@expiren/opencode-antigravity-auth 1.6.19 → 1.6.21

package/dist/index.js

@@ -27,17 +27,14 @@ var ANTIGRAVITY_SCOPES = [
 ];
 var ANTIGRAVITY_REDIRECT_URI = "http://localhost:51121/oauth-callback";
 var ANTIGRAVITY_ENDPOINT_DAILY = "https://daily-cloudcode-pa.sandbox.googleapis.com";
-var ANTIGRAVITY_ENDPOINT_AUTOPUSH = "https://autopush-cloudcode-pa.sandbox.googleapis.com";
 var ANTIGRAVITY_ENDPOINT_PROD = "https://cloudcode-pa.googleapis.com";
 var ANTIGRAVITY_ENDPOINT_FALLBACKS = [
   ANTIGRAVITY_ENDPOINT_DAILY,
-  ANTIGRAVITY_ENDPOINT_AUTOPUSH,
   ANTIGRAVITY_ENDPOINT_PROD
 ];
 var ANTIGRAVITY_LOAD_ENDPOINTS = [
   ANTIGRAVITY_ENDPOINT_PROD,
-  ANTIGRAVITY_ENDPOINT_DAILY,
-  ANTIGRAVITY_ENDPOINT_AUTOPUSH
+  ANTIGRAVITY_ENDPOINT_DAILY
 ];
 var ANTIGRAVITY_ENDPOINT = ANTIGRAVITY_ENDPOINT_DAILY;
 var GEMINI_CLI_ENDPOINT = ANTIGRAVITY_ENDPOINT_PROD;
@@ -2032,7 +2029,10 @@ var QUOTA_GROUP_BY_MODEL_ID = {
   "gemini-3.1-pro-high": "gemini-pro",
   "gemini-3-flash": "gemini-flash",
   "gemini-3-flash-agent": "gemini-flash",
-  "gemini-3.5-flash-low": "gemini-flash"
+  "gemini-3.5-flash-low": "gemini-flash",
+  "gemini-3.1-flash-image": "gemini-flash",
+  "gpt-oss-120b": "gpt-oss",
+  "gpt-oss-120b-medium": "gpt-oss"
 };
 var OPENCODE_MODEL_DEFINITIONS = PUBLIC_MODEL_DEFINITIONS;
 function getResolverAliasMap() {
@@ -3020,9 +3020,21 @@ var AntigravityConfigSchema = z2.object({
    * @default false
    */
   keep_thinking: z2.boolean().default(false),
+  /**
+   * Enable thinking warmup requests for Claude thinking models.
+   * When enabled, sends a separate API call at session start to warm up
+   * thinking signature validation. Costs 1 full API call per session.
+   *
+   * Disable to save quota — most users don't need this unless using
+   * keep_thinking with signature caching.
+   *
+   * Env override: OPENCODE_ANTIGRAVITY_THINKING_WARMUP=1
+   * @default false
+   */
+  thinking_warmup: z2.boolean().default(false),
   // =========================================================================
   // Session Recovery
-  // =========================================================================
+  // =========================================================================  
   /**
    * Enable automatic session recovery from tool_result_missing errors.
    * When enabled, shows a toast notification when recoverable errors occur.
@@ -3062,9 +3074,9 @@ var AntigravityConfigSchema = z2.object({
    * Maximum retry attempts when Antigravity returns an empty response.
    * Empty responses occur when no candidates/choices are returned.
    * 
-   * @default 4
+   * @default 2
    */
-  empty_response_max_attempts: z2.number().min(1).max(10).default(4),
+  empty_response_max_attempts: z2.number().min(1).max(10).default(2),
   /**
    * Delay in milliseconds between empty response retries.
    * 
@@ -3185,8 +3197,28 @@ var AntigravityConfigSchema = z2.object({
      */
   switch_on_first_rate_limit: z2.boolean().default(true),
   /**
-   * Scheduling mode for rate limit behavior.
+   * Maximum number of account switches per request before giving up.
+   * Each switch re-sends the full request payload, consuming quota on the new account.
+   * Lower values reduce quota waste from cascading rate limits across accounts.
+   * 
+   * Env override: OPENCODE_ANTIGRAVITY_MAX_ACCOUNT_SWITCHES
+   * @default 2
+   */
+  max_account_switches: z2.number().min(0).max(50).default(2),
+  /**
+   * Allow falling back between quota pools (antigravity ↔ gemini-cli) when rate-limited.
+   * When enabled, if one quota pool is exhausted the plugin re-sends the SAME request
+   * using the alternate header style, consuming tokens from BOTH pools.
+   * Disable to prevent double-spending quota across pools — the plugin will only
+   * rotate accounts instead.
+   * Only applies to Gemini models (Claude always uses antigravity).
    * 
+   * Env override: OPENCODE_ANTIGRAVITY_QUOTA_STYLE_FALLBACK
+   * @default false
+   */
+  quota_style_fallback: z2.boolean().default(false),
+  /**
+   * Scheduling mode for rate limit behavior.     * 
    * - `cache_first`: Wait for same account to recover (preserves prompt cache). Default.
    * - `balance`: Switch account immediately on rate limit. Maximum availability.
    * - `performance_first`: Round-robin distribution for maximum throughput.
@@ -3250,7 +3282,7 @@ var AntigravityConfigSchema = z2.object({
    * 
    * @default 15
    */
-  quota_refresh_interval_minutes: z2.number().min(0).max(60).default(15),
+  quota_refresh_interval_minutes: z2.number().min(0).max(120).default(30),
   /**
    * How long quota cache is considered fresh for threshold checks (in minutes).
    * After this time, cache is stale and account is allowed (fail-open).
@@ -3298,10 +3330,11 @@ var DEFAULT_CONFIG = {
   debug: false,
   debug_tui: false,
   keep_thinking: false,
+  thinking_warmup: false,
   session_recovery: true,
   auto_resume: true,
   resume_text: "continue",
-  empty_response_max_attempts: 4,
+  empty_response_max_attempts: 2,
   empty_response_retry_delay_ms: 2e3,
   tool_id_recovery: true,
   claude_tool_hardening: true,
@@ -3315,6 +3348,8 @@ var DEFAULT_CONFIG = {
   account_selection_strategy: "hybrid",
   pid_offset_enabled: false,
   switch_on_first_rate_limit: true,
+  max_account_switches: 2,
+  quota_style_fallback: false,
   scheduling_mode: "cache_first",
   max_cache_first_wait_seconds: 60,
   failure_ttl_seconds: 3600,
@@ -3322,7 +3357,7 @@ var DEFAULT_CONFIG = {
   max_backoff_seconds: 60,
   request_jitter_max_ms: 0,
   soft_quota_threshold_percent: 90,
-  quota_refresh_interval_minutes: 15,
+  quota_refresh_interval_minutes: 30,
   soft_quota_cache_ttl_minutes: "auto",
   auto_update: true,
   signature_cache: {
@@ -5603,6 +5638,12 @@ function needsThinkingRecovery(state) {
 
 // src/plugin/transform/claude.ts
 var CLAUDE_THINKING_MAX_OUTPUT_TOKENS = 64e3;
+function computeClaudeMaxOutputTokens(thinkingBudget) {
+  if (typeof thinkingBudget !== "number" || thinkingBudget <= 0) {
+    return CLAUDE_THINKING_MAX_OUTPUT_TOKENS;
+  }
+  return Math.min(Math.max(thinkingBudget * 2, 32e3), CLAUDE_THINKING_MAX_OUTPUT_TOKENS);
+}
 function isClaudeModel(model) {
   return model.toLowerCase().includes("claude");
 }
@@ -6135,7 +6176,7 @@ function resolveModelWithTier(requestedModel, options = {}) {
     if (isClaudeThinking) {
       return {
         actualModel: resolvedModel,
-        thinkingBudget: THINKING_TIER_BUDGETS.claude.high,
+        thinkingBudget: THINKING_TIER_BUDGETS.claude.medium,
         isThinkingModel: true,
         quotaPreference,
         explicitQuota
@@ -7507,7 +7548,7 @@ function prepareAntigravityRequest(input2, init, accessToken, projectId, endpoin
               if (isClaudeThinking && typeof thinkingBudget === "number" && thinkingBudget > 0) {
                 const currentMax = rawGenerationConfig.maxOutputTokens ?? rawGenerationConfig.max_output_tokens;
                 if (!currentMax || currentMax <= thinkingBudget) {
-                  rawGenerationConfig.maxOutputTokens = CLAUDE_THINKING_MAX_OUTPUT_TOKENS;
+                  rawGenerationConfig.maxOutputTokens = computeClaudeMaxOutputTokens(thinkingBudget);
                   if (rawGenerationConfig.max_output_tokens !== void 0) {
                     delete rawGenerationConfig.max_output_tokens;
                   }
@@ -7517,7 +7558,7 @@ function prepareAntigravityRequest(input2, init, accessToken, projectId, endpoin
             } else {
               const generationConfig = { thinkingConfig };
               if (isClaudeThinking && typeof thinkingBudget === "number" && thinkingBudget > 0) {
-                generationConfig.maxOutputTokens = CLAUDE_THINKING_MAX_OUTPUT_TOKENS;
+                generationConfig.maxOutputTokens = computeClaudeMaxOutputTokens(thinkingBudget);
               }
               requestPayload.generationConfig = generationConfig;
             }
@@ -7827,7 +7868,7 @@ function buildThinkingWarmupBody(bodyText, isClaudeThinking) {
       include_thoughts: true,
       thinking_budget: DEFAULT_THINKING_BUDGET
     };
-    generationConfig.maxOutputTokens = CLAUDE_THINKING_MAX_OUTPUT_TOKENS;
+    generationConfig.maxOutputTokens = computeClaudeMaxOutputTokens(DEFAULT_THINKING_BUDGET);
     req.generationConfig = generationConfig;
   };
   if (parsed.request && typeof parsed.request === "object") {
@@ -10287,7 +10328,7 @@ function aggregateGeminiCliQuota(response) {
       continue;
     }
     const modelId = bucket.modelId;
-    const isRelevantModel = modelId.startsWith("gemini-3-") || modelId === "gemini-2.5-pro";
+    const isRelevantModel = modelId.startsWith("gemini-3-") || modelId.startsWith("gemini-3.") || modelId.startsWith("gemini-2.5-");
     if (!isRelevantModel) {
       continue;
     }
@@ -11914,6 +11955,8 @@ var createAntigravityPlugin = (providerId) => async ({ client, directory }) => {
               if (family !== "gemini") return false;
               return accountManager.hasOtherAccountWithAntigravityAvailable(currentAccount.index, family, model);
             };
+            let accountSwitchCount = 0;
+            const maxAccountSwitches = config.max_account_switches ?? 2;
             while (true) {
               checkAborted();
               const accountCount = accountManager.getAccountCount();
@@ -12134,6 +12177,9 @@ var createAntigravityPlugin = (providerId) => async ({ client, directory }) => {
                 }
               }
               const runThinkingWarmup = async (prepared, projectId) => {
+                if (!config.thinking_warmup) {
+                  return;
+                }
                 if (!prepared.needsSignedThinkingWarmup || !prepared.sessionId) {
                   return;
                 }
@@ -12245,6 +12291,7 @@ var createAntigravityPlugin = (providerId) => async ({ client, directory }) => {
                 let tokenConsumed = false;
                 let capacityRetryCount = 0;
                 let lastEndpointIndex = -1;
+                let apiRequestCount = 0;
                 for (let i = 0; i < ANTIGRAVITY_ENDPOINT_FALLBACKS.length; i++) {
                   if (i !== lastEndpointIndex) {
                     capacityRetryCount = 0;
@@ -12307,7 +12354,8 @@ var createAntigravityPlugin = (providerId) => async ({ client, directory }) => {
                       tokenConsumed = getTokenTracker().consume(account.index);
                     }
                     const response = await fetch(prepared.request, prepared.init);
-                    pushDebug(`status=${response.status} ${response.statusText}`);
+                    apiRequestCount++;
+                    pushDebug(`status=${response.status} ${response.statusText} (api_request #${apiRequestCount})`);
                     if (response.status === 429 || response.status === 503 || response.status === 529) {
                       if (tokenConsumed) {
                         getTokenTracker().refund(account.index);
@@ -12332,12 +12380,12 @@ var createAntigravityPlugin = (providerId) => async ({ client, directory }) => {
                           "warning"
                         );
                         await sleep(waitMs, abortSignal);
-                        if (capacityRetryCount < 3) {
+                        if (capacityRetryCount < 1) {
                           capacityRetryCount++;
                           i -= 1;
                           continue;
                         } else {
-                          pushDebug(`Max capacity retries (3) exhausted for endpoint ${currentEndpoint}, regenerating fingerprint...`);
+                          pushDebug(`Max capacity retries (1) exhausted for endpoint ${currentEndpoint}, regenerating fingerprint...`);
                           const newFingerprint = accountManager.regenerateAccountFingerprint(account.index);
                           if (newFingerprint) {
                             pushDebug(`Fingerprint regenerated for account ${account.index}`);
@@ -12589,6 +12637,9 @@ Alternatively, you can:
                         );
                       }
                     }
+                    if (apiRequestCount > 1) {
+                      pushDebug(`[Quota] Total API requests for this user message: ${apiRequestCount} (${apiRequestCount - 1} retries)`);
+                    }
                     return transformedResponse;
                   } catch (error) {
                     if (tokenConsumed) {
@@ -12635,6 +12686,27 @@ Alternatively, you can:
                 }
               }
               if (shouldSwitchAccount) {
+                accountSwitchCount++;
+                if (accountSwitchCount > maxAccountSwitches) {
+                  pushDebug(`account-switch-cap: exceeded max_account_switches=${maxAccountSwitches}, giving up`);
+                  if (lastFailure) {
+                    return transformAntigravityResponse(
+                      lastFailure.response,
+                      lastFailure.streaming,
+                      lastFailure.debugContext,
+                      lastFailure.requestedModel,
+                      lastFailure.projectId,
+                      lastFailure.endpoint,
+                      lastFailure.effectiveModel,
+                      lastFailure.sessionId,
+                      lastFailure.toolDebugMissing,
+                      lastFailure.toolDebugSummary,
+                      lastFailure.toolDebugPayload,
+                      debugLines
+                    );
+                  }
+                  throw lastError || new Error(`Exceeded max account switches (${maxAccountSwitches}). All accounts rate-limited.`);
+                }
                 if (accountCount <= 1) {
                   if (lastFailure) {
                     return transformAntigravityResponse(
@@ -13478,7 +13550,7 @@ function resolveHeaderRoutingDecision(urlString, family, config) {
     cliFirst,
     preferredHeaderStyle,
     explicitQuota,
-    allowQuotaFallback: family === "gemini"
+    allowQuotaFallback: family === "gemini" && !!(config.quota_style_fallback ?? false)
   };
 }
 function getCliFirst(config) {