fix: issue #400 header-style selection for gemini-cli

src/plugin.ts

@@ -1107,6 +1107,8 @@ export const createAntigravityPlugin = (providerId: string) => async (
           while (true) {
             // Check for abort at the start of each iteration
             checkAborted();
+
+            const requestedHeaderStyle = getHeaderStyleFromUrl(urlString, family);
 
             const accountCount = accountManager.getAccountCount();
 
@@ -1123,7 +1125,7 @@ export const createAntigravityPlugin = (providerId: string) => async (
               family, 
               model, 
               config.account_selection_strategy,
-              'antigravity',
+              requestedHeaderStyle,
               config.pid_offset_enabled,
               config.soft_quota_threshold_percent,
               softQuotaCacheTtlMs,
@@ -1160,7 +1162,7 @@ export const createAntigravityPlugin = (providerId: string) => async (
                 continue;
               }
 
-              const headerStyle = getHeaderStyleFromUrl(urlString, family);
+              const headerStyle = requestedHeaderStyle;
               const explicitQuota = isExplicitQuotaFromUrl(urlString);
               // All accounts are rate-limited - wait and retry
               const waitMs = accountManager.getMinWaitTimeForFamily(
@@ -1411,9 +1413,10 @@ export const createAntigravityPlugin = (providerId: string) => async (
             let shouldSwitchAccount = false;
 
             // Determine header style from model suffix:
-            // - Gemini models default to Antigravity
-            // - Claude models always use Antigravity
-            let headerStyle = getHeaderStyleFromUrl(urlString, family);
+            // - Claude models -> always use Antigravity
+            // - Gemini defaults to Antigravity (including explicit gemini-cli suffix)
+            // - Gemini CLI is used as fallback when Antigravity is exhausted
+            let headerStyle = requestedHeaderStyle;
             const explicitQuota = isExplicitQuotaFromUrl(urlString);
             const cliFirst = getCliFirst(config);
             pushDebug(`headerStyle=${headerStyle} explicit=${explicitQuota}`);

src/plugin/accounts.test.ts

@@ -1056,6 +1056,51 @@ describe("AccountManager", () => {
     });
   });
 
+  describe("Issue #400: hybrid strategy respects requested header style", () => {
+    it("skips account when requested antigravity style is rate-limited", () => {
+      const stored: AccountStorageV3 = {
+        version: 3,
+        accounts: [
+          { refreshToken: "r1", projectId: "p1", addedAt: 1, lastUsed: 0 },
+          { refreshToken: "r2", projectId: "p2", addedAt: 1, lastUsed: 0 },
+        ],
+        activeIndex: 0,
+        activeIndexByFamily: { claude: 0, gemini: 0 },
+      };
+
+      const manager = new AccountManager(undefined, stored);
+      const firstAccount = manager.getCurrentOrNextForFamily("gemini", null, "hybrid", "antigravity");
+
+      manager.markRateLimited(firstAccount!, 60000, "gemini", "antigravity");
+
+      expect(manager.isRateLimitedForHeaderStyle(firstAccount!, "gemini", "antigravity")).toBe(true);
+      expect(manager.isRateLimitedForHeaderStyle(firstAccount!, "gemini", "gemini-cli")).toBe(false);
+
+      const nextAccount = manager.getCurrentOrNextForFamily("gemini", null, "hybrid", "antigravity");
+      expect(nextAccount?.index).toBe(1);
+    });
+
+    it("keeps account for gemini-cli when only antigravity style is rate-limited", () => {
+      const stored: AccountStorageV3 = {
+        version: 3,
+        accounts: [
+          { refreshToken: "r1", projectId: "p1", addedAt: 1, lastUsed: 0 },
+        ],
+        activeIndex: 0,
+        activeIndexByFamily: { claude: 0, gemini: 0 },
+      };
+
+      const manager = new AccountManager(undefined, stored);
+      const firstAccount = manager.getCurrentOrNextForFamily("gemini", null, "hybrid", "gemini-cli");
+
+      manager.markRateLimited(firstAccount!, 60000, "gemini", "antigravity");
+
+      const nextAccount = manager.getCurrentOrNextForFamily("gemini", null, "hybrid", "gemini-cli");
+      expect(nextAccount).not.toBeNull();
+      expect(nextAccount?.parts.refreshToken).toBe("r1");
+    });
+  });
+
   describe("Issue #174: saveToDisk throttling", () => {
     it("requestSaveToDisk coalesces multiple calls into one write", async () => {
       vi.useFakeTimers();
@@ -1608,6 +1653,38 @@ describe("AccountManager", () => {
       expect(account?.parts.refreshToken).toBe("r1");
     });
 
+    it("does not apply soft quota threshold to gemini-cli in sticky mode", () => {
+      const stored: AccountStorageV3 = {
+        version: 3,
+        accounts: [
+          { refreshToken: "r1", projectId: "p1", addedAt: 1, lastUsed: 0 },
+        ],
+        activeIndex: 0,
+      };
+
+      const manager = new AccountManager(undefined, stored);
+      manager.updateQuotaCache(0, { "gemini-pro": { remainingFraction: 0.01, modelCount: 1 } });
+
+      const account = manager.getCurrentOrNextForFamily("gemini", "gemini-2.5-pro", "sticky", "gemini-cli", false, 90);
+      expect(account?.parts.refreshToken).toBe("r1");
+    });
+
+    it("does not apply soft quota threshold to gemini-cli in hybrid mode", () => {
+      const stored: AccountStorageV3 = {
+        version: 3,
+        accounts: [
+          { refreshToken: "r1", projectId: "p1", addedAt: 1, lastUsed: 0 },
+        ],
+        activeIndex: 0,
+      };
+
+      const manager = new AccountManager(undefined, stored);
+      manager.updateQuotaCache(0, { "gemini-pro": { remainingFraction: 0.01, modelCount: 1 } });
+
+      const account = manager.getCurrentOrNextForFamily("gemini", "gemini-2.5-pro", "hybrid", "gemini-cli", false, 90);
+      expect(account?.parts.refreshToken).toBe("r1");
+    });
+
     it("returns null when all accounts over threshold", () => {
       const stored: AccountStorageV3 = {
         version: 3,

src/plugin/accounts.ts

@@ -516,20 +516,22 @@ export class AccountManager {
     if (strategy === 'hybrid') {
       const healthTracker = getHealthTracker();
       const tokenTracker = getTokenTracker();
-
-      const accountsWithMetrics: AccountWithMetrics[] = this.accounts
-        .filter(acc => acc.enabled !== false)
-        .map(acc => {
-          clearExpiredRateLimits(acc);
-          return {
-            index: acc.index,
-            lastUsed: acc.lastUsed,
-            healthScore: healthTracker.getScore(acc.index),
-            isRateLimited: isRateLimitedForFamily(acc, family, model) || 
-                          isOverSoftQuotaThreshold(acc, family, softQuotaThresholdPercent, softQuotaCacheTtlMs, model),
-            isCoolingDown: this.isAccountCoolingDown(acc),
-          };
-        });
+
+      const shouldApplySoftQuota = headerStyle !== "gemini-cli";
+      const accountsWithMetrics: AccountWithMetrics[] = this.accounts.map(acc => {
+        clearExpiredRateLimits(acc);
+        const isOverThreshold = shouldApplySoftQuota &&
+          isOverSoftQuotaThreshold(acc, family, softQuotaThresholdPercent, softQuotaCacheTtlMs, model);
+        return {
+          index: acc.index,
+          lastUsed: acc.lastUsed,
+          healthScore: healthTracker.getScore(acc.index),
+          isRateLimited: acc.enabled === false ||
+            isRateLimitedForHeaderStyle(acc, family, headerStyle, model) ||
+            isOverThreshold,
+          isCoolingDown: this.isAccountCoolingDown(acc),
+        };
+      });
 
       // Get current account index for stickiness
       const currentIndex = this.currentAccountIndexByFamily[family] ?? null;
@@ -564,7 +566,8 @@ export class AccountManager {
     if (current) {
       clearExpiredRateLimits(current);
       const isLimitedForRequestedStyle = isRateLimitedForHeaderStyle(current, family, headerStyle, model);
-      const isOverThreshold = isOverSoftQuotaThreshold(current, family, softQuotaThresholdPercent, softQuotaCacheTtlMs, model);
+      const isOverThreshold = headerStyle !== "gemini-cli" &&
+        isOverSoftQuotaThreshold(current, family, softQuotaThresholdPercent, softQuotaCacheTtlMs, model);
       if (!isLimitedForRequestedStyle && !isOverThreshold && !this.isAccountCoolingDown(current)) {
         this.markTouchedForQuota(current, quotaKey);
         return current;
@@ -582,9 +585,11 @@ export class AccountManager {
   getNextForFamily(family: ModelFamily, model?: string | null, headerStyle: HeaderStyle = "antigravity", softQuotaThresholdPercent: number = 100, softQuotaCacheTtlMs: number = 10 * 60 * 1000): ManagedAccount | null {
     const available = this.accounts.filter((a) => {
       clearExpiredRateLimits(a);
+      const isOverThreshold = headerStyle !== "gemini-cli" &&
+        isOverSoftQuotaThreshold(a, family, softQuotaThresholdPercent, softQuotaCacheTtlMs, model);
       return a.enabled !== false && 
              !isRateLimitedForHeaderStyle(a, family, headerStyle, model) && 
-             !isOverSoftQuotaThreshold(a, family, softQuotaThresholdPercent, softQuotaCacheTtlMs, model) &&
+             !isOverThreshold &&
              !this.isAccountCoolingDown(a);
     });