agent-shim 11.83.5 → 11.83.6

package/dist/cli.cjs

@@ -39986,7 +39986,7 @@ var init_package = __esm({
   "package.json"() {
     package_default = {
       name: "agent-shim",
-      version: "11.83.5",
+      version: "11.83.6",
       description: "A profile manager and launcher for Claude Code that lets one person run multiple logins from one machine while controlling what gets shared between them.",
       license: "MIT",
       author: "Joseph Mearman <joseph@mearman.co.uk>",
@@ -68743,9 +68743,12 @@ var init_cloudUpstream = __esm({
     AntigravityUpstreamError = class extends Error {
       status;
       retryAfter;
+      /** The backend's own answer, as it sent it, which carries more than the message says (the model a 429 names and when its allowance lifts). */
+      body;
       constructor(status, message, retryAfter) {
         super(`Google Antigravity API error (${String(status)}): ${message}`);
         this.name = "AntigravityUpstreamError";
+        this.body = message;
         this.status = status;
         this.retryAfter = retryAfter;
       }
@@ -68794,6 +68797,60 @@ function antigravityQuotaPressure(quota, model) {
   const pressures = [windowPressure(fiveHour, FIVE_HOUR_PERIOD_SECONDS), windowPressure(weekly, WEEK_PERIOD_SECONDS)].filter((pressure) => pressure !== void 0);
   return pressures.length === 0 ? void 0 : Math.min(...pressures);
 }
+function parseQuotaExhaustion(body, nowMs) {
+  let parsed;
+  try {
+    parsed = JSON.parse(body);
+  } catch {
+    return void 0;
+  }
+  const details = isRecord7(parsed) && isRecord7(parsed.error) && Array.isArray(parsed.error.details) ? parsed.error.details : [];
+  let model;
+  let resetAtMs;
+  for (const detail of details) {
+    if (!isRecord7(detail)) {
+      continue;
+    }
+    if (detail.reason === "QUOTA_EXHAUSTED" && isRecord7(detail.metadata)) {
+      const named = detail.metadata.model;
+      const stamp = typeof detail.metadata.quotaResetTimeStamp === "string" ? Date.parse(detail.metadata.quotaResetTimeStamp) : Number.NaN;
+      model = typeof named === "string" && named !== "" ? named : void 0;
+      resetAtMs = Number.isNaN(stamp) ? void 0 : stamp;
+    } else if (typeof detail.retryDelay === "string" && resetAtMs === void 0) {
+      const seconds = RETRY_DELAY.exec(detail.retryDelay)?.groups?.seconds;
+      resetAtMs = seconds === void 0 ? void 0 : nowMs + Math.ceil(Number(seconds) * MS_PER_SECOND9);
+    }
+  }
+  return model === void 0 || resetAtMs === void 0 ? void 0 : { model, resetAtMs };
+}
+function bucketKeyOf(model, resetInSeconds) {
+  const lower = model.toLowerCase();
+  if (lower.includes("argon")) {
+    return "gemini-argon-wtu-weekly";
+  }
+  const withinFiveHours = resetInSeconds <= FIVE_HOUR_PERIOD_SECONDS;
+  if (lower.includes("gemini")) {
+    return withinFiveHours ? "gemini-5h" : "gemini-weekly";
+  }
+  return withinFiveHours ? "3p-5h" : "3p-weekly";
+}
+function quotaWithExhaustions(quota, exhaustions, nowMs) {
+  const spent = [...exhaustions].filter(([, resetAtMs]) => resetAtMs > nowMs);
+  if (spent.length === 0) {
+    return quota;
+  }
+  const merged = { ...quota };
+  for (const [model, resetAtMs] of spent) {
+    const resetIn = Math.ceil((resetAtMs - nowMs) / MS_PER_SECOND9);
+    const key = bucketKeyOf(model, resetIn);
+    const known = merged[key];
+    if (known?.remaining_fraction === 0 && (known.reset_in_seconds ?? 0) >= resetIn) {
+      continue;
+    }
+    merged[key] = { remaining_fraction: 0, reset_in_seconds: resetIn };
+  }
+  return merged;
+}
 function calculateUtilization(remainingFraction) {
   if (remainingFraction === void 0 || !Number.isFinite(remainingFraction)) {
     return void 0;
@@ -68975,7 +69032,7 @@ function ageQuotaMap(quota, elapsedMs) {
     Object.entries(quota).map(([key, bucket]) => [key, bucket?.reset_in_seconds === void 0 ? bucket : { ...bucket, reset_in_seconds: Math.max(0, bucket.reset_in_seconds - elapsedSec) }])
   );
 }
-var MS_PER_SECOND9, DEFAULT_RETRY_AFTER_SECONDS, FRACTION_DIGITS, GEMINI_5H_KEYS, GEMINI_WEEKLY_KEYS, ARGON_WEEKLY_KEYS, THIRD_PARTY_5H_KEYS, THIRD_PARTY_WEEKLY_KEYS, QUOTA_FAMILY_WINDOWS, SECONDS_PER_HOUR, HOURS_PER_FIVE_HOUR_WINDOW, DAYS_PER_WEEK, HOURS_PER_DAY4, FIVE_HOUR_PERIOD_SECONDS, WEEK_PERIOD_SECONDS, ANTIGRAVITY_QUOTA_REFRESH_MS;
+var MS_PER_SECOND9, DEFAULT_RETRY_AFTER_SECONDS, FRACTION_DIGITS, GEMINI_5H_KEYS, GEMINI_WEEKLY_KEYS, ARGON_WEEKLY_KEYS, THIRD_PARTY_5H_KEYS, THIRD_PARTY_WEEKLY_KEYS, QUOTA_FAMILY_WINDOWS, SECONDS_PER_HOUR, HOURS_PER_FIVE_HOUR_WINDOW, DAYS_PER_WEEK, HOURS_PER_DAY4, FIVE_HOUR_PERIOD_SECONDS, WEEK_PERIOD_SECONDS, RETRY_DELAY, ANTIGRAVITY_QUOTA_REFRESH_MS;
 var init_quota = __esm({
   "src/antigravity/quota.ts"() {
     "use strict";
@@ -69002,6 +69059,7 @@ var init_quota = __esm({
     HOURS_PER_DAY4 = 24;
     FIVE_HOUR_PERIOD_SECONDS = HOURS_PER_FIVE_HOUR_WINDOW * SECONDS_PER_HOUR;
     WEEK_PERIOD_SECONDS = DAYS_PER_WEEK * HOURS_PER_DAY4 * SECONDS_PER_HOUR;
+    RETRY_DELAY = /^(?<seconds>\d+(?:\.\d+)?)s$/u;
     ANTIGRAVITY_QUOTA_REFRESH_MS = 6e4;
   }
 });
@@ -71168,7 +71226,13 @@ var init_providerQuota = __esm({
       /** The account's paid subscription; absent when the account has none. */
       paidTier: external_exports.object({ name: external_exports.string().optional(), id: external_exports.string().optional() }).optional()
     });
-    ANTIGRAVITY_WINDOW_MS = { "gemini-5h": FIVE_HOURS_MS, "3p-5h": FIVE_HOURS_MS, "gemini-weekly": WEEK_MS, "3p-weekly": WEEK_MS };
+    ANTIGRAVITY_WINDOW_MS = {
+      "gemini-5h": FIVE_HOURS_MS,
+      "3p-5h": FIVE_HOURS_MS,
+      "gemini-weekly": WEEK_MS,
+      "3p-weekly": WEEK_MS,
+      "gemini-argon-wtu-weekly": WEEK_MS
+    };
     antigravityQuotaAdapter = {
       source: ANTIGRAVITY_SOURCE,
       matches: (baseUrl) => URL.canParse(baseUrl) && new URL(baseUrl).hostname === ANTIGRAVITY_HOST,
@@ -141846,6 +141910,15 @@ function exhaustedMessage(exhaustion) {
 function createAntigravityRoute(ports) {
   const fallbackId = ports.fallbackId ?? (() => (0, import_node_crypto24.randomUUID)());
   const lastRouted = ports.lastRouted ?? /* @__PURE__ */ new Map();
+  const exhaustedModels = ports.exhaustedModels ?? /* @__PURE__ */ new Map();
+  const liveQuota = (nowMs) => {
+    for (const [model, resetAtMs] of exhaustedModels) {
+      if (resetAtMs <= nowMs) {
+        exhaustedModels.delete(model);
+      }
+    }
+    return quotaWithExhaustions(ports.quotaSource?.(), exhaustedModels, nowMs);
+  };
   return async (request7) => {
     await Promise.resolve();
     if (request7.method !== "POST") {
@@ -141899,71 +141972,90 @@ function createAntigravityRoute(ports) {
     const requestedModel = parsed.data.model ?? config2.defaultModel;
     const nowMs = ports.now?.() ?? Date.now();
     const sessionKey = parsed.data.metadata?.user_id === void 0 ? void 0 : `${parsed.data.metadata.user_id}\0${parsed.data.model ?? ""}`;
-    const translated = translateRequest(parsed.data, config2, { quota: ports.quotaSource?.(), nowMs, previous: sessionKey === void 0 ? void 0 : lastRouted.get(sessionKey) });
-    if (sessionKey !== void 0) {
-      lastRouted.delete(sessionKey);
-      lastRouted.set(sessionKey, translated.model);
-      if (lastRouted.size > ROUTING_MEMORY_SESSIONS) {
-        const oldest = lastRouted.keys().next();
-        if (oldest.done !== true) {
-          lastRouted.delete(oldest.value);
+    const routeRequest = () => {
+      const routed = translateRequest(parsed.data, config2, { quota: liveQuota(nowMs), nowMs, previous: sessionKey === void 0 ? void 0 : lastRouted.get(sessionKey) });
+      if (sessionKey !== void 0) {
+        lastRouted.delete(sessionKey);
+        lastRouted.set(sessionKey, routed.model);
+        if (lastRouted.size > ROUTING_MEMORY_SESSIONS) {
+          const oldest = lastRouted.keys().next();
+          if (oldest.done !== true) {
+            lastRouted.delete(oldest.value);
+          }
         }
       }
-    }
-    const relay = createAntigravityRelay({
-      // The frame carries the model that actually served the request, as the real API echoes the concrete model behind an alias: a usage record read from the response then attributes quota to the backend model, not the tier alias the session typed.
-      servedModel: translated.model,
-      toolNames: translated.toolNames,
-      fallbackId
-    });
-    const upstreamChunks2 = config2.mode === "local" ? callLocalAntigravity({
-      config: config2,
-      model: translated.model,
-      request: translated.request,
-      signal: request7.signal,
-      connector: ports.localConnector
-    }) : callCloudAntigravity(
-      {
+      return routed;
+    };
+    let translated = routeRequest();
+    const tried = /* @__PURE__ */ new Set();
+    let iterator;
+    let firstResult;
+    for (; ; ) {
+      tried.add(translated.model);
+      const upstreamChunks2 = config2.mode === "local" ? callLocalAntigravity({
         config: config2,
         model: translated.model,
         request: translated.request,
-        token,
-        credentialTarget,
-        signal: request7.signal
-      },
-      ports.fetchFn
-    );
-    const iterator = upstreamChunks2[Symbol.asyncIterator]();
-    let firstResult;
-    try {
-      firstResult = await iterator.next();
-    } catch (error63) {
-      if (error63 instanceof AntigravityUpstreamError) {
-        const is429 = error63.status === HTTP_STATUS.tooManyRequests;
-        const quota = ports.quotaSource?.();
-        const quotaHeaders3 = unifiedAntigravityQuotaHeaders({ quota, rejected: is429, nowMs, model: translated.model });
-        const retryAfter = is429 && quotaHeaders3["retry-after"] === void 0 ? parseRetryAfterSeconds(error63.retryAfter) : void 0;
-        const exhaustion = is429 ? antigravityQuotaExhaustion(quota, translated.model, nowMs) : void 0;
-        const errorType = is429 ? "rate_limit_error" : "api_error";
+        signal: request7.signal,
+        connector: ports.localConnector
+      }) : callCloudAntigravity(
+        {
+          config: config2,
+          model: translated.model,
+          request: translated.request,
+          token,
+          credentialTarget,
+          signal: request7.signal
+        },
+        ports.fetchFn
+      );
+      iterator = upstreamChunks2[Symbol.asyncIterator]();
+      try {
+        firstResult = await iterator.next();
+        break;
+      } catch (error63) {
+        if (error63 instanceof AntigravityUpstreamError) {
+          const is429 = error63.status === HTTP_STATUS.tooManyRequests;
+          const learned = is429 ? parseQuotaExhaustion(error63.body, nowMs) : void 0;
+          if (learned !== void 0) {
+            exhaustedModels.set(learned.model, learned.resetAtMs);
+            const next = routeRequest();
+            if (!tried.has(next.model)) {
+              translated = next;
+              continue;
+            }
+          }
+          const quota = liveQuota(nowMs);
+          const quotaHeaders3 = unifiedAntigravityQuotaHeaders({ quota, rejected: is429, nowMs, model: translated.model });
+          const retryAfter = is429 && quotaHeaders3["retry-after"] === void 0 ? parseRetryAfterSeconds(error63.retryAfter) : void 0;
+          const exhaustion = is429 ? antigravityQuotaExhaustion(quota, translated.model, nowMs) : void 0;
+          const errorType = is429 ? "rate_limit_error" : "api_error";
+          return {
+            status: error63.status,
+            headers: {
+              "Content-Type": "application/json",
+              ...quotaHeaders3,
+              ...retryAfter !== void 0 ? { "retry-after": String(retryAfter) } : {}
+            },
+            body: JSON.stringify({ type: "error", error: { type: errorType, message: exhaustion === void 0 ? error63.message : exhaustedMessage(exhaustion) } })
+          };
+        }
+        const errMsg = error63 instanceof Error ? error63.message : String(error63);
         return {
-          status: error63.status,
-          headers: {
-            "Content-Type": "application/json",
-            ...quotaHeaders3,
-            ...retryAfter !== void 0 ? { "retry-after": String(retryAfter) } : {}
-          },
-          body: JSON.stringify({ type: "error", error: { type: errorType, message: exhaustion === void 0 ? error63.message : exhaustedMessage(exhaustion) } })
+          status: HTTP_STATUS.badGateway,
+          headers: { "Content-Type": "application/json" },
+          body: JSON.stringify({ type: "error", error: { type: "api_error", message: errMsg } })
         };
       }
-      const errMsg = error63 instanceof Error ? error63.message : String(error63);
-      return {
-        status: HTTP_STATUS.badGateway,
-        headers: { "Content-Type": "application/json" },
-        body: JSON.stringify({ type: "error", error: { type: "api_error", message: errMsg } })
-      };
     }
+    const relay = createAntigravityRelay({
+      // The frame carries the model that actually served the request, as the real API echoes the concrete model behind an alias: a usage record read from the response then attributes quota to the backend model, not the tier alias the session typed.
+      servedModel: translated.model,
+      toolNames: translated.toolNames,
+      fallbackId
+    });
     const quotaHeaders2 = unifiedAntigravityQuotaHeaders({
-      quota: ports.quotaSource?.(),
+      quota: liveQuota(nowMs),
       rejected: false,
       nowMs,
       model: translated.model
@@ -142427,6 +142519,7 @@ function createProviderRouteResolver(deps) {
   const resolveCredential2 = shareInFlightResolutions(resolveCredentialAsync);
   const antigravityQuotaSources = /* @__PURE__ */ new Map();
   const antigravityRoutingMemory = /* @__PURE__ */ new Map();
+  const antigravityExhaustedModels = /* @__PURE__ */ new Map();
   const definesProvider = (name) => {
     try {
       return loadProvider(deps.providersDir, name, deps.fs) !== void 0;
@@ -142445,6 +142538,15 @@ function createProviderRouteResolver(deps) {
     antigravityRoutingMemory.set(provider, created);
     return created;
   };
+  const exhaustedModelsFor = (provider) => {
+    const existing = antigravityExhaustedModels.get(provider);
+    if (existing !== void 0) {
+      return existing;
+    }
+    const created = /* @__PURE__ */ new Map();
+    antigravityExhaustedModels.set(provider, created);
+    return created;
+  };
   const antigravityQuotaOf = (provider, definition, config2) => {
     const existing = antigravityQuotaSources.get(provider);
     if (existing !== void 0) {
@@ -142540,7 +142642,8 @@ function createProviderRouteResolver(deps) {
               };
             },
             quotaSource: antigravityQuotaOf(provider, antigravityProvider, config2),
-            lastRouted: memoryFor(provider)
+            lastRouted: memoryFor(provider),
+            exhaustedModels: exhaustedModelsFor(provider)
           },
           directOrigin2(deps.directPort()),
           provider
@@ -146837,20 +146940,21 @@ function windowsOf(states, nowMs) {
   let fiveHourResetAt;
   let sevenDayUsed;
   let sevenDayResetAt;
+  let bindingUsed;
   for (const state of states) {
     const unified = state.rateLimit?.unified;
     if (unified === void 0) {
       continue;
     }
     const five = livePercentOf(unified.fiveHour, nowMs);
-    if (five !== void 0 && (fiveHourUsed === void 0 || five > fiveHourUsed)) {
-      fiveHourUsed = five;
-      fiveHourResetAt = unified.fiveHour?.resetsAt;
-    }
     const seven = livePercentOf(unified.sevenDay, nowMs);
-    if (seven !== void 0 && (sevenDayUsed === void 0 || seven > sevenDayUsed)) {
+    const max = five === void 0 ? seven : seven === void 0 ? five : Math.max(five, seven);
+    if (max !== void 0 && (bindingUsed === void 0 || max > bindingUsed)) {
+      bindingUsed = max;
+      fiveHourUsed = five;
+      fiveHourResetAt = five === void 0 ? void 0 : unified.fiveHour?.resetsAt;
       sevenDayUsed = seven;
-      sevenDayResetAt = unified.sevenDay?.resetsAt;
+      sevenDayResetAt = seven === void 0 ? void 0 : unified.sevenDay?.resetsAt;
     }
   }
   return { fiveHourUsed, fiveHourResetAt, sevenDayUsed, sevenDayResetAt };
@@ -146896,8 +147000,9 @@ function antigravityQuotaMapOf(payload) {
   );
 }
 function bucketView(bucket, nowMs) {
+  const expired = bucket?.reset_in_seconds !== void 0 && Number.isFinite(bucket.reset_in_seconds) && bucket.reset_in_seconds <= 0;
   return {
-    used: bucket?.remaining_fraction === void 0 || !Number.isFinite(bucket.remaining_fraction) ? void 0 : (1 - Math.min(1, Math.max(0, bucket.remaining_fraction))) * FULL_PERCENT2,
+    used: bucket?.remaining_fraction === void 0 || !Number.isFinite(bucket.remaining_fraction) ? void 0 : expired ? 0 : (1 - Math.min(1, Math.max(0, bucket.remaining_fraction))) * FULL_PERCENT2,
     resetAt: bucket?.reset_in_seconds === void 0 || !Number.isFinite(bucket.reset_in_seconds) ? void 0 : isoOf(nowMs + bucket.reset_in_seconds * MS_PER_SECOND18)
   };
 }
@@ -146947,10 +147052,12 @@ function enrichStatuslinePayload(params) {
     ...params.modelChoice === void 0 ? {} : { modelChoice: params.modelChoice },
     ...params.doorCostUsd === void 0 ? {} : { doorCostUsd: params.doorCostUsd }
   };
+  let scopedName;
   if (params.snapshot !== void 0) {
-    const scopedName = providerScopedState(params.snapshot, params.providerName, params.env.AGENT_SHIM_PROVIDER);
+    const routedProvider = routedProviderOf(params);
+    scopedName = providerScopedState(params.snapshot, params.providerName, params.env.AGENT_SHIM_PROVIDER) ?? providerScopedState(params.snapshot, routedProvider, routedProvider);
     const scopedState = scopedName === void 0 ? void 0 : params.snapshot.providers[scopedName];
-    const states = summary.provider === void 0 ? Object.values(params.snapshot.providers) : scopedState === void 0 ? [] : [scopedState];
+    const states = summary.provider === void 0 && scopedState === void 0 ? Object.values(params.snapshot.providers) : scopedState === void 0 ? [] : [scopedState];
     let email3;
     let sub;
     let plan;
@@ -146981,7 +147088,7 @@ function enrichStatuslinePayload(params) {
     );
     summary = { ...summary, ...windowsOf(states, params.nowMs), providerAccountEmail: email3, providerAccountSub: sub, providerPlan: plan, providerQuotaWindows, quotaFamilies: quotaFamiliesOf(providerBuckets, params.nowMs) };
   }
-  if (summary.provider === void 0) {
+  if (summary.provider === void 0 && scopedName === void 0) {
     const agyBuckets = antigravityQuotaMapOf(params.payload);
     if (agyBuckets !== void 0) {
       summary = { ...summary, quotaFamilies: quotaFamiliesOf(agyBuckets, params.nowMs) };
@@ -147003,6 +147110,29 @@ function enrichStatuslinePayload(params) {
   }
   return { payload: enriched, summary };
 }
+function routedProviderOf(params) {
+  if (params.modelChoice?.provider !== void 0 && params.modelChoice.provider !== "") {
+    return params.modelChoice.provider;
+  }
+  const targetPrefix = prefixBeforeSlash(params.routedTarget);
+  if (targetPrefix !== void 0) {
+    return targetPrefix;
+  }
+  const model = isObject4(params.payload) && isObject4(params.payload.model) ? params.payload.model : void 0;
+  return prefixBeforeSlash(typeof model?.id === "string" ? model.id : void 0) ?? prefixBeforeSlash(typeof model?.display_name === "string" ? model.display_name : void 0);
+}
+function prefixBeforeSlash(value3) {
+  if (value3 === void 0) {
+    return void 0;
+  }
+  const slash = value3.indexOf("/");
+  if (slash <= 0) {
+    return void 0;
+  }
+  const prefix = value3.slice(0, slash).trim();
+  const rest = value3.slice(slash + 1).trim();
+  return prefix === "" || rest === "" ? void 0 : prefix;
+}
 function providerScopedState(snapshot, providerName, providerDisplayName) {
   if (providerName !== void 0 && providerName in snapshot.providers) {
     return providerName;

package/dist/index.cjs

@@ -4853,7 +4853,7 @@ var init_package = __esm({
   "package.json"() {
     package_default = {
       name: "agent-shim",
-      version: "11.83.5",
+      version: "11.83.6",
       description: "A profile manager and launcher for Claude Code that lets one person run multiple logins from one machine while controlling what gets shared between them.",
       license: "MIT",
       author: "Joseph Mearman <joseph@mearman.co.uk>",

package/dist/index.mjs

@@ -4829,7 +4829,7 @@ var init_package = __esm({
   "package.json"() {
     package_default = {
       name: "agent-shim",
-      version: "11.83.5",
+      version: "11.83.6",
       description: "A profile manager and launcher for Claude Code that lets one person run multiple logins from one machine while controlling what gets shared between them.",
       license: "MIT",
       author: "Joseph Mearman <joseph@mearman.co.uk>",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "agent-shim",
-  "version": "11.83.5",
+  "version": "11.83.6",
   "description": "A profile manager and launcher for Claude Code that lets one person run multiple logins from one machine while controlling what gets shared between them.",
   "license": "MIT",
   "author": "Joseph Mearman <joseph@mearman.co.uk>",