claude-use 11.31.0 → 11.32.0

package/dist/cli.cjs

@@ -34323,7 +34323,7 @@ var init_package = __esm({
   "package.json"() {
     package_default = {
       name: "claude-use",
-      version: "11.31.0",
+      version: "11.32.0",
       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>",
@@ -34936,7 +34936,7 @@ function destinationOf(path65, sessionProvider) {
 }
 function buildRecord(event, context) {
   const { scan } = context;
-  const { provider, endpoint } = destinationOf(event.path, event.session.provider);
+  const { provider, endpoint } = destinationOf(event.servedPath, event.session.provider);
   const rateLimitHeaders = rateLimitHeadersOf(event.headers);
   const limit = classifyLimit({ status: event.status, headers: rateLimitHeaders ?? {}, error: scan?.error, nowMs: context.endedAt });
   const requestId = headerValue(event.headers, "request-id");
@@ -63767,6 +63767,158 @@ var init_launchReport = __esm({
   }
 });
 
+// src/selection/engine.ts
+function compareMaybe(left, right, preferHigh) {
+  if (left === void 0 && right === void 0) {
+    return 0;
+  }
+  if (left === void 0) {
+    return 1;
+  }
+  if (right === void 0) {
+    return -1;
+  }
+  return preferHigh ? right - left : left - right;
+}
+function by(comparator, candidates) {
+  return [...candidates].sort((left, right) => comparator(left, right) || left.order - right.order || (left.ref < right.ref ? -1 : left.ref > right.ref ? 1 : 0));
+}
+function selectOrdered(candidates) {
+  const reasons = candidates.map((candidate) => `${candidate.ref}${availability(candidate)}`);
+  const unspent = candidates.filter((candidate) => candidate.available);
+  const current = unspent.find((candidate) => candidate.isCurrent);
+  if (current !== void 0) {
+    return { ref: current.ref, reasons: [...reasons, `${current.ref} is the model this session is already on and still has quota`] };
+  }
+  const first = by((left, right) => left.order - right.order, candidates)[0];
+  if (first?.available === true) {
+    return { ref: first.ref, reasons: [...reasons, `${first.ref} is the first member with quota`] };
+  }
+  const byPressure = by((left, right) => compareMaybe(left.pressure, right.pressure, true), unspent);
+  const top = byPressure[0];
+  if (top !== void 0) {
+    return { ref: top.ref, reasons: [...reasons, `${top.ref} carries the most allowance its reset would waste`] };
+  }
+  const byReset = by((left, right) => compareMaybe(left.hoursToReset, right.hoursToReset, false), candidates);
+  return { ref: byReset[0]?.ref, reasons: [...reasons, "every member is spent; serving from the one whose window lifts soonest"] };
+}
+function selectRanked(candidates, comparator, what, why) {
+  const ordered = selectOrdered(candidates);
+  const available = candidates.filter((candidate) => candidate.available);
+  if (available.length === 0) {
+    return ordered;
+  }
+  const ranked = by(comparator, available);
+  const best = ranked[0];
+  if (best === void 0) {
+    return ordered;
+  }
+  const current = available.find((candidate) => candidate.isCurrent);
+  const chosen = current !== void 0 && (comparator(current, best) === 0 || best.isCurrent) ? current : best;
+  if (chosen.isCurrent || current === void 0) {
+    return { ref: chosen.ref, reasons: [...ordered.reasons, `${chosen.ref} ${what}`] };
+  }
+  return { ref: best.ref, reasons: [...ordered.reasons, `${best.ref} ${what}; the switch from the current member is worth re-sending the conversation because ${why}`] };
+}
+function selectMember(mode, candidates, options = {}) {
+  const plain = candidates.map((entry) => entry.candidate);
+  switch (mode) {
+    case "ordered":
+      return selectOrdered(plain);
+    case "most-allowance":
+      return selectRanked(plain, (left, right) => compareMaybe(left.remainingAllowance, right.remainingAllowance, true), "holds the most remaining allowance", "its allowance is the largest");
+    case "soonest-reset":
+      return selectRanked(plain, (left, right) => compareMaybe(left.hoursToReset, right.hoursToReset, false), "resets soonest, so its allowance is the most use-it-or-lose-it", "its reset is the soonest");
+    case "lowest-latency":
+      return selectRanked(plain, (left, right) => compareMaybe(left.latencyMs, right.latencyMs, false), "has answered fastest of late", "its latency is the lowest observed");
+    case "highest-success":
+      return selectRanked(plain, (left, right) => compareMaybe(left.successRate, right.successRate, true), "has succeeded most reliably of late", "its success rate is the highest observed");
+    case "cheapest":
+      return selectRanked(plain, (left, right) => compareMaybe(left.pricePerMTok, right.pricePerMTok, false), "is the cheapest for the tokens this session sends", "its blended price is the lowest");
+    case "spread": {
+      const ordered = selectOrdered(plain);
+      const available = plain.filter((candidate) => candidate.available);
+      if (available.length === 0) {
+        return ordered;
+      }
+      const rotation = options.rotation ?? 0;
+      const orderedByList = by((left, right) => left.order - right.order, available);
+      const chosen = orderedByList[rotation % orderedByList.length];
+      return chosen === void 0 ? ordered : { ref: chosen.ref, reasons: [...ordered.reasons, `${chosen.ref} is turn ${String(rotation % orderedByList.length + 1)} of the spread across the members with quota`] };
+    }
+    case "adaptive":
+      return selectAdaptive(candidates, options.needs);
+  }
+  return selectOrdered(plain);
+}
+function selectAdaptive(candidates, needs) {
+  const plain = candidates.map((entry) => entry.candidate);
+  const ruledOut = candidates.map((entry) => {
+    const { candidate, capabilities } = entry;
+    if (needs?.hasImage === true && !capabilities.vision) {
+      return `${candidate.ref} does not accept images`;
+    }
+    if (needs?.hasTools === true && !capabilities.tools) {
+      return `${candidate.ref} does not accept tools`;
+    }
+    if (needs?.thinking === true && !capabilities.thinking) {
+      return `${candidate.ref} does not think`;
+    }
+    if (candidate.available && !candidate.feasible) {
+      return `${candidate.ref} would run dry before its window resets`;
+    }
+    if (candidate.successRate !== void 0 && candidate.successRate < ADAPTIVE_HEALTH_FLOOR) {
+      return `${candidate.ref} has been failing`;
+    }
+    return void 0;
+  }).filter((line) => line !== void 0);
+  const standing = candidates.filter((entry) => {
+    const { candidate, capabilities } = entry;
+    if (!candidate.available) {
+      return false;
+    }
+    if (needs?.hasImage === true && !capabilities.vision) {
+      return false;
+    }
+    if (needs?.hasTools === true && !capabilities.tools) {
+      return false;
+    }
+    if (needs?.thinking === true && !capabilities.thinking) {
+      return false;
+    }
+    if (!candidate.feasible) {
+      return false;
+    }
+    return !(candidate.successRate !== void 0 && candidate.successRate < ADAPTIVE_HEALTH_FLOOR);
+  });
+  const reasons = [...ruledOut];
+  if (standing.length === 0) {
+    const fallback = selectOrdered(plain);
+    return { ref: fallback.ref, reasons: [...reasons, "nothing stands: serving by order"] };
+  }
+  const current = standing.find((entry) => entry.candidate.isCurrent);
+  if (current !== void 0) {
+    return { ref: current.candidate.ref, reasons: [...reasons, `${current.candidate.ref} stands and is the model this session is already on, so the conversation is not re-sent`] };
+  }
+  const ranked = by(
+    (left, right) => compareMaybe(left.pressure, right.pressure, true) || compareMaybe(left.hoursToReset, right.hoursToReset, false) || compareMaybe(left.successRate, right.successRate, true),
+    standing.map((entry) => entry.candidate)
+  );
+  const best = ranked[0];
+  if (best === void 0) {
+    return { ref: void 0, reasons };
+  }
+  return { ref: best.ref, reasons: [...reasons, `${best.ref} stands and carries the most allowance its reset would waste`] };
+}
+var ADAPTIVE_HEALTH_FLOOR, availability;
+var init_engine = __esm({
+  "src/selection/engine.ts"() {
+    "use strict";
+    ADAPTIVE_HEALTH_FLOOR = 0.5;
+    availability = (candidate) => candidate.available ? "" : "quota spent";
+  }
+});
+
 // src/antigravity/cloudUpstream.ts
 function isCcpaEndpoint(endpoint) {
   return endpoint.includes("cloudcode-pa.googleapis.com");
@@ -64309,32 +64461,24 @@ function chooseCandidate(candidates, routing) {
   if (routing === void 0 || candidates.length === 1) {
     return first;
   }
-  if (routing.previous !== void 0 && candidates.includes(routing.previous) && antigravityQuotaExhaustion(routing.quota, routing.previous, routing.nowMs) === void 0) {
-    return routing.previous;
-  }
-  const unspent = candidates.filter((model) => antigravityQuotaExhaustion(routing.quota, model, routing.nowMs) === void 0);
-  if (unspent[0] === first) {
-    return first;
-  }
-  const neutral = 1;
-  let best;
-  for (const model of unspent) {
-    const pressure = antigravityQuotaPressure(routing.quota, model) ?? neutral;
-    if (best === void 0 || pressure > best.pressure) {
-      best = { model, pressure };
-    }
-  }
-  if (best !== void 0) {
-    return best.model;
-  }
-  let soonest;
-  for (const model of candidates) {
-    const exhaustion = antigravityQuotaExhaustion(routing.quota, model, routing.nowMs);
-    if (exhaustion !== void 0 && (soonest === void 0 || exhaustion.resetSec < soonest.resetSec)) {
-      soonest = { model, resetSec: exhaustion.resetSec };
-    }
-  }
-  return soonest?.model ?? first;
+  const exhaustionOf = (model) => antigravityQuotaExhaustion(routing.quota, model, routing.nowMs);
+  const evaluated = candidates.map((model, index) => {
+    const exhaustion = exhaustionOf(model);
+    return {
+      candidate: {
+        ref: model,
+        order: index,
+        available: exhaustion === void 0,
+        // A quota the map does not measure competes at the neutral pressure the old in-place ranking gave it, not as an unknown that always loses.
+        pressure: antigravityQuotaPressure(routing.quota, model) ?? NEUTRAL_PRESSURE,
+        hoursToReset: exhaustion === void 0 ? void 0 : exhaustion.resetSec / SECONDS_PER_HOUR2,
+        feasible: true,
+        isCurrent: routing.previous === model
+      },
+      capabilities: { vision: true, tools: true, thinking: true }
+    };
+  });
+  return selectMember("ordered", evaluated).ref ?? first;
 }
 function mapEffort(request7, config2) {
   const requested = request7.output_config?.effort;
@@ -64526,15 +64670,18 @@ function translateRequest(request7, config2, routing) {
     toolNames
   };
 }
-var import_node_crypto16, AUTO_ALIAS, CONTEXT_MARKER_1M, BACKEND_FAMILY_PREFIXES, BACKEND_CLAUDE_NAME, EFFORT_TO_BUDGET, BILLING_HEADER_LINE, ANTIGRAVITY_TOOL_NAME_LIMIT, TOOL_NAME_HASH_CHARS, NON_TEXT_TOOL_RESULT, EMPTY_SCHEMA, EFFORT_SUFFIXES;
+var import_node_crypto16, SECONDS_PER_HOUR2, NEUTRAL_PRESSURE, AUTO_ALIAS, CONTEXT_MARKER_1M, BACKEND_FAMILY_PREFIXES, BACKEND_CLAUDE_NAME, EFFORT_TO_BUDGET, BILLING_HEADER_LINE, ANTIGRAVITY_TOOL_NAME_LIMIT, TOOL_NAME_HASH_CHARS, NON_TEXT_TOOL_RESULT, EMPTY_SCHEMA, EFFORT_SUFFIXES;
 var init_translate = __esm({
   "src/antigravity/translate.ts"() {
     "use strict";
     import_node_crypto16 = require("node:crypto");
+    init_engine();
     init_quota();
     init_signatures();
     init_schema();
     init_anthropic();
+    SECONDS_PER_HOUR2 = 3600;
+    NEUTRAL_PRESSURE = 1;
     AUTO_ALIAS = "auto";
     CONTEXT_MARKER_1M = "[1m]";
     BACKEND_FAMILY_PREFIXES = ["gemini-", "antigravity-", "gpt-oss-"];
@@ -64812,7 +64959,7 @@ async function fetchCatalogue(params) {
   }
   throw new CatalogueFetchError(`no catalogue from ${outcomes.join("; ")}`);
 }
-var import_node_path42, import_picomatch3, DEFAULT_CANDIDATE_URLS, MS_PER_SECOND8, SECONDS_PER_HOUR2, HOURS_PER_DAY4, MS_PER_HOUR3, HOURS_BETWEEN_RETRIES, HTTP_OK, HTTP_MULTIPLE_CHOICES, CATALOGUE_MAX_AGE_MS, CATALOGUE_RETRY_MS, CATALOGUE_REQUEST_SECONDS, CATALOGUE_REQUEST_TIMEOUT_MS, CATALOGUE_REFRESH_TIMEOUT_MS, CatalogueModelSchema, CatalogueCacheSchema, CatalogueFetchError;
+var import_node_path42, import_picomatch3, DEFAULT_CANDIDATE_URLS, MS_PER_SECOND8, SECONDS_PER_HOUR3, HOURS_PER_DAY4, MS_PER_HOUR3, HOURS_BETWEEN_RETRIES, HTTP_OK, HTTP_MULTIPLE_CHOICES, CATALOGUE_MAX_AGE_MS, CATALOGUE_RETRY_MS, CATALOGUE_REQUEST_SECONDS, CATALOGUE_REQUEST_TIMEOUT_MS, CATALOGUE_REFRESH_TIMEOUT_MS, CatalogueModelSchema, CatalogueCacheSchema, CatalogueFetchError;
 var init_providerCatalogue = __esm({
   "src/providerCatalogue.ts"() {
     "use strict";
@@ -64821,9 +64968,9 @@ var init_providerCatalogue = __esm({
     init_zod();
     DEFAULT_CANDIDATE_URLS = 2;
     MS_PER_SECOND8 = 1e3;
-    SECONDS_PER_HOUR2 = 3600;
+    SECONDS_PER_HOUR3 = 3600;
     HOURS_PER_DAY4 = 24;
-    MS_PER_HOUR3 = SECONDS_PER_HOUR2 * MS_PER_SECOND8;
+    MS_PER_HOUR3 = SECONDS_PER_HOUR3 * MS_PER_SECOND8;
     HOURS_BETWEEN_RETRIES = 1;
     HTTP_OK = 200;
     HTTP_MULTIPLE_CHOICES = 300;
@@ -65823,7 +65970,7 @@ function quotaFreshnessMs(quota) {
   const lengths = quota.windows.flatMap((window) => window.periodMs === void 0 ? [] : [window.periodMs]);
   return lengths.length === 0 ? Number.POSITIVE_INFINITY : Math.min(...lengths) / PERCENT3;
 }
-var ProviderQuotaError, MS_PER_SECOND9, SECONDS_PER_HOUR3, HOURS_PER_DAY5, DAYS_PER_MONTH, PERCENT3, HTTP_OK2, ZAI_SOURCE, ZAI_HOST, ZAI_QUOTA_URL, ZAI_UNIT_HOURS, ZAI_UNIT_MONTHS, ZAI_UNIT_MS, ZAI_MEASURES, ZaiLimitSchema, ZaiResponseSchema, zaiQuotaAdapter, OPENROUTER_SOURCE, OPENROUTER_HOST, OPENROUTER_KEY_URL, OPENROUTER_CREDITS_URL, HTTP_FORBIDDEN, MS_PER_DAY2, OpenRouterKeySchema, OpenRouterCreditsSchema, openRouterQuotaAdapter, ANTIGRAVITY_SOURCE, ANTIGRAVITY_HOST, ANTIGRAVITY_QUOTA_URL, ANTIGRAVITY_USER_URL, GOOGLE_TOKENINFO_URL, FIVE_HOURS2, DAYS_PER_WEEK2, FIVE_HOURS_MS, WEEK_MS, AntigravityBucketSchema, AntigravitySummarySchema, GoogleTokenInfoSchema, AntigravityUserSchema, ANTIGRAVITY_WINDOW_MS, antigravityQuotaAdapter, PROVIDER_QUOTA_ADAPTERS;
+var ProviderQuotaError, MS_PER_SECOND9, SECONDS_PER_HOUR4, HOURS_PER_DAY5, DAYS_PER_MONTH, PERCENT3, HTTP_OK2, ZAI_SOURCE, ZAI_HOST, ZAI_QUOTA_URL, ZAI_UNIT_HOURS, ZAI_UNIT_MONTHS, ZAI_UNIT_MS, ZAI_MEASURES, ZaiLimitSchema, ZaiResponseSchema, zaiQuotaAdapter, OPENROUTER_SOURCE, OPENROUTER_HOST, OPENROUTER_KEY_URL, OPENROUTER_CREDITS_URL, HTTP_FORBIDDEN, MS_PER_DAY2, OpenRouterKeySchema, OpenRouterCreditsSchema, openRouterQuotaAdapter, ANTIGRAVITY_SOURCE, ANTIGRAVITY_HOST, ANTIGRAVITY_QUOTA_URL, ANTIGRAVITY_USER_URL, GOOGLE_TOKENINFO_URL, FIVE_HOURS2, DAYS_PER_WEEK2, FIVE_HOURS_MS, WEEK_MS, AntigravityBucketSchema, AntigravitySummarySchema, GoogleTokenInfoSchema, AntigravityUserSchema, ANTIGRAVITY_WINDOW_MS, antigravityQuotaAdapter, PROVIDER_QUOTA_ADAPTERS;
 var init_providerQuota = __esm({
   "src/usage/providerQuota.ts"() {
     "use strict";
@@ -65837,7 +65984,7 @@ var init_providerQuota = __esm({
       }
     };
     MS_PER_SECOND9 = 1e3;
-    SECONDS_PER_HOUR3 = 3600;
+    SECONDS_PER_HOUR4 = 3600;
     HOURS_PER_DAY5 = 24;
     DAYS_PER_MONTH = 30;
     PERCENT3 = 100;
@@ -65848,8 +65995,8 @@ var init_providerQuota = __esm({
     ZAI_UNIT_HOURS = 3;
     ZAI_UNIT_MONTHS = 5;
     ZAI_UNIT_MS = /* @__PURE__ */ new Map([
-      [ZAI_UNIT_HOURS, SECONDS_PER_HOUR3 * MS_PER_SECOND9],
-      [ZAI_UNIT_MONTHS, DAYS_PER_MONTH * HOURS_PER_DAY5 * SECONDS_PER_HOUR3 * MS_PER_SECOND9]
+      [ZAI_UNIT_HOURS, SECONDS_PER_HOUR4 * MS_PER_SECOND9],
+      [ZAI_UNIT_MONTHS, DAYS_PER_MONTH * HOURS_PER_DAY5 * SECONDS_PER_HOUR4 * MS_PER_SECOND9]
     ]);
     ZAI_MEASURES = { TOKENS_LIMIT: "tokens", TIME_LIMIT: "tool-calls" };
     ZaiLimitSchema = external_exports.object({
@@ -65901,7 +66048,7 @@ var init_providerQuota = __esm({
     OPENROUTER_KEY_URL = `https://${OPENROUTER_HOST}/api/v1/key`;
     OPENROUTER_CREDITS_URL = `https://${OPENROUTER_HOST}/api/v1/credits`;
     HTTP_FORBIDDEN = 403;
-    MS_PER_DAY2 = HOURS_PER_DAY5 * SECONDS_PER_HOUR3 * MS_PER_SECOND9;
+    MS_PER_DAY2 = HOURS_PER_DAY5 * SECONDS_PER_HOUR4 * MS_PER_SECOND9;
     OpenRouterKeySchema = external_exports.object({
       data: external_exports.object({
         limit: external_exports.number().nonnegative().nullable(),
@@ -65949,8 +66096,8 @@ var init_providerQuota = __esm({
     GOOGLE_TOKENINFO_URL = "https://oauth2.googleapis.com/tokeninfo";
     FIVE_HOURS2 = 5;
     DAYS_PER_WEEK2 = 7;
-    FIVE_HOURS_MS = FIVE_HOURS2 * SECONDS_PER_HOUR3 * MS_PER_SECOND9;
-    WEEK_MS = DAYS_PER_WEEK2 * HOURS_PER_DAY5 * SECONDS_PER_HOUR3 * MS_PER_SECOND9;
+    FIVE_HOURS_MS = FIVE_HOURS2 * SECONDS_PER_HOUR4 * MS_PER_SECOND9;
+    WEEK_MS = DAYS_PER_WEEK2 * HOURS_PER_DAY5 * SECONDS_PER_HOUR4 * MS_PER_SECOND9;
     AntigravityBucketSchema = external_exports.object({
       bucketId: external_exports.string(),
       remainingFraction: external_exports.number().optional(),
@@ -67064,7 +67211,7 @@ function Lp(e14) {
 function my(e14) {
   return e14.replace(/[\u200c-\u200f\u202a-\u202e\u206a-\u206f\ufeff]/g, "").toUpperCase().toLowerCase();
 }
-function by(e14, t10) {
+function by2(e14, t10) {
   if (Qo(e14)) return true;
   if (Yn(e14)) {
     let n10 = hc(e14);
@@ -67672,7 +67819,7 @@ function ri(e14) {
   return po(e14);
 }
 function Vp(e14, t10) {
-  return t10 !== void 0 && t10.some((n10) => po(n10) && !by(e14, n10) && (Xy(e14, n10) || Yn(e14)));
+  return t10 !== void 0 && t10.some((n10) => po(n10) && !by2(e14, n10) && (Xy(e14, n10) || Yn(e14)));
 }
 function Ck(e14, t10) {
   return t10 !== void 0 && t10.some((n10) => po(n10) && Xy(n10, e14));
@@ -67682,7 +67829,7 @@ function Xy(e14, t10) {
   return r10 === o || r10.startsWith(o === "/" ? o : o + "/");
 }
 function Zp(e14, t10) {
-  return t10 === void 0 || t10.every((n10) => by(e14, n10));
+  return t10 === void 0 || t10.every((n10) => by2(e14, n10));
 }
 function vk(e14, t10) {
   return t10 === void 0 || t10.every((n10) => zO(e14, n10));
@@ -129168,7 +129315,7 @@ function createRoutedResponse(response, context) {
     start: (status, headers) => {
       for (const observe of deps.responseObservers) {
         try {
-          const body = observe({ session, route, method: request7.method, path: request7.path, receivedAt: request7.receivedAt, viaHeadroom: request7.viaHeadroom, lineage: request7.thread.lineage, status, headers });
+          const body = observe({ session, route, method: request7.method, path: request7.path, servedPath: context.servedPath, receivedAt: request7.receivedAt, viaHeadroom: request7.viaHeadroom, lineage: request7.thread.lineage, status, headers });
           if (body !== void 0) {
             bodyObservers.push(body);
           }
@@ -129213,7 +129360,7 @@ async function serveRouted(request7, deps) {
   const identified = identifyRequest(request7.headers);
   const admission = deps.admit({ url: request7.url, headers: request7.headers, forwardable: identified.forwardableHeaders });
   if (!admission.ok) {
-    const response2 = createRoutedResponse(request7.response, { deps, session: identified.session, route: "(unauthorized)", request: observed });
+    const response2 = createRoutedResponse(request7.response, { deps, session: identified.session, route: "(unauthorized)", request: observed, servedPath: observed.path });
     response2.start(HTTP_STATUS.unauthorized, { "Content-Type": "application/json" });
     await response2.write(errorBody(HTTP_STATUS.unauthorized, admission.message));
     response2.end();
@@ -129229,7 +129376,7 @@ async function serveRouted(request7, deps) {
   };
   const resolution = await deps.resolveRoute(routed);
   if (!resolution.ok) {
-    const response2 = createRoutedResponse(request7.response, { deps, session: identified.session, route: "(unrouted)", request: observed });
+    const response2 = createRoutedResponse(request7.response, { deps, session: identified.session, route: "(unrouted)", request: observed, servedPath: observed.path });
     response2.start(resolution.status, { "Content-Type": "application/json" });
     await response2.write(errorBody(resolution.status, resolution.message));
     response2.end();
@@ -129238,10 +129385,10 @@ async function serveRouted(request7, deps) {
   const { headroom } = deps;
   const wantsHeadroom = headroom !== void 0 && await deps.headroomDecision.decide(identified.session.sessionId, identified.session.headroom, deps.now(), async () => await headroomReady(headroom, request7.signal));
   const viaHeadroom = wantsHeadroom && resolution.route.headroomEligible && deps.headroom !== void 0;
-  const response = createRoutedResponse(request7.response, { deps, session: identified.session, route: resolution.route.name, request: { ...observed, viaHeadroom } });
+  const response = createRoutedResponse(request7.response, { deps, session: identified.session, route: resolution.route.name, request: { ...observed, viaHeadroom }, servedPath: pathOf(resolution.request?.url ?? request7.url) });
   try {
     if (viaHeadroom) {
-      await applyHeadroomHop(routed, response, resolution.route.headroomUpstream, deps.headroom, (lineage) => {
+      await applyHeadroomHop(resolution.request ?? routed, response, resolution.route.headroomUpstream, deps.headroom, (lineage) => {
         observed.thread.lineage = lineage;
       });
     } else {
@@ -130236,7 +130383,7 @@ function exhaustedMessage(exhaustion) {
 }
 function createAntigravityRoute(ports) {
   const fallbackId = ports.fallbackId ?? (() => (0, import_node_crypto23.randomUUID)());
-  const lastRouted = /* @__PURE__ */ new Map();
+  const lastRouted = ports.lastRouted ?? /* @__PURE__ */ new Map();
   return async (request7) => {
     await Promise.resolve();
     if (request7.method !== "POST") {
@@ -130614,10 +130761,6 @@ async function scanRequestHead(request7, wholeBody = false) {
       text2 += chunk2.toString("utf8");
       if (!wholeBody && (modelIn() !== void 0 || seen >= REQUEST_SCAN_CAP_BYTES)) {
         decided = true;
-        body.pause();
-        replay.once("resume", () => {
-          body.resume();
-        });
         settle2();
       }
     });
@@ -130658,6 +130801,16 @@ function scopeSessionProviderPath(request7) {
 function createProviderRouteResolver(deps) {
   const oauthRoute = createPassthroughRoute("anthropic", { baseUrl: `https://${CONNECT_INTERCEPT_HOST}:${String(HTTPS_PORT)}`, stripPrefix: void 0, headroomUpstream: void 0 });
   const antigravityQuotaSources = /* @__PURE__ */ new Map();
+  const antigravityRoutingMemory = /* @__PURE__ */ new Map();
+  const memoryFor = (provider) => {
+    const existing = antigravityRoutingMemory.get(provider);
+    if (existing !== void 0) {
+      return existing;
+    }
+    const created = /* @__PURE__ */ new Map();
+    antigravityRoutingMemory.set(provider, created);
+    return created;
+  };
   const antigravityQuotaOf = (provider, definition, config2) => {
     const existing = antigravityQuotaSources.get(provider);
     if (existing !== void 0) {
@@ -130749,7 +130902,8 @@ function createProviderRouteResolver(deps) {
                 credentialTarget: antigravityProvider.credential.target
               };
             },
-            quotaSource: antigravityQuotaOf(provider, antigravityProvider, config2)
+            quotaSource: antigravityQuotaOf(provider, antigravityProvider, config2),
+            lastRouted: memoryFor(provider)
           },
           directOrigin2(deps.directPort()),
           provider
@@ -130803,7 +130957,7 @@ function createProviderRouteResolver(deps) {
       }
       if (routes === void 0 || routes.length === 0) {
         const resolution = resolve2(current);
-        return resolution.ok && replay !== void 0 ? { ok: true, route: servingRewritten(resolution.route, { ...replay, url: current.url }) } : resolution.ok ? serveScoped(resolution) : resolution;
+        return resolution.ok && replay !== void 0 ? { ok: true, route: servingRewritten(resolution.route, { ...replay, url: current.url }), request: { ...replay, url: current.url } } : resolution.ok ? serveScoped(resolution) : resolution;
       }
       const whole = routes.some((entry) => entry.model !== void 0 || referencesFact(entry.when, "request.hasImage"));
       const snapshot = deps.usageSnapshotOf?.(current.session.identity);
@@ -130815,7 +130969,7 @@ function createProviderRouteResolver(deps) {
       const scan = await scanRequestHead(current, whole);
       if (scan.replayed === void 0) {
         const resolution = resolve2(current);
-        return resolution.ok && replay !== void 0 ? { ok: true, route: servingRewritten(resolution.route, { ...replay, url: current.url }) } : resolution.ok ? serveScoped(resolution) : resolution;
+        return resolution.ok && replay !== void 0 ? { ok: true, route: servingRewritten(resolution.route, { ...replay, url: current.url }), request: { ...replay, url: current.url } } : resolution.ok ? serveScoped(resolution) : resolution;
       }
       replay = scan.replayed;
       const requestFacts = {
@@ -130837,7 +130991,7 @@ function createProviderRouteResolver(deps) {
       }
       if (matched === void 0) {
         const resolution = resolve2(current);
-        return resolution.ok ? { ok: true, route: servingRewritten(resolution.route, { ...replay, url: current.url }) } : resolution;
+        return resolution.ok ? { ok: true, route: servingRewritten(resolution.route, { ...replay, url: current.url }), request: { ...replay, url: current.url } } : resolution;
       }
       if (matched.model !== void 0 && scan.text !== void 0 && scan.model !== void 0) {
         const rewritten = scan.text.replace(new RegExp('"model":\\s*"' + scan.model.replace(/[.*+?^${}()|[\]\\]/g, "\\$&") + '"'), '"model":"' + matched.model + '"');
@@ -133868,6 +134022,67 @@ function createRealQuotaRefresher(params) {
 
 // src/frontdoor/commands.ts
 init_store2();
+
+// src/selection/targetStore.ts
+init_read();
+
+// src/selection/signals.ts
+function targetOf(record2) {
+  return record2.model === void 0 ? void 0 : `${record2.provider}/${record2.model}`;
+}
+function median(values) {
+  if (values.length === 0) {
+    return void 0;
+  }
+  const sorted = [...values].sort((left, right) => left - right);
+  const middle = Math.floor(sorted.length / 2);
+  const above = sorted[middle];
+  const below = sorted[middle - 1];
+  return sorted.length % 2 === 0 && above !== void 0 && below !== void 0 ? (above + below) / 2 : sorted[middle];
+}
+var OBSERVATION_WINDOW = 20;
+function observeRecord(observations, record2) {
+  const target = targetOf(record2);
+  if (target === void 0) {
+    return new Map(observations);
+  }
+  const kept = [...observations.get(target) ?? [], record2].slice(-OBSERVATION_WINDOW);
+  return new Map(observations).set(target, kept);
+}
+function observationOf(records) {
+  const completed = records.filter((record2) => record2.outcome === "completed");
+  const latencies = completed.map((record2) => record2.latencyMs);
+  const usage = records.flatMap((record2) => record2.usage === void 0 ? [] : [record2.usage]);
+  const input2 = usage.reduce((sum, entry) => sum + (entry.inputTokens ?? 0), 0);
+  const output2 = usage.reduce((sum, entry) => sum + (entry.outputTokens ?? 0), 0);
+  const cacheRead = usage.reduce((sum, entry) => sum + (entry.cacheReadInputTokens ?? 0), 0);
+  return {
+    latencyMs: median(latencies),
+    successRate: records.length === 0 ? void 0 : completed.length / records.length,
+    tokenMix: input2 + output2 + cacheRead === 0 ? void 0 : { input: input2, output: output2, cacheRead }
+  };
+}
+
+// src/selection/targetStore.ts
+var SEED_LOOKBACK_RECORDS = 500;
+function createTargetObservationStore(fs26, paths) {
+  let observations = /* @__PURE__ */ new Map();
+  const seeded = readUsageLog(fs26, paths.usageLogDir, {}).records.slice(-SEED_LOOKBACK_RECORDS);
+  for (const record2 of seeded) {
+    observations = observeRecord(observations, record2);
+  }
+  return {
+    observe: (record2) => {
+      observations = observeRecord(observations, record2);
+    },
+    observation: (target) => {
+      const records = observations.get(target);
+      return records === void 0 ? void 0 : observationOf(records);
+    }
+  };
+}
+
+// src/frontdoor/commands.ts
 function respawnInvocation(paths) {
   const env = {};
   for (const [name, value2] of Object.entries({ ...process.env, AGENT_SHIM_HOME: paths.root })) {
@@ -134054,7 +134269,11 @@ function realFrontDoorSupervisorPorts(paths) {
   catalogueKeeper.tick();
   setInterval(catalogueKeeper.tick, ANTHROPIC_USAGE_FRESHNESS_MS).unref();
   const headroomDecision = createHeadroomDecision(createRoutingSettingReader(paths));
-  const recordUsage = withQuotaRefresh(usageStore.record, quotaRefresher);
+  const targetObservations = createTargetObservationStore(realFarmFs, paths);
+  const recordUsage = (record2) => {
+    withQuotaRefresh(usageStore.record, quotaRefresher)(record2);
+    targetObservations.observe(record2);
+  };
   const usageMiddleware = createUsageMiddleware({
     // Each finished response also tells the per-session headroom decision when the session's cache was last touched and whether it read from it.
     record: (record2, event) => {
@@ -146603,6 +146822,188 @@ function registerProviderCommand(program2, deps) {
   );
 }
 
+// src/models.ts
+init_commandDeps();
+init_providersStore();
+init_modelPicker();
+init_providerCatalogue();
+init_realPorts();
+
+// src/models/identity.ts
+var NON_IDENTITY_BRACKET_RE = /\[[^\]]*\]/g;
+var DATE_STAMP_RE = /-(?:19|20)\d{2}(?:0[1-9]|1[0-2])(?:0[1-9]|[12]\d|3[01])$/;
+var KNOWN_VARIANTS = /* @__PURE__ */ new Set(["high", "medium", "low", "thinking", "xhigh", "max", "highspeed", "sol", "terra", "luna", "mini", "flash-lite", "turbo"]);
+var VENDOR_BY_NAMESPACE = Object.freeze({
+  anthropic: "anthropic",
+  google: "google",
+  openai: "openai",
+  zai: "zai",
+  "z-ai": "zai",
+  minimax: "minimax",
+  qwen: "qwen",
+  deepseek: "deepseek",
+  moonshot: "moonshot",
+  meta: "meta",
+  mistral: "mistral"
+});
+var VENDOR_BY_TOKEN = Object.freeze({
+  claude: "anthropic",
+  gemini: "google",
+  antigravity: "google",
+  gpt: "openai",
+  glm: "zai",
+  minimax: "minimax",
+  qwen: "qwen",
+  deepseek: "deepseek",
+  kimi: "moonshot",
+  llama: "meta",
+  mistral: "mistral",
+  mixtral: "mistral"
+});
+function canonicalText(name) {
+  return name.trim().toLowerCase().replace(NON_IDENTITY_BRACKET_RE, "").replace(/[\s_]+/g, "-").replace(/([a-z])(\d+(?:\.\d+)+)/g, "$1-$2").replace(/-+/g, "-").replace(/^-|-$/g, "");
+}
+function splitVariants(tokens) {
+  const variants = [];
+  let end = tokens.length;
+  for (; ; ) {
+    const token = tokens[end - 1];
+    if (token === void 0 || !KNOWN_VARIANTS.has(token)) {
+      return { familyTokens: tokens.slice(0, end), variants };
+    }
+    variants.unshift(token);
+    end -= 1;
+  }
+}
+function versionOf(tokens) {
+  for (let index = 0; index < tokens.length; index += 1) {
+    const token = tokens[index];
+    const next = tokens[index + 1];
+    if (token === void 0) {
+      return "";
+    }
+    if (/^\d+$/.test(token) && next !== void 0 && /^\d+$/.test(next)) {
+      return `${token}.${next}`;
+    }
+    const dotted = /^\d+(?:\.\d+)+$/.exec(token);
+    if (dotted !== null) {
+      return dotted[0];
+    }
+  }
+  return "";
+}
+function modelIdentity(rawName) {
+  const segments = rawName.trim().toLowerCase().split("/");
+  const stripped = canonicalText(segments[segments.length - 1] ?? "").split(":")[0] ?? "";
+  const withoutDate = stripped.replace(DATE_STAMP_RE, "");
+  const tokens = withoutDate.split("-").filter((token) => token !== "");
+  const { familyTokens, variants } = splitVariants(tokens);
+  const version2 = versionOf(familyTokens);
+  const versionless = familyTokens.filter((token, index) => {
+    if (version2 === "") {
+      return true;
+    }
+    const dotted = /^\d+(?:\.\d+)+$/.test(token);
+    const previous = familyTokens[index - 1];
+    const next = familyTokens[index + 1];
+    const insideVersionRun = dotted || /^\d+$/.test(token) && (previous !== void 0 && /^\d+$/.test(previous) || next !== void 0 && /^\d+$/.test(next));
+    return !insideVersionRun;
+  });
+  const family = versionless.join("-");
+  const tokenVendor = familyTokens.find((token) => Object.keys(VENDOR_BY_TOKEN).some((key) => token === key || token.startsWith(key)));
+  const namespace = segments.length >= 2 ? (segments[0] ?? "").split(":").pop() ?? "" : "";
+  const namespaceVendor = VENDOR_BY_NAMESPACE[namespace] ?? "";
+  const tokenVendorName = tokenVendor === void 0 ? "" : VENDOR_BY_TOKEN[Object.keys(VENDOR_BY_TOKEN).find((key) => tokenVendor === key || tokenVendor.startsWith(key)) ?? ""] ?? "";
+  return {
+    vendor: tokenVendorName === "" ? namespaceVendor : tokenVendorName,
+    vendorConflict: tokenVendorName !== "" && namespaceVendor !== "" && tokenVendorName !== namespaceVendor,
+    family,
+    version: version2,
+    variant: variants.join("-"),
+    familyAndVersion: version2 === "" ? family : `${family}-${version2.replace(/\./g, "-")}`
+  };
+}
+function relateModels(left, right) {
+  const a10 = modelIdentity(left);
+  const b = modelIdentity(right);
+  if (a10.familyAndVersion !== b.familyAndVersion) {
+    return sameFamilyOtherVersion(a10, b) ? "similar" : "distinct";
+  }
+  if (a10.variant !== b.variant || a10.vendorConflict || b.vendorConflict || a10.vendor !== "" && b.vendor !== "" && a10.vendor !== b.vendor) {
+    return "similar";
+  }
+  return "same";
+}
+function sameFamilyOtherVersion(a10, b) {
+  if (a10.family !== b.family) {
+    return false;
+  }
+  if (a10.vendor !== "" && b.vendor !== "" && a10.vendor !== b.vendor) {
+    return false;
+  }
+  return true;
+}
+
+// src/models.ts
+function providerModels(paths) {
+  return listProviders(paths).flatMap((entry) => {
+    const catalogue = parseCatalogueCache(realFsPort.readFileUtf8(catalogueCachePath(paths, entry.name)));
+    return providerModelRows(entry.provider, catalogue?.models).map((row) => ({ provider: entry.name, model: row.model }));
+  });
+}
+function matchModels(models) {
+  const same = [];
+  for (const model of models) {
+    const group = same.find((candidates) => candidates.some((other) => relateModels(other.model, model.model) === "same" && other.provider !== model.provider));
+    if (group === void 0) {
+      same.push([model]);
+    } else {
+      group.push(model);
+    }
+  }
+  const similar = [];
+  for (let outer = 0; outer < models.length; outer += 1) {
+    for (let inner = outer + 1; inner < models.length; inner += 1) {
+      const left = models[outer];
+      const right = models[inner];
+      if (left === void 0 || right === void 0 || left.provider === right.provider) {
+        continue;
+      }
+      if (relateModels(left.model, right.model) === "similar" && !same.some((group) => group.includes(left) && group.includes(right))) {
+        similar.push([left, right]);
+      }
+    }
+  }
+  return { same: same.filter((group) => group.length > 1), similar };
+}
+function registerModelsCommand(program2, deps) {
+  const { paths } = deps;
+  const modelsCommand = program2.command("model").description("Inspect the models the defined providers offer: which names are the same model across providers, and which only look alike.");
+  withExamples(modelsCommand, ["agent-shim model match"]);
+  withExamples(
+    modelsCommand.command("match").description("List every model every provider offers, grouped into the ones that are the same model across providers, with the pairs that are merely similar named beside them.").option("--json", "Print the match as JSON."),
+    ["agent-shim model match"]
+  ).action((options) => {
+    const models = providerModels(paths);
+    const { same, similar } = matchModels(models);
+    if (options.json === true) {
+      printJson({ models, same, similar });
+      return;
+    }
+    console.log(`Models of ${String(models.length)} across the defined providers:`);
+    if (same.length === 0 && similar.length === 0) {
+      console.log("  no two names are the same model, and none look alike");
+      return;
+    }
+    for (const group of same) {
+      console.log(`  same: ${group.map((entry) => `${entry.provider}/${entry.model}`).join(" = ")}`);
+    }
+    for (const [left, right] of similar) {
+      console.log(`  similar: ${left.provider}/${left.model} ~ ${right.provider}/${right.model}`);
+    }
+  });
+}
+
 // src/directoryRules.ts
 init_commandDeps();
 init_cliError();
@@ -147097,6 +147498,7 @@ function buildProgram(deps) {
   registerJobCommand(program2, deps);
   registerHeadlessCommands(program2, deps);
   registerProviderCommand(program2, deps);
+  registerModelsCommand(program2, deps);
   registerRuleCommand(program2, deps);
   registerCredentialCommand(program2, deps, deps.credentialPorts);
   registerCheckCommand(program2, deps);

package/dist/index.cjs

@@ -2983,7 +2983,7 @@ var init_package = __esm({
   "package.json"() {
     package_default = {
       name: "claude-use",
-      version: "11.31.0",
+      version: "11.32.0",
       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>",
@@ -8186,6 +8186,13 @@ var init_launchReport = __esm({
   }
 });
 
+// src/selection/engine.ts
+var init_engine = __esm({
+  "src/selection/engine.ts"() {
+    "use strict";
+  }
+});
+
 // src/antigravity/cloudUpstream.ts
 var init_cloudUpstream = __esm({
   "src/antigravity/cloudUpstream.ts"() {
@@ -8320,6 +8327,7 @@ var AUTO_ALIAS;
 var init_translate = __esm({
   "src/antigravity/translate.ts"() {
     "use strict";
+    init_engine();
     init_quota();
     init_signatures();
     init_schema();

package/dist/index.mjs

@@ -2959,7 +2959,7 @@ var init_package = __esm({
   "package.json"() {
     package_default = {
       name: "claude-use",
-      version: "11.31.0",
+      version: "11.32.0",
       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>",
@@ -8158,6 +8158,13 @@ var init_launchReport = __esm({
   }
 });
 
+// src/selection/engine.ts
+var init_engine = __esm({
+  "src/selection/engine.ts"() {
+    "use strict";
+  }
+});
+
 // src/antigravity/cloudUpstream.ts
 var init_cloudUpstream = __esm({
   "src/antigravity/cloudUpstream.ts"() {
@@ -8292,6 +8299,7 @@ var AUTO_ALIAS;
 var init_translate = __esm({
   "src/antigravity/translate.ts"() {
     "use strict";
+    init_engine();
     init_quota();
     init_signatures();
     init_schema();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claude-use",
-  "version": "11.31.0",
+  "version": "11.32.0",
   "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>",