@goodandready/dsh-subscriptions 0.6.68 → 0.6.69

package/README.md

@@ -207,6 +207,11 @@ const res = await ctx.subscriptions.request('codex', '/backend-api/codex/images/
 * **In-App Toast Alerts (`enableToasts`)**: Lightweight floating toast alerts in the DSH chat interface notifying you when accounts enter quarantine, recover from cooldowns, or exceed quota safety thresholds.
 * **Automated Webhook Integration**: Instant event notifications dispatched via `notifyQuarantineEntered` and `notifyQuarantineReleased`.
 
+### 23. 🧠 Antigravity Tiered Models, Tool Sentinel & Stream Resilience (v0.6.69)
+* **Gemini 3 Tool Replay Sentinel (#534)**: Injects `thoughtSignature: 'skip_thought_signature_validator'` on reconstructed tool calls, enabling reliable multi-turn tool calling without Gemini 3 validation failures.
+* **Tiered Flash Models & Thinking Levels (#534)**: Exposes Gemini 3.8 and 3.7 Flash High, Medium, and Low thinking levels with upstream `*-flash-tiered` mapping.
+* **Stream Resilience & Empty Stream Retry (#534)**: Idle SSE timeout (30s), empty response detection, and transient stream retry with automatic account rotation.
+
 ### 22. ⚡ xAI Grok RFC 8628 Device Authorization Flow (v0.6.68)
 * **RFC 8628 Device Code Authorization (GH #17, #532)**: Native device authorization grant support for xAI Grok. Solves the issue where xAI presents a Grok Build user-code verification page for CLI client IDs without redirecting back to loopback callbacks. Provides direct user code display and automatic token polling.
 

package/README.ru.md

@@ -183,6 +183,11 @@ const res = await ctx.subscriptions.request('codex', '/backend-api/codex/images/
 * **Всплывающие уведомления (Toast Alerts)**: Легковесные всплывающие тосты в веб-интерфейсе DSH при событиях карантина (с отображением причины), восстановлении доступа и критическом расходе квоты. Включение управляется параметром `enableToasts` (по умолчанию `true`).
 * **Интеграция с вебхуками**: Автоматическая отправка уведомлений через `notifyQuarantineEntered` и `notifyQuarantineReleased`.
 
+### 19. 🧠 Antigravity Tiered-модели, сентинел тулов и устойчивость стримов (v0.6.69)
+* **Сентинел реплея тулов Gemini 3 (#534)**: Автоматическая вставка `thoughtSignature: 'skip_thought_signature_validator'` для реплея истории вызовов тулов без сбоев валидации Google API.
+* **Tiered-модели и уровни мышления (#534)**: Выбор уровней High/Medium/Low для Gemini 3.8 и 3.7 Flash с маппингом в upstream `*-flash-tiered` и `thinkingLevel`.
+* **Устойчивость стримов и авто-ретрай (#534)**: Защита от зависания SSE (таймаут 30с), детекция пустых ответов и авто-ротация аккаунтов по `EMPTY_RESPONSE`/`TIMEOUT`.
+
 ### 18. ⚡ RFC 8628 Device Code авторизация для xAI Grok (v0.6.68)
 * **Авторизация по коду устройства (GH #17, #532)**: Нативная поддержка RFC 8628 device code flow для xAI Grok. Устраняет проблему, когда xAI для CLI client ID открывает страницу ввода кода для Grok Build без обратного редиректа на loopback-порт. Отображает одноразовый код на карточке и автоматически поллит токены в фоне.
 

package/README.zh.md

@@ -132,6 +132,11 @@ graph LR
   queueTimeoutMs: 30000     # v0.6.62: 队列最大等待超时时间（毫秒）
 * **自动化 Webhook 告警集成**：通过 `notifyQuarantineEntered` 与 `notifyQuarantineReleased` 自动向外部通知系统派发隔离生命周期事件。
 
+### Antigravity 分档思考模型、工具回放哨兵与流式自愈 (v0.6.69)
+* **Gemini 3 工具回放哨兵 (#534)**：在重建的工具调用中注入 `thoughtSignature: 'skip_thought_signature_validator'`，彻底解决多轮工具调用历史重放时 Gemini 3 的校验失败问题。
+* **分档 Flash 模型与思考档位 (#534)**：将 `gemini-3.8-flash-tiered` 与 `3.7` 自动解析展开为 High/Medium/Low 档位选择，并在调用时自动组装 `thinkingConfig.thinkingLevel`。
+* **流式空响应自愈与账号轮换 (#534)**：引入 30 秒 SSE 空闲超时保护与空响应检测，在未输出前自动重试一次，失败时通过 `EMPTY_RESPONSE` / `TIMEOUT` 自动轮换至下一个可用账号。
+
 ### xAI Grok RFC 8628 设备码授权流支持 (v0.6.68)
 * **RFC 8628 设备码授权流 (GH #17, #532)**：为 xAI Grok 新增原生设备码登录授权支持。彻底解决 xAI 针对 CLI 客户端标识返回 Grok Build 验证码页面且不触发回环重定向回调导致的死锁问题。直接在槽位卡片展示用户授权码，并自动在后台轮询兑换会话令牌。
 

package/lib/code-assist.js

@@ -156,21 +156,28 @@ function catalogRows(json) {
   return []
 }
 
+export function antigravityTieredAlias(model) {
+  const match = /^gemini-(3\.[78])-flash-(low|medium|high)$/.exec(model || '')
+  return match ? { model: `gemini-${match[1]}-flash-tiered`, thinkingLevel: match[2] } : null
+}
+
 export async function fetchAvailableModels(fetchImpl, token, extraHeaders, fallbackIds, provider) {
   try {
     const json = await postAssist(fetchImpl, 'fetchAvailableModels', token, {}, extraHeaders)
     const out = []
-    for (const row of catalogRows(json)) {
+    const catalog = catalogRows(json)
+    const publishedIds = new Set(catalog.map((row) => String(row?.id || '').replace(/^models\//, '')))
+    for (const row of catalog) {
       if (!row) continue
       if (typeof row === 'string') {
-        if (row && !row.startsWith('chat_')) out.push({ provider, id: row, name: row })
+        if (row && !row.startsWith('chat_') && !row.startsWith('tab_')) out.push({ provider, id: row, name: row })
         continue
       }
       if (row.isInternal) continue
       const raw = String(row.id || row.name || row.model || '').replace(/^models\//, '')
-      if (!raw || raw.startsWith('chat_')) continue
+      if (!raw || raw.startsWith('chat_') || raw.startsWith('tab_')) continue
       const display = row.displayName ?? row.display_name
-      if (display === '') continue
+      if (display === '' && !raw.endsWith('-tiered')) continue
       out.push({
         provider,
         id: raw,
@@ -178,6 +185,19 @@ export async function fetchAvailableModels(fetchImpl, token, extraHeaders, fallb
         ...(row.recommended ? { description: 'Recommended' } : {}),
         ...(row.maxTokens ? { contextWindow: row.maxTokens } : {}),
       })
+      const tiered = /^gemini-(3\.[78])-flash-tiered$/.exec(raw)
+      if (tiered) {
+        for (const level of ['high', 'medium', 'low']) {
+          const id = `gemini-${tiered[1]}-flash-${level}`
+          if (publishedIds.has(id)) continue
+          out.push({
+            provider,
+            id,
+            name: `Gemini ${tiered[1]} Flash (${level[0].toUpperCase()}${level.slice(1)})`,
+            ...(row.maxTokens ? { contextWindow: row.maxTokens } : {}),
+          })
+        }
+      }
     }
     if (out.length) {
       out.sort((a, b) => Number(Boolean(b.description)) - Number(Boolean(a.description)) || a.name.localeCompare(b.name))

package/lib/messages.js

@@ -478,7 +478,13 @@ export function googleContents(options) {
       }))),
     ]
     for (const call of calls) {
-      parts.push({ functionCall: { name: call.name, args: safeJson(call.arguments) } })
+      // DSH provider-neutral tool history cannot retain Gemini's opaque thoughtSignature.
+      // Gemini 3 requires one on replay after a tool result; Google's documented sentinel
+      // permits reconstructed/foreign history.
+      parts.push({
+        functionCall: { name: call.name, args: safeJson(call.arguments) },
+        thoughtSignature: 'skip_thought_signature_validator',
+      })
     }
     for (const result of toolResults(message.content)) {
       const name = result.toolName || result.name || 'tool'

package/lib/rotate.js

@@ -18,6 +18,7 @@ export function isRegionError(err) {
 
 export function isSwitchableError(err) {
   if (!err || typeof err !== 'object') return false
+  if (err.name === 'TimeoutError' || err.code === 'TIMEOUT' || err.code === 'EMPTY_RESPONSE') return true
   if (err.name === 'AbortError' || err.code === 'ABORT_ERR' || err.code === 'ERR_ABORTED' || (err.message && /aborted/i.test(err.message))) return false
   const code = err.code || (err.failure && err.failure.code)
   if (SWITCH_CODES.has(code)) return true

package/lib/vendors/antigravity.js

@@ -2,6 +2,7 @@ import { randomUUID } from 'node:crypto'
 import { buildAuthorizeUrl } from '../oauth.js'
 import { googleContents } from '../messages.js'
 import { formTokenRequest, googleStream, throwHttpError, tokenBlobFromOAuth } from '../wire.js'
+import { fetchWithTimeout } from '../http.js'
 import { emailFromToken } from '../jwt.js'
 import {
   CODE_ASSIST_STREAM,
@@ -9,6 +10,7 @@ import {
   antigravityIdentityHeaders,
   discoverProject,
   fetchAvailableModels,
+  antigravityTieredAlias,
   retrieveQuotaPercent,
   streamEnvelope,
   assistHeaders,
@@ -38,14 +40,22 @@ export function defaults() {
     clientSecret: '',
     redirectUri: 'http://localhost:8085/oauth/callback',
     models: [
+      'gemini-3.8-flash-high',
       'gemini-3.8-flash-medium',
+      'gemini-3.8-flash-low',
+      'gemini-3.8-flash-tiered',
+      'gemini-3.7-flash-high',
       'gemini-3.7-flash-medium',
+      'gemini-3.7-flash-low',
+      'gemini-3.7-flash-tiered',
+      'gemini-3.6-flash-high',
       'gemini-3.6-flash-medium',
+      'gemini-3.6-flash-low',
       'gemini-3.1-pro-low',
-      'claude-sonnet-4.6-thinking',
-      'claude-opus-4.6-thinking',
+      'gemini-3.1-pro-high',
+      'claude-sonnet-4-6',
+      'claude-opus-4-6-thinking',
       'gpt-oss-120b-medium',
-      'gemini-3.1-pro-high-vertex',
       'gemini-3-flash',
       'gemini-2.5-pro',
       'gemini-2.5-flash',
@@ -183,30 +193,72 @@ export async function usage(blob, _cfg, fetchImpl) {
   return retrieveQuotaPercent(fetchImpl || fetch, blob.accessToken, headersFor(blob.projectId || ''))
 }
 
+const RESPONSE_TIMEOUT_MS = 30000
+
+function isTransientFailure(err) {
+  return err?.name === 'TimeoutError' || err?.code === 'TIMEOUT' || err?.code === 'EMPTY_RESPONSE' || err instanceof TypeError
+}
+
 export async function* streamOnce({ blob, options, fetchImpl, headers, signal, saveBlob }) {
+  const impl = fetchImpl || fetch
   // Abort must cover the project pre-flight as well as the stream.
-  const preflight = (u, i) => fetchImpl(u, { ...(i || {}), signal })
-  const ready = await withProject(blob, preflight, saveBlob)
-  const { envelope, sessionId } = streamEnvelope({
-    projectId: ready.projectId,
-    model: options.model,
-    request: googleContents(options),
-    userAgent: 'antigravity',
-    sessionId: ready.sessionId,
-    paidTierId: ready.paidTierId,
-  })
-  const res = await fetchImpl(`${CODE_ASSIST_STREAM}:streamGenerateContent?alt=sse`, {
-    method: 'POST',
-    headers: {
-      ...headers,
-      ...assistHeaders(ready.accessToken, { ...headersFor(ready.projectId || ''), Accept: 'text/event-stream' }),
-    },
-    body: JSON.stringify(envelope),
-    signal,
-  })
-  if (!res.ok) throwHttpError(res.status, await res.text())
-  if (saveBlob && sessionId && sessionId !== ready.sessionId) {
-    await saveBlob({ ...ready, sessionId })
+  const preflight = (u, i) => fetchWithTimeout(impl, u, { ...(i || {}), signal }, { timeoutMs: RESPONSE_TIMEOUT_MS })
+  for (let attempt = 0; attempt < 2; attempt += 1) {
+    let emitted = false
+    try {
+      const ready = await withProject(blob, preflight, saveBlob)
+      const tiered = antigravityTieredAlias(options.model)
+      const request = googleContents(options)
+      if (tiered) {
+        if (!request.generationConfig) request.generationConfig = {}
+        request.generationConfig.thinkingConfig = /** @type {any} */ ({ thinkingLevel: tiered.thinkingLevel })
+      }
+      const { envelope, sessionId } = streamEnvelope({
+        projectId: ready.projectId,
+        model: tiered ? tiered.model : options.model,
+        request,
+        userAgent: 'antigravity',
+        sessionId: ready.sessionId,
+        paidTierId: ready.paidTierId,
+      })
+      const res = await fetchWithTimeout(impl, `${CODE_ASSIST_STREAM}:streamGenerateContent?alt=sse`, {
+        method: 'POST',
+        headers: {
+          ...headers,
+          ...assistHeaders(ready.accessToken, { ...headersFor(ready.projectId || ''), Accept: 'text/event-stream' }),
+        },
+        body: JSON.stringify(envelope),
+        signal,
+      }, { timeoutMs: RESPONSE_TIMEOUT_MS })
+      if (!res.ok) throwHttpError(res.status, await res.text())
+      if (saveBlob && sessionId && sessionId !== ready.sessionId) {
+        await saveBlob({ ...ready, sessionId })
+      }
+      const pending = []
+      let hasContent = false
+      for await (const event of googleStream(res.body, { signal, idleTimeoutMs: RESPONSE_TIMEOUT_MS })) {
+        if (!hasContent) {
+          if (event.type === 'text-delta' || event.type === 'tool-call-delta' || event.type === 'reasoning-delta') {
+            hasContent = true
+            for (const item of pending) { emitted = true; yield item }
+          } else if (event.type !== 'finish') {
+            pending.push(event)
+            continue
+          } else {
+            continue
+          }
+        }
+        emitted = true
+        yield event
+      }
+      if (!hasContent) {
+        const err = /** @type {any} */ (new Error('Antigravity returned no content'))
+        err.code = 'EMPTY_RESPONSE'
+        throw err
+      }
+      return
+    } catch (err) {
+      if (emitted || signal?.aborted || attempt === 1 || !isTransientFailure(err)) throw err
+    }
   }
-  yield* googleStream(res.body)
 }

package/lib/wire.js

@@ -328,14 +328,18 @@ export async function* anthropicStream(body) {
   yield { type: 'finish', reason: finish }
 }
 
-export async function* googleStream(body) {
+/**
+ * @param {any} body
+ * @param {{ signal?: AbortSignal, idleTimeoutMs?: number }} [options]
+ */
+export async function* googleStream(body, { signal, idleTimeoutMs } = {}) {
   let index = 0
   let started = false
   let callCount = 0
   const functionCallIndexes = new Map()
   let finish = { kind: 'stop' }
   let sawTerminal = false
-  for await (const data of iterateSse(body)) {
+  for await (const data of iterateSse(body, { signal, idleTimeoutMs })) {
     if (data === '[DONE]') {
       sawTerminal = true
       break

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@goodandready/dsh-subscriptions",
-  "version": "0.6.68",
+  "version": "0.6.69",
   "description": "Use ChatGPT Codex, Claude, Grok, Antigravity, Kimi, GLM, Cursor, Kiro, Copilot, Qwen, ERNIE, Spark, JetBrains, Perplexity, Replit, and Cody subscriptions as DeepSeek Harness LLM providers via OAuth.",
   "license": "MIT",
   "type": "module",