peerLabel = String(route.best).slice(4).trim();
+  if (!peerLabel) return { handled: false };
+  // 找到对应该 label 的 peer（仅走 best 单路径，不并发）
+  const ordered = (() => { try { return peers.ordered(); } catch { return []; } })();
+  const byErr = (() => { try { return peers.orderedByLastError(); } catch { return []; } })();
+  const all = [...ordered, ...byErr];
+  const uniq = [];
+  const seen = new Set();
+  for (const p of all) { const k = p.url; if (!seen.has(k)) { seen.add(k); uniq.push(p); } }
+  let peer = uniq.find((p) => shortLabel(p) === peerLabel || String(p.url || "").includes(peerLabel) || String(p.id || "") === peerLabel || String(p.name || "") === peerLabel);
+  if (!peer) {
+    try {
+      const { loadPeers } = await import("../../state.js");
+      const disk = loadPeers() || [];
+      peer = disk.find((p) => shortLabel(p) === peerLabel || String(p.url || "").includes(peerLabel));
+    } catch {}
+  }
+  if (!peer) {
+    evt("via-route-miss", { reqId: handlerCtx.reqId, model, peerLabel, reason: "peer not found" });
+    return { handled: false };
+  }
+  evt("via-route-hit", { reqId: handlerCtx.reqId, model, peer: peer.url, peerLabel, routeBest: route.best, at: route.at });
+  // 单路径转发，不并发：直接打 best peer
+  const hops = handlerCtx.hops || 0;
+  const providerId = String(model).split("/")[0] || "";
+  let shareHeader = null;
+  try {
+    const keys = loadProviderKeys(providerId) || [];
+    if (keys.length) shareHeader = `${providerId}=${keys.join(",")}`;
+  } catch {}
+  const controller = new AbortController();
+  const timer = setTimeout(() => controller.abort(new Error("via-route timeout 30000ms")), 30000);
+  let upRes;
+  try {
+    const headers = {
+      "Content-Type": "application/json",
+      "Authorization": `Bearer ${peer.token || ""}`,
+      "x-mslxdff-hops": String(hops + 1),
+      "x-mslxdff-model-lock": model,
+      "Accept": "text/event-stream",
+    };
+    if (shareHeader) headers[SHARE_KEYS_HEADER] = shareHeader;
+    // workbuddyUid 透传
+    if (handlerCtx.workbuddyUid) headers["x-mslxdff-workbuddy-uid"] = handlerCtx.workbuddyUid;
+    evt("via-route-request", { reqId: handlerCtx.reqId, peer: peer.url, model, hops: hops + 1, hasShare: Boolean(shareHeader) });
+    upRes = await fetch(`${String(peer.url).replace(/\/+$/, "")}/v1/chat/completions`, {
+      method: "POST",
+      headers,
+      body: JSON.stringify({ ...body, model }),
+      signal: controller.signal,
+    });
+  } catch (e) {
+    clearTimeout(timer);
+    const msg = errMsg(e);
+    evt("via-route-error", { reqId: handlerCtx.reqId, peer: peer.url, model, error: msg });
+    try { await peers.recordError(peer.url); } catch {}
+    try { await peers.recordResult(peer.url, { ok: false }); } catch {}
+    return { handled: false, lastErr: { model, status: 502, message: msg } };
+  }
+  clearTimeout(timer);
+  const failed = upRes instanceof Error || upRes.status >= 400;
+  if (failed) {
+    const status = upRes instanceof Error ? 502 : upRes.status;
+    let bodyText = "";
+    try { bodyText = await upRes.clone().text(); } catch {}
+    const msg = bodyText.slice(0, 300) || errMsg(upRes) || `peer ${status}`;
+    evt("via-route-peer-error", { reqId: handlerCtx.reqId, peer: peer.url, model, status, message: msg.slice(0, 200) });
+    try { await peers.recordError(peer.url); } catch {}
+    try { await peers.recordResult(peer.url, { ok: false }); } catch {}
+    // 502/429 等可 fallback 到 direct
+    return { handled: false, lastErr: { model, upstream: upRes, status, message: msg } };
+  }
+  // 成功：记录并走 pipeline 中继（复用 peer 的 streaming 逻辑）
+  try { await peers.recordResult(peer.url, { ok: true, latencyMs: 0, model }); } catch {}
+  evt("via-route-win", { reqId: handlerCtx.reqId, peer: peer.url, model });
+  const pipeline = createRelayPipeline({
+    relay,
+    buildFallbackInfo,
+    auto,
+    plugins,
+    evt,
+    mark,
+    logCall,
+    logError: logError || (() => {}),
+    constants: { STREAM_TIMEOUT_MS, SLOW_TOTAL_MS, STALL_TIMEOUT_MS, SCORE_STALL_MS },
+    startedAt,
+    stages,
+  });
+  await pipeline.execute({
+    res,
+    upRes,
+    body,
+    requested: requested ?? model,
+    actual: model,
+    lastErr: null,
+    via: "peer",
+    lockModel: lockModel || model,
+    useAuto: Boolean(useAuto),
+    handlerCtx: { ...handlerCtx, model },
+    mark,
+    perf0,
+    stages,
+    startedAt,
+  });
+  return { handled: true };
+}