Merge pull request #7247 from Qiaogun/codex/seedance-native-api

feat: 支持 Seedance / Ark 原生视频任务 API
Merge pull request #7258 from HuntercodeT/fix/antigravity-gemini-bare-model-thinking-variant
2026-10-17 08:04:00 +08:00 · 2026-09-19 12:11:17 +08:00 · 2026-09-19 12:01:22 +08:00 · 2026-09-19 12:00:18 +08:00 · 2026-09-19 11:59:46 +08:00 · 2026-09-19 11:59:21 +08:00
154 changed files with 8003 additions and 454 deletions
@@ -133,6 +133,7 @@ deploy/docker-compose.override.yml
 .gocache/
 vite.config.js
 docs/*
+!docs/ANTIGRAVITY_ATTRIBUTION_429.md
 !docs/PAYMENT.md
 !docs/PAYMENT_CN.md
 !docs/ADMIN_PAYMENT_INTEGRATION_API.md
@@ -49,8 +49,8 @@ Please read the following carefully before using this project:
 </tr>

 <tr>
-<td width="180"><a href="https://apikey.fun/register?aff=SUB2API"><img src="assets/partners/logos/apikey-fun.png" alt="APIKEY.FUN" width="150"></a></td>
-<td>Thanks to APIKEY.FUN for sponsoring this project! <a href="https://apikey.fun/register?aff=SUB2API">APIKEY.FUN</a> is one of the core contributors to the sub2api open-source project, dedicated to providing open, stable, and cost-effective AI API access. The platform supports API relay services for Claude, OpenAI, Gemini, and other popular models, with pricing starting from as low as 7% of the original rate. Register via the exclusive link: <a href="https://apikey.fun/register?aff=SUB2API">APIKEY</a> to enjoy up to 5% off on all recharges.</td>
+<td width="180"><a href="https://apikey.fan/register?aff=SUB2API"><img src="assets/partners/logos/apikey-fun.png" alt="APIKEY.FUN" width="150"></a></td>
+<td>Thanks to APIKEY.FUN for sponsoring this project! <a href="https://apikey.fan/register?aff=SUB2API">APIKEY.FUN</a> is one of the core contributors to the sub2api open-source project, dedicated to providing open, stable, and cost-effective AI API access. The platform supports API relay services for Claude, OpenAI, Gemini, and other popular models, with pricing starting from as low as 7% of the original rate. Register via the exclusive link: <a href="https://apikey.fan/register?aff=SUB2API">APIKEY</a> to enjoy up to 5% off on all recharges.</td>
 </tr>

 <tr>
@@ -102,14 +102,6 @@ Please read the following carefully before using this project:
 </td>
 </tr>

-<tr>
-<td width="180"><a href="http://www.fastaitoken.com/register"><img src="assets/partners/logos/fastaitoken.jpg" alt="fastaitoken" width="150"></a></td>
-<td>🎉 Thanks to FastAIToken for sponsoring this project! <a href="http://www.fastaitoken.com/register">FastAIToken</a> is an AI API aggregation platform for developers, supporting mainstream large models such as OpenAI, Claude, and Gemini. Top-up at 1:1 — 1 CNY = 1 USD of API credit — letting developers use the world's leading large model services at lower cost and with greater convenience.<br>
-
-🚀 The platform offers a variety of channels to choose from: an ultra-low-price 0.02x OpenAI promotional group (limited time), groups as low as 0.25x OpenAI, 0.7x Claude with 95% fixed cache, and a 1.2x Claude Max channel. It also provides a public status page showing real-time availability, latency, and operating status of each group for transparent and reliable service, plus 7×24 human technical support (not bots) with fast responses to developer needs.
-</td>
-</tr>
-
 <tr>
 <td width="180"><a href="http://aimzoon.com"><img src="assets/partners/logos/aimzoon.jpg" alt="aimzoon" width="150"></a></td>
 <td>Thanks to Aimzoon for sponsoring this project! <a href="http://aimzoon.com">Aimzoon</a> provides stable, cost-effective AI API access services, enabling developers to quickly connect popular AI services to coding tools such as Codex, Claude Code, and Gemini CLI. No complex configuration — faster onboarding, more stable calls, and lower costs. Ongoing promotions including discounted Codex rates and special pricing, with free trial credits upon registration, bringing AI coding into your daily workflow. <a href="http://aimzoon.com">Click here</a> to register and try it out!
@@ -50,8 +50,8 @@
 </tr>

 <tr>
-<td width="180"><a href="https://apikey.fun/register?aff=SUB2API"><img src="assets/partners/logos/apikey-fun.png" alt="APIKEY.FUN" width="150"></a></td>
-<td>感谢 APIKEY.FUN 赞助了本项目！<a href="https://apikey.fun/register?aff=SUB2API">APIKEY.FUN</a> 是 sub2api 开源项目的核心贡献者之一，致力于提供开放、稳定、高性价比的 AI API 接入服务。平台支持 Claude、OpenAI、Gemini 等热门模型的 API 中转服务，价格低至官方原价的 7%。通过专属链接 <a href="https://apikey.fun/register?aff=SUB2API">APIKEY</a> 注册，可享受充值最高 95 折优惠。</td>
+<td width="180"><a href="https://apikey.fan/register?aff=SUB2API"><img src="assets/partners/logos/apikey-fun.png" alt="APIKEY.FUN" width="150"></a></td>
+<td>感谢 APIKEY.FUN 赞助了本项目！<a href="https://apikey.fan/register?aff=SUB2API">APIKEY.FUN</a> 是 sub2api 开源项目的核心贡献者之一，致力于提供开放、稳定、高性价比的 AI API 接入服务。平台支持 Claude、OpenAI、Gemini 等热门模型的 API 中转服务，价格低至官方原价的 7%。通过专属链接 <a href="https://apikey.fan/register?aff=SUB2API">APIKEY</a> 注册，可享受充值最高 95 折优惠。</td>
 </tr>

 <tr>
@@ -105,14 +105,6 @@
 </td>
 </tr>

-<tr>
-<td width="180"><a href="http://www.fastaitoken.com/register"><img src="assets/partners/logos/fastaitoken.jpg" alt="fastaitoken" width="150"></a></td>
-<td>🎉 感谢 FastAIToken 对本项目的赞助！ <a href="http://www.fastaitoken.com/register">FastAIToken</a> 是面向开发者的 AI API 聚合平台，支持 OpenAI、Claude、Gemini 等主流大模型，充值 1:1，1 元 = 1 美元 API 额度，让开发者以更低成本、更便捷地使用全球领先的大模型服务。<br>
-
-🚀 平台提供多种渠道自由选择：超级低价的0.02x OpenAI 福利分组（限时）、低至 0.25x OpenAI 分组、0.7x Claude 95%固定缓存、1.2x Claude Max 渠道；同时提供公开状态页，实时展示各分组的可用率、延迟及运行状态，服务透明可靠，并提供 7×24 小时真人技术支持（非机器人），快速响应开发者需求。
-</td>
-</tr>
-
 <tr>
 <td width="180"><a href="http://aimzoon.com"><img src="assets/partners/logos/aimzoon.jpg" alt="aimzoon" width="150"></a></td>
 <td>感谢 Aimzoon 对本项目的赞助！ <a href="http://aimzoon.com">Aimzoon</a> 提供稳定、高性价比的 AI API 接入服务，支持开发者将常用 AI 服务快速接入 Codex、Claude Code、Gemini CLI 等编程工具。无需复杂配置，更快接入，更稳调用，更省成本。codex倍率优惠，特价倍率等促销不断，注册即送免费体验额度，让 AI 编程真正进入日常工作流。<a href="http://aimzoon.com">点击这里</a>注册体验！
@@ -49,8 +49,8 @@
 </tr>

 <tr>
-<td width="180"><a href="https://apikey.fun/register?aff=SUB2API"><img src="assets/partners/logos/apikey-fun.png" alt="APIKEY.FUN" width="150"></a></td>
-<td>APIKEY.FUN のご支援に感謝します！<a href="https://apikey.fun/register?aff=SUB2API">APIKEY.FUN</a> は sub2api オープンソースプロジェクトのコアコントリビューターの一つであり、オープンで安定した、コストパフォーマンスに優れた AI API アクセスサービスの提供に取り組んでいます。プラットフォームは Claude、OpenAI、Gemini など人気モデルの API 中継サービスをサポートし、価格は公式料金のわずか 7% から。専用リンク <a href="https://apikey.fun/register?aff=SUB2API">APIKEY</a> から登録すると、すべてのチャージで永久 5% 割引をご利用いただけます。</td>
+<td width="180"><a href="https://apikey.fan/register?aff=SUB2API"><img src="assets/partners/logos/apikey-fun.png" alt="APIKEY.FUN" width="150"></a></td>
+<td>APIKEY.FUN のご支援に感謝します！<a href="https://apikey.fan/register?aff=SUB2API">APIKEY.FUN</a> は sub2api オープンソースプロジェクトのコアコントリビューターの一つであり、オープンで安定した、コストパフォーマンスに優れた AI API アクセスサービスの提供に取り組んでいます。プラットフォームは Claude、OpenAI、Gemini など人気モデルの API 中継サービスをサポートし、価格は公式料金のわずか 7% から。専用リンク <a href="https://apikey.fan/register?aff=SUB2API">APIKEY</a> から登録すると、すべてのチャージで永久 5% 割引をご利用いただけます。</td>
 </tr>

 <tr>
@@ -104,14 +104,6 @@
 </td>
 </tr>

-<tr>
-<td width="180"><a href="http://www.fastaitoken.com/register"><img src="assets/partners/logos/fastaitoken.jpg" alt="fastaitoken" width="150"></a></td>
-<td>🎉 FastAIToken のご支援に感謝します！<a href="http://www.fastaitoken.com/register">FastAIToken</a> は開発者向けの AI API アグリゲーションプラットフォームで、OpenAI、Claude、Gemini などの主要な大規模モデルに対応しています。チャージは 1:1（1 元 = 1 米ドル分の API クレジット）で、開発者がより低コスト・より手軽に世界トップクラスの大規模モデルサービスを利用できます。<br>
-
-🚀 プラットフォームでは多彩なチャネルを自由に選択できます：超低価格の 0.02x OpenAI 特典グループ（期間限定）、最低 0.25x の OpenAI グループ、0.7x Claude（95% 固定キャッシュ）、1.2x Claude Max チャネル。さらに、各グループの可用率・レイテンシ・稼働状況をリアルタイムで表示する公開ステータスページを提供し、透明で信頼できるサービスを実現。7×24 時間の有人テクニカルサポート（ボットではありません）により、開発者のニーズに迅速に対応します。
-</td>
-</tr>
-
 <tr>
 <td width="180"><a href="http://aimzoon.com"><img src="assets/partners/logos/aimzoon.jpg" alt="aimzoon" width="150"></a></td>
 <td>Aimzoon のご支援に感謝します！<a href="http://aimzoon.com">Aimzoon</a> は安定してコストパフォーマンスに優れた AI API 接続サービスを提供し、開発者が主要な AI サービスを Codex、Claude Code、Gemini CLI などのコーディングツールへ素早く接続できるようにします。複雑な設定は不要で、より速い接続、より安定した呼び出し、より低いコストを実現。Codex レート割引や特価レートなどのキャンペーンも随時開催中、登録するだけで無料お試しクレジットをプレゼント。AI コーディングを日常のワークフローへ。<a href="http://aimzoon.com">こちら</a>から登録してお試しください！
@@ -1 +1 @@
-0.2.4
+0.2.5
@@ -204,7 +204,9 @@ func initializeApplication(buildInfo handler.BuildInfo) (*Application, error) {
 	accountUsageService := service.ProvideAccountUsageService(accountRepository, usageLogRepository, claudeUsageFetcher, geminiQuotaService, antigravityQuotaFetcher, grokQuotaFetcher, grokQuotaService, openAIQuotaService, usageCache, identityCache, tlsFingerprintProfileService, openAIGatewayService)
 	pluginRepository := repository.NewPluginRepository(db)
 	pluginHostInfo := providePluginHostInfo(buildInfo)
-	pluginManager := service.NewPluginManager(pluginRepository, secretEncryptor, configConfig, pluginHostInfo)
+	pluginKVStore := repository.NewPluginKVStore(redisClient)
+	pluginManager := service.NewPluginManager(pluginRepository, secretEncryptor, configConfig, pluginHostInfo, pluginKVStore)
+	pluginManager.SetAccountDirectory(openAIGatewayService)
 	accountTestService := service.ProvideAccountTestService(accountRepository, geminiTokenProvider, claudeTokenProvider, grokTokenProvider, antigravityGatewayService, httpUpstream, configConfig, tlsFingerprintProfileService, openAIGatewayService, settingService, pluginManager)
 	crsSyncService := service.NewCRSSyncService(accountRepository, proxyRepository, oAuthService, openAIOAuthService, geminiOAuthService, configConfig)
 	accountHandler := admin.ProvideAccountHandler(configConfig, adminService, oAuthService, openAIOAuthService, geminiOAuthService, antigravityOAuthService, grokOAuthService, rateLimitService, accountUsageService, accountTestService, concurrencyService, crsSyncService, sessionLimitCache, rpmCache, compositeTokenCacheInvalidator, grokQuotaService)
@@ -50,13 +50,13 @@ require (
 	github.com/wechatpay-apiv3/wechatpay-go v0.2.21
 	github.com/zeromicro/go-zero v1.9.4
 	go.uber.org/zap v1.24.0
-	golang.org/x/crypto v0.53.0
+	golang.org/x/crypto v0.55.0
 	golang.org/x/image v0.41.0
-	golang.org/x/mod v0.37.0
-	golang.org/x/net v0.56.0
-	golang.org/x/sync v0.21.0
-	golang.org/x/term v0.44.0
-	google.golang.org/grpc v1.82.1
+	golang.org/x/mod v0.38.0
+	golang.org/x/net v0.58.0
+	golang.org/x/sync v0.22.0
+	golang.org/x/term v0.45.0
+	google.golang.org/grpc v1.83.2
 	google.golang.org/protobuf v1.36.11
 	gopkg.in/natefinch/lumberjack.v2 v2.2.1
 	gopkg.in/yaml.v3 v3.0.1
@@ -192,20 +192,20 @@ require (
 	github.com/zclconf/go-cty v1.14.4 // indirect
 	github.com/zclconf/go-cty-yaml v1.1.0 // indirect
 	go.opentelemetry.io/auto/sdk v1.2.1 // indirect
-	go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp v0.49.0 // indirect
-	go.opentelemetry.io/otel v1.43.0 // indirect
-	go.opentelemetry.io/otel/metric v1.43.0 // indirect
-	go.opentelemetry.io/otel/trace v1.43.0 // indirect
+	go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp v0.61.0 // indirect
+	go.opentelemetry.io/otel v1.44.0 // indirect
+	go.opentelemetry.io/otel/metric v1.44.0 // indirect
+	go.opentelemetry.io/otel/trace v1.44.0 // indirect
 	go.uber.org/atomic v1.11.0 // indirect
 	go.uber.org/automaxprocs v1.6.0 // indirect
 	go.uber.org/multierr v1.9.0 // indirect
 	golang.org/x/arch v0.3.0 // indirect
 	golang.org/x/exp v0.0.0-20251023183803-a4bb9ffd2546 // indirect
-	golang.org/x/sys v0.46.0 // indirect
-	golang.org/x/text v0.39.0 // indirect
+	golang.org/x/sys v0.47.0 // indirect
+	golang.org/x/text v0.41.0 // indirect
 	golang.org/x/time v0.12.0 // indirect
-	golang.org/x/tools v0.47.0 // indirect
-	google.golang.org/genproto/googleapis/rpc v0.0.0-20260414002931-afd174a4e478 // indirect
+	golang.org/x/tools v0.48.0 // indirect
+	google.golang.org/genproto/googleapis/rpc v0.0.0-20260526163538-3dc84a4a5aaa // indirect
 	gopkg.in/ini.v1 v1.67.0 // indirect
 	modernc.org/libc v1.67.6 // indirect
 	modernc.org/mathutil v1.7.1 // indirect
@@ -253,8 +253,6 @@ github.com/google/go-tpm-tools v0.3.13-0.20230620182252-4639ecce2aba/go.mod h1:E
 github.com/google/gofuzz v1.0.0/go.mod h1:dBl0BpW6vV/+mYPU4Po3pmUjxk6FQPldtuIdl/M65Eg=
 github.com/google/pprof v0.0.0-20250317173921-a4b03ec1a45e h1:ijClszYn+mADRFY17kjQEVQ1XRhq2/JR1M3sGqeJoxs=
 github.com/google/pprof v0.0.0-20250317173921-a4b03ec1a45e/go.mod h1:boTsfXsheKC2y+lKOCMpSfarhxDeIzfZG1jqGcPl3cA=
-github.com/google/subcommands v1.2.0 h1:vWQspBTo2nEqTUFita5/KeEWlUL8kQObDFbub/EN9oE=
-github.com/google/subcommands v1.2.0/go.mod h1:ZjhPrFU+Olkh9WazFPsl27BQ4UPiG37m3yTrtFlrHVk=
 github.com/google/uuid v1.6.0 h1:NIvaJDMOsjHA8n1jAhLSgzrAzy1Hgr+hNrb57e+94F0=
 github.com/google/uuid v1.6.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
 github.com/google/wire v0.7.0 h1:JxUKI6+CVBgCO2WToKy/nQk0sS+amI9z9EjVmdaocj4=
@@ -300,8 +298,6 @@ github.com/kisielk/sqlstruct v0.0.0-20201105191214-5f3e10d3ab46/go.mod h1:yyMNCy
 github.com/klauspost/compress v1.18.2 h1:iiPHWW0YrcFgpBYhsA6D1+fqHssJscY/Tm/y2Uqnapk=
 github.com/klauspost/compress v1.18.2/go.mod h1:R0h/fSBs8DE4ENlcrlib3PsXS61voFxhIs2DeRhCvJ4=
 github.com/klauspost/cpuid/v2 v2.0.9/go.mod h1:FInQzS24/EEf25PyTYn52gqo7WaD8xa0213Md/qVLRg=
-github.com/klauspost/cpuid/v2 v2.2.4 h1:acbojRNwl3o09bUq+yDCtZFc1aiwaAAxtcn8YkZXnvk=
-github.com/klauspost/cpuid/v2 v2.2.4/go.mod h1:RVVoqg1df56z8g3pUjL/3lE5UfnlrJX8tyFgg4nqhuY=
 github.com/klauspost/cpuid/v2 v2.2.10 h1:tBs3QSyvjDyFTq3uoc/9xFpCuOsJQFNPiAhYdw2skhE=
 github.com/klauspost/cpuid/v2 v2.2.10/go.mod h1:hqwkgyIinND0mEev00jJYCxPNVRVXFQeu1XKlok6oO0=
 github.com/kr/pretty v0.3.1 h1:flRD4NNwYAUpkphVc1HcthR4KEIFJ65n8Mw5qdRn3LE=
@@ -329,8 +325,6 @@ github.com/mattn/go-isatty v0.0.14/go.mod h1:7GGIvUiUoEMVVmxf/4nioHXj79iQHKdU27k
 github.com/mattn/go-isatty v0.0.16/go.mod h1:kYGgaQfpe5nmfYZH+SKPsOc2e4SrIfOl2e/yFXSvRLM=
 github.com/mattn/go-isatty v0.0.20 h1:xfD0iDuEKnDkl03q4limB+vH+GxLEtL/jb4xVJSWWEY=
 github.com/mattn/go-isatty v0.0.20/go.mod h1:W+V8PltTTMOvKvAeJH7IuucS94S2C6jfK/D7dTCTo3Y=
-github.com/mattn/go-runewidth v0.0.15 h1:UNAjwbU9l54TA3KzvqLGxwWjHmMgBUVhBiTjelZgg3U=
-github.com/mattn/go-runewidth v0.0.15/go.mod h1:Jdepj2loyihRzMpdS35Xk/zdY8IAYHsh153qUoGf23w=
 github.com/mattn/go-sqlite3 v1.14.17 h1:mCRHCLDUBXgpKAqIKsaAaAsrAlbkeomtRFKXh2L6YIM=
 github.com/mattn/go-sqlite3 v1.14.17/go.mod h1:2eHXhiwb8IkHr+BDWZGa96P6+rkvnG63S2DGjv9HUNg=
 github.com/mdelapenya/tlscert v0.2.0 h1:7H81W6Z/4weDvZBNOfQte5GpIMo0lGYEeWbkGp5LJHI=
@@ -369,8 +363,6 @@ github.com/ncruces/go-strftime v1.0.0/go.mod h1:Fwc5htZGVVkseilnfgOVb9mKy6w1naJm
 github.com/niemeyer/pretty v0.0.0-20200227124842-a10e7caefd8e/go.mod h1:zD1mROLANZcx1PVRCS0qkT7pwLkGfwJo4zjcN/Tysno=
 github.com/oklog/run v1.1.0 h1:GEenZ1cK0+q0+wsJew9qUg/DyD8k3JzYsZAi5gYi2mA=
 github.com/oklog/run v1.1.0/go.mod h1:sVPdnTZT1zYwAJeCMu2Th4T21pA3FPOQRfWjQlk7DVU=
-github.com/olekukonko/tablewriter v0.0.5 h1:P2Ga83D34wi1o9J6Wh1mRuqd4mF/x/lgBS7N7AbDhec=
-github.com/olekukonko/tablewriter v0.0.5/go.mod h1:hPp6KlRPjbx+hW8ykQs1w3UBbZlj6HuIJcUGPhkA7kY=
 github.com/opencontainers/go-digest v1.0.0 h1:apOUWs51W5PlhuyGyz9FCeeBIOUDA/6nW8Oi/yOhh5U=
 github.com/opencontainers/go-digest v1.0.0/go.mod h1:0JzlMkj0TRzQZfJkVvzbP0HBR3IKzErnv2BNG4W4MAM=
 github.com/opencontainers/image-spec v1.1.1 h1:y0fUlFfIZhPF1W537XOLg0/fcx6zcHCJwooC2xJA040=
@@ -405,8 +397,6 @@ github.com/refraction-networking/utls v1.8.2 h1:j4Q1gJj0xngdeH+Ox/qND11aEfhpgoEv
 github.com/refraction-networking/utls v1.8.2/go.mod h1:jkSOEkLqn+S/jtpEHPOsVv/4V4EVnelwbMQl4vCWXAM=
 github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec h1:W09IVJc94icq4NjY3clb7Lk8O1qJ8BdBEF8z0ibU0rE=
 github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec/go.mod h1:qqbHyh8v60DhA7CoWK5oRCqLrMHRGoxYCSS9EjAz6Eo=
-github.com/rivo/uniseg v0.2.0 h1:S1pD9weZBuJdFmowNwbpi7BJ8TNftyUImj/0WQi72jY=
-github.com/rivo/uniseg v0.2.0/go.mod h1:J6wj4VEh+S6ZtnVlnTBMWIodfgj8LQOQFoIToxlJtxc=
 github.com/robfig/cron/v3 v3.0.1 h1:WdRxkvbJztn8LMz/QEvLN5sBU+xKpSqwwUO1Pjr4qDs=
 github.com/robfig/cron/v3 v3.0.1/go.mod h1:eQICP3HwyT7UooqI/z+Ov+PtYAWygg1TEWWzGIFLtro=
 github.com/rogpeppe/go-internal v1.14.1 h1:UQB4HGPB6osV0SQTLymcB4TgvyWu6ZyliaW0tI/otEQ=
@@ -442,8 +432,6 @@ github.com/spf13/afero v1.11.0 h1:WJQKhtpdm3v2IzqG8VMqrr6Rf3UYpEF239Jy9wNepM8=
 github.com/spf13/afero v1.11.0/go.mod h1:GH9Y3pIexgf1MTIWtNGyogA5MwRIDXGUr+hbWNoBjkY=
 github.com/spf13/cast v1.6.0 h1:GEiTHELF+vaR5dhz3VqZfFSzZjYbgeKDpBxQVS4GYJ0=
 github.com/spf13/cast v1.6.0/go.mod h1:ancEpBxwJDODSW/UG4rDrAqiKolqNNh2DX3mk86cAdo=
-github.com/spf13/cobra v1.7.0 h1:hyqWnYt1ZQShIddO5kBpj3vu05/++x6tJ6dg8EC572I=
-github.com/spf13/cobra v1.7.0/go.mod h1:uLxZILRyS/50WlhOIKD7W6V5bgeIt+4sICxh6uRMrb0=
 github.com/spf13/pflag v1.0.5 h1:iy+VFUOCP1a+8yFto/drg2CJ5u0yRoB7fZw3DKv/JXA=
 github.com/spf13/pflag v1.0.5/go.mod h1:McXfInJRrz4CZXVZOBLb0bTZqETkiAhM9Iw0y3An2Bg=
 github.com/spf13/viper v1.18.2 h1:LUXCnvUvSM6FXAsj6nnfc8Q2tp1dIgUfY9Kc8GsSOiQ=
@@ -521,30 +509,30 @@ github.com/zclconf/go-cty v1.14.4 h1:uXXczd9QDGsgu0i/QFR/hzI5NYCHLf6NQw/atrbnhq8
 github.com/zclconf/go-cty v1.14.4/go.mod h1:VvMs5i0vgZdhYawQNq5kePSpLAoz8u1xvZgrPIxfnZE=
 github.com/zclconf/go-cty-yaml v1.1.0 h1:nP+jp0qPHv2IhUVqmQSzjvqAWcObN0KBkUl2rWBdig0=
 github.com/zclconf/go-cty-yaml v1.1.0/go.mod h1:9YLUH4g7lOhVWqUbctnVlZ5KLpg7JAprQNgxSZ1Gyxs=
+github.com/zeebo/xxh3 v1.1.0 h1:s7DLGDK45Dyfg7++yxI0khrfwq9661w9EN78eP/UZVs=
+github.com/zeebo/xxh3 v1.1.0/go.mod h1:IisAie1LELR4xhVinxWS5+zf1lA4p0MW4T+w+W07F5s=
 github.com/zeromicro/go-zero v1.9.4 h1:aRLFoISqAYijABtkbliQC5SsI5TbizJpQvoHc9xup8k=
 github.com/zeromicro/go-zero v1.9.4/go.mod h1:a17JOTch25SWxBcUgJZYps60hygK3pIYdw7nGwlcS38=
 go.opentelemetry.io/auto/sdk v1.2.1 h1:jXsnJ4Lmnqd11kwkBV2LgLoFMZKizbCi5fNZ/ipaZ64=
 go.opentelemetry.io/auto/sdk v1.2.1/go.mod h1:KRTj+aOaElaLi+wW1kO/DZRXwkF4C5xPbEe3ZiIhN7Y=
-go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp v0.49.0 h1:jq9TW8u3so/bN+JPT166wjOI6/vQPF6Xe7nMNIltagk=
-go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp v0.49.0/go.mod h1:p8pYQP+m5XfbZm9fxtSKAbM6oIllS7s2AfxrChvc7iw=
-go.opentelemetry.io/otel v1.43.0 h1:mYIM03dnh5zfN7HautFE4ieIig9amkNANT+xcVxAj9I=
-go.opentelemetry.io/otel v1.43.0/go.mod h1:JuG+u74mvjvcm8vj8pI5XiHy1zDeoCS2LB1spIq7Ay0=
+go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp v0.61.0 h1:F7Jx+6hwnZ41NSFTO5q4LYDtJRXBf2PD0rNBkeB/lus=
+go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp v0.61.0/go.mod h1:UHB22Z8QsdRDrnAtX4PntOl36ajSxcdUMt1sF7Y6E7Q=
+go.opentelemetry.io/otel v1.44.0 h1:JjwHmHpA4iZ3wBxluu2fbbE7j4kqlE8jXyAyPXH7HqU=
+go.opentelemetry.io/otel v1.44.0/go.mod h1:BMgjTHL9WPRlRjL2oZCBTL4whCGtXch2H4BhOPIAyYc=
 go.opentelemetry.io/otel/exporters/otlp/otlptrace v1.24.0 h1:t6wl9SPayj+c7lEIFgm4ooDBZVb01IhLB4InpomhRw8=
 go.opentelemetry.io/otel/exporters/otlp/otlptrace v1.24.0/go.mod h1:iSDOcsnSA5INXzZtwaBPrKp/lWu/V14Dd+llD0oI2EA=
 go.opentelemetry.io/otel/exporters/otlp/otlptrace/otlptracehttp v1.24.0 h1:Xw8U6u2f8DK2XAkGRFV7BBLENgnTGX9i4rQRxJf+/vs=
 go.opentelemetry.io/otel/exporters/otlp/otlptrace/otlptracehttp v1.24.0/go.mod h1:6KW1Fm6R/s6Z3PGXwSJN2K4eT6wQB3vXX6CVnYX9NmM=
-go.opentelemetry.io/otel/metric v1.43.0 h1:d7638QeInOnuwOONPp4JAOGfbCEpYb+K6DVWvdxGzgM=
-go.opentelemetry.io/otel/metric v1.43.0/go.mod h1:RDnPtIxvqlgO8GRW18W6Z/4P462ldprJtfxHxyKd2PY=
-go.opentelemetry.io/otel/sdk v1.43.0 h1:pi5mE86i5rTeLXqoF/hhiBtUNcrAGHLKQdhg4h4V9Dg=
-go.opentelemetry.io/otel/sdk v1.43.0/go.mod h1:P+IkVU3iWukmiit/Yf9AWvpyRDlUeBaRg6Y+C58QHzg=
-go.opentelemetry.io/otel/sdk/metric v1.43.0 h1:S88dyqXjJkuBNLeMcVPRFXpRw2fuwdvfCGLEo89fDkw=
-go.opentelemetry.io/otel/sdk/metric v1.43.0/go.mod h1:C/RJtwSEJ5hzTiUz5pXF1kILHStzb9zFlIEe85bhj6A=
-go.opentelemetry.io/otel/trace v1.43.0 h1:BkNrHpup+4k4w+ZZ86CZoHHEkohws8AY+WTX09nk+3A=
-go.opentelemetry.io/otel/trace v1.43.0/go.mod h1:/QJhyVBUUswCphDVxq+8mld+AvhXZLhe+8WVFxiFff0=
+go.opentelemetry.io/otel/metric v1.44.0 h1:1w0gILTcHdr3YI+ixLyjemwrVnsMURbTZFrSYCdDdmc=
+go.opentelemetry.io/otel/metric v1.44.0/go.mod h1:8O7hanEPBNgEMmybD3s2VBKcgWOCsA6tzHBPODAiquo=
+go.opentelemetry.io/otel/sdk v1.44.0 h1:nHYwb9lK+fJPU/dnT6s7W7Z8itMWyqrnVfbheVYrZ58=
+go.opentelemetry.io/otel/sdk v1.44.0/go.mod h1:Osuydd3Se74nqjAKxid74N5eC+jfEqfTegHRnq58oK0=
+go.opentelemetry.io/otel/sdk/metric v1.44.0 h1:3LlKgI+VjbVsjNRFZJZAJ30WjXC5VkNRks6si09iEfI=
+go.opentelemetry.io/otel/sdk/metric v1.44.0/go.mod h1:5B5pMARnXxKhltooO4xUuCBorl65a4EpnTalObqOigA=
+go.opentelemetry.io/otel/trace v1.44.0 h1:jxF5CsGYCe74MCRx2X4g7WsY/VBKRqqpNvXlX/6gtIk=
+go.opentelemetry.io/otel/trace v1.44.0/go.mod h1:oLl1jrMQAVo6v3GAggN+1VH9VIz9iUSvW53sW1Q8PIE=
 go.opentelemetry.io/proto/otlp v1.3.1 h1:TrMUixzpM0yuc/znrFTP9MMRh8trP93mkCiDVeXrui0=
 go.opentelemetry.io/proto/otlp v1.3.1/go.mod h1:0X1WI4de4ZsLrrJNLAQbFeLCm3T7yBkR0XqQ7niQU+8=
-go.uber.org/atomic v1.10.0 h1:9qC72Qh0+3MqyJbAn8YU5xVq1frD8bn3JtD2oXtafVQ=
-go.uber.org/atomic v1.10.0/go.mod h1:LUxbIzbOniOlMKjJjyPfpl4v+PKK2cNJn91OQbhoJI0=
 go.uber.org/atomic v1.11.0 h1:ZvwS0R+56ePWxUNi+Atn9dWONBPp/AUETXlHW0DxSjE=
 go.uber.org/atomic v1.11.0/go.mod h1:LUxbIzbOniOlMKjJjyPfpl4v+PKK2cNJn91OQbhoJI0=
 go.uber.org/automaxprocs v1.6.0 h1:O3y2/QNTOdbF+e/dpXNNW7Rx2hZ4sTIPyybbxyNqTUs=
@@ -574,8 +562,8 @@ golang.org/x/crypto v0.19.0/go.mod h1:Iy9bg/ha4yyC70EfRS8jz+B6ybOBKMaSxLj6P6oBDf
 golang.org/x/crypto v0.21.0/go.mod h1:0BP7YvVV9gBbVKyeTG0Gyn+gZm94bibOW5BjDEYAOMs=
 golang.org/x/crypto v0.23.0/go.mod h1:CKFgDieR+mRhux2Lsu27y0fO304Db0wZe70UKqHu0v8=
 golang.org/x/crypto v0.24.0/go.mod h1:Z1PMYSOR5nyMcyAVAIQSKCDwalqy85Aqn1x3Ws4L5DM=
-golang.org/x/crypto v0.53.0 h1:QZ4Muo8THX6CizN2vPPd5fBGHyogrdK9fG4wLPFUsto=
-golang.org/x/crypto v0.53.0/go.mod h1:DNLU434OwVakk9PzuwV8w62mAJpRJL3vsgcfp4Qnsio=
+golang.org/x/crypto v0.55.0 h1:+KWHjbgOaAQ66dh/YlkZKHlz9ZUlq61AFirAR9ntP8M=
+golang.org/x/crypto v0.55.0/go.mod h1:uq0V9dE/fzQuJtbnL+2EhWOE63vo164FY8xqEnV9xis=
 golang.org/x/exp v0.0.0-20190121172915-509febef88a4/go.mod h1:CJ0aWSM057203Lf6IL+f9T1iT9GByDxfZKAQTCR3kQA=
 golang.org/x/exp v0.0.0-20251023183803-a4bb9ffd2546 h1:mgKeJMpvi0yx/sU5GsxQ7p6s2wtOnGAHZWCHUM4KGzY=
 golang.org/x/exp v0.0.0-20251023183803-a4bb9ffd2546/go.mod h1:j/pmGrbnkbPtQfxEe5D0VQhZC6qKbfKifgD0oM7sR70=
@@ -590,8 +578,8 @@ golang.org/x/mod v0.8.0/go.mod h1:iBbtSCu2XBx23ZKBPSOrRkjjQPZFPuis4dIYUhu/chs=
 golang.org/x/mod v0.12.0/go.mod h1:iBbtSCu2XBx23ZKBPSOrRkjjQPZFPuis4dIYUhu/chs=
 golang.org/x/mod v0.15.0/go.mod h1:hTbmBsO62+eylJbnUtE2MGJUyE7QWk4xUqPFrRgJ+7c=
 golang.org/x/mod v0.17.0/go.mod h1:hTbmBsO62+eylJbnUtE2MGJUyE7QWk4xUqPFrRgJ+7c=
-golang.org/x/mod v0.37.0 h1:vF1DjpVEshcIqoEaauuHebaLk1O1forxjxBaVn884JQ=
-golang.org/x/mod v0.37.0/go.mod h1:m8S8VeM9r4dzDwjrKO0a1sZP3YjeMamRRlD+fmR2Q/0=
+golang.org/x/mod v0.38.0 h1:MECBjubtXD7yj4HrhIUcywNaGeNVUdfVnxmPajOk4yk=
+golang.org/x/mod v0.38.0/go.mod h1:V6Xz0pq8TQ3dGqVQ1FVHuelZpAL0uNhSkk9ogYP3c40=
 golang.org/x/net v0.0.0-20180724234803-3673e40ba225/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
 golang.org/x/net v0.0.0-20180826012351-8a410e7b638d/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
 golang.org/x/net v0.0.0-20190213061140-3a22650c66bd/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
@@ -613,8 +601,8 @@ golang.org/x/net v0.21.0/go.mod h1:bIjVDfnllIU7BJ2DNgfnXvpSvtn8VRwhlsaeUTyUS44=
 golang.org/x/net v0.23.0/go.mod h1:JKghWKKOSdJwpW2GEx0Ja7fmaKnMsbu+MWVZTokSYmg=
 golang.org/x/net v0.25.0/go.mod h1:JkAGAh7GEvH74S6FOH42FLoXpXbE/aqXSrIQjXgsiwM=
 golang.org/x/net v0.26.0/go.mod h1:5YKkiSynbBIh3p6iOc/vibscux0x38BZDkn8sCUPxHE=
-golang.org/x/net v0.56.0 h1:Rw8j/hFzGvJUZwNBXnAtf5sVDVt+65SK2C7IxCxZt5o=
-golang.org/x/net v0.56.0/go.mod h1:D3Ku6r+V6JROoZK144D2XfMHFcMq/0zSfLelVTCFKec=
+golang.org/x/net v0.58.0 h1:ynWG7rqYi4ccpTEuPZ2QGWHktVEM9DMCj9yzDE0Q7To=
+golang.org/x/net v0.58.0/go.mod h1:YwCddHnFlT7eLQqVprV19OnhLGtc5xOKgE0RyqgfWAU=
 golang.org/x/oauth2 v0.0.0-20180821212333-d2e6202438be/go.mod h1:N/0e6XlmueqKjAGxoOufVs8QHGRruUQn6yWY3a++T0U=
 golang.org/x/sync v0.0.0-20180314180146-1d60e4601c6f/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
 golang.org/x/sync v0.0.0-20181108010431-42b317875d0f/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
@@ -626,8 +614,8 @@ golang.org/x/sync v0.1.0/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
 golang.org/x/sync v0.3.0/go.mod h1:FU7BRWz2tNW+3quACPkgCx/L+uEAv1htQ0V83Z9Rj+Y=
 golang.org/x/sync v0.6.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
 golang.org/x/sync v0.7.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
-golang.org/x/sync v0.21.0 h1:HLII4xRRTtCRkxYp4HNFF0Js/Og6q2i++KXbg0gHCwM=
-golang.org/x/sync v0.21.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
+golang.org/x/sync v0.22.0 h1:SZjpbeLmrCk4xhRSZFNZW5gFUeCeFgjekvI/+gfScek=
+golang.org/x/sync v0.22.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
 golang.org/x/sys v0.0.0-20180830151530-49385e6e1522/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20190412213103-97732733099d/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
@@ -645,7 +633,6 @@ golang.org/x/sys v0.0.0-20210630005230-0f9fa26af87c/go.mod h1:oPkhp1MJrh7nUepCBc
 golang.org/x/sys v0.0.0-20210927094055-39ccf1dd6fa6/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.0.0-20220503163025-988cb79eb6c6/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.0.0-20220520151302-bc2c85ada10a/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.0.0-20220704084225-05e143d24a9e/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.0.0-20220715151400-c0bba94af5f8/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.0.0-20220722155257-8c9f86f7a55f/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.0.0-20220811171246-fbc7d0a398ab/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
@@ -660,8 +647,8 @@ golang.org/x/sys v0.17.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
 golang.org/x/sys v0.18.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
 golang.org/x/sys v0.20.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
 golang.org/x/sys v0.21.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
-golang.org/x/sys v0.46.0 h1:noSf2Fq6F8DBgS+LysIkx7rIExoNHJsxOAtPp4rthXw=
-golang.org/x/sys v0.46.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
+golang.org/x/sys v0.47.0 h1:o7XGOvZQCADBQQ4Y7VNq2dRWQR7JmOUW8Kxx4ZsNgWs=
+golang.org/x/sys v0.47.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
 golang.org/x/telemetry v0.0.0-20240228155512-f48c80bd79b2/go.mod h1:TeRTkGYfJXctD9OcfyVLyj2J3IxLnKwHJR8f4D8a3YE=
 golang.org/x/term v0.0.0-20201126162022-7de9c90e9dd1/go.mod h1:bj7SfCRtBDWHUb9snDiAeCFNEtKQo2Wmx5Cou7ajbmo=
 golang.org/x/term v0.0.0-20210927222741-03fcf44c2211/go.mod h1:jbD1KX2456YbFQfuXm/mYQcufACuNUgVhRMnK/tPxf8=
@@ -674,8 +661,8 @@ golang.org/x/term v0.17.0/go.mod h1:lLRBjIVuehSbZlaOtGMbcMncT+aqLLLmKrsjNrUguwk=
 golang.org/x/term v0.18.0/go.mod h1:ILwASektA3OnRv7amZ1xhE/KTR+u50pbXfZ03+6Nx58=
 golang.org/x/term v0.20.0/go.mod h1:8UkIAJTvZgivsXaD6/pH6U9ecQzZ45awqEOzuCvwpFY=
 golang.org/x/term v0.21.0/go.mod h1:ooXLefLobQVslOqselCNF4SxFAaoS6KujMbsGzSDmX0=
-golang.org/x/term v0.44.0 h1:0rLvDRCtNj0gZkyIXhCyOb2OAzEhLVqc4B+hrsBhrmc=
-golang.org/x/term v0.44.0/go.mod h1:7ze4MdzUzLXpSAoFP1H0bOI9aXDqveSvatT5vKcFh2Y=
+golang.org/x/term v0.45.0 h1:NwWyBmoJCbfTHpxrWoZ9C6/VxOf7ic219I8xZZFdrf0=
+golang.org/x/term v0.45.0/go.mod h1:9aqxs0blBcrm/n0L9QW0aRVD+ktan8ssZromtqJC43w=
 golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
 golang.org/x/text v0.3.2/go.mod h1:bEr9sfX3Q8Zfm5fL9x+3itogRgK3+ptLWKqgva+5dAk=
 golang.org/x/text v0.3.3/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
@@ -686,8 +673,8 @@ golang.org/x/text v0.13.0/go.mod h1:TvPlkZtksWOMsz7fbANvkp4WM8x/WCo/om8BMLbz+aE=
 golang.org/x/text v0.14.0/go.mod h1:18ZOQIKpY8NJVqYksKHtTdi31H5itFRjB5/qKTNYzSU=
 golang.org/x/text v0.15.0/go.mod h1:18ZOQIKpY8NJVqYksKHtTdi31H5itFRjB5/qKTNYzSU=
 golang.org/x/text v0.16.0/go.mod h1:GhwF1Be+LQoKShO3cGOHzqOgRrGaYc9AvblQOmPVHnI=
-golang.org/x/text v0.39.0 h1:UbZz4pLOvn600D6Oh6GGEI6VAmndrEBLv8/6BEXzyus=
-golang.org/x/text v0.39.0/go.mod h1:3UwRclnC2g0TU9x8PZiyfOajCd1zaUNHF9cvqcQZ+ZM=
+golang.org/x/text v0.41.0 h1:vz/seA0lnX87Othu2f/0L24RcgrXD9/YFTSuGjj3rH8=
+golang.org/x/text v0.41.0/go.mod h1:jvf1O8ajNzZqhSrQBPbutR/EB83Cc0CFrezNQIwbb5M=
 golang.org/x/time v0.12.0 h1:ScB/8o8olJvc+CQPWrK3fPZNfh7qgwCrY0zJmoEQLSE=
 golang.org/x/time v0.12.0/go.mod h1:CDIdPxbZBQxdj6cxyCIdrNogrJKMJ7pr37NYpMcMDSg=
 golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
@@ -702,8 +689,8 @@ golang.org/x/tools v0.1.12/go.mod h1:hNGJHUnrk76NpqgfD5Aqm5Crs+Hm0VOH/i9J2+nxYbc
 golang.org/x/tools v0.6.0/go.mod h1:Xwgl3UAJ/d3gWutnCtw505GrjyAbvKui8lOU390QaIU=
 golang.org/x/tools v0.13.0/go.mod h1:HvlwmtVNQAhOuCjW7xxvovg8wbNq7LwfXh/k7wXUl58=
 golang.org/x/tools v0.21.1-0.20240508182429-e35e4ccd0d2d/go.mod h1:aiJjzUbINMkxbQROHiO6hDPo2LHcIPhhQsa9DLh0yGk=
-golang.org/x/tools v0.47.0 h1:7Kn5x/d1svx/PzryTsqeoZN4TZwqeH5pGWjefhLi/1Q=
-golang.org/x/tools v0.47.0/go.mod h1:dFHnyTvFWY212G+h7ZY4Vsp/K3U4/7W9TyVaAul8uCA=
+golang.org/x/tools v0.48.0 h1:3+hClM1aLL5mjMKm5ovokw9epgRXPuu2tILgismM6RE=
+golang.org/x/tools v0.48.0/go.mod h1:08xX0orndb/F7jJxGDicx061tyd5pcMto75YMAXr6lk=
 golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
@@ -714,16 +701,16 @@ google.golang.org/appengine v1.4.0/go.mod h1:xpcJRLb0r/rnEns0DIKYYv+WjYCduHsrkT7
 google.golang.org/genproto v0.0.0-20180817151627-c66870c02cf8/go.mod h1:JiN7NxoALGmiZfu7CAH4rXhgtRTLTxftemlI0sWmxmc=
 google.golang.org/genproto v0.0.0-20190819201941-24fa4b261c55/go.mod h1:DMBHOl98Agz4BDEuKkezgsaosCRResVns1a3J2ZsMNc=
 google.golang.org/genproto v0.0.0-20231106174013-bbf56f31fb17 h1:wpZ8pe2x1Q3f2KyT5f8oP/fa9rHAKgFPr/HZdNuS+PQ=
-google.golang.org/genproto/googleapis/api v0.0.0-20260414002931-afd174a4e478 h1:yQugLulqltosq0B/f8l4w9VryjV+N/5gcW0jQ3N8Qec=
-google.golang.org/genproto/googleapis/api v0.0.0-20260414002931-afd174a4e478/go.mod h1:C6ADNqOxbgdUUeRTU+LCHDPB9ttAMCTff6auwCVa4uc=
-google.golang.org/genproto/googleapis/rpc v0.0.0-20260414002931-afd174a4e478 h1:RmoJA1ujG+/lRGNfUnOMfhCy5EipVMyvUE+KNbPbTlw=
-google.golang.org/genproto/googleapis/rpc v0.0.0-20260414002931-afd174a4e478/go.mod h1:4Hqkh8ycfw05ld/3BWL7rJOSfebL2Q+DVDeRgYgxUU8=
+google.golang.org/genproto/googleapis/api v0.0.0-20260526163538-3dc84a4a5aaa h1:Kjn0N0tCrDgiAFW+lGO4JZ3ck44CehvJQMAwj9QF0G8=
+google.golang.org/genproto/googleapis/api v0.0.0-20260526163538-3dc84a4a5aaa/go.mod h1:q4lMZS6kskjT5HvCPrnnypcDPVJqT/f4nfxmkE7gryY=
+google.golang.org/genproto/googleapis/rpc v0.0.0-20260526163538-3dc84a4a5aaa h1:mZHHdPZl0dbGHCflZgAq/Q468DWVFcU2whhB2KAo8fk=
+google.golang.org/genproto/googleapis/rpc v0.0.0-20260526163538-3dc84a4a5aaa/go.mod h1:4Hqkh8ycfw05ld/3BWL7rJOSfebL2Q+DVDeRgYgxUU8=
 google.golang.org/grpc v1.19.0/go.mod h1:mqu4LbDTu4XGKhr4mRzUsmM4RtVoemTSY81AxZiDr8c=
 google.golang.org/grpc v1.23.0/go.mod h1:Y5yQAOtifL1yxbo5wqy6BxZv8vAUGQwXBOALyacEbxg=
 google.golang.org/grpc v1.25.1/go.mod h1:c3i+UQWmh7LiEpx4sFZnkU36qjEYZ0imhYfXVyQciAY=
 google.golang.org/grpc v1.31.0/go.mod h1:N36X2cJ7JwdamYAgDz+s+rVMFjt3numwzf/HckM8pak=
-google.golang.org/grpc v1.82.1 h1:NnAxzGRA0677vCa4BUkOAnO5+FfQqVl9iUXeD0IqcGE=
-google.golang.org/grpc v1.82.1/go.mod h1:yzTZ1TB1Z3SG+LIYaI+WiE8D5+PZ3ArnrSp8zF3+/ZA=
+google.golang.org/grpc v1.83.2 h1:EManeRomTObA0BU7I8vXgg/78uE5MJ9M8B39EX2WscU=
+google.golang.org/grpc v1.83.2/go.mod h1:YPI1hK3kDked6iHvgX3tR0y+nX/qpMFKhPgFsokw1S8=
 google.golang.org/protobuf v0.0.0-20200109180630-ec00e32a8dfd/go.mod h1:DFci5gLYBciE7Vtevhsrf46CRTquxDuWsQurQQe4oz8=
 google.golang.org/protobuf v0.0.0-20200221191635-4d8936d0db64/go.mod h1:kwYJMbMJ01Woi6D6+Kah6886xMZcty6N08ah7+eCXa0=
 google.golang.org/protobuf v0.0.0-20200228230310-ab0ca4ff8a60/go.mod h1:cfTl7dwQJ+fmap5saPgwCLgHXTUD7jkjRqWcaiX5VyM=
@@ -183,6 +183,22 @@ func (h *PluginHandler) Test(c *gin.Context) {
 	response.Success(c, result)
 }

+// Status returns the plugin's passive runtime status for the config UI. It is
+// read-only (no config apply, no upstream call) and therefore not step-up gated,
+// so a status UI can poll it without a 2FA prompt or a "test" side effect.
+func (h *PluginHandler) Status(c *gin.Context) {
+	id, ok := pluginIDParam(c)
+	if !ok {
+		return
+	}
+	result, err := h.manager.Status(c.Request.Context(), id)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	response.Success(c, result)
+}
+
 func (h *PluginHandler) CreateUISession(c *gin.Context) {
 	id, ok := pluginIDParam(c)
 	if !ok {
@@ -29,6 +29,7 @@ const (
 	EndpointVideosEdits          = "/v1/videos/edits"
 	EndpointVideosExtensions     = "/v1/videos/extensions"
 	EndpointVideos               = "/v1/videos"
+	EndpointSeedanceTasks        = "/api/v3/contents/generations/tasks"
 	EndpointGeminiModels         = "/v1beta/models"
 )

@@ -81,6 +82,8 @@ const (
 func NormalizeInboundEndpoint(path string) string {
 	path = strings.TrimSpace(path)
 	switch {
+	case strings.Contains(path, "/contents/generations/tasks"):
+		return EndpointSeedanceTasks
 	case strings.Contains(path, EndpointResponsesInputTokens) || isResponsesInputTokensAliasPath(path):
 		return EndpointResponsesInputTokens
 	case strings.Contains(path, EndpointEmbeddings):
@@ -91,7 +91,7 @@ func TestGeminiV1BetaListModels_FiltersFallbackByAllowlist(t *testing.T) {
 	repo := &geminiAllowlistAccountRepoStub{gatewayModelsAccountRepoStub: gatewayModelsAccountRepoStub{
 		byGroup: map[int64][]service.Account{
 			// 只有 antigravity 账号：Gemini 选号失败后回落静态模型列表。
-			41: {{ID: 2, Platform: service.PlatformAntigravity, Status: service.StatusActive, Schedulable: true}},
+			41: {{ID: 2, Platform: service.PlatformAntigravity, Extra: map[string]any{"mixed_scheduling": true}, Status: service.StatusActive, Schedulable: true}},
 		},
 	}}
 	h := &GatewayHandler{
@@ -109,7 +109,7 @@ func TestGeminiV1BetaListModels_FiltersFallbackByAllowlist(t *testing.T) {
 			Platform: service.PlatformGemini,
 			ModelAllowlist: service.GroupModelAllowlist{
 				Enabled: true,
-				Models:  []string{"gemini-2.5-pro", "gemini-3-*"},
+				Models:  []string{"gemini-3.8-flash-high", "gemini-3-*"},
 			},
 		},
 	})
@@ -129,8 +129,8 @@ func TestGeminiV1BetaListModels_FiltersFallbackByAllowlist(t *testing.T) {
 		names = append(names, model.Name)
 	}
 	// models/ 前缀的候选形式也应命中条目。
-	require.Contains(t, names, "models/gemini-2.5-pro")
-	require.Contains(t, names, "models/gemini-3-pro-preview")
+	require.Contains(t, names, "models/gemini-3.8-flash-high")
+	require.Contains(t, names, "models/gemini-3-flash")
 	require.NotContains(t, names, "models/gemini-2.5-flash")
 	require.NotContains(t, names, "models/gemini-2.0-flash")
 }
@@ -0,0 +1,136 @@
+package handler
+
+import (
+	"encoding/json"
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/gemini"
+	"github.com/Wei-Shaw/sub2api/internal/server/middleware"
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/require"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+)
+
+func TestGeminiNativeModelsUsesAccountMappings(t *testing.T) {
+	for _, tt := range []struct {
+		name                     string
+		forced, mixed, allowlist bool
+		status                   int
+	}{
+		{"mixed", false, true, false, 200},
+		{"mixed allowlist", false, true, true, 200},
+		{"disabled mixed", false, false, false, 503},
+		{"forced without mixed opt-in", true, false, false, 200},
+		{"forced allowlist", true, false, true, 200},
+	} {
+		t.Run(tt.name, func(t *testing.T) {
+			groupID := int64(45)
+			repo := &geminiAllowlistAccountRepoStub{gatewayModelsAccountRepoStub: gatewayModelsAccountRepoStub{byGroup: map[int64][]service.Account{
+				groupID: {{ID: 1, Platform: service.PlatformAntigravity,
+					Extra:       map[string]any{"mixed_scheduling": tt.mixed},
+					Credentials: map[string]any{"model_mapping": map[string]any{"gemini-synced-custom": "gemini-3.8-flash-high", "claude-custom": "claude-sonnet-4-6"}}}},
+			}}}
+			h := &GatewayHandler{geminiCompatService: service.NewGeminiMessagesCompatService(repo, nil, nil, nil, nil, nil, nil, nil, nil)}
+			rec := httptest.NewRecorder()
+			c, _ := gin.CreateTestContext(rec)
+			c.Request = httptest.NewRequest(http.MethodGet, "/v1beta/models", nil)
+			c.Set(string(middleware.ContextKeyAPIKey), &service.APIKey{GroupID: &groupID, Group: &service.Group{ID: groupID, Platform: service.PlatformGemini,
+				ModelAllowlist: service.GroupModelAllowlist{Enabled: tt.allowlist, Models: []string{"gemini-synced-custom"}},
+			}})
+			if tt.forced {
+				c.Set(string(middleware.ContextKeyForcePlatform), service.PlatformAntigravity)
+			}
+			h.GeminiV1BetaListModels(c)
+			require.Equal(t, tt.status, rec.Code, rec.Body.String())
+			if tt.status != 200 {
+				return
+			}
+			var got gemini.ModelsListResponse
+			require.NoError(t, json.Unmarshal(rec.Body.Bytes(), &got))
+			names := []string{}
+			for _, model := range got.Models {
+				names = append(names, model.Name)
+				require.Contains(t, model.SupportedGenerationMethods, "generateContent")
+			}
+			require.Contains(t, names, "models/gemini-synced-custom")
+			require.NotContains(t, names, "models/claude-custom")
+			require.NotContains(t, names, "models/gemini-2.0-flash")
+			if tt.allowlist {
+				require.Equal(t, []string{"models/gemini-synced-custom"}, names)
+			}
+		})
+	}
+}
+
+func TestAppendUpstreamGeminiModelsPreservesMetadata(t *testing.T) {
+	body := []byte(`{"models":[{"name":"models/gemini-native","inputTokenLimit":123,"custom":{"a":true}}],"nextPageToken":"next","other":42}`)
+	extra := []gemini.Model{gemini.FallbackModel("gemini-native"), gemini.FallbackModel("gemini-synced-custom"), gemini.FallbackModel("gemini-synced-custom")}
+	merged, ok := appendUpstreamGeminiModels(body, extra)
+	require.True(t, ok)
+	require.JSONEq(t, `{"models":[{"name":"models/gemini-native","inputTokenLimit":123,"custom":{"a":true}},{"name":"models/gemini-synced-custom","supportedGenerationMethods":["generateContent","streamGenerateContent"]}],"nextPageToken":"next","other":42}`, string(merged))
+	filtered, dropped, ok := filterUpstreamGeminiModelsBody(merged, service.GroupModelAllowlist{Enabled: true, Models: []string{"gemini-synced-*"}})
+	require.True(t, ok)
+	require.True(t, dropped)
+	require.NotContains(t, string(filtered), "gemini-native")
+	require.Contains(t, string(filtered), "gemini-synced-custom")
+	for _, invalid := range []string{`null`, `not-json`, `{"error":"bad"}`, `{"models":{}}`} {
+		got, ok := appendUpstreamGeminiModels([]byte(invalid), extra)
+		require.False(t, ok)
+		require.Equal(t, invalid, string(got))
+	}
+}
+
+// Exercise the real handler's native-upstream branch, including scope fallback.
+type geminiMixedModelsUpstream struct {
+	service.HTTPUpstream
+	status int
+	body   string
+}
+
+func (u *geminiMixedModelsUpstream) Do(_ *http.Request, _ string, _ int64, _ int) (*http.Response, error) {
+	return &http.Response{StatusCode: u.status, Header: http.Header{"Content-Type": {"application/json"}, "X-Request-Id": {"native-id"}}, Body: io.NopCloser(strings.NewReader(u.body))}, nil
+}
+func TestGeminiNativeModelsMergesNativeUpstream(t *testing.T) {
+	for _, tt := range []struct {
+		name           string
+		status         int
+		body           string
+		expectedNative string
+	}{
+		{"native", 200, `{"models":[{"name":"models/gemini-native","inputTokenLimit":123}],"nextPageToken":"next"}`, "models/gemini-native"},
+		{"scope fallback", 403, `{"error":"insufficient authentication scopes"}`, "models/gemini-2.5-pro"},
+		{"upstream error", 429, `{"error":"rate limited"}`, ""},
+	} {
+		t.Run(tt.name, func(t *testing.T) {
+			id := int64(46)
+			repo := &geminiAllowlistAccountRepoStub{gatewayModelsAccountRepoStub: gatewayModelsAccountRepoStub{byGroup: map[int64][]service.Account{id: {
+				{ID: 1, Platform: service.PlatformGemini, Type: service.AccountTypeAPIKey, Credentials: map[string]any{"api_key": "test"}},
+				{ID: 2, Platform: service.PlatformAntigravity, Extra: map[string]any{"mixed_scheduling": true}, Credentials: map[string]any{"model_mapping": map[string]any{"gemini-synced-custom": "gemini-3.8-flash-high"}}},
+			}}}}
+			upstream := &geminiMixedModelsUpstream{status: tt.status, body: tt.body}
+			h := &GatewayHandler{geminiCompatService: service.NewGeminiMessagesCompatService(repo, nil, nil, nil, nil, nil, upstream, nil, &config.Config{})}
+			rec := httptest.NewRecorder()
+			c, _ := gin.CreateTestContext(rec)
+			c.Request = httptest.NewRequest(http.MethodGet, "/v1beta/models", nil)
+			c.Set(string(middleware.ContextKeyAPIKey), &service.APIKey{GroupID: &id, Group: &service.Group{ID: id, Platform: service.PlatformGemini}})
+			h.GeminiV1BetaListModels(c)
+			if tt.status == 429 {
+				require.Equal(t, 429, rec.Code)
+				require.JSONEq(t, tt.body, rec.Body.String())
+				return
+			}
+			require.Equal(t, 200, rec.Code, rec.Body.String())
+			require.Contains(t, rec.Body.String(), tt.expectedNative)
+			require.Contains(t, rec.Body.String(), "models/gemini-synced-custom")
+			if tt.status == 200 {
+				require.Contains(t, rec.Body.String(), `"inputTokenLimit":123`)
+				require.Contains(t, rec.Body.String(), `"nextPageToken":"next"`)
+				require.Equal(t, "native-id", rec.Header().Get("X-Request-Id"))
+			}
+		})
+	}
+}
@@ -60,30 +60,24 @@ func (h *GatewayHandler) GeminiV1BetaListModels(c *gin.Context) {
 		return filtered
 	}

-	// 强制 antigravity 模式：返回 antigravity 支持的模型列表
+	agModelIDs, err := h.geminiCompatService.AntigravityGeminiModelIDs(c.Request.Context(), apiKey.GroupID, forcePlatform != service.PlatformAntigravity)
+	if err != nil {
+		googleError(c, http.StatusServiceUnavailable, "Unable to list Antigravity models")
+		return
+	}
+	agModels := make([]gemini.Model, 0, len(agModelIDs))
+	for _, id := range agModelIDs {
+		agModels = append(agModels, gemini.FallbackModel(id))
+	}
 	if forcePlatform == service.PlatformAntigravity {
-		if apiKey.Group != nil && apiKey.Group.ModelAllowlistEnabled() {
-			agModels := antigravity.DefaultGeminiModels()
-			filtered := make([]antigravity.GeminiModel, 0, len(agModels))
-			for _, model := range agModels {
-				if apiKey.Group.ModelAllowlist.Allows(model.Name) {
-					filtered = append(filtered, model)
-				}
-			}
-			c.JSON(http.StatusOK, antigravity.GeminiModelsListResponse{Models: filtered})
-			return
-		}
-		c.JSON(http.StatusOK, antigravity.FallbackGeminiModelsList())
+		c.JSON(http.StatusOK, gemini.ModelsListResponse{Models: filterGeminiModels(agModels)})
 		return
 	}

 	account, err := h.geminiCompatService.SelectAccountForAIStudioEndpoints(c.Request.Context(), apiKey.GroupID)
 	if err != nil {
-		// 没有 gemini 账户，检查是否有 antigravity 账户可用
-		hasAntigravity, _ := h.geminiCompatService.HasAntigravityAccounts(c.Request.Context(), apiKey.GroupID)
-		if hasAntigravity {
-			// antigravity 账户使用静态模型列表
-			c.JSON(http.StatusOK, gemini.ModelsListResponse{Models: filterGeminiModels(gemini.DefaultModels())})
+		if len(agModels) > 0 {
+			c.JSON(http.StatusOK, gemini.ModelsListResponse{Models: filterGeminiModels(agModels)})
 			return
 		}
 		markOpsRoutingCapacityLimitedIfNoAvailable(c, err)
@@ -97,9 +91,15 @@ func (h *GatewayHandler) GeminiV1BetaListModels(c *gin.Context) {
 		return
 	}
 	if shouldFallbackGeminiModels(res) {
-		c.JSON(http.StatusOK, gemini.ModelsListResponse{Models: filterGeminiModels(gemini.DefaultModels())})
+		c.JSON(http.StatusOK, gemini.ModelsListResponse{Models: filterGeminiModels(mergeGeminiModelLists(gemini.DefaultModels(), agModels))})
 		return
 	}
+	if res.StatusCode == http.StatusOK && len(agModels) > 0 {
+		if merged, ok := appendUpstreamGeminiModels(res.Body, agModels); ok {
+			res.Body = merged
+		}
+	}
+
 	if apiKey.Group != nil && apiKey.Group.ModelAllowlistEnabled() {
 		if filtered, dropped, ok := filterUpstreamGeminiModelsBody(res.Body, apiKey.Group.ModelAllowlist); ok && dropped {
 			// 只在确有条目被过滤时替换响应体；全命中或解析失败时保持原始响应，
@@ -110,6 +110,64 @@ func (h *GatewayHandler) GeminiV1BetaListModels(c *gin.Context) {
 	writeUpstreamResponse(c, res)
 }

+// mergeGeminiModelLists keeps native metadata when both sources advertise a model.
+func mergeGeminiModelLists(native, extra []gemini.Model) []gemini.Model {
+	result := append([]gemini.Model{}, native...)
+	seen := make(map[string]bool, len(native))
+	for _, model := range native {
+		seen[model.Name] = true
+	}
+	for _, model := range extra {
+		if !seen[model.Name] {
+			result = append(result, model)
+			seen[model.Name] = true
+		}
+	}
+	return result
+}
+
+// appendUpstreamGeminiModels preserves unknown model metadata and envelope fields.
+func appendUpstreamGeminiModels(body []byte, extra []gemini.Model) ([]byte, bool) {
+	var envelope map[string]json.RawMessage
+	if json.Unmarshal(body, &envelope) != nil || envelope == nil {
+		return body, false
+	}
+	var models []json.RawMessage
+	raw, exists := envelope["models"]
+	if !exists || json.Unmarshal(raw, &models) != nil {
+		return body, false
+	}
+	seen := make(map[string]bool, len(models))
+	for _, raw := range models {
+		var model struct {
+			Name string `json:"name"`
+		}
+		if json.Unmarshal(raw, &model) != nil {
+			return body, false
+		}
+		seen[model.Name] = true
+	}
+	changed := false
+	for _, model := range extra {
+		if seen[model.Name] {
+			continue
+		}
+		raw, err := json.Marshal(model)
+		if err != nil {
+			return body, false
+		}
+		models = append(models, raw)
+		seen[model.Name] = true
+		changed = true
+	}
+	if !changed {
+		return body, true
+	}
+	envelope["models"], _ = json.Marshal(models)
+	merged, err := json.Marshal(envelope)
+	return merged, err == nil
+}
+
 // filterUpstreamGeminiModelsBody 按白名单过滤上游 /v1beta/models 响应中的
 // models[].name，其余信封字段（如 nextPageToken）原样保留。
 // 返回值：filtered 为过滤后的响应体；dropped 表示是否有条目被移除（全命中时
@@ -39,7 +39,11 @@ func TestGeminiV1BetaListModels_ForcedAntigravityAppliesAllowlist(t *testing.T)
 	})
 	c.Set(string(middleware.ContextKeyForcePlatform), service.PlatformAntigravity)

-	(&GatewayHandler{}).GeminiV1BetaListModels(c)
+	groupID := int64(42)
+	key, _ := middleware.GetAPIKeyFromContext(c)
+	key.GroupID = &groupID
+	repo := &geminiAllowlistAccountRepoStub{}
+	(&GatewayHandler{geminiCompatService: service.NewGeminiMessagesCompatService(repo, nil, nil, nil, nil, nil, nil, nil, nil)}).GeminiV1BetaListModels(c)

 	require.Equal(t, http.StatusOK, rec.Code)
 	var got antigravity.GeminiModelsListResponse
@@ -52,6 +52,12 @@ func (h *OpenAIGatewayHandler) GrokVideoContent(c *gin.Context) {
 }

 func (h *OpenAIGatewayHandler) handleGrokMedia(c *gin.Context, endpoint service.GrokMediaEndpoint, requestID string) {
+	platform := service.PlatformGrok
+	noAccountCode, noAccountMessage := "grok_media_no_eligible_account", "No eligible Grok media accounts"
+	if endpoint.IsSeedance() {
+		platform = service.PlatformOpenAI
+		noAccountCode, noAccountMessage = "seedance_no_eligible_account", "No eligible Seedance accounts"
+	}
 	streamStarted := false
 	defer h.recoverResponsesPanic(c, &streamStarted)

@@ -99,6 +105,13 @@ func (h *OpenAIGatewayHandler) handleGrokMedia(c *gin.Context, endpoint service.

 	contentType := c.GetHeader("Content-Type")
 	requestInfo := service.ParseGrokMediaRequest(contentType, body)
+	if endpoint == service.SeedanceEndpointCreate {
+		requestInfo, err = service.ParseSeedanceRequest(body)
+		if err != nil {
+			h.errorResponse(c, http.StatusBadRequest, "invalid_request_error", err.Error())
+			return
+		}
+	}
 	requestModel := requestInfo.Model
 	routingModel := service.NormalizeGrokMediaModelForEndpoint(endpoint, requestModel, requestInfo.HasInputImage())
 	if endpoint.IsGenerationRequest() && strings.TrimSpace(requestModel) == "" {
@@ -215,8 +228,8 @@ func (h *OpenAIGatewayHandler) handleGrokMedia(c *gin.Context, endpoint service.
 		var selection *service.AccountSelectionResult
 		var scheduleDecision service.OpenAIAccountScheduleDecision
 		if boundLookupAccountID > 0 {
-			selection, scheduleDecision, err = h.gatewayService.SelectGrokMediaVideoRequestAccount(
-				requestCtx, apiKey.GroupID, sessionHash, boundLookupAccountID, routingModel,
+			selection, scheduleDecision, err = h.gatewayService.SelectMediaVideoRequestAccount(
+				requestCtx, apiKey.GroupID, sessionHash, boundLookupAccountID, routingModel, platform,
 			)
 		} else {
 			selection, scheduleDecision, err = h.gatewayService.SelectAccountWithSchedulerForCapability(
@@ -231,7 +244,7 @@ func (h *OpenAIGatewayHandler) handleGrokMedia(c *gin.Context, endpoint service.
 				false,
 				false,
 				false,
-				service.PlatformGrok,
+				platform,
 			)
 		}
 		// Own an eagerly acquired slot before any rejection or eligibility probe.
@@ -256,11 +269,11 @@ func (h *OpenAIGatewayHandler) handleGrokMedia(c *gin.Context, endpoint service.
 			if endpoint.IsGenerationRequest() && errors.Is(err, service.ErrNoAvailableAccounts) &&
 				(len(failedAccountIDs) == 0 || (mediaEligibilityRejected && lastFailoverErr == nil)) {
 				markOpsRoutingCapacityLimitedIfNoAvailable(c, err)
-				h.errorResponse(c, http.StatusServiceUnavailable, "grok_media_no_eligible_account", "No eligible Grok media accounts")
+				h.errorResponse(c, http.StatusServiceUnavailable, noAccountCode, noAccountMessage)
 				return
 			}
 			if len(failedAccountIDs) == 0 {
-				cls := classifyNoAccountErrorFromGin(c, h.gatewayService, apiKey, requestModel, routingModel, service.PlatformGrok)
+				cls := classifyNoAccountErrorFromGin(c, h.gatewayService, apiKey, requestModel, routingModel, platform)
 				if !cls.ModelNotFound {
 					markOpsRoutingCapacityLimitedIfNoAvailable(c, err)
 				}
@@ -277,10 +290,10 @@ func (h *OpenAIGatewayHandler) handleGrokMedia(c *gin.Context, endpoint service.
 		if selection == nil || selection.Account == nil {
 			if endpoint.IsGenerationRequest() {
 				markOpsRoutingCapacityLimited(c)
-				h.errorResponse(c, http.StatusServiceUnavailable, "grok_media_no_eligible_account", "No eligible Grok media accounts")
+				h.errorResponse(c, http.StatusServiceUnavailable, noAccountCode, noAccountMessage)
 				return
 			}
-			cls := classifyNoAccountErrorFromGin(c, h.gatewayService, apiKey, requestModel, routingModel, service.PlatformGrok)
+			cls := classifyNoAccountErrorFromGin(c, h.gatewayService, apiKey, requestModel, routingModel, platform)
 			if !cls.ModelNotFound {
 				markOpsRoutingCapacityLimited(c)
 			}
@@ -309,7 +322,7 @@ func (h *OpenAIGatewayHandler) handleGrokMedia(c *gin.Context, endpoint service.
 		)

 		account := selection.Account
-		if endpoint.IsGenerationRequest() {
+		if endpoint.IsGenerationRequest() && !endpoint.IsSeedance() {
 			eligible, eligibilityReason, eligibilityErr := h.ensureGrokMediaAccountEligibility(requestCtx, account)
 			if !eligible {
 				releaseAccount()
@@ -360,6 +373,9 @@ func (h *OpenAIGatewayHandler) handleGrokMedia(c *gin.Context, endpoint service.
 		writerSizeBeforeForward := c.Writer.Size()
 		result, err := func() (*service.OpenAIForwardResult, error) {
 			defer releaseAccount()
+			if endpoint.IsSeedance() {
+				return h.gatewayService.ForwardSeedance(requestCtx, c, account, endpoint, requestID, body)
+			}
 			return h.gatewayService.ForwardGrokMedia(requestCtx, c, account, endpoint, requestID, body, contentType)
 		}()

@@ -490,7 +506,11 @@ func (h *OpenAIGatewayHandler) handleGrokMedia(c *gin.Context, endpoint service.
 		}
 		// Status poll OR content download can observe official done+video.url.
 		// Both paths share the same claim key so the customer is charged once.
-		if endpoint == service.GrokMediaEndpointVideoStatus || endpoint == service.GrokMediaEndpointVideoContent {
+		if endpoint == service.SeedanceEndpointStatus {
+			if billResult := prepareSeedanceCompletionBilling(requestCtx, h, apiKey, subject, requestID, result); billResult != nil {
+				recordGrokMediaUsage(c, h, reqLog, apiKey, subject, subscription, account, billResult, billResult.Model, body, requestID)
+			}
+		} else if endpoint == service.GrokMediaEndpointVideoStatus || endpoint == service.GrokMediaEndpointVideoContent {
 			taskID := strings.TrimSpace(requestID)
 			if billResult := prepareGrokVideoCompletionBilling(requestCtx, h, reqLog, apiKey, subject, taskID, result); billResult != nil {
 				recordGrokMediaUsage(c, h, reqLog, apiKey, subject, subscription, account, billResult, billResult.Model, body, taskID)
@@ -521,6 +541,9 @@ func (h *OpenAIGatewayHandler) ensureGrokMediaAccountEligibility(ctx context.Con
 }

 func grokMediaRequiredCapability(endpoint service.GrokMediaEndpoint) service.OpenAIEndpointCapability {
+	if endpoint.IsSeedance() {
+		return service.OpenAIEndpointCapabilitySeedance
+	}
 	if endpoint.IsGenerationRequest() {
 		return service.OpenAIEndpointCapabilityGrokMediaGeneration
 	}
@@ -539,7 +562,7 @@ func grokMediaScheduleModel(account *service.Account, routingModel string, resul

 func isGrokVideoCreateEndpoint(endpoint service.GrokMediaEndpoint) bool {
 	switch endpoint {
-	case service.GrokMediaEndpointVideosGenerations,
+	case service.SeedanceEndpointCreate, service.GrokMediaEndpointVideosGenerations,
 		service.GrokMediaEndpointVideosEdits,
 		service.GrokMediaEndpointVideosExtensions:
 		return true
@@ -712,7 +735,7 @@ func recordGrokMediaUsage(
 	}
 	// Async video: force durable task request id and release claim if billing fails.
 	videoTaskID := ""
-	if result != nil && result.VideoCount > 0 {
+	if result != nil && (result.VideoCount > 0 || strings.HasPrefix(result.ResponseID, "seedance:")) {
 		videoTaskID = strings.TrimSpace(firstNonEmptyString(requestID, result.ResponseID))
 		if stable := service.StableGrokVideoBillingRequestID(firstNonEmptyString(result.ResponseID, requestID)); stable != "" {
 			result.RequestID = stable
@@ -120,10 +120,22 @@ func (s *grokMediaSlotsCache) assertReleased(t *testing.T) {

 type grokMediaSlotBindings struct {
 	testutil.StubGatewayCache
-	owner  int64
-	writes int
-	key    string
-	billed map[string]bool
+	owner   int64
+	writes  int
+	key     string
+	billed  map[string]bool
+	pending map[string][]byte
+}
+
+func (s *grokMediaSlotBindings) SetGrokVideoPendingBilling(_ context.Context, key string, body []byte, _ time.Duration) error {
+	if s.pending == nil {
+		s.pending = make(map[string][]byte)
+	}
+	s.pending[key] = append([]byte(nil), body...)
+	return nil
+}
+func (s *grokMediaSlotBindings) GetGrokVideoPendingBilling(_ context.Context, key string) ([]byte, error) {
+	return s.pending[key], nil
 }

 func (s *grokMediaSlotBindings) GetSessionAccountID(_ context.Context, groupID int64, key string) (int64, error) {
@@ -186,7 +198,7 @@ func (p grokMediaSlotProber) ProbeMediaEligibility(ctx context.Context, id int64
 	return p(ctx, id)
 }

-func newGrokMediaSlotHandler(t *testing.T, oauth, mismatch bool) (*OpenAIGatewayHandler, *grokMediaSlotsCache, *grokMediaSlotBindings, *grokMediaSlotUpstream) {
+func newGrokMediaSlotHandler(t *testing.T, oauth, mismatch bool, platforms ...string) (*OpenAIGatewayHandler, *grokMediaSlotsCache, *grokMediaSlotBindings, *grokMediaSlotUpstream) {
 	t.Helper()
 	gin.SetMode(gin.TestMode)
 	accounts := make([]service.Account, 3)
@@ -194,6 +206,11 @@ func newGrokMediaSlotHandler(t *testing.T, oauth, mismatch bool) (*OpenAIGateway
 		accounts[i] = service.Account{ID: int64(i + 1), Platform: service.PlatformGrok, Type: service.AccountTypeAPIKey,
 			Status: service.StatusActive, Schedulable: true, Concurrency: 50, Priority: i,
 			GroupIDs: []int64{24}, Credentials: map[string]any{"api_key": "test-key", "access_token": "test-token"}}
+		if len(platforms) > 0 {
+			accounts[i].Platform = platforms[0]
+			accounts[i].Credentials["base_url"] = "https://ark.cn-beijing.volces.com/api/v3"
+			accounts[i].Credentials["openai_capabilities"] = []string{"seedance"}
+		}
 		if oauth {
 			accounts[i].Type = service.AccountTypeOAuth
 			accounts[i].Credentials["refresh_token"] = "test-refresh"
@@ -1,7 +1,10 @@
 package handler

 import (
+	"strconv"
+
 	"github.com/Wei-Shaw/sub2api/internal/handler/dto"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/pagination"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/response"
 	middleware2 "github.com/Wei-Shaw/sub2api/internal/server/middleware"
 	"github.com/Wei-Shaw/sub2api/internal/service"
@@ -68,6 +71,39 @@ func (h *RedeemHandler) GetHistory(c *gin.Context) {
 		return
 	}

+	// Keep the legacy array response for clients that do not request pagination.
+	if c.Request.URL.Query().Has("page") || c.Request.URL.Query().Has("page_size") {
+		page, pageSize := 1, 20
+		for name, target := range map[string]*int{"page": &page, "page_size": &pageSize} {
+			if raw, exists := c.GetQuery(name); exists {
+				value, err := strconv.Atoi(raw)
+				if err != nil || value < 1 {
+					response.BadRequest(c, "Invalid "+name)
+					return
+				}
+				*target = value
+			}
+		}
+		if pageSize > 100 {
+			pageSize = 100
+		}
+		if page-1 > int(^uint(0)>>1)/pageSize {
+			response.BadRequest(c, "Invalid page")
+			return
+		}
+		codes, result, err := h.redeemService.GetUserHistoryPaginated(c.Request.Context(), subject.UserID, pagination.PaginationParams{Page: page, PageSize: pageSize})
+		if err != nil {
+			response.ErrorFrom(c, err)
+			return
+		}
+		out := make([]dto.RedeemCode, 0, len(codes))
+		for i := range codes {
+			out = append(out, *dto.RedeemCodeFromService(&codes[i]))
+		}
+		response.Paginated(c, out, result.Total, page, pageSize)
+		return
+	}
+
 	// Default limit is 25
 	limit := 25

@@ -0,0 +1,89 @@
+package handler
+
+import (
+	"context"
+	"encoding/json"
+	"net/http/httptest"
+	"testing"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/pagination"
+	middleware "github.com/Wei-Shaw/sub2api/internal/server/middleware"
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/require"
+)
+
+type historyRepo struct {
+	service.RedeemCodeRepository
+	userID      int64
+	params      pagination.PaginationParams
+	legacyLimit int
+}
+
+func (r *historyRepo) ListByUser(_ context.Context, userID int64, limit int) ([]service.RedeemCode, error) {
+	r.userID, r.legacyLimit = userID, limit
+	return []service.RedeemCode{}, nil
+}
+func (r *historyRepo) ListByUserPaginated(_ context.Context, userID int64, params pagination.PaginationParams, codeType string) ([]service.RedeemCode, *pagination.PaginationResult, error) {
+	r.userID, r.params = userID, params
+	return []service.RedeemCode{}, &pagination.PaginationResult{Total: 101}, nil
+}
+func TestRedeemHistory(t *testing.T) {
+	for _, tt := range []struct {
+		name, query        string
+		userID             int64
+		status, page, size int
+	}{
+		{"legacy", "", 7, 200, 0, 0},
+		{"default", "?page=1", 7, 200, 1, 20},
+		{"size only", "?page_size=50", 7, 200, 1, 50},
+		{"second user", "?page=2&page_size=100&user_id=7", 8, 200, 2, 100},
+		{"cap", "?page_size=101", 7, 200, 1, 100},
+		{"beyond last", "?page=100&page_size=20", 7, 200, 100, 20},
+		{"zero", "?page=0", 7, 400, 0, 0},
+		{"negative", "?page_size=-1", 7, 400, 0, 0},
+		{"invalid", "?page=abc", 7, 400, 0, 0},
+		{"overflow", "?page=9223372036854775807", 7, 400, 0, 0},
+		{"unauthenticated", "?page=1", 0, 401, 0, 0},
+	} {
+		t.Run(tt.name, func(t *testing.T) {
+			repo := &historyRepo{}
+			h := NewRedeemHandler(service.NewRedeemService(repo, nil, nil, nil, nil, nil, nil, nil))
+			w := httptest.NewRecorder()
+			c, _ := gin.CreateTestContext(w)
+			c.Request = httptest.NewRequest("GET", "/api/v1/redeem/history"+tt.query, nil)
+			if tt.userID != 0 {
+				c.Set(string(middleware.ContextKeyUser), middleware.AuthSubject{UserID: tt.userID})
+			}
+			h.GetHistory(c)
+			require.Equal(t, tt.status, w.Code)
+			if tt.status != 200 {
+				require.Zero(t, repo.userID)
+				return
+			}
+			require.Equal(t, tt.userID, repo.userID)
+			var body struct {
+				Data json.RawMessage `json:"data"`
+			}
+			require.NoError(t, json.Unmarshal(w.Body.Bytes(), &body))
+			if tt.query == "" {
+				require.Equal(t, 25, repo.legacyLimit)
+				require.JSONEq(t, "[]", string(body.Data))
+				return
+			}
+			require.Equal(t, tt.page, repo.params.Page)
+			require.Equal(t, tt.size, repo.params.PageSize)
+			var data struct {
+				Items []service.RedeemCode `json:"items"`
+				Total int                  `json:"total"`
+				Page  int                  `json:"page"`
+				Size  int                  `json:"page_size"`
+			}
+			require.NoError(t, json.Unmarshal(body.Data, &data))
+			require.NotNil(t, data.Items)
+			require.Equal(t, 101, data.Total)
+			require.Equal(t, tt.page, data.Page)
+			require.Equal(t, tt.size, data.Size)
+		})
+	}
+}
@@ -0,0 +1,63 @@
+package handler
+
+import (
+	"context"
+	"mime"
+	"net/http"
+	"time"
+
+	middleware "github.com/Wei-Shaw/sub2api/internal/server/middleware"
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/gin-gonic/gin"
+)
+
+// SeedanceTasks exposes Ark's native asynchronous video task protocol.
+func (h *OpenAIGatewayHandler) SeedanceTasks(c *gin.Context) {
+	if c.Request.Method == http.MethodPost && c.GetHeader("Content-Type") != "" {
+		mediaType, _, err := mime.ParseMediaType(c.GetHeader("Content-Type"))
+		if err != nil || mediaType != "application/json" {
+			h.errorResponse(c, http.StatusUnsupportedMediaType, "invalid_request_error", "Seedance requires application/json")
+			return
+		}
+	}
+	key, ok := middleware.GetAPIKeyFromContext(c)
+	if !ok || key.Group == nil || (key.Group.Platform != service.PlatformOpenAI && key.Group.Platform != service.PlatformComposite) {
+		h.errorResponse(c, http.StatusForbidden, "permission_error", "Seedance requires an OpenAI or composite group")
+		return
+	}
+	endpoint := service.SeedanceEndpointCreate
+	setActualUpstreamEndpoint(c, EndpointSeedanceTasks)
+	taskID := ""
+	if c.Request.Method != http.MethodPost {
+		taskID = service.SeedanceTaskKey(c.Param("task_id"))
+		endpoint = service.SeedanceEndpointStatus
+		if c.Request.Method == http.MethodDelete {
+			endpoint = service.SeedanceEndpointDelete
+		}
+	}
+	h.handleGrokMedia(c, endpoint, taskID)
+}
+
+// Ark reports actual completion tokens. Never infer tokens from duration or use
+// Grok's per-second video tariff. Repeated polls share the durable task dedup key.
+func prepareSeedanceCompletionBilling(ctx context.Context, h *OpenAIGatewayHandler, key *service.APIKey, subject middleware.AuthSubject, taskID string, result *service.OpenAIForwardResult) *service.OpenAIForwardResult {
+	if result == nil || result.Usage.OutputTokens <= 0 {
+		return nil
+	}
+	pending, err := h.gatewayService.LoadGrokVideoPendingBilling(ctx, taskID, subject.UserID, key.ID)
+	if err != nil || pending == nil {
+		return nil
+	}
+	claimed, err := h.gatewayService.ClaimGrokVideoBilling(ctx, taskID, subject.UserID, key.ID)
+	if err != nil || !claimed {
+		return nil
+	}
+	merged := *result
+	merged.Model = pending.Model
+	merged.BillingModel = firstNonEmptyString(pending.BillingModel, pending.Model)
+	merged.UpstreamModel = firstNonEmptyString(pending.UpstreamModel, result.UpstreamModel)
+	merged.RequestID = service.StableGrokVideoBillingRequestID(taskID)
+	merged.ResponseID = taskID
+	merged.Duration = service.GrokVideoE2EDuration(pending.CreatedAt, time.Now())
+	return &merged
+}
@@ -0,0 +1,97 @@
+//go:build unit
+
+package handler
+
+import (
+	"context"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	middleware "github.com/Wei-Shaw/sub2api/internal/server/middleware"
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/require"
+)
+
+func TestSeedanceHandlerLifecycleAndOwnership(t *testing.T) {
+	h, slots, bindings, upstream := newGrokMediaSlotHandler(t, false, false, service.PlatformOpenAI)
+	var owner int64
+	upstream.call = func(req *http.Request, id int64) (*http.Response, error) {
+		body := `{"id":"task-ark","status":"queued"}`
+		if req.Method == http.MethodPost {
+			owner = id
+		} else {
+			require.Equal(t, owner, id)
+		}
+		return &http.Response{StatusCode: 200, Header: http.Header{"Content-Type": []string{"application/json"}}, Body: io.NopCloser(strings.NewReader(body))}, nil
+	}
+	newContext := func(method string) (*gin.Context, *httptest.ResponseRecorder) {
+		c, w := grokMediaSlotContext(context.Background(), method == http.MethodPost)
+		key, _ := middleware.GetAPIKeyFromContext(c)
+		key.Group.Platform = service.PlatformOpenAI
+		body := ""
+		if method == http.MethodPost {
+			body = `{"model":"doubao-seedance","content":[{"type":"text","text":"waves"}]}`
+		}
+		c.Request = httptest.NewRequest(method, "/api/v3/contents/generations/tasks", strings.NewReader(body))
+		c.Params = gin.Params{{Key: "task_id", Value: "task-ark"}}
+		return c, w
+	}
+	c, w := newContext(http.MethodPost)
+	h.SeedanceTasks(c)
+	require.Equal(t, 200, w.Code, w.Body.String())
+	require.Positive(t, owner)
+	require.Len(t, bindings.pending, 1)
+	slots.assertReleased(t)
+	for _, method := range []string{http.MethodGet, http.MethodDelete} {
+		c, w = newContext(method)
+		h.SeedanceTasks(c)
+		require.Equal(t, 200, w.Code, w.Body.String())
+		slots.assertReleased(t)
+	}
+	for _, other := range []string{"user", "key", "group", "task", "provider"} {
+		c, w = newContext(http.MethodGet)
+		key, _ := middleware.GetAPIKeyFromContext(c)
+		switch other {
+		case "user":
+			c.Set(string(middleware.ContextKeyUser), middleware.AuthSubject{UserID: 11, Concurrency: 5})
+		case "key":
+			key.ID = 21
+		case "group":
+			group := int64(25)
+			key.GroupID = &group
+		case "task":
+			c.Params = gin.Params{{Key: "task_id", Value: "other"}}
+		case "provider":
+			c.Params = gin.Params{{Key: "request_id", Value: "task-ark"}}
+		}
+		before := upstream.calls
+		if other == "provider" {
+			h.GrokVideoStatus(c)
+		} else {
+			h.SeedanceTasks(c)
+		}
+		require.Equal(t, 404, w.Code, other+": "+w.Body.String())
+		require.Equal(t, before, upstream.calls)
+		slots.assertReleased(t)
+	}
+	c, _ = newContext(http.MethodGet)
+	key, _ := middleware.GetAPIKeyFromContext(c)
+	subject, _ := middleware.GetAuthSubjectFromContext(c)
+	result := &service.OpenAIForwardResult{Usage: service.OpenAIUsage{OutputTokens: 12345}, ResponseID: "seedance:task-ark"}
+	for i := range 20 {
+		billed := prepareSeedanceCompletionBilling(context.Background(), h, key, subject, result.ResponseID, result)
+		if i == 0 {
+			require.NotNil(t, billed)
+			require.Equal(t, "doubao-seedance", billed.BillingModel)
+			require.Equal(t, 12345, billed.Usage.OutputTokens)
+			require.Zero(t, billed.VideoCount)
+		} else {
+			require.Nil(t, billed)
+		}
+	}
+	require.Len(t, bindings.billed, 1)
+}
@@ -33,10 +33,13 @@ type responsesFailedBody struct {
 }

 // responsesFailedEvent 是写入 SSE data 行的顶层结构。
-// 故意不带 sequence_number：spec 标记可选，且本函数被调用时无法可靠拿到 last seq。
+// SequenceNumber 不带 omitempty：grok-build 等 Rust serde 客户端把它当必填字段，
+// 缺失即 `missing field 'sequence_number'`。拿不到上一帧序号时写 0——这是终止事件，
+// 后面不会再续流。
 type responsesFailedEvent struct {
-	Type     string              `json:"type"`
-	Response responsesFailedBody `json:"response"`
+	Type           string              `json:"type"`
+	SequenceNumber int                 `json:"sequence_number"`
+	Response       responsesFailedBody `json:"response"`
 }

 // writeResponsesFailedSSE emits a `response.failed` SSE event in the OpenAI
@@ -49,8 +52,7 @@ type responsesFailedEvent struct {
 // 而抛出 "stream closed before response.completed"。
 //
 // 字段集对齐 apicompat.makeResponsesCompletedEvent：id/object/model/status/output/error。
-// 故意不写 sequence_number：本函数被调用时无法可靠拿到当前流的 last sequence，
-// 而 OpenAI spec 将 sequence_number 设为可选；省略避免破坏单调性约束。
+// sequence_number 始终写出（未知时为 0）：OpenAI spec 标可选，但 grok-build 当必填。
 //
 // 返回 true 表示已尝试 SSE 写出（不论 Write 是否成功，caller 都应直接 return）。
 // 返回 false 表示 writer 不支持 Flusher，无法以 SSE 形式回报错误；
@@ -48,9 +48,12 @@ func parseResponsesFailedSSE(t *testing.T, body string) (map[string]any, map[str
 	require.NoError(t, json.Unmarshal([]byte(jsonStr), &parsed), "data must be valid JSON: %s", jsonStr)

 	assert.Equal(t, "response.failed", parsed["type"])
-	// 故意不发 sequence_number，避免与后续真实事件的序号冲突。
-	_, hasSeq := parsed["sequence_number"]
-	assert.False(t, hasSeq, "synthetic event must not emit sequence_number")
+	// grok-build 把 sequence_number 当必填；合成终止事件未知上一帧时写 0。
+	rawSeq, hasSeq := parsed["sequence_number"]
+	assert.True(t, hasSeq, "synthetic event must emit sequence_number")
+	seq, ok := rawSeq.(float64)
+	assert.True(t, ok, "sequence_number must be a number, got %T", rawSeq)
+	assert.GreaterOrEqual(t, seq, float64(0))

 	resp, ok := parsed["response"].(map[string]any)
 	require.True(t, ok, "response object missing")
@@ -316,7 +319,20 @@ func TestOpenAIHandleStreamingAwareError_BareResponsesRouteEmitsResponseFailed(t
 	assert.Equal(t, "rate_limit_exceeded", errObj["code"])
 }

-// Synthesized response.failed id falls back to uuid when no request_id is present.
+// issue #7128：grok-build 把顶层 sequence_number 当必填，合成的 response.failed
+// 必须写出该字段，否则整轮反序列化失败。
+func TestOpenAIHandleStreamingAwareError_ResponsesStreamingCarriesSequenceNumber(t *testing.T) {
+	c, w := newGinContextForEndpoint(t, EndpointResponses)
+	h := &OpenAIGatewayHandler{}
+	h.handleStreamingAwareError(c, http.StatusBadGateway, "upstream_error", "boom", true)
+
+	require.True(t, strings.HasPrefix(w.Body.String(), "event: response.failed\n"))
+	_, payload, found := strings.Cut(w.Body.String(), "data: ")
+	require.True(t, found)
+	require.True(t, gjson.Get(payload, "sequence_number").Exists(), "response.failed 必须带 sequence_number")
+	require.GreaterOrEqual(t, gjson.Get(payload, "sequence_number").Int(), int64(0))
+}
+
 // issue #5601：严格的 Responses 客户端把 created_at 当必填字段，缺失即
 // `missing field 'created_at'`。合成的终止事件若解析不了，本文件存在的意义
 // （给客户端一个可识别的终止事件而不是盲重连）就落空了。
@@ -333,6 +349,7 @@ func TestOpenAIHandleStreamingAwareError_ResponsesStreamingCarriesCreatedAt(t *t
 	assert.Greater(t, int64(createdAt), int64(0), "created_at 必须是有效的 unix 时间戳")
 }

+// Synthesized response.failed id falls back to uuid when no request_id is present.
 func TestSynthesizeResponseID_FallbackUUID(t *testing.T) {
 	c, _ := newGinContextForEndpoint(t, EndpointResponses)
 	id := synthesizeResponseID(c)
@@ -0,0 +1,88 @@
+package antigravity
+
+import (
+	"encoding/json"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+func TestTransformClaudeToGemini_AttributionSystemText(t *testing.T) {
+	const attribution = "x-anthropic-billing-header: cc_version=2.1.271.4bf; cc_entrypoint=claude-desktop-3p;"
+	tests := []struct {
+		name string
+		text string
+		want string
+	}{
+		{"metadata only", attribution, ""},
+		{"metadata with newline", attribution + "\n", ""},
+		{"leading whitespace", " \t\n" + attribution, ""},
+		{"following instructions", attribution + "\nKeep these instructions.", "Keep these instructions."},
+		{"CRLF", attribution + "\r\n  Keep indentation.\n", "  Keep indentation.\n"},
+		{"CR", attribution + "\rKeep these instructions.", "Keep these instructions."},
+		{"blank lines preserved", attribution + "\n\nKeep these instructions.", "\nKeep these instructions."},
+		{"ordinary text", "  Keep these instructions.\n", "  Keep these instructions.\n"},
+		{"no colon", "x-anthropic-billing-header keep", "x-anthropic-billing-header keep"},
+		{"different field", "x-anthropic-billing-header-extra: keep", "x-anthropic-billing-header-extra: keep"},
+		{"quoted within instructions", "Explain this metadata: " + attribution, "Explain this metadata: " + attribution},
+		{"later line", "Example:\n" + attribution, "Example:\n" + attribution},
+	}
+	for _, tt := range tests {
+		for _, arrayForm := range []bool{false, true} {
+			name := tt.name + "/string"
+			var system any = tt.text
+			if arrayForm {
+				name = tt.name + "/array"
+				system = []SystemBlock{{Type: "text", Text: tt.text}, {Type: "text", Text: "Keep the next block."}}
+			}
+			t.Run(name, func(t *testing.T) {
+				systemJSON, err := json.Marshal(system)
+				require.NoError(t, err)
+				userJSON, err := json.Marshal(attribution)
+				require.NoError(t, err)
+				input := &ClaudeRequest{
+					Model: "gemini-3.8-flash-high", System: systemJSON,
+					Messages: []ClaudeMessage{{Role: "user", Content: userJSON}},
+					Tools: []ClaudeTool{{Name: "example", Description: attribution,
+						InputSchema: map[string]any{"type": "object", "properties": map[string]any{"query": map[string]any{"type": "string"}}}}},
+				}
+				before, err := json.Marshal(input)
+				require.NoError(t, err)
+				body, err := TransformClaudeToGeminiWithOptions(input, "test-project", input.Model, TransformOptions{})
+				require.NoError(t, err)
+				var got V1InternalRequest
+				require.NoError(t, json.Unmarshal(body, &got))
+				var texts []string
+				for _, part := range got.Request.SystemInstruction.Parts {
+					if part.Text != "\n--- [SYSTEM_PROMPT_END] ---" {
+						texts = append(texts, part.Text)
+					}
+				}
+				var want []string
+				if tt.want != "" {
+					want = append(want, tt.want)
+				}
+				if arrayForm {
+					want = append(want, "Keep the next block.")
+				}
+				require.Equal(t, want, texts)
+				// The same literal text in user content and tool descriptions is not metadata.
+				require.Equal(t, attribution, got.Request.Contents[0].Parts[0].Text)
+				require.Equal(t, attribution, got.Request.Tools[0].FunctionDeclarations[0].Description)
+				after, err := json.Marshal(input)
+				require.NoError(t, err)
+				require.JSONEq(t, string(before), string(after), "do not mutate the incoming request")
+			})
+		}
+	}
+}
+
+func TestTransformClaudeToGemini_AttributionDoesNotHideIdentity(t *testing.T) {
+	system, err := json.Marshal("x-anthropic-billing-header: cc_version=example;\nYou are Antigravity. Keep this identity.")
+	require.NoError(t, err)
+	got := buildSystemInstruction(system, "gemini-3.8-flash-high", DefaultTransformOptions(), nil)
+	require.Len(t, got.Parts, 1)
+	require.Equal(t, "You are Antigravity. Keep this identity.", got.Parts[0].Text)
+	require.False(t, strings.Contains(got.Parts[0].Text, "cc_version="))
+}
@@ -293,6 +293,26 @@ func filterOpenCodePrompt(text string) string {
 	return ""
 }

+// stripClaudeAttribution removes the leading Claude Code attribution metadata line
+// from Antigravity system text. This is prompt metadata, not an HTTP header; it
+// can trigger RESOURCE_EXHAUSTED on the Google upstream. Keep this scoped to the
+// Antigravity transformer: native Anthropic OAuth may require the attribution.
+func stripClaudeAttribution(text string) string {
+	trimmed := strings.TrimLeft(text, " \t\r\n")
+	if !strings.HasPrefix(trimmed, "x-anthropic-billing-header:") {
+		return text
+	}
+	end := strings.IndexAny(trimmed, "\r\n")
+	if end < 0 {
+		return ""
+	}
+	rest := trimmed[end+1:]
+	if trimmed[end] == '\r' {
+		rest = strings.TrimPrefix(rest, "\n")
+	}
+	return rest
+}
+
 // buildSystemInstruction 构建 systemInstruction（与 Antigravity-Manager 保持一致）
 func buildSystemInstruction(system json.RawMessage, modelName string, opts TransformOptions, tools []ClaudeTool) *GeminiContent {
 	var parts []GeminiPart
@@ -305,6 +325,7 @@ func buildSystemInstruction(system json.RawMessage, modelName string, opts Trans
 		// 尝试解析为字符串
 		var sysStr string
 		if err := json.Unmarshal(system, &sysStr); err == nil {
+			sysStr = stripClaudeAttribution(sysStr)
 			if strings.TrimSpace(sysStr) != "" {
 				if strings.Contains(sysStr, "You are Antigravity") {
 					userHasAntigravityIdentity = true
@@ -320,6 +341,7 @@ func buildSystemInstruction(system json.RawMessage, modelName string, opts Trans
 			var sysBlocks []SystemBlock
 			if err := json.Unmarshal(system, &sysBlocks); err == nil {
 				for _, block := range sysBlocks {
+					block.Text = stripClaudeAttribution(block.Text)
 					if block.Type == "text" && strings.TrimSpace(block.Text) != "" {
 						if strings.Contains(block.Text, "You are Antigravity") {
 							userHasAntigravityIdentity = true
@@ -132,6 +132,31 @@ func TestWire_UnknownEventFallsBackToDefault(t *testing.T) {
 	require.Contains(t, m, "response")
 }

+// grok-build 把 sequence_number 当必填。response.created 从 0 起号，
+// omitempty 会把 0 整段丢掉，第一帧就反序列化失败。
+func TestWire_SequenceNumberPresentAtZero(t *testing.T) {
+	created := marshalEvent(t, ResponsesStreamEvent{
+		Type:     "response.created",
+		Response: &ResponsesResponse{ID: "resp_1", Object: "response", Status: "in_progress"},
+	})
+	require.Contains(t, created, "sequence_number")
+	require.EqualValues(t, 0, created["sequence_number"])
+
+	completed := marshalEvent(t, ResponsesStreamEvent{
+		Type:           "response.completed",
+		SequenceNumber: 0,
+		Response:       &ResponsesResponse{ID: "resp_1", Object: "response", Status: "completed"},
+	})
+	require.Contains(t, completed, "sequence_number")
+	require.EqualValues(t, 0, completed["sequence_number"])
+
+	delta := marshalEvent(t, ResponsesStreamEvent{
+		Type: "response.output_text.delta", OutputIndex: 0, ContentIndex: 0, ItemID: "msg_1", Delta: "hi",
+	})
+	require.Contains(t, delta, "sequence_number")
+	require.EqualValues(t, 0, delta["sequence_number"])
+}
+
 func TestResponsesOutputUnmarshal_ToolSearchObjectArguments(t *testing.T) {
 	var item ResponsesOutput
 	require.NoError(t, json.Unmarshal([]byte(`{
@@ -646,6 +646,9 @@ func convertResponsesToAnthropicTools(tools []ResponsesTool) []AnthropicTool {
 }

 // normalizeAnthropicInputSchema ensures input_schema is a valid object schema.
+// Codex 会把部分内置工具（例如 codex_app 的 automation_update）的 parameters
+// 根节点声明成对象分支的 oneOf/anyOf，Anthropic 只接受 object 根节点，这里把
+// 顶层联合摊平成单个 object schema。
 func normalizeAnthropicInputSchema(schema json.RawMessage) json.RawMessage {
 	const emptyObjectSchema = `{"type":"object","properties":{}}`

@@ -659,6 +662,8 @@ func normalizeAnthropicInputSchema(schema json.RawMessage) json.RawMessage {
 		return json.RawMessage(`{"type":"object","properties":{}}`)
 	}

+	flattenAnthropicRootUnions(m)
+
 	typeRaw, ok := m["type"]
 	if !ok || strings.TrimSpace(string(typeRaw)) == "" || string(typeRaw) == "null" {
 		m["type"] = json.RawMessage(`"object"`)
@@ -0,0 +1,213 @@
+package apicompat
+
+import (
+	"bytes"
+	"encoding/json"
+)
+
+// anthropicInputSchemaUnionKeywords 是 Anthropic 在工具 input_schema 顶层拒绝的
+// JSON Schema 联合关键字；嵌套层不受限制，因此只摊平根节点。
+var anthropicInputSchemaUnionKeywords = []string{"oneOf", "anyOf", "allOf"}
+
+// flattenAnthropicRootUnions 用单一 object schema 替换顶层的 oneOf/anyOf/allOf：
+// 属性取各分支并集，同名属性用嵌套联合保留两侧约束；required 在 oneOf/anyOf
+// 下取分支交集、在 allOf 下取分支并集。非对象分支无法用 Anthropic 的 object
+// schema 表达，视为不贡献属性与必填字段。
+func flattenAnthropicRootUnions(schema map[string]json.RawMessage) {
+	present := make([]string, 0, len(anthropicInputSchemaUnionKeywords))
+	for _, keyword := range anthropicInputSchemaUnionKeywords {
+		if _, ok := schema[keyword]; ok {
+			present = append(present, keyword)
+		}
+	}
+	if len(present) == 0 {
+		return
+	}
+
+	properties := anthropicSchemaProperties(schema)
+	rootRequired := anthropicBranchRequired(schema)
+	var (
+		intersections [][]string
+		allOfRequired []string
+	)
+	for _, keyword := range present {
+		raw := schema[keyword]
+		delete(schema, keyword)
+
+		branches, ok := decodeAnthropicUnionBranches(raw)
+		if !ok {
+			continue
+		}
+		mergeAnthropicObjectBranchProperties(properties, branches)
+		if keyword == "allOf" {
+			for _, branch := range branches {
+				allOfRequired = appendAnthropicRequired(allOfRequired, anthropicBranchRequired(branch))
+			}
+			continue
+		}
+		intersections = append(intersections, intersectAnthropicRequired(branches))
+	}
+
+	if len(properties) > 0 {
+		if encoded, err := json.Marshal(properties); err == nil {
+			schema["properties"] = encoded
+		}
+	}
+	required := intersectAnthropicRequiredLists(intersections)
+	required = appendAnthropicRequired(required, allOfRequired)
+	required = appendAnthropicRequired(required, rootRequired)
+	if len(required) > 0 {
+		if encoded, err := json.Marshal(required); err == nil {
+			schema["required"] = encoded
+		}
+	}
+}
+
+func anthropicSchemaProperties(schema map[string]json.RawMessage) map[string]json.RawMessage {
+	properties := make(map[string]json.RawMessage)
+	raw, ok := schema["properties"]
+	if !ok {
+		return properties
+	}
+	if err := json.Unmarshal(raw, &properties); err != nil || properties == nil {
+		return make(map[string]json.RawMessage)
+	}
+	return properties
+}
+
+// decodeAnthropicUnionBranches 解析联合分支并递归摊平分支自身的顶层联合。
+// 非对象 JSON 分支以 nil 占位：它不贡献属性，也不要求任何字段。
+func decodeAnthropicUnionBranches(raw json.RawMessage) ([]map[string]json.RawMessage, bool) {
+	var items []json.RawMessage
+	if err := json.Unmarshal(raw, &items); err != nil || len(items) == 0 {
+		return nil, false
+	}
+	branches := make([]map[string]json.RawMessage, 0, len(items))
+	for _, item := range items {
+		var branch map[string]json.RawMessage
+		if err := json.Unmarshal(item, &branch); err != nil || branch == nil {
+			branches = append(branches, nil)
+			continue
+		}
+		flattenAnthropicRootUnions(branch)
+		branches = append(branches, branch)
+	}
+	return branches, true
+}
+
+func mergeAnthropicObjectBranchProperties(into map[string]json.RawMessage, branches []map[string]json.RawMessage) {
+	for _, branch := range branches {
+		if !anthropicSchemaBranchIsObject(branch) {
+			continue
+		}
+		raw, ok := branch["properties"]
+		if !ok {
+			continue
+		}
+		var properties map[string]json.RawMessage
+		if err := json.Unmarshal(raw, &properties); err != nil {
+			continue
+		}
+		for name, property := range properties {
+			existing, seen := into[name]
+			if !seen {
+				into[name] = property
+				continue
+			}
+			if anthropicSchemaJSONEqual(existing, property) {
+				continue
+			}
+			merged, err := json.Marshal(map[string][]json.RawMessage{"anyOf": {existing, property}})
+			if err != nil {
+				continue
+			}
+			into[name] = merged
+		}
+	}
+}
+
+// anthropicSchemaBranchIsObject 判断分支是否可能是对象：显式 type=object，或省略
+// type 但声明了 properties（缺失 type 等价于不约束，不能据此排除对象分支）。
+func anthropicSchemaBranchIsObject(branch map[string]json.RawMessage) bool {
+	if branch == nil {
+		return false
+	}
+	if raw, ok := branch["type"]; ok {
+		var typ string
+		return json.Unmarshal(raw, &typ) == nil && typ == "object"
+	}
+	_, ok := branch["properties"]
+	return ok
+}
+
+func anthropicBranchRequired(branch map[string]json.RawMessage) []string {
+	if branch == nil {
+		return nil
+	}
+	raw, ok := branch["required"]
+	if !ok {
+		return nil
+	}
+	var required []string
+	if err := json.Unmarshal(raw, &required); err != nil {
+		return nil
+	}
+	return required
+}
+
+// intersectAnthropicRequired 返回所有分支共同要求的字段，顺序沿用第一个分支。
+func intersectAnthropicRequired(branches []map[string]json.RawMessage) []string {
+	lists := make([][]string, 0, len(branches))
+	for _, branch := range branches {
+		lists = append(lists, anthropicBranchRequired(branch))
+	}
+	return intersectAnthropicRequiredLists(lists)
+}
+
+func intersectAnthropicRequiredLists(lists [][]string) []string {
+	if len(lists) == 0 {
+		return nil
+	}
+	result := lists[0]
+	for _, list := range lists[1:] {
+		if len(result) == 0 {
+			return nil
+		}
+		next := make([]string, 0, len(result))
+		for _, name := range result {
+			if containsAnthropicName(list, name) {
+				next = append(next, name)
+			}
+		}
+		result = next
+	}
+	return result
+}
+
+func appendAnthropicRequired(required []string, names []string) []string {
+	merged := make([]string, 0, len(required))
+	merged = append(merged, required...)
+	for _, name := range names {
+		if !containsAnthropicName(merged, name) {
+			merged = append(merged, name)
+		}
+	}
+	return merged
+}
+
+func containsAnthropicName(names []string, target string) bool {
+	for _, name := range names {
+		if name == target {
+			return true
+		}
+	}
+	return false
+}
+
+func anthropicSchemaJSONEqual(left, right json.RawMessage) bool {
+	var leftBuf, rightBuf bytes.Buffer
+	if json.Compact(&leftBuf, left) != nil || json.Compact(&rightBuf, right) != nil {
+		return false
+	}
+	return leftBuf.String() == rightBuf.String()
+}
@@ -138,3 +138,180 @@ func TestResponsesToAnthropic_DefaultToolNormalizesInputSchema(t *testing.T) {
 	assert.Equal(t, "shell", tools[0].Name)
 	assert.JSONEq(t, `{"type":"object","properties":{}}`, string(tools[0].InputSchema))
 }
+
+// Codex 的 codex_app 命名空间工具（如 automation_update）把 parameters 根节点声明为
+// 对象分支的 oneOf/anyOf；Anthropic 拒绝 input_schema 顶层的 oneOf/anyOf/allOf，
+// 转换时必须摊平成单个 object schema。
+func TestResponsesToAnthropic_ObjectUnionRootFlattened(t *testing.T) {
+	tools := convertResponsesToAnthropicTools([]ResponsesTool{{
+		Type: "function",
+		Name: "codex_app__automation_update",
+		Parameters: json.RawMessage(`{
+			"oneOf": [
+				{"type":"object","properties":{"id":{"type":"string"}}},
+				{"anyOf":[{"type":"object"},{"type":"object","properties":{}}]}
+			]
+		}`),
+	}})
+
+	require.Len(t, tools, 1)
+	schema := requireObjectInputSchema(t, tools[0].InputSchema)
+	assert.NotContains(t, schema, "oneOf")
+	assert.NotContains(t, schema, "anyOf")
+	assert.NotContains(t, schema, "allOf")
+	assert.JSONEq(t, `{"id":{"type":"string"}}`, string(schema["properties"]))
+	assert.NotContains(t, schema, "required")
+}
+
+func TestResponsesToAnthropic_ObjectUnionRootMergesBranches(t *testing.T) {
+	tools := convertResponsesToAnthropicTools([]ResponsesTool{{
+		Type: "function",
+		Name: "codex_app__automation_update",
+		Parameters: json.RawMessage(`{
+			"anyOf": [
+				{"type":"object","properties":{"mode":{"enum":["view"]},"id":{"type":"string"}},"required":["mode","id"]},
+				{"type":"object","properties":{"mode":{"enum":["update"]},"prompt":{"type":"string"}},"required":["mode","prompt"]}
+			]
+		}`),
+	}})
+
+	require.Len(t, tools, 1)
+	schema := requireObjectInputSchema(t, tools[0].InputSchema)
+	assert.NotContains(t, schema, "anyOf")
+
+	// 属性取各分支并集；同名属性保留两侧约束（嵌套联合本身是允许的）。
+	var properties map[string]json.RawMessage
+	require.NoError(t, json.Unmarshal(schema["properties"], &properties))
+	require.Contains(t, properties, "id")
+	require.Contains(t, properties, "prompt")
+	var mode map[string]json.RawMessage
+	require.NoError(t, json.Unmarshal(properties["mode"], &mode))
+	assert.Contains(t, mode, "anyOf")
+
+	// required 取各分支交集，只保留所有分支都要求的字段。
+	assert.JSONEq(t, `["mode"]`, string(schema["required"]))
+}
+
+func TestResponsesToAnthropic_AllOfRootKeepsUnionRequired(t *testing.T) {
+	tools := convertResponsesToAnthropicTools([]ResponsesTool{{
+		Type: "function",
+		Name: "strict_tool",
+		Parameters: json.RawMessage(`{
+			"allOf": [
+				{"type":"object","properties":{"path":{"type":"string"}},"required":["path"]},
+				{"type":"object","properties":{"encoding":{"type":"string"}},"required":["encoding"]}
+			]
+		}`),
+	}})
+
+	require.Len(t, tools, 1)
+	schema := requireObjectInputSchema(t, tools[0].InputSchema)
+	assert.NotContains(t, schema, "allOf")
+	assert.JSONEq(t, `["path","encoding"]`, string(schema["required"]))
+}
+
+// 非对象分支无法用 Anthropic 的 object schema 表达，直接丢弃分支但保留属性词汇。
+func TestResponsesToAnthropic_RootUnionDropsNonObjectBranches(t *testing.T) {
+	tools := convertResponsesToAnthropicTools([]ResponsesTool{{
+		Type:       "function",
+		Name:       "mixed_tool",
+		Parameters: json.RawMessage(`{"oneOf":[{"type":"object","properties":{"path":{"type":"string"}}},{"type":"string"}]}`),
+	}})
+
+	require.Len(t, tools, 1)
+	schema := requireObjectInputSchema(t, tools[0].InputSchema)
+	assert.NotContains(t, schema, "oneOf")
+	var properties map[string]json.RawMessage
+	require.NoError(t, json.Unmarshal(schema["properties"], &properties))
+	assert.Contains(t, properties, "path")
+}
+
+func TestResponsesToAnthropic_NestedUnionPreserved(t *testing.T) {
+	parameters := json.RawMessage(`{"type":"object","properties":{"value":{"anyOf":[{"type":"string"},{"type":"array","items":{"type":"string"}}]}}}`)
+	tools := convertResponsesToAnthropicTools([]ResponsesTool{{
+		Type:       "function",
+		Name:       "get_value",
+		Parameters: parameters,
+	}})
+
+	require.Len(t, tools, 1)
+	assert.JSONEq(t, string(parameters), string(tools[0].InputSchema))
+}
+
+// 根节点自带 properties/required 时，与分支约束合并而不是被覆盖。
+func TestResponsesToAnthropic_RootUnionKeepsRootPropertiesAndRequired(t *testing.T) {
+	tools := convertResponsesToAnthropicTools([]ResponsesTool{{
+		Type: "function",
+		Name: "merge_root_tool",
+		Parameters: json.RawMessage(`{
+			"type":"object",
+			"properties":{"shared":{"type":"string"}},
+			"required":["shared"],
+			"oneOf":[{"type":"object","properties":{"extra":{"type":"integer"}}}]
+		}`),
+	}})
+
+	require.Len(t, tools, 1)
+	schema := requireObjectInputSchema(t, tools[0].InputSchema)
+	assert.NotContains(t, schema, "oneOf")
+	var properties map[string]json.RawMessage
+	require.NoError(t, json.Unmarshal(schema["properties"], &properties))
+	assert.Contains(t, properties, "shared")
+	assert.Contains(t, properties, "extra")
+	assert.JSONEq(t, `["shared"]`, string(schema["required"]))
+}
+
+// 端到端：Codex 命名空间工具经降级/摊平后，最终 Anthropic 工具 schema 顶部不再有联合关键字。
+func TestResponsesToAnthropic_CodexNamespaceRootUnionTool(t *testing.T) {
+	body := []byte(`{
+		"model": "claude-opus-5",
+		"input": "create an automation",
+		"tools": [{
+			"type": "namespace",
+			"name": "codex_app",
+			"tools": [{
+				"type": "function",
+				"name": "automation_update",
+				"description": "Create, update, view, or delete recurring automations",
+				"parameters": {
+					"oneOf": [
+						{"type":"object","properties":{"mode":{"enum":["view"]},"id":{"type":"string"}},"required":["mode","id"]},
+						{"type":"object","properties":{"mode":{"enum":["update"]},"id":{"type":"string"}},"required":["mode","id"]}
+					]
+				}
+			}]
+		}]
+	}`)
+
+	var requestBody map[string]any
+	require.NoError(t, json.Unmarshal(body, &requestBody))
+
+	_, changed, err := AdaptResponsesClientTools(requestBody)
+	require.NoError(t, err)
+	require.True(t, changed)
+
+	adapted, err := json.Marshal(requestBody)
+	require.NoError(t, err)
+
+	var responsesReq ResponsesRequest
+	require.NoError(t, json.Unmarshal(adapted, &responsesReq))
+
+	anthropicReq, err := ResponsesToAnthropicRequest(&responsesReq)
+	require.NoError(t, err)
+	require.Len(t, anthropicReq.Tools, 1)
+
+	tool := anthropicReq.Tools[0]
+	assert.Equal(t, "codex_app__automation_update", tool.Name)
+	schema := requireObjectInputSchema(t, tool.InputSchema)
+	assert.NotContains(t, schema, "oneOf")
+	assert.NotContains(t, schema, "anyOf")
+	assert.NotContains(t, schema, "allOf")
+	assert.JSONEq(t, `["mode","id"]`, string(schema["required"]))
+
+	wire, err := json.Marshal(tool)
+	require.NoError(t, err)
+	assert.NotContains(t, string(wire), `"oneOf"`)
+	assert.NotContains(t, string(wire), `"allOf"`)
+	// 同名属性两侧的约束用嵌套联合保留，只允许出现在 properties 内部。
+	assert.Contains(t, string(wire), `"mode":{"anyOf":[{"enum":["view"]},{"enum":["update"]}]}`)
+}
@@ -0,0 +1,167 @@
+package apicompat
+
+import (
+	"encoding/json"
+	"fmt"
+	"strings"
+)
+
+type responsesToolOutputMedia struct {
+	callID   string
+	imageURL string
+}
+
+// LiftResponsesToolOutputMedia moves image parts out of Responses tool outputs
+// and into a following user message. Native Responses endpoints such as
+// DeepSeek accept function_call_output.output as a string, but Codex view_image
+// returns an array containing input_image. Keeping the image inside the tool
+// output makes the upstream report "No tool output found for tool call ...".
+func LiftResponsesToolOutputMedia(input any) (any, bool) {
+	items, ok := input.([]any)
+	if !ok {
+		return input, false
+	}
+
+	rewritten := make([]any, 0, len(items)+1)
+	changed := false
+
+	// DeepSeek validates parallel tool outputs as one contiguous run. Codex can
+	// inject developer notices between image outputs, so keep those notices
+	// pending until the full batch and its lifted media have been emitted.
+	for index := 0; index < len(items); {
+		item, ok := items[index].(map[string]any)
+		if !ok || !isResponsesToolOutputItem(item) {
+			rewritten = append(rewritten, items[index])
+			index++
+			continue
+		}
+
+		batchStart := index
+		outputs := make([]any, 0)
+		trailing := make([]any, 0)
+		pending := make([]responsesToolOutputMedia, 0)
+		batchChanged := false
+
+		for index < len(items) {
+			rawItem := items[index]
+			item, ok := rawItem.(map[string]any)
+			if !ok {
+				break
+			}
+			if isResponsesToolOutputItem(item) {
+				rewrittenItem, media, didRewrite := liftResponsesToolOutputMediaItem(item)
+				if didRewrite {
+					batchChanged = true
+					pending = append(pending, media...)
+				}
+				outputs = append(outputs, rewrittenItem)
+				index++
+				continue
+			}
+			if isResponsesToolBatchInstruction(item) {
+				trailing = append(trailing, rawItem)
+				index++
+				continue
+			}
+			break
+		}
+
+		if !batchChanged {
+			rewritten = append(rewritten, items[batchStart:index]...)
+			continue
+		}
+
+		rewritten = append(rewritten, outputs...)
+		if len(pending) > 0 {
+			rewritten = append(rewritten, buildResponsesToolOutputMediaMessage(pending))
+		}
+		rewritten = append(rewritten, trailing...)
+		changed = true
+	}
+
+	if !changed {
+		return input, false
+	}
+	return rewritten, true
+}
+
+func liftResponsesToolOutputMediaItem(item map[string]any) (any, []responsesToolOutputMedia, bool) {
+	output, exists := item["output"]
+	if !exists {
+		return item, nil, false
+	}
+	outputRaw, err := json.Marshal(output)
+	if err != nil {
+		return item, nil, false
+	}
+
+	outputText, media, didRewrite := extractToolOutputMedia(outputRaw)
+	if !didRewrite {
+		return item, nil, false
+	}
+
+	item["output"] = outputText
+	callID := strings.TrimSpace(stringValue(item["call_id"]))
+	lifted := make([]responsesToolOutputMedia, 0, len(media))
+	for _, part := range media {
+		if part.ImageURL == nil {
+			continue
+		}
+		imageURL := strings.TrimSpace(part.ImageURL.URL)
+		if imageURL == "" {
+			continue
+		}
+		lifted = append(lifted, responsesToolOutputMedia{
+			callID:   callID,
+			imageURL: imageURL,
+		})
+	}
+	return item, lifted, true
+}
+
+func buildResponsesToolOutputMediaMessage(pending []responsesToolOutputMedia) map[string]any {
+	content := make([]map[string]any, 0, len(pending)*2)
+	lastCallID := ""
+	for _, media := range pending {
+		if media.callID != lastCallID {
+			text := "Tool output media"
+			if media.callID != "" {
+				text = fmt.Sprintf(toolOutputMediaAttribution, media.callID)
+			}
+			content = append(content, map[string]any{
+				"type": "input_text",
+				"text": text,
+			})
+			lastCallID = media.callID
+		}
+		content = append(content, map[string]any{
+			"type":      "input_image",
+			"image_url": media.imageURL,
+		})
+	}
+
+	return map[string]any{
+		"type":    "message",
+		"role":    "user",
+		"content": content,
+	}
+}
+
+func isResponsesToolBatchInstruction(item map[string]any) bool {
+	itemType := strings.TrimSpace(stringValue(item["type"]))
+	if itemType != "" && itemType != "message" {
+		return false
+	}
+	role := strings.TrimSpace(stringValue(item["role"]))
+	return role == "developer" || role == "system"
+}
+
+func isResponsesToolOutputItem(item map[string]any) bool {
+	switch strings.TrimSpace(stringValue(item["type"])) {
+	case "function_call_output", "custom_tool_call_output",
+		"tool_search_output", "tool_search_call_output", "mcp_tool_call_output":
+		return true
+	default:
+		return false
+	}
+}
@@ -0,0 +1,116 @@
+package apicompat
+
+import (
+	"encoding/json"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+func TestLiftResponsesToolOutputMedia(t *testing.T) {
+	var input any
+	require.NoError(t, json.Unmarshal([]byte(`[
+		{"type":"function_call","call_id":"call_image","name":"view_image","arguments":"{}"},
+		{"type":"function_call_output","call_id":"call_image","output":[{"type":"input_image","image_url":"data:image/png;base64,AQID"}]}
+	]`), &input))
+
+	lifted, changed := LiftResponsesToolOutputMedia(input)
+	require.True(t, changed)
+
+	items, ok := lifted.([]any)
+	require.True(t, ok)
+	require.Len(t, items, 3)
+
+	outputItem, ok := items[1].(map[string]any)
+	require.True(t, ok)
+	outputText, ok := outputItem["output"].(string)
+	require.True(t, ok)
+	require.Contains(t, outputText, toolOutputMediaMarker)
+	require.NotContains(t, outputText, "data:image/png")
+
+	mediaMessage, ok := items[2].(map[string]any)
+	require.True(t, ok)
+	require.Equal(t, "message", mediaMessage["type"])
+	require.Equal(t, "user", mediaMessage["role"])
+
+	parts, ok := mediaMessage["content"].([]map[string]any)
+	require.True(t, ok)
+	require.Len(t, parts, 2)
+	require.Equal(t, "input_text", parts[0]["type"])
+	require.Equal(t, "[Tool output media for call call_image]", parts[0]["text"])
+	require.Equal(t, "input_image", parts[1]["type"])
+	require.Equal(t, "data:image/png;base64,AQID", parts[1]["image_url"])
+}
+
+func TestLiftResponsesToolOutputMediaLeavesPlainOutputUntouched(t *testing.T) {
+	input := []any{
+		map[string]any{
+			"type":    "function_call_output",
+			"call_id": "call_text",
+			"output":  "plain output",
+		},
+	}
+
+	lifted, changed := LiftResponsesToolOutputMedia(input)
+	require.False(t, changed)
+	require.Equal(t, input, lifted)
+}
+
+func TestLiftResponsesToolOutputMediaKeepsParallelBatchContiguous(t *testing.T) {
+	var input any
+	require.NoError(t, json.Unmarshal([]byte(`[
+		{"type":"function_call","call_id":"call_A","name":"view_image","arguments":"{}"},
+		{"type":"function_call","call_id":"call_B","name":"view_image","arguments":"{}"},
+		{"type":"function_call","call_id":"call_C","name":"view_image","arguments":"{}"},
+		{"type":"function_call_output","call_id":"call_A","output":[{"type":"input_image","image_url":"data:image/png;base64,QQ=="}]},
+		{"type":"message","role":"developer","content":[{"type":"input_text","text":"resized A"}]},
+		{"type":"function_call_output","call_id":"call_B","output":[{"type":"input_image","image_url":"data:image/png;base64,Qg=="}]},
+		{"type":"message","role":"developer","content":[{"type":"input_text","text":"resized B"}]},
+		{"type":"function_call_output","call_id":"call_C","output":[{"type":"input_image","image_url":"data:image/png;base64,Qw=="}]},
+		{"type":"message","role":"developer","content":[{"type":"input_text","text":"resized C"}]}
+	]`), &input))
+
+	lifted, changed := LiftResponsesToolOutputMedia(input)
+	require.True(t, changed)
+
+	items, ok := lifted.([]any)
+	require.True(t, ok)
+	require.Len(t, items, 10)
+
+	for index, callID := range []string{"call_A", "call_B", "call_C"} {
+		item, ok := items[index+3].(map[string]any)
+		require.True(t, ok)
+		require.Equal(t, "function_call_output", item["type"])
+		require.Equal(t, callID, item["call_id"])
+		outputText, ok := item["output"].(string)
+		require.True(t, ok)
+		require.Contains(t, outputText, toolOutputMediaMarker)
+		require.NotContains(t, outputText, "data:image/png")
+	}
+
+	mediaMessage, ok := items[6].(map[string]any)
+	require.True(t, ok)
+	require.Equal(t, "message", mediaMessage["type"])
+	require.Equal(t, "user", mediaMessage["role"])
+	parts, ok := mediaMessage["content"].([]map[string]any)
+	require.True(t, ok)
+	require.Len(t, parts, 6)
+	require.Equal(t, "[Tool output media for call call_A]", parts[0]["text"])
+	require.Equal(t, "data:image/png;base64,QQ==", parts[1]["image_url"])
+	require.Equal(t, "[Tool output media for call call_B]", parts[2]["text"])
+	require.Equal(t, "data:image/png;base64,Qg==", parts[3]["image_url"])
+	require.Equal(t, "[Tool output media for call call_C]", parts[4]["text"])
+	require.Equal(t, "data:image/png;base64,Qw==", parts[5]["image_url"])
+
+	for index, text := range []string{"resized A", "resized B", "resized C"} {
+		notice, ok := items[index+7].(map[string]any)
+		require.True(t, ok)
+		require.Equal(t, "developer", notice["role"])
+		content, ok := notice["content"].([]any)
+		require.True(t, ok)
+		require.Len(t, content, 1)
+		contentPart, ok := content[0].(map[string]any)
+		require.True(t, ok)
+		require.Equal(t, text, contentPart["text"])
+	}
+}
@@ -642,8 +642,11 @@ type ResponsesStreamEvent struct {
 	Code  string `json:"code,omitempty"`
 	Param string `json:"param,omitempty"`

-	// Sequence number for ordering events
-	SequenceNumber int `json:"sequence_number,omitempty"`
+	// SequenceNumber orders streamed events. Strict Responses clients (Grok Build,
+	// Codex CLI) declare it required and abort with `missing field 'sequence_number'`
+	// when it is absent, so it is always emitted — no omitempty. Same rule as
+	// ResponsesResponse.CreatedAt. Zero is a valid first-event value.
+	SequenceNumber int `json:"sequence_number"`
 }

 // ---------------------------------------------------------------------------
@@ -1211,11 +1211,20 @@ func (r *accountRepository) ListOAuthRefreshCandidatePage(ctx context.Context, o
 	// NOT (a AND b) 在 PG 三值逻辑下会把 a 或 b 为 NULL 的行（即绝大多数
 	// 健康账号：temp_unschedulable_until=NULL）也排除，导致后台 token
 	// 刷新工作器漏掉所有正常账号 → access_token 到期后请求开始 401。
+	//
+	// Deliberately NO `schedulable = TRUE` filter here: paused accounts
+	// (schedulable=false, status=active) still hold valid refresh tokens and
+	// their stored access_token must keep working for the admin usage-window
+	// probe. Excluding them lets the token silently expire, after which the
+	// dashboard reports a false "needs re-auth" even though Test Connection
+	// (which refreshes on demand) succeeds. Permanent rejection is already
+	// covered by the status = 'active' filter (error accounts drop out), and
+	// accounts whose refresh actually fails are rate-limited by the
+	// ExcludeRetryCooldown clause below.
 	query := `
 		SELECT id
 		FROM accounts
 		WHERE deleted_at IS NULL
-			AND schedulable = TRUE
 			AND platform = ANY($1)
 			AND id > $2`
 	if options.ActiveOnly {
@@ -327,14 +327,16 @@ func (s *AccountRepoSuite) TestListOAuthRefreshCandidatePage_GrokCursorAndExclus
 			"expires_at":    now.Add(30 * time.Minute).Format(time.RFC3339),
 		},
 	})
-	unschedulable := mustCreateAccount(s.T(), s.client, &service.Account{
-		Name:        "grok-oauth-unschedulable-excluded",
+	// Paused but active OAuth accounts (schedulable=false) must remain refresh
+	// candidates so their stored access_token does not silently expire.
+	paused := mustCreateAccount(s.T(), s.client, &service.Account{
+		Name:        "grok-oauth-paused-included",
 		Platform:    service.PlatformGrok,
 		Type:        service.AccountTypeOAuth,
 		Status:      service.StatusActive,
-		Credentials: map[string]any{"refresh_token": "refresh-unschedulable"},
+		Credentials: map[string]any{"refresh_token": "refresh-paused"},
 	})
-	s.Require().NoError(s.client.Account.UpdateOneID(unschedulable.ID).SetSchedulable(false).Exec(s.ctx))
+	s.Require().NoError(s.client.Account.UpdateOneID(paused.ID).SetSchedulable(false).Exec(s.ctx))
 	mustCreateAccount(s.T(), s.client, &service.Account{
 		Name:     "grok-api-key-excluded",
 		Platform: service.PlatformGrok,
@@ -393,15 +395,14 @@ func (s *AccountRepoSuite) TestListOAuthRefreshCandidatePage_GrokCursorAndExclus
 	s.Require().NoError(err)
 	first := firstPage.Accounts
 	s.Require().Len(first, 2)
-	s.Require().Equal([]int64{valid1.ID, valid2.ID}, []int64{first[0].ID, first[1].ID})
-	s.Require().NotContains([]int64{first[0].ID, first[1].ID}, unschedulable.ID)
+	s.Require().Equal([]int64{valid1.ID, paused.ID}, []int64{first[0].ID, first[1].ID})

 	options.AfterID = first[len(first)-1].ID
 	secondPage, err := s.repo.ListOAuthRefreshCandidatePage(s.ctx, options)
 	s.Require().NoError(err)
 	second := secondPage.Accounts
-	s.Require().Len(second, 1)
-	s.Require().Equal(valid3.ID, second[0].ID)
+	s.Require().Len(second, 2)
+	s.Require().Equal([]int64{valid2.ID, valid3.ID}, []int64{second[0].ID, second[1].ID})
 	s.Require().NotContains([]int64{first[0].ID, first[1].ID}, second[0].ID)
 }

@@ -288,8 +288,8 @@ func TestAccountRepository_ListOAuthRefreshCandidatePage_SQLFilter(t *testing.T)

 	normalized := normalizeSQLWhitespace(capturedSQL)
 	require.Contains(t, normalized, "deleted_at IS NULL")
-	require.Contains(t, normalized, "schedulable = TRUE",
-		"permanently unschedulable accounts must not remain OAuth refresh candidates")
+	require.NotContains(t, normalized, "schedulable = TRUE",
+		"paused (schedulable=false, status=active) accounts must remain OAuth refresh candidates: excluding them lets their stored access_token expire and the usage-window probe then reports a false 'needs re-auth'; permanent rejection is already covered by status = 'active' and refresh failures are bounded by the retry-cooldown exclusion")
 	require.Contains(t, normalized, "status = 'active'")
 	// setup-token 的 access_token 同为 8h 短期令牌，必须与 oauth 一起纳入后台刷新候选
 	require.Contains(t, normalized, "type IN ('oauth', 'setup-token')")
@@ -10,6 +10,62 @@ import (
 	"github.com/Wei-Shaw/sub2api/internal/service"
 )

+// groupUsageRollupSnapshot 是一次汇总水位的快照。
+// closedBefore / retainedFrom 在水位无效时退化为 1970-01-01，
+// 与此前 SQL 里 CASE WHEN valid 的语义一一对应。
+type groupUsageRollupSnapshot struct {
+	valid        bool
+	closedBefore string    // date，历史日桶的右开边界
+	retainedDate string    // retained_from 在服务端时区内的日期，历史日桶的左闭边界
+	tailStart    time.Time // 尾段（还没进日桶的那部分 usage_logs）的起点
+}
+
+// readGroupUsageRollupSnapshot 读一次汇总水位并在 Go 侧判定有效性。
+//
+// 判定条件与原先 SQL 里的 state_values CTE 完全一致：
+// 恰好一行、时区名与当前服务端配置相同、且水位不在未来。
+func (r *usageLogRepository) readGroupUsageRollupSnapshot(ctx context.Context, timezoneName, todayDate string) (groupUsageRollupSnapshot, error) {
+	epoch := time.Unix(0, 0).UTC()
+	invalid := groupUsageRollupSnapshot{
+		closedBefore: "1970-01-01",
+		retainedDate: "1970-01-01",
+		tailStart:    epoch,
+	}
+
+	var rowCount int
+	var closedBefore sql.NullString
+	var retainedFrom sql.NullTime
+	var stateTimezone sql.NullString
+	if err := scanSingleRow(ctx, r.sql, `
+		SELECT
+			COUNT(*),
+			MAX(closed_before)::text,
+			MAX(retained_from),
+			MAX(timezone_name)
+		FROM usage_group_rollup_state
+		WHERE id = 1
+	`, nil, &rowCount, &closedBefore, &retainedFrom, &stateTimezone); err != nil {
+		return groupUsageRollupSnapshot{}, fmt.Errorf("读取分组用量汇总水位: %w", err)
+	}
+	if rowCount != 1 || !closedBefore.Valid || !retainedFrom.Valid ||
+		!stateTimezone.Valid || stateTimezone.String != timezoneName ||
+		closedBefore.String > todayDate {
+		return invalid, nil
+	}
+
+	tailStart, err := service.ParseGroupUsageDate(closedBefore.String)
+	if err != nil {
+		// 水位值解析不出来，按无效处理：宁可多扫一次，也不要算错钱。
+		return invalid, nil //nolint:nilerr // 与上面的 valid 判定同语义，均降级为全量重算
+	}
+	return groupUsageRollupSnapshot{
+		valid:        true,
+		closedBefore: closedBefore.String,
+		retainedDate: service.GroupUsageDate(retainedFrom.Time),
+		tailStart:    tailStart.UTC(),
+	}, nil
+}
+
 func (r *usageLogRepository) getAllGroupUsageSummaryFromRollups(ctx context.Context, todayStart time.Time) (results []usagestats.GroupUsageSummary, err error) {
 	todayStart = service.GroupUsageTodayStart(todayStart)
 	yesterdayStart := service.GroupUsageYesterdayStart(todayStart)
@@ -17,40 +73,34 @@ func (r *usageLogRepository) getAllGroupUsageSummaryFromRollups(ctx context.Cont
 	todayDate := service.GroupUsageDate(todayStart)
 	yesterdayDate := service.GroupUsageDate(yesterdayStart)

+	state, err := r.readGroupUsageRollupSnapshot(ctx, timezoneName, todayDate)
+	if err != nil {
+		return nil, err
+	}
+
+	// 尾段起点必须以**查询参数**的形式给进来。
+	//
+	// 此前它是在同一条 SQL 里由 state CTE 算出、再 CROSS JOIN 给 tail 用的，
+	// 于是 created_at 的下界对 planner 来说是个运行期才知道的值：既进不了
+	// index cond，也估不准选择性，只能退化成 usage_logs 全表扫。
+	// 生产实测（1721 万行 / PostgreSQL 17）：
+	//   Seq Scan on usage_logs … rows=17212870, Rows Removed by Join Filter: 17178082
+	//   Execution Time: 48720 ms
+	// 换成参数之后走 idx_usage_logs_created_at：
+	//   Index Scan … Index Cond: (created_at >= $7)
+	//   Execution Time: 34 ms
 	const query = `
-		WITH state_values AS (
-			SELECT
-				COUNT(*) = 1
-					AND MAX(timezone_name) = $3
-					AND MAX(closed_before) <= $4::date AS valid,
-				MAX(closed_before) AS closed_before,
-				MAX(retained_from) AS retained_from
-			FROM usage_group_rollup_state
-			WHERE id = 1
-		),
-		state AS (
-			SELECT
-				CASE WHEN valid THEN closed_before ELSE DATE '1970-01-01' END AS closed_before,
-				CASE WHEN valid THEN retained_from ELSE TIMESTAMPTZ '1970-01-01 00:00:00+00' END AS retained_from,
-				CASE
-					WHEN valid THEN closed_before::timestamp AT TIME ZONE $3::text
-					ELSE TIMESTAMPTZ '1970-01-01 00:00:00+00'
-				END AS tail_start,
-				valid
-			FROM state_values
-		),
-		historical AS (
+		WITH historical AS (
 			SELECT
 				rollup.group_id,
 				COALESCE(SUM(rollup.actual_cost), 0) AS actual_cost,
 				COALESCE(SUM(rollup.actual_cost) FILTER (
-					WHERE rollup.bucket_date = $5::date
+					WHERE rollup.bucket_date = $3::date
 				), 0) AS yesterday_cost
 			FROM usage_group_daily_rollups rollup
-			CROSS JOIN state
-			WHERE state.valid
-				AND rollup.bucket_date >= (state.retained_from AT TIME ZONE $3::text)::date
-				AND rollup.bucket_date < state.closed_before
+			WHERE $4::boolean
+				AND rollup.bucket_date >= $5::date
+				AND rollup.bucket_date < $6::date
 			GROUP BY rollup.group_id
 		),
 		tail AS (
@@ -63,8 +113,7 @@ func (r *usageLogRepository) getAllGroupUsageSummaryFromRollups(ctx context.Cont
 						AND ul.created_at < $1
 				), 0) AS yesterday_cost
 			FROM usage_logs ul
-			CROSS JOIN state
-			WHERE ul.created_at >= state.tail_start
+			WHERE ul.created_at >= $7
 			GROUP BY ul.group_id
 		)
 		SELECT
@@ -83,9 +132,11 @@ func (r *usageLogRepository) getAllGroupUsageSummaryFromRollups(ctx context.Cont
 		query,
 		todayStart,
 		yesterdayStart,
-		timezoneName,
-		todayDate,
 		yesterdayDate,
+		state.valid,
+		state.retainedDate,
+		state.closedBefore,
+		state.tailStart,
 	)
 	if err != nil {
 		return nil, err
@@ -0,0 +1,99 @@
+package repository
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"strings"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/redis/go-redis/v9"
+)
+
+// pluginKVKeyPrefix 是所有插件通用键值存储在 Redis 中的根前缀。版本段 v1 便于未来
+// 变更内部键结构时平滑迁移。
+const pluginKVKeyPrefix = "plugin:kv:v1:"
+
+// pluginKVScanCount 是 SCAN 每轮的 COUNT 提示，兼顾单轮开销与往返次数。
+const pluginKVScanCount = 200
+
+type pluginKVStore struct {
+	rdb *redis.Client
+}
+
+// NewPluginKVStore 返回 Redis 支撑的通用插件键值存储。使用 Redis 而非进程内存，使得
+// 插件状态可跨副本、跨重启共享，这正是有状态插件所需的通用能力。
+func NewPluginKVStore(rdb *redis.Client) service.PluginKVStore {
+	return &pluginKVStore{rdb: rdb}
+}
+
+// namespacePrefix 拼出某插件某命名空间的内部键前缀。pluginKey / namespace 均已由宿主
+// 服务端按 [A-Za-z0-9._-] 校验，不含 ':' 或 glob 元字符，键结构因此无歧义。
+func (s *pluginKVStore) namespacePrefix(pluginKey, namespace string) string {
+	return pluginKVKeyPrefix + pluginKey + ":" + namespace + ":"
+}
+
+func (s *pluginKVStore) fullKey(pluginKey, namespace, key string) string {
+	return s.namespacePrefix(pluginKey, namespace) + key
+}
+
+func (s *pluginKVStore) Get(ctx context.Context, pluginKey, namespace, key string) ([]byte, bool, error) {
+	value, err := s.rdb.Get(ctx, s.fullKey(pluginKey, namespace, key)).Bytes()
+	if errors.Is(err, redis.Nil) {
+		return nil, false, nil
+	}
+	if err != nil {
+		return nil, false, fmt.Errorf("get plugin kv: %w", err)
+	}
+	return value, true, nil
+}
+
+func (s *pluginKVStore) Set(ctx context.Context, pluginKey, namespace, key string, value []byte, ttl time.Duration) error {
+	// go-redis 将 ttl==0 视为不过期，正好对应“持久保存”语义。
+	if err := s.rdb.Set(ctx, s.fullKey(pluginKey, namespace, key), value, ttl).Err(); err != nil {
+		return fmt.Errorf("set plugin kv: %w", err)
+	}
+	return nil
+}
+
+func (s *pluginKVStore) Delete(ctx context.Context, pluginKey, namespace, key string) error {
+	if err := s.rdb.Del(ctx, s.fullKey(pluginKey, namespace, key)).Err(); err != nil {
+		return fmt.Errorf("delete plugin kv: %w", err)
+	}
+	return nil
+}
+
+func (s *pluginKVStore) List(ctx context.Context, pluginKey, namespace, keyPrefix string, limit int) ([]string, error) {
+	if limit <= 0 {
+		return nil, nil
+	}
+	internalPrefix := s.namespacePrefix(pluginKey, namespace)
+	// keyPrefix 已按 [A-Za-z0-9._-] 校验，不含 glob 元字符，可安全拼接进 MATCH。
+	pattern := internalPrefix + keyPrefix + "*"
+	seen := make(map[string]struct{}, limit)
+	keys := make([]string, 0, limit)
+	var cursor uint64
+	for {
+		batch, next, err := s.rdb.Scan(ctx, cursor, pattern, pluginKVScanCount).Result()
+		if err != nil {
+			return nil, fmt.Errorf("scan plugin kv: %w", err)
+		}
+		for _, full := range batch {
+			// SCAN 可能返回重复键，去重后再裁剪前缀返回面向插件的键名。
+			if _, ok := seen[full]; ok {
+				continue
+			}
+			seen[full] = struct{}{}
+			keys = append(keys, strings.TrimPrefix(full, internalPrefix))
+			if len(keys) >= limit {
+				return keys, nil
+			}
+		}
+		cursor = next
+		if cursor == 0 {
+			break
+		}
+	}
+	return keys, nil
+}
@@ -0,0 +1,148 @@
+package repository
+
+import (
+	"context"
+	"testing"
+	"time"
+
+	"github.com/alicebob/miniredis/v2"
+	"github.com/redis/go-redis/v9"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func newTestPluginKVStore(t *testing.T) (*pluginKVStore, *miniredis.Miniredis) {
+	t.Helper()
+	mr := miniredis.RunT(t)
+	rdb := redis.NewClient(&redis.Options{Addr: mr.Addr()})
+	store, ok := NewPluginKVStore(rdb).(*pluginKVStore)
+	require.True(t, ok)
+	return store, mr
+}
+
+func TestPluginKVStore_SetGetDelete(t *testing.T) {
+	store, _ := newTestPluginKVStore(t)
+	ctx := context.Background()
+
+	value, found, err := store.Get(ctx, "p1", "state", "k")
+	require.NoError(t, err)
+	assert.False(t, found)
+	assert.Nil(t, value)
+
+	require.NoError(t, store.Set(ctx, "p1", "state", "k", []byte("hello"), 0))
+	value, found, err = store.Get(ctx, "p1", "state", "k")
+	require.NoError(t, err)
+	assert.True(t, found)
+	assert.Equal(t, []byte("hello"), value)
+
+	require.NoError(t, store.Delete(ctx, "p1", "state", "k"))
+	_, found, err = store.Get(ctx, "p1", "state", "k")
+	require.NoError(t, err)
+	assert.False(t, found)
+}
+
+func TestPluginKVStore_TTLExpires(t *testing.T) {
+	store, mr := newTestPluginKVStore(t)
+	ctx := context.Background()
+
+	require.NoError(t, store.Set(ctx, "p1", "state", "k", []byte("v"), 30*time.Second))
+	full := store.fullKey("p1", "state", "k")
+	ttl := mr.TTL(full)
+	assert.Equal(t, 30*time.Second, ttl)
+
+	mr.FastForward(31 * time.Second)
+	_, found, err := store.Get(ctx, "p1", "state", "k")
+	require.NoError(t, err)
+	assert.False(t, found)
+}
+
+func TestPluginKVStore_PluginIsolation(t *testing.T) {
+	store, _ := newTestPluginKVStore(t)
+	ctx := context.Background()
+
+	require.NoError(t, store.Set(ctx, "p1", "state", "shared", []byte("one"), 0))
+	require.NoError(t, store.Set(ctx, "p2", "state", "shared", []byte("two"), 0))
+
+	v1, found, err := store.Get(ctx, "p1", "state", "shared")
+	require.NoError(t, err)
+	require.True(t, found)
+	assert.Equal(t, []byte("one"), v1)
+
+	v2, found, err := store.Get(ctx, "p2", "state", "shared")
+	require.NoError(t, err)
+	require.True(t, found)
+	assert.Equal(t, []byte("two"), v2)
+}
+
+func TestPluginKVStore_ListStripsPrefixAndHonorsLimit(t *testing.T) {
+	store, _ := newTestPluginKVStore(t)
+	ctx := context.Background()
+	for _, key := range []string{"account-1", "account-2", "account-3", "other"} {
+		require.NoError(t, store.Set(ctx, "p1", "state", key, []byte("x"), 0))
+	}
+	// 另一命名空间/插件的键不应混入结果。
+	require.NoError(t, store.Set(ctx, "p1", "other-ns", "account-9", []byte("x"), 0))
+	require.NoError(t, store.Set(ctx, "p2", "state", "account-9", []byte("x"), 0))
+
+	keys, err := store.List(ctx, "p1", "state", "account-", 100)
+	require.NoError(t, err)
+	assert.ElementsMatch(t, []string{"account-1", "account-2", "account-3"}, keys)
+
+	keys, err = store.List(ctx, "p1", "state", "", 100)
+	require.NoError(t, err)
+	assert.ElementsMatch(t, []string{"account-1", "account-2", "account-3", "other"}, keys)
+
+	keys, err = store.List(ctx, "p1", "state", "account-", 2)
+	require.NoError(t, err)
+	assert.Len(t, keys, 2)
+
+	keys, err = store.List(ctx, "p1", "state", "account-", 0)
+	require.NoError(t, err)
+	assert.Empty(t, keys)
+}
+
+// namespace 是另一 namespace 的前缀时（state vs state2），List 不能串味——内部键的 ':'
+// 分隔符 + namespace 字符集排除 ':' 共同保证消歧。
+func TestPluginKVStore_ListNamespacePrefixDisambiguation(t *testing.T) {
+	store, _ := newTestPluginKVStore(t)
+	ctx := context.Background()
+	require.NoError(t, store.Set(ctx, "p1", "state", "k1", []byte("a"), 0))
+	require.NoError(t, store.Set(ctx, "p1", "state2", "k2", []byte("b"), 0))
+
+	keys, err := store.List(ctx, "p1", "state", "", 100)
+	require.NoError(t, err)
+	assert.Equal(t, []string{"k1"}, keys)
+
+	keys, err = store.List(ctx, "p1", "state2", "", 100)
+	require.NoError(t, err)
+	assert.Equal(t, []string{"k2"}, keys)
+}
+
+func TestPluginKVStore_OverwriteResetsValueAndTTL(t *testing.T) {
+	store, mr := newTestPluginKVStore(t)
+	ctx := context.Background()
+	require.NoError(t, store.Set(ctx, "p1", "state", "k", []byte("v1"), 10*time.Second))
+	require.NoError(t, store.Set(ctx, "p1", "state", "k", []byte("v2"), 0))
+
+	value, found, err := store.Get(ctx, "p1", "state", "k")
+	require.NoError(t, err)
+	require.True(t, found)
+	assert.Equal(t, []byte("v2"), value)
+	// ttl=0 覆盖后应变为不过期。
+	assert.Equal(t, time.Duration(0), mr.TTL(store.fullKey("p1", "state", "k")))
+}
+
+func TestPluginKVStore_EmptyValueIsFound(t *testing.T) {
+	store, _ := newTestPluginKVStore(t)
+	ctx := context.Background()
+	require.NoError(t, store.Set(ctx, "p1", "state", "k", []byte{}, 0))
+	value, found, err := store.Get(ctx, "p1", "state", "k")
+	require.NoError(t, err)
+	assert.True(t, found)
+	assert.Empty(t, value)
+}
+
+func TestPluginKVStore_DeleteMissingIsNoError(t *testing.T) {
+	store, _ := newTestPluginKVStore(t)
+	require.NoError(t, store.Delete(context.Background(), "p1", "state", "missing"))
+}
@@ -347,7 +347,7 @@ func (r *redeemCodeRepository) ListByUser(ctx context.Context, userID int64, lim
 	codes, err := r.client.RedeemCode.Query().
 		Where(redeemcode.UsedByEQ(userID)).
 		WithGroup().
-		Order(dbent.Desc(redeemcode.FieldUsedAt)).
+		Order(dbent.Desc(redeemcode.FieldUsedAt), dbent.Desc(redeemcode.FieldID)).
 		Limit(limit).
 		All(ctx)
 	if err != nil {
@@ -377,7 +377,7 @@ func (r *redeemCodeRepository) ListByUserPaginated(ctx context.Context, userID i
 		WithGroup().
 		Offset(params.Offset()).
 		Limit(params.Limit()).
-		Order(dbent.Desc(redeemcode.FieldUsedAt)).
+		Order(dbent.Desc(redeemcode.FieldUsedAt), dbent.Desc(redeemcode.FieldID)).
 		All(ctx)
 	if err != nil {
 		return nil, nil, err
@@ -3,8 +3,10 @@
 package repository

 import (
+	"fmt"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/pagination"
 	"github.com/Wei-Shaw/sub2api/internal/service"
+	"time"
 )

 func (s *RedeemCodeRepoSuite) TestListWithFilters_SortByValueAsc() {
@@ -22,3 +24,50 @@ func (s *RedeemCodeRepoSuite) TestListWithFilters_SortByValueAsc() {
 	s.Require().Equal("VALUE-10", codes[0].Code)
 	s.Require().Equal("VALUE-20", codes[1].Code)
 }
+
+func (s *RedeemCodeRepoSuite) TestHistoryPaginationIsolationAndOrdering() {
+	user := s.createUser("history@example.com")
+	other := s.createUser("other-history@example.com")
+	now := time.Now().UTC().Truncate(time.Second)
+	var ids []int64
+	types := []string{service.RedeemTypeBalance, service.RedeemTypeConcurrency, service.RedeemTypeSubscription, "admin_balance", "admin_concurrency"}
+	for i := 0; i < 105; i++ {
+		usedAt := now
+		if i == 0 {
+			usedAt = now.Add(time.Hour)
+		}
+		code, err := s.client.RedeemCode.Create().SetCode(fmt.Sprintf("HISTORY-%d", i)).SetType(types[i%len(types)]).SetStatus(service.StatusUsed).SetUsedBy(user.ID).SetUsedAt(usedAt).Save(s.ctx)
+		s.Require().NoError(err)
+		ids = append(ids, code.ID)
+	}
+	_, err := s.client.RedeemCode.Create().SetCode("OTHER-HISTORY").SetType(service.RedeemTypeSubscription).SetStatus(service.StatusUsed).SetUsedBy(other.ID).SetUsedAt(now.Add(2 * time.Hour)).Save(s.ctx)
+	s.Require().NoError(err)
+	expected := []int64{ids[0]}
+	for i := len(ids) - 1; i > 0; i-- {
+		expected = append(expected, ids[i])
+	}
+	for _, size := range []int{20, 50, 100} {
+		var got []int64
+		for page := 1; page <= (105+size-1)/size+1; page++ {
+			codes, result, err := s.repo.ListByUserPaginated(s.ctx, user.ID, pagination.PaginationParams{Page: page, PageSize: size}, "")
+			s.Require().NoError(err)
+			s.Equal(int64(105), result.Total)
+			s.LessOrEqual(len(codes), size)
+			for _, code := range codes {
+				s.Equal(user.ID, *code.UsedBy)
+				got = append(got, code.ID)
+			}
+		}
+		s.Equal(expected, got)
+	}
+	codes, result, err := s.repo.ListByUserPaginated(s.ctx, other.ID, pagination.PaginationParams{Page: 1, PageSize: 20}, "")
+	s.Require().NoError(err)
+	s.Equal(int64(1), result.Total)
+	s.Require().Len(codes, 1)
+	s.Equal("OTHER-HISTORY", codes[0].Code)
+	empty := s.createUser("empty-history@example.com")
+	codes, result, err = s.repo.ListByUserPaginated(s.ctx, empty.ID, pagination.PaginationParams{Page: 1, PageSize: 20}, "")
+	s.Require().NoError(err)
+	s.Empty(codes)
+	s.Zero(result.Total)
+}
@@ -18,8 +18,23 @@ func TestUsageLogRepositoryGetAllGroupUsageSummaryUsesRollupTail(t *testing.T) {
 	todayStart := time.Date(2026, 3, 9, 4, 0, 0, 0, time.UTC)
 	yesterdayStart := time.Date(2026, 3, 8, 5, 0, 0, 0, time.UTC)

-	mock.ExpectQuery(`(?s)usage_group_rollup_state.*usage_group_daily_rollups.*created_at >= state\.tail_start`).
-		WithArgs(todayStart, yesterdayStart, "America/New_York", "2026-03-09", "2026-03-08").
+	// 水位有效：closed_before = 2026-03-07（≤ today），时区与配置一致。
+	mock.ExpectQuery(`(?s)COUNT\(\*\).*usage_group_rollup_state.*WHERE id = 1`).
+		WillReturnRows(sqlmock.NewRows([]string{"count", "closed_before", "retained_from", "timezone_name"}).
+			AddRow(1, "2026-03-07", time.Date(2026, 2, 1, 5, 0, 0, 0, time.UTC), "America/New_York"))
+
+	// 尾段起点必须以参数形式进 WHERE —— 这正是把 usage_logs 全表扫换回
+	// idx_usage_logs_created_at 索引扫的那一步。
+	mock.ExpectQuery(`(?s)usage_group_daily_rollups.*FROM usage_logs ul\s+WHERE ul\.created_at >= \$7`).
+		WithArgs(
+			todayStart,
+			yesterdayStart,
+			"2026-03-08", // yesterdayDate
+			true,         // 水位有效
+			"2026-02-01", // retained_from 在配置时区内的日期
+			"2026-03-07", // closed_before
+			time.Date(2026, 3, 7, 5, 0, 0, 0, time.UTC), // tail_start = closed_before 在配置时区的零点
+		).
 		WillReturnRows(sqlmock.NewRows([]string{"group_id", "total_cost", "today_cost", "yesterday_cost"}).
 			AddRow(int64(7), 12.5, 1.25, 2.5))

@@ -31,3 +46,51 @@ func TestUsageLogRepositoryGetAllGroupUsageSummaryUsesRollupTail(t *testing.T) {
 	require.InDelta(t, 2.5, result[0].YesterdayCost, 0.0000001)
 	require.NoError(t, mock.ExpectationsWereMet())
 }
+
+// 水位无效（时区变了 / 水位在未来 / 行缺失）时必须退化成"从 epoch 起全量重算"，
+// 与改动前 SQL 里 CASE WHEN valid 的 ELSE 分支逐字对应：
+// 历史日桶整段不参与，尾段起点回到 1970-01-01。
+func TestUsageLogRepositoryGetAllGroupUsageSummaryFallsBackWhenWatermarkInvalid(t *testing.T) {
+	epoch := time.Unix(0, 0).UTC()
+	todayStart := time.Date(2026, 3, 9, 4, 0, 0, 0, time.UTC)
+	yesterdayStart := time.Date(2026, 3, 8, 5, 0, 0, 0, time.UTC)
+
+	cases := []struct {
+		name string
+		rows *sqlmock.Rows
+	}{
+		{
+			name: "水位行缺失",
+			rows: sqlmock.NewRows([]string{"count", "closed_before", "retained_from", "timezone_name"}).
+				AddRow(0, nil, nil, nil),
+		},
+		{
+			name: "时区与当前配置不一致",
+			rows: sqlmock.NewRows([]string{"count", "closed_before", "retained_from", "timezone_name"}).
+				AddRow(1, "2026-03-07", time.Date(2026, 2, 1, 5, 0, 0, 0, time.UTC), "Asia/Shanghai"),
+		},
+		{
+			name: "水位位于未来",
+			rows: sqlmock.NewRows([]string{"count", "closed_before", "retained_from", "timezone_name"}).
+				AddRow(1, "2026-03-10", time.Date(2026, 2, 1, 5, 0, 0, 0, time.UTC), "America/New_York"),
+		},
+	}
+
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			db, mock := newSQLMock(t)
+			repo := newUsageLogRepositoryWithSQL(nil, db)
+			useGroupUsageRepositoryTestTimezone(t, "America/New_York")
+
+			mock.ExpectQuery(`(?s)COUNT\(\*\).*usage_group_rollup_state.*WHERE id = 1`).
+				WillReturnRows(tc.rows)
+			mock.ExpectQuery(`(?s)usage_group_daily_rollups.*FROM usage_logs ul\s+WHERE ul\.created_at >= \$7`).
+				WithArgs(todayStart, yesterdayStart, "2026-03-08", false, "1970-01-01", "1970-01-01", epoch).
+				WillReturnRows(sqlmock.NewRows([]string{"group_id", "total_cost", "today_cost", "yesterday_cost"}))
+
+			_, err := repo.GetAllGroupUsageSummary(context.Background(), todayStart)
+			require.NoError(t, err)
+			require.NoError(t, mock.ExpectationsWereMet())
+		})
+	}
+}
@@ -97,6 +97,7 @@ var ProviderSet = wire.NewSet(
 	NewErrorPassthroughRepository,
 	NewTLSFingerprintProfileRepository,
 	NewPluginRepository,
+	NewPluginKVStore,
 	NewChannelRepository,
 	NewChannelMonitorRepository,
 	NewChannelMonitorV2Repository,
@@ -759,6 +759,7 @@ func registerPluginRoutes(admin *gin.RouterGroup, h *handler.Handlers, stepUpAut
 		plugins.POST("/:id/disable", gin.HandlerFunc(stepUpAuth), h.Admin.Plugin.Disable)
 		plugins.DELETE("/:id", gin.HandlerFunc(stepUpAuth), h.Admin.Plugin.Delete)
 		plugins.GET("/:id/config", h.Admin.Plugin.GetConfig)
+		plugins.GET("/:id/status", h.Admin.Plugin.Status)
 		plugins.PUT("/:id/config", gin.HandlerFunc(stepUpAuth), h.Admin.Plugin.SaveConfig)
 		plugins.POST("/:id/test", gin.HandlerFunc(stepUpAuth), h.Admin.Plugin.Test)
 		plugins.POST("/:id/ui-session", h.Admin.Plugin.CreateUISession)
@@ -368,6 +368,11 @@ func RegisterGatewayRoutes(
 	rootRoute := func(method, path string, limit gin.HandlerFunc, handler gin.HandlerFunc) {
 		r.Handle(method, path, limit, clientRequestID, opsErrorLogger, endpointNorm, gin.HandlerFunc(apiKeyAuth), groupModelAllowlist, compositeTarget, requireGroupAnthropic, handler)
 	}
+	for _, prefix := range []string{"/api/v3", "/v3", "/v1", ""} {
+		rootRoute(http.MethodPost, prefix+"/contents/generations/tasks", bodyLimit, h.OpenAIGateway.SeedanceTasks)
+		rootRoute(http.MethodGet, prefix+"/contents/generations/tasks/:task_id", bodyLimit, h.OpenAIGateway.SeedanceTasks)
+		rootRoute(http.MethodDelete, prefix+"/contents/generations/tasks/:task_id", bodyLimit, h.OpenAIGateway.SeedanceTasks)
+	}
 	rootRoute(http.MethodPost, "/responses", bodyLimit, responsesHandler)
 	rootRoute(http.MethodPost, "/responses/*subpath", bodyLimit, guardResponsesSubpath(responsesHandler))
 	rootRoute(http.MethodPost, "/alpha/search", textBodyLimit, h.OpenAIGateway.AlphaSearch)
@@ -157,6 +157,10 @@ func TestGatewayRoutesGroupModelAllowlistCoversRootAliasRoutes(t *testing.T) {
 		{http.MethodPost, "/v1/embeddings", `{"model":"gpt-4.1","input":"hi"}`},
 		{http.MethodPost, "/v1/images/generations", `{"model":"gpt-4.1"}`},
 		{http.MethodPost, "/v1/videos/generations", `{"model":"gpt-4.1"}`},
+		{http.MethodPost, "/api/v3/contents/generations/tasks", `{"model":"gpt-4.1","content":[{"type":"text","text":"waves"}]}`},
+		{http.MethodPost, "/v3/contents/generations/tasks", `{"model":"gpt-4.1","content":[{"type":"text","text":"waves"}]}`},
+		{http.MethodPost, "/v1/contents/generations/tasks", `{"model":"gpt-4.1","content":[{"type":"text","text":"waves"}]}`},
+		{http.MethodPost, "/contents/generations/tasks", `{"model":"gpt-4.1","content":[{"type":"text","text":"waves"}]}`},
 		{http.MethodPost, "/v1/live", `{"session":{"model":"gpt-4.1"},"sdp":"v=0"}`},
 		{http.MethodPost, "/backend-api/codex/responses", `{"model":"gpt-4.1"}`},
 		{http.MethodPost, "/backend-api/codex/realtime/calls", `{"session":{"model":"gpt-4.1"},"sdp":"v=0"}`},
@@ -0,0 +1,34 @@
+package routes
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/stretchr/testify/require"
+)
+
+func TestSeedanceNativeRoutes(t *testing.T) {
+	router := newGatewayRoutesTestRouter()
+	for _, prefix := range []string{"/api/v3", "/v3", "/v1", ""} {
+		for _, method := range []string{http.MethodPost, http.MethodGet, http.MethodDelete} {
+			path := prefix + "/contents/generations/tasks"
+			if method != http.MethodPost {
+				path += "/task-1"
+			}
+			w := httptest.NewRecorder()
+			router.ServeHTTP(w, httptest.NewRequest(method, path, strings.NewReader(`{"model":"seedance","content":[{"type":"text","text":"waves"}]}`)))
+			require.NotEqual(t, http.StatusNotFound, w.Code, method+" "+path)
+		}
+	}
+}
+
+func TestSeedanceRejectsOtherPlatforms(t *testing.T) {
+	for _, platform := range []string{service.PlatformGrok, service.PlatformAnthropic, service.PlatformGemini} {
+		w := httptest.NewRecorder()
+		newGatewayRoutesTestRouter(platform).ServeHTTP(w, httptest.NewRequest(http.MethodPost, "/api/v3/contents/generations/tasks", strings.NewReader(`{"model":"seedance","content":[{}]}`)))
+		require.Equal(t, http.StatusForbidden, w.Code)
+	}
+}
@@ -848,6 +848,10 @@ func resolveRequestedModelInMapping(mapping map[string]string, requestedModel st
 // 会把未知模型原样透传，Codex 上游对这类模型必然返回不可重试的 400，导致
 // 请求卡死在该账号上、无法 failover 到真正支持该模型的 API Key 账号（#3662）。
 // 未知/自定义别名仍保持允许（兼容渠道级映射），见 isOpenAIOAuthServableModel。
+//
+// 例外：DeepSeek 平台的空映射不再是「允许所有」，改按官方模型白名单判定
+// （isDeepseekServableModel）——未知模型名透传上游只会得到 404/400，并误触发
+// per-(账号,模型) 30 分钟冷却；带 [1m] 上下文后缀的写法先归一化再比对。
 func (a *Account) IsModelSupported(requestedModel string) bool {
 	// 透传模式仅替换认证、模型语义完全交由上游决定，因此放行所有模型。
 	// 该短路必须在 model_mapping 判定之前：账号从"白名单模式"切换到透传后，
@@ -861,6 +865,9 @@ func (a *Account) IsModelSupported(requestedModel string) bool {
 		if a.IsOpenAIOAuth() {
 			return isOpenAIOAuthServableModel(requestedModel)
 		}
+		if a.Platform == PlatformDeepseek {
+			return isDeepseekServableModel(requestedModel)
+		}
 		return true // 无映射 = 允许所有
 	}
 	if mappingSupportsRequestedModel(mapping, requestedModel) {
@@ -1821,6 +1828,11 @@ func (a *Account) SupportsOpenAIEndpointCapability(capability OpenAIEndpointCapa
 	if a == nil {
 		return false
 	}
+	if capability == OpenAIEndpointCapabilitySeedance {
+		configured, _ := a.openAIEndpointCapabilitySet()
+		return configured["seedance"] && a.Platform == PlatformOpenAI && a.Type == AccountTypeAPIKey &&
+			strings.TrimSpace(a.GetCredential("base_url")) != ""
+	}
 	if capability == "" {
 		return true
 	}
@@ -231,6 +231,82 @@ func TestAccountIsModelSupported(t *testing.T) {
 			requestedModel: "gemini-3-flash",
 			expected:       false,
 		},
+
+		// DeepSeek 空映射：官方模型白名单（不再是「允许所有」）
+		{
+			name:           "deepseek empty mapping allows official flash",
+			platform:       PlatformDeepseek,
+			credentials:    map[string]any{},
+			requestedModel: "deepseek-flash",
+			expected:       true,
+		},
+		{
+			name:           "deepseek empty mapping normalizes claude code long context suffix",
+			platform:       PlatformDeepseek,
+			credentials:    map[string]any{},
+			requestedModel: "deepseek-flash[1m]",
+			expected:       true,
+		},
+		{
+			name:           "deepseek empty mapping matches case-insensitively",
+			platform:       PlatformDeepseek,
+			credentials:    map[string]any{},
+			requestedModel: "deepseek-FLASH",
+			expected:       true,
+		},
+		{
+			name:           "deepseek empty mapping allows versioned pro name",
+			platform:       PlatformDeepseek,
+			credentials:    map[string]any{},
+			requestedModel: "deepseek-v4-pro-0813",
+			expected:       true,
+		},
+		{
+			name:           "deepseek empty mapping rejects retired chat model",
+			platform:       PlatformDeepseek,
+			credentials:    map[string]any{},
+			requestedModel: "deepseek-chat",
+			expected:       false,
+		},
+		{
+			name:           "deepseek empty mapping rejects foreign claude model",
+			platform:       PlatformDeepseek,
+			credentials:    map[string]any{},
+			requestedModel: "claude-sonnet-4-6",
+			expected:       false,
+		},
+		{
+			name:           "deepseek empty mapping rejects unknown gpt model",
+			platform:       PlatformDeepseek,
+			credentials:    map[string]any{},
+			requestedModel: "gpt-future-model",
+			expected:       false,
+		},
+		{
+			name:           "deepseek empty mapping rejects typo",
+			platform:       PlatformDeepseek,
+			credentials:    map[string]any{},
+			requestedModel: "deepseek-flas",
+			expected:       false,
+		},
+		{
+			name:     "deepseek explicit mapping wins over whitelist",
+			platform: PlatformDeepseek,
+			credentials: map[string]any{
+				"model_mapping": map[string]any{
+					"foo-bar": "deepseek-flash",
+				},
+			},
+			requestedModel: "foo-bar",
+			expected:       true,
+		},
+		{
+			name:           "non deepseek empty mapping still allows all",
+			platform:       PlatformAnthropic,
+			credentials:    map[string]any{},
+			requestedModel: "any-model",
+			expected:       true,
+		},
 	}

 	for _, tt := range tests {
@@ -86,7 +86,14 @@ func (s *AntigravityGatewayService) ForwardGemini(ctx context.Context, c *gin.Co
 		return nil, s.writeGoogleError(c, http.StatusNotFound, "Unsupported action: "+action)
 	}

-	mappedModel := s.getMappedModel(account, originalModel)
+	// 裸模型名（gemini-3.8-flash）+ thinkingConfig → 账号映射表里对应的 -low/-medium/-high 变体；
+	// 裸名有显式映射时保持原行为。
+	mappedModel, variantResolved := resolveGeminiThinkingVariant(account, originalModel, body)
+	if !variantResolved {
+		mappedModel = s.getMappedModel(account, originalModel)
+	} else {
+		logger.LegacyPrintf("service.antigravity_gateway", "%s resolved bare Gemini model %s to thinking variant %s", prefix, originalModel, mappedModel)
+	}
 	if mappedModel == "" {
 		MarkOpsClientBusinessLimited(c, OpsClientBusinessLimitedReasonLocalFeatureGate)
 		return nil, s.writeGoogleError(c, http.StatusForbidden, fmt.Sprintf("model %s not in whitelist", originalModel))
@@ -1387,6 +1387,49 @@ func TestHandleClaudeStreamingResponse_NormalComplete(t *testing.T) {

 // TestHandleGeminiStreamingResponse_ThoughtsTokenCount
 // 验证：Gemini 流式转发时 thoughtsTokenCount 被计入 OutputTokens
+// 回归：上游事件之间的空分隔行不能再透传，否则下游看到的是 "data: ...\n\n\n"。
+// google-genai Go SDK 按 "\n\n" 切事件，多出的 "\n" 会让第二个事件的前缀变成
+// "\ndata" 而报 invalid stream chunk（Antigravity CLI 每条流都在第二个事件中断）。
+func TestHandleGeminiStreamingResponse_EventSeparatorIsExactlyOneBlankLine(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	svc := newAntigravityTestService(&config.Config{
+		Gateway: config.GatewayConfig{MaxLineSize: defaultMaxLineSize},
+	})
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/", nil)
+
+	pr, pw := io.Pipe()
+	resp := &http.Response{StatusCode: http.StatusOK, Body: pr, Header: http.Header{}}
+
+	first := `{"candidates":[{"content":{"role":"model","parts":[{"text":"Hello"}]}}],"usageMetadata":{"promptTokenCount":10,"candidatesTokenCount":3}}`
+	second := `{"candidates":[{"content":{"role":"model","parts":[{"thoughtSignature":"sig","text":""}]},"finishReason":"STOP"}],"usageMetadata":{"promptTokenCount":10,"candidatesTokenCount":3,"thoughtsTokenCount":5}}`
+
+	go func() {
+		defer func() { _ = pw.Close() }()
+		// 上游原样：每个 data 行后跟一个空分隔行（第二个事件用 CRLF，覆盖两种换行）
+		fmt.Fprintf(pw, "data: %s\n\n", first)
+		fmt.Fprintf(pw, "data: %s\r\n\r\n", second)
+	}()
+
+	result, err := svc.handleGeminiStreamingResponse(c, resp, time.Now())
+	_ = pr.Close()
+
+	require.NoError(t, err)
+	require.NotNil(t, result)
+
+	body := rec.Body.String()
+	require.Equal(t, "data: "+first+"\n\ndata: "+second+"\n\n", body)
+	require.NotContains(t, body, "\n\n\n", "events must be separated by exactly one blank line")
+
+	// 模拟 google-genai 的切帧方式：按 "\n\n" 切，每个非空 token 都必须以 "data" 为前缀
+	for _, token := range strings.Split(strings.TrimSuffix(body, "\n\n"), "\n\n") {
+		prefix, _, _ := strings.Cut(token, ":")
+		require.Equal(t, "data", prefix, "token %q would be rejected by a \\n\\n-delimited SSE parser", token)
+	}
+}
+
 func TestHandleGeminiStreamingResponse_ThoughtsTokenCount(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 	svc := newAntigravityTestService(&config.Config{
@@ -197,6 +197,11 @@ func (s *AntigravityGatewayService) handleGeminiStreamingResponse(c *gin.Context
 	if s.settingService.cfg != nil && s.settingService.cfg.Gateway.StreamKeepaliveInterval > 0 {
 		keepaliveInterval = time.Duration(s.settingService.cfg.Gateway.StreamKeepaliveInterval) * time.Second
 	}
+	// go-genai / python-genai 不会忽略 SSE 注释行，收到 ":\n\n" 会直接把整个流判成
+	// invalid stream chunk 而中断（Antigravity CLI 在用 go-genai）。对这类客户端宁可不发心跳。
+	if keepaliveInterval > 0 && downstreamRejectsSSEComments(c) {
+		keepaliveInterval = 0
+	}
 	var keepaliveTicker *time.Ticker
 	if keepaliveInterval > 0 {
 		keepaliveTicker = time.NewTicker(keepaliveInterval)
@@ -288,6 +293,14 @@ func (s *AntigravityGatewayService) handleGeminiStreamingResponse(c *gin.Context
 				continue
 			}

+			// 上游每个 data 事件后面跟一个空行作为事件分隔。上面已经把 data 行写成
+			// "data: ...\n\n"，若再把这个空行透传出去，事件之间就会变成 "\n\n\n"。
+			// google-genai 的 Go SDK（Antigravity CLI 在用）按 "\n\n" 切事件，多出的
+			// "\n" 会粘到下一个事件开头，前缀变成 "\ndata" 而被判成 invalid stream chunk。
+			if trimmed == "" {
+				continue
+			}
+
 			cw.Fprintf("%s\n", line)

 		case <-intervalCh:
@@ -0,0 +1,137 @@
+package service
+
+import (
+	"encoding/json"
+	"strings"
+)
+
+// Gemini 原生请求（/v1beta/models/{model}:generateContent 等）经 Antigravity 账号转发时，
+// 客户端（如 Antigravity CLI / go-genai）习惯发"裸"模型名（gemini-3.8-flash）并用
+// generationConfig.thinkingConfig 表达思考深度；而 Antigravity 上游的模型目录只有
+// gemini-3.8-flash-low / -medium / -high / -tiered 这类带后缀的变体，裸名直接转发会被上游
+// 以 404 "Requested entity was not found." 拒绝。
+//
+// 这里在账号 model_mapping 没有为裸名配置显式条目、但配置了它的后缀变体时，
+// 按 thinkingConfig 自动挑一个变体：
+//   - thinkingLevel: "low" / "medium" / "high" → 同名后缀
+//   - thinkingBudget: -1（动态）→ high；0 → low；1..1024 → low；1025..8192 → medium；>8192 → high
+//   - 未携带 thinkingConfig → high（与 Gemini 3 系列默认开启动态思考一致）
+// 选中的后缀在映射表里不存在时按 high → medium → low → tiered 的顺序降级到存在的变体。
+// 显式映射（裸名本身在 model_mapping 里）始终优先，行为与现在完全一致。
+
+var geminiThinkingVariantSuffixes = []string{"-low", "-medium", "-high", "-tiered"}
+
+const (
+	geminiThinkingBudgetLowMax    = 1024
+	geminiThinkingBudgetMediumMax = 8192
+)
+
+type geminiThinkingConfigProbe struct {
+	GenerationConfig struct {
+		ThinkingConfig *struct {
+			ThinkingBudget *json.Number `json:"thinkingBudget"`
+			ThinkingLevel  string       `json:"thinkingLevel"`
+		} `json:"thinkingConfig"`
+	} `json:"generationConfig"`
+}
+
+// hasGeminiThinkingVariantSuffix 判断模型名是否已经带了思考深度后缀。
+func hasGeminiThinkingVariantSuffix(model string) bool {
+	for _, suffix := range geminiThinkingVariantSuffixes {
+		if strings.HasSuffix(model, suffix) {
+			return true
+		}
+	}
+	return false
+}
+
+// geminiThinkingLevelFromBody 从 Gemini 原生请求体推导期望的思考档位（low/medium/high）。
+// 解析失败或未携带 thinkingConfig 时返回 "high"。
+func geminiThinkingLevelFromBody(body []byte) string {
+	if len(body) == 0 {
+		return "high"
+	}
+	var probe geminiThinkingConfigProbe
+	if err := json.Unmarshal(body, &probe); err != nil || probe.GenerationConfig.ThinkingConfig == nil {
+		return "high"
+	}
+	tc := probe.GenerationConfig.ThinkingConfig
+	switch strings.ToLower(strings.TrimSpace(tc.ThinkingLevel)) {
+	case "low":
+		return "low"
+	case "medium":
+		return "medium"
+	case "high":
+		return "high"
+	}
+	if tc.ThinkingBudget == nil {
+		return "high"
+	}
+	budget, err := tc.ThinkingBudget.Float64()
+	if err != nil {
+		return "high"
+	}
+	switch {
+	case budget < 0:
+		return "high" // -1 = 动态思考
+	case budget <= geminiThinkingBudgetLowMax:
+		return "low" // 含 0（关闭思考）：上游没有"无思考"变体，取最浅档，budget 本身仍原样透传
+	case budget <= geminiThinkingBudgetMediumMax:
+		return "medium"
+	default:
+		return "high"
+	}
+}
+
+// accountRawModelMappingHasKey 判断 key 是否由用户写在账号 credentials.model_mapping 里
+// （而不是 resolveModelMapping 运行时补进来的默认透传条目）。
+func accountRawModelMappingHasKey(account *Account, key string) bool {
+	if account == nil || account.Credentials == nil {
+		return false
+	}
+	raw, _ := account.Credentials["model_mapping"].(map[string]any)
+	_, ok := raw[key]
+	return ok
+}
+
+// resolveGeminiThinkingVariant 为裸 Gemini 模型名挑选账号映射表里存在的思考深度变体。
+// 返回 (映射后的上游模型名, 是否命中)。未命中时调用方应回退到常规 getMappedModel 流程。
+func resolveGeminiThinkingVariant(account *Account, requestedModel string, body []byte) (string, bool) {
+	if account == nil {
+		return "", false
+	}
+	model := strings.TrimSpace(strings.TrimPrefix(requestedModel, "models/"))
+	if !strings.HasPrefix(model, "gemini-") || hasGeminiThinkingVariantSuffix(model) {
+		return "", false
+	}
+	mapping := account.GetModelMapping()
+	if len(mapping) == 0 {
+		return "", false
+	}
+	// 裸名本身已有"真正的"映射 → 尊重现有配置，不做推导。判定分两层：
+	//   1. 映射目标不是自己（如 gemini-3.8-flash → gemini-3.8-flash-tiered）：用户明确指定了目标；
+	//   2. 映射目标是自己（原样透传）：仅当这条是用户亲手写进 credentials.model_mapping 的才尊重。
+	//      resolveModelMapping 会给每个 Antigravity 账号自动补 gemini-3.x-flash 裸名自映射
+	//      （ensureAntigravityDefaultPassthroughs），而上游目录里并没有裸名模型，那条自映射
+	//      正是导致 404 的来源，不能算作用户意图。
+	if mapped, matched := resolveRequestedModelInMapping(mapping, model); matched {
+		if strings.TrimSpace(mapped) != model || accountRawModelMappingHasKey(account, model) {
+			return "", false
+		}
+	}
+
+	preferred := geminiThinkingLevelFromBody(body)
+	order := []string{preferred}
+	for _, level := range []string{"high", "medium", "low", "tiered"} {
+		if level != preferred {
+			order = append(order, level)
+		}
+	}
+	for _, level := range order {
+		candidate := model + "-" + level
+		if mapped, matched := resolveRequestedModelInMapping(mapping, candidate); matched && strings.TrimSpace(mapped) != "" {
+			return mapped, true
+		}
+	}
+	return "", false
+}
@@ -0,0 +1,118 @@
+package service
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+func newAntigravityAccountWithMapping(mapping map[string]string) *Account {
+	mappingAny := make(map[string]any, len(mapping))
+	for k, v := range mapping {
+		mappingAny[k] = v
+	}
+	return &Account{
+		ID:       1,
+		Name:     "ag",
+		Platform: PlatformAntigravity,
+		Credentials: map[string]any{
+			"model_mapping": mappingAny,
+		},
+	}
+}
+
+func TestGeminiThinkingLevelFromBody(t *testing.T) {
+	tests := []struct {
+		name string
+		body string
+		want string
+	}{
+		{"empty body", ``, "high"},
+		{"no thinkingConfig", `{"contents":[]}`, "high"},
+		{"budget -1 dynamic", `{"generationConfig":{"thinkingConfig":{"thinkingBudget":-1}}}`, "high"},
+		{"budget 0 off", `{"generationConfig":{"thinkingConfig":{"thinkingBudget":0}}}`, "low"},
+		{"budget 1000 (agy low)", `{"generationConfig":{"thinkingConfig":{"thinkingBudget":1000}}}`, "low"},
+		{"budget 4000 (agy medium)", `{"generationConfig":{"thinkingConfig":{"thinkingBudget":4000}}}`, "medium"},
+		{"budget 24576", `{"generationConfig":{"thinkingConfig":{"thinkingBudget":24576}}}`, "high"},
+		{"thinkingLevel wins over budget", `{"generationConfig":{"thinkingConfig":{"thinkingBudget":24576,"thinkingLevel":"low"}}}`, "low"},
+		{"thinkingLevel medium", `{"generationConfig":{"thinkingConfig":{"thinkingLevel":"MEDIUM"}}}`, "medium"},
+		{"invalid json", `{"generationConfig":`, "high"},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			require.Equal(t, tt.want, geminiThinkingLevelFromBody([]byte(tt.body)))
+		})
+	}
+}
+
+func TestResolveGeminiThinkingVariant(t *testing.T) {
+	// 与生产 Antigravity 账号从上游同步来的目录一致：只有带后缀的变体，没有裸名。
+	catalog := map[string]string{
+		"gemini-3.8-flash-low":    "gemini-3.8-flash-low",
+		"gemini-3.8-flash-medium": "gemini-3.8-flash-medium",
+		"gemini-3.8-flash-high":   "gemini-3.8-flash-high",
+		"gemini-3.8-flash-tiered": "gemini-3.8-flash-tiered",
+		"gemini-3.6-flash-high":   "gemini-3.6-flash-high",
+		"gemini-3.1-pro-high":     "gemini-3.1-pro-high",
+		"gemini-2.5-flash":        "gemini-2.5-flash",
+	}
+	budget := func(v string) []byte {
+		return []byte(`{"contents":[],"generationConfig":{"thinkingConfig":{"thinkingBudget":` + v + `}}}`)
+	}
+
+	tests := []struct {
+		name      string
+		mapping   map[string]string
+		model     string
+		body      []byte
+		want      string
+		wantMatch bool
+	}{
+		{"agy high (-1) -> -high", catalog, "gemini-3.8-flash", budget("-1"), "gemini-3.8-flash-high", true},
+		{"agy medium (4000) -> -medium", catalog, "gemini-3.8-flash", budget("4000"), "gemini-3.8-flash-medium", true},
+		{"agy low (1000) -> -low", catalog, "gemini-3.8-flash", budget("1000"), "gemini-3.8-flash-low", true},
+		{"models/ prefix stripped", catalog, "models/gemini-3.8-flash", budget("1000"), "gemini-3.8-flash-low", true},
+		{"no thinkingConfig -> -high", catalog, "gemini-3.8-flash", []byte(`{"contents":[]}`), "gemini-3.8-flash-high", true},
+		// gemini-3.6/3.7/3.8-flash 的四个变体会被 resolveModelMapping 自动补齐，所以降级用一个不在默认表里的型号
+		{"only -high exists: low request degrades to high", map[string]string{
+			"gemini-3.5-flash-high": "gemini-3.5-flash-high",
+		}, "gemini-3.5-flash", budget("1000"), "gemini-3.5-flash-high", true},
+		{"injected default variants are usable", catalog, "gemini-3.6-flash", budget("1000"), "gemini-3.6-flash-low", true},
+		{"already suffixed: untouched", catalog, "gemini-3.8-flash-low", budget("-1"), "", false},
+		{"bare name explicitly mapped: untouched", catalog, "gemini-2.5-flash", budget("-1"), "", false},
+		{"no variants in mapping: untouched", catalog, "gemini-9.9-flash", budget("-1"), "", false},
+		{"non-gemini model: untouched", catalog, "claude-sonnet-4-6", budget("-1"), "", false},
+		{"explicit bare alias wins over variants", map[string]string{
+			"gemini-3.8-flash":      "gemini-3.8-flash-tiered",
+			"gemini-3.8-flash-high": "gemini-3.8-flash-high",
+		}, "gemini-3.8-flash", budget("-1"), "", false},
+		{"variant value is itself a mapping target", map[string]string{
+			"gemini-3.8-flash-high": "gemini-3.8-flash-tiered",
+		}, "gemini-3.8-flash", budget("-1"), "gemini-3.8-flash-tiered", true},
+		{"user-written bare identity passthrough: respected", map[string]string{
+			"gemini-3.8-flash":      "gemini-3.8-flash",
+			"gemini-3.8-flash-high": "gemini-3.8-flash-high",
+		}, "gemini-3.8-flash", budget("1000"), "", false},
+		{"runtime-injected bare passthrough (not in credentials): still resolved", map[string]string{
+			// resolveModelMapping 会补 gemini-3.7-flash → gemini-3.7-flash 的默认透传，
+			// 但 credentials 里没有这条，应当继续推导变体。
+			"gemini-3.7-flash-medium": "gemini-3.7-flash-medium",
+		}, "gemini-3.7-flash", budget("4000"), "gemini-3.7-flash-medium", true},
+		// 空 credentials 映射 → 走 DefaultAntigravityModelMapping，其中同样只有带后缀的 3.8 flash
+		{"empty mapping falls back to default catalog", map[string]string{}, "gemini-3.8-flash", budget("-1"), "gemini-3.8-flash-high", true},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			account := newAntigravityAccountWithMapping(tt.mapping)
+			got, matched := resolveGeminiThinkingVariant(account, tt.model, tt.body)
+			require.Equal(t, tt.wantMatch, matched)
+			require.Equal(t, tt.want, got)
+		})
+	}
+
+	t.Run("nil account", func(t *testing.T) {
+		got, matched := resolveGeminiThinkingVariant(nil, "gemini-3.8-flash", budget("-1"))
+		require.False(t, matched)
+		require.Empty(t, got)
+	})
+}
@@ -834,6 +834,16 @@ func TestNormalizeDeepSeekResponsesRequestBody(t *testing.T) {
 	require.False(t, gjson.GetBytes(adaptiveNormalized, "store").Bool())
 	require.False(t, gjson.GetBytes(adaptiveNormalized, "previous_response_id").Exists())

+	mediaBody := []byte(`{"model":"deepseek-v4.1-flash","store":true,"input":[{"type":"function_call","call_id":"call_image","name":"view_image","arguments":"{}"},{"type":"function_call_output","call_id":"call_image","output":[{"type":"input_image","image_url":"data:image/png;base64,AQID"}]}]}`)
+	mediaNormalized := normalizeDeepSeekResponsesRequestBody(deepseekResponses, mediaBody)
+	require.False(t, gjson.GetBytes(mediaNormalized, "store").Bool())
+	require.Equal(t, gjson.String, gjson.GetBytes(mediaNormalized, "input.1.output").Type)
+	require.NotContains(t, gjson.GetBytes(mediaNormalized, "input.1.output").String(), "data:image/png")
+	require.Equal(t, "message", gjson.GetBytes(mediaNormalized, "input.2.type").String())
+	require.Equal(t, "user", gjson.GetBytes(mediaNormalized, "input.2.role").String())
+	require.Equal(t, "input_image", gjson.GetBytes(mediaNormalized, "input.2.content.1.type").String())
+	require.Equal(t, "data:image/png;base64,AQID", gjson.GetBytes(mediaNormalized, "input.2.content.1.image_url").String())
+
 	// 非 responses 协议（deepseek CC 账号）原样返回
 	deepseekCC := &Account{Platform: PlatformDeepseek, Type: AccountTypeAPIKey}
 	require.Equal(t, string(body), string(normalizeDeepSeekResponsesRequestBody(deepseekCC, body)))
@@ -74,6 +74,53 @@ func TestAdaptResponsesClientToolsForAnthropic_LiftsAdditionalTools(t *testing.T
 	require.Equal(t, "message", input[0].(map[string]any)["type"])
 }

+// Codex 的 codex_app 工具（如 automation_update）把 parameters 根节点声明成对象
+// 联合；Responses→Anthropic 转换后 input_schema 顶部不能再出现联合关键字。
+func TestAdaptResponsesClientToolsForAnthropic_FlattensRootUnionSchema(t *testing.T) {
+	t.Parallel()
+
+	body := []byte(`{
+		"model":"claude-opus-5",
+		"input":[{"type":"message","role":"user","content":[{"type":"input_text","text":"create an automation"}]}],
+		"tools":[{"type":"namespace","name":"codex_app","tools":[{
+			"type":"function",
+			"name":"automation_update",
+			"description":"Create, update, view, or delete recurring automations",
+			"parameters":{"oneOf":[
+				{"type":"object","properties":{"mode":{"enum":["view"]},"id":{"type":"string"}},"required":["mode","id"]},
+				{"type":"object","properties":{"mode":{"enum":["update"]},"id":{"type":"string"}},"required":["mode","id"]}
+			]}
+		}]}]
+	}`)
+
+	adapted, _, err := adaptResponsesClientToolsForAnthropic(body)
+	require.NoError(t, err)
+
+	var responsesReq apicompat.ResponsesRequest
+	require.NoError(t, json.Unmarshal(adapted, &responsesReq))
+
+	claudeReq, err := apicompat.ResponsesToAnthropicRequest(&responsesReq)
+	require.NoError(t, err)
+	require.Len(t, claudeReq.Tools, 1)
+
+	tool := claudeReq.Tools[0]
+	require.Equal(t, "codex_app__automation_update", tool.Name)
+
+	var schema map[string]json.RawMessage
+	require.NoError(t, json.Unmarshal(tool.InputSchema, &schema))
+	require.JSONEq(t, `"object"`, string(schema["type"]))
+	require.NotContains(t, schema, "oneOf")
+	require.NotContains(t, schema, "anyOf")
+	require.NotContains(t, schema, "allOf")
+	require.JSONEq(t, `["mode","id"]`, string(schema["required"]))
+
+	wire, err := json.Marshal(tool)
+	require.NoError(t, err)
+	require.NotContains(t, string(wire), `"input_schema":{"oneOf"`)
+	require.NotContains(t, string(wire), `"input_schema":{"anyOf"`)
+	require.NotContains(t, string(wire), `"input_schema":{"allOf"`)
+}
+
 func namespaceToolAnthropicStream() string {
 	return strings.Join([]string{
 		`event: message_start`,
@@ -16,6 +16,7 @@ import (
 	mathrand "math/rand"
 	"net/http"
 	"regexp"
+	"sort"
 	"strings"
 	"time"

@@ -483,6 +484,32 @@ func (s *GeminiMessagesCompatService) validateUpstreamBaseURL(raw string) (strin
 	return normalized, nil
 }

+// AntigravityGeminiModelIDs returns the effective Gemini mappings of schedulable
+// Antigravity accounts. Mixed routes must additionally respect the account opt-in.
+func (s *GeminiMessagesCompatService) AntigravityGeminiModelIDs(ctx context.Context, groupID *int64, requireMixed bool) ([]string, error) {
+	accounts, err := s.listSchedulableAccountsOnce(ctx, groupID, PlatformAntigravity, false)
+	if err != nil {
+		return nil, err
+	}
+	seen := make(map[string]struct{})
+	for _, account := range accounts {
+		if account.Platform != PlatformAntigravity || (requireMixed && !account.IsMixedSchedulingEnabled()) {
+			continue
+		}
+		for model := range account.GetModelMapping() {
+			if isAntigravityGeminiModel(model) {
+				seen[model] = struct{}{}
+			}
+		}
+	}
+	models := make([]string, 0, len(seen))
+	for model := range seen {
+		models = append(models, model)
+	}
+	sort.Strings(models)
+	return models, nil
+}
+
 // HasAntigravityAccounts 检查是否有可用的 antigravity 账户
 func (s *GeminiMessagesCompatService) HasAntigravityAccounts(ctx context.Context, groupID *int64) (bool, error) {
 	accounts, err := s.listSchedulableAccountsOnce(ctx, groupID, PlatformAntigravity, false)
@@ -0,0 +1,35 @@
+package service
+
+import (
+	"strings"
+
+	"github.com/gin-gonic/gin"
+)
+
+// geminiClientRejectsSSEComments 判断下游客户端是否是「无法容忍 SSE 注释行」的 Google GenAI SDK。
+//
+// Gemini 原生流里我们用 ":\n\n" 作为空闲心跳（防止代理/Cloudflare Tunnel 断开），但官方 SDK 里
+// 有两个实现不按 SSE 规范忽略注释行：
+//   - go-genai（Antigravity CLI 在用）：api_client.go iterateResponseStream 对任何非 "data:" 前缀的
+//     事件直接报 "invalid stream chunk: :"，整个流中断；
+//   - python-genai：_api_client.py 把非 "data:" 行当作错误 JSON 拼接后 json.loads，同样抛错。
+//
+// js-genai 用的是规范的 event-stream 解析器，能正确忽略注释。Google 官方 API 从不下发注释行，
+// 所以这两个 SDK 从未处理过这种输入。识别依据是 SDK 固定携带的 User-Agent / X-Goog-Api-Client，
+// 形如 "google-genai-sdk/1.71.0 gl-go/go1.28 ..." 或 "google-genai-sdk/1.x gl-python/3.12"。
+func geminiClientRejectsSSEComments(clientHint string) bool {
+	hint := strings.ToLower(strings.TrimSpace(clientHint))
+	if !strings.Contains(hint, "google-genai-sdk/") {
+		return false
+	}
+	return strings.Contains(hint, "gl-go/") || strings.Contains(hint, "gl-python/")
+}
+
+// downstreamRejectsSSEComments 从请求头识别下游客户端（同时看 User-Agent 与 X-Goog-Api-Client）。
+func downstreamRejectsSSEComments(c *gin.Context) bool {
+	if c == nil || c.Request == nil {
+		return false
+	}
+	return geminiClientRejectsSSEComments(c.GetHeader("User-Agent")) ||
+		geminiClientRejectsSSEComments(c.GetHeader("X-Goog-Api-Client"))
+}
@@ -0,0 +1,95 @@
+package service
+
+import (
+	"io"
+	"net/http"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/require"
+)
+
+func TestGeminiClientRejectsSSEComments(t *testing.T) {
+	cases := []struct {
+		name string
+		hint string
+		want bool
+	}{
+		{"go-genai (Antigravity CLI)", "google-genai-sdk/1.71.0 gl-go/go1.28-20260721-RC03 cl/951519500 +3ebc191975 X:fieldtrack,boringcrypto", true},
+		{"python-genai", "google-genai-sdk/1.20.0 gl-python/3.12.4", true},
+		{"js-genai tolerates comments", "google-genai-sdk/1.9.0 gl-node/22.3.0", false},
+		{"gemini-cli", "GeminiCLI/0.60.0 (darwin; arm64)", false},
+		{"curl", "curl/8.7.1", false},
+		{"empty", "", false},
+		{"case-insensitive", "Google-GenAI-SDK/1.0.0 GL-Go/go1.27", true},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			require.Equal(t, tc.want, geminiClientRejectsSSEComments(tc.hint))
+		})
+	}
+}
+
+func TestDownstreamRejectsSSECommentsReadsBothHeaders(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	c, _ := newAntigravityCompatContext(http.MethodPost, "/v1beta/models/gemini-3.8-flash:streamGenerateContent", nil)
+	require.False(t, downstreamRejectsSSEComments(c))
+
+	c.Request.Header.Set("X-Goog-Api-Client", "google-genai-sdk/1.71.0 gl-go/go1.28")
+	require.True(t, downstreamRejectsSSEComments(c))
+
+	c.Request.Header.Del("X-Goog-Api-Client")
+	c.Request.Header.Set("User-Agent", "google-genai-sdk/1.71.0 gl-go/go1.28")
+	require.True(t, downstreamRejectsSSEComments(c))
+
+	require.False(t, downstreamRejectsSSEComments(nil))
+}
+
+// runAntigravityGeminiStreamWithIdle 起一条上游流：先发一个 data 事件，然后空闲 idle 时长再关闭，
+// 返回写给下游的全部字节。用来观察空闲期间网关是否发了 ":\n\n" 心跳。
+func runAntigravityGeminiStreamWithIdle(t *testing.T, userAgent string, idle time.Duration) string {
+	t.Helper()
+	gin.SetMode(gin.TestMode)
+	svc := newAntigravityCompatService(
+		config.GatewayConfig{MaxLineSize: defaultMaxLineSize, StreamKeepaliveInterval: 1},
+		nil,
+	)
+	c, recorder := newAntigravityCompatContext(http.MethodPost, "/v1beta/models/gemini-3.8-flash:streamGenerateContent", nil)
+	if userAgent != "" {
+		c.Request.Header.Set("User-Agent", userAgent)
+	}
+	reader, writer := io.Pipe()
+	resp := &http.Response{StatusCode: http.StatusOK, Header: http.Header{}, Body: reader}
+	done := make(chan error, 1)
+	go func() {
+		_, err := svc.handleGeminiStreamingResponse(c, resp, time.Now())
+		done <- err
+	}()
+	_, err := io.WriteString(
+		writer,
+		`data: {"response":{"responseId":"resp_1","candidates":[{"content":{"parts":[{"text":"partial"}]}}],"usageMetadata":{"promptTokenCount":8,"candidatesTokenCount":1}}}`+"\n\n",
+	)
+	require.NoError(t, err)
+	time.Sleep(idle)
+	require.NoError(t, writer.Close())
+	require.NoError(t, <-done)
+	require.NoError(t, reader.Close())
+	return recorder.Body.String()
+}
+
+func TestAntigravityGeminiStreamKeepsCommentKeepaliveForOrdinaryClients(t *testing.T) {
+	out := runAntigravityGeminiStreamWithIdle(t, "curl/8.7.1", 1200*time.Millisecond)
+	require.Contains(t, out, ":\n\n", "ordinary clients should still get the idle keepalive")
+	require.Contains(t, out, `"text":"partial"`)
+}
+
+func TestAntigravityGeminiStreamSkipsCommentKeepaliveForGoGenai(t *testing.T) {
+	out := runAntigravityGeminiStreamWithIdle(t, "google-genai-sdk/1.71.0 gl-go/go1.28-20260721-RC03", 1200*time.Millisecond)
+	require.Contains(t, out, `"text":"partial"`)
+	for _, event := range strings.Split(out, "\n\n") {
+		require.False(t, strings.HasPrefix(event, ":"), "go-genai must never receive an SSE comment event, got %q", event)
+	}
+}
@@ -41,10 +41,13 @@ func (e GrokMediaEndpoint) RequiresRequestBody() bool {
 }

 func (e GrokMediaEndpoint) IsVideoLookupRequest() bool {
-	return e == GrokMediaEndpointVideoStatus || e == GrokMediaEndpointVideoContent
+	return e == GrokMediaEndpointVideoStatus || e == GrokMediaEndpointVideoContent || e == SeedanceEndpointStatus || e == SeedanceEndpointDelete
 }

 func (e GrokMediaEndpoint) IsGenerationRequest() bool {
+	if e == SeedanceEndpointCreate {
+		return true
+	}
 	switch e {
 	case GrokMediaEndpointImagesGenerations, GrokMediaEndpointImagesEdits, GrokMediaEndpointVideosGenerations, GrokMediaEndpointVideosEdits, GrokMediaEndpointVideosExtensions:
 		return true
@@ -337,6 +340,12 @@ func (s *OpenAIGatewayService) ResolveGrokMediaVideoRequestAccount(
 // the ownership key; video lookups must neither escape nor refresh that key.
 func (s *OpenAIGatewayService) SelectGrokMediaVideoRequestAccount(
 	ctx context.Context, groupID *int64, sessionHash string, accountID int64, requestedModel string,
+) (*AccountSelectionResult, OpenAIAccountScheduleDecision, error) {
+	return s.SelectMediaVideoRequestAccount(ctx, groupID, sessionHash, accountID, requestedModel, PlatformGrok)
+}
+
+func (s *OpenAIGatewayService) SelectMediaVideoRequestAccount(
+	ctx context.Context, groupID *int64, sessionHash string, accountID int64, requestedModel, platform string,
 ) (*AccountSelectionResult, OpenAIAccountScheduleDecision, error) {
 	decision := OpenAIAccountScheduleDecision{Layer: openAIAccountScheduleLayerSessionSticky}
 	if accountID <= 0 || strings.TrimSpace(sessionHash) == "" {
@@ -345,7 +354,7 @@ func (s *OpenAIGatewayService) SelectGrokMediaVideoRequestAccount(
 	ctx = s.withOpenAIGroupPrivacyRequirement(WithOpenAIProfitControlSuppressed(ctx), groupID)
 	scheduler := &defaultOpenAIAccountScheduler{service: s}
 	selection, _, err := scheduler.selectBySessionHash(ctx, OpenAIAccountScheduleRequest{
-		GroupID: groupID, Platform: PlatformGrok, SessionHash: sessionHash,
+		GroupID: groupID, Platform: platform, SessionHash: sessionHash,
 		StickyAccountID: accountID, PreserveStickyBinding: true, DisableStickyEscape: true,
 		RequestedModel: requestedModel, RequiredTransport: OpenAIUpstreamTransportHTTPSSE,
 		RequirePrivacySet: s.openAIGroupRequiresPrivacySet(ctx, groupID),
@@ -961,15 +961,15 @@ func (s *defaultOpenAIAccountScheduler) buildOpenAIAccountLoadPlan(
 		}
 	}

-	// Reset 因子（use-it-or-lose-it）：在拥有「未来会话窗口结束时间」的账号中，
+	// Reset 因子（use-it-or-lose-it）：优先读取 Codex 5h 重置时间，回退到会话窗口。
 	// 剩余时间越短 → 因子越接近 1（越早重置越优先用尽）。无活跃窗口的账号因子为 0。
 	// 仅在 weights.Reset > 0 时计算，默认关闭不影响原有行为。
 	minResetRemaining, maxResetRemaining := 0.0, 0.0
 	hasResetSample := false
 	if weights.Reset > 0 {
 		for _, candidate := range candidates {
-			end := candidate.account.SessionWindowEnd
-			if end == nil || !now.Before(*end) {
+			end, ok := openAISchedulingResetWindowEnd(candidate.account, now)
+			if !ok {
 				continue
 			}
 			remaining := end.Sub(now).Seconds()
@@ -1002,7 +1002,7 @@ func (s *defaultOpenAIAccountScheduler) buildOpenAIAccountLoadPlan(
 		}
 		resetFactor := 0.0
 		if weights.Reset > 0 && hasResetSample {
-			if end := item.account.SessionWindowEnd; end != nil && now.Before(*end) {
+			if end, ok := openAISchedulingResetWindowEnd(item.account, now); ok {
 				if maxResetRemaining > minResetRemaining {
 					resetFactor = 1 - clamp01((end.Sub(now).Seconds()-minResetRemaining)/(maxResetRemaining-minResetRemaining))
 				} else {
@@ -2732,8 +2732,8 @@ func buildOpenAIAccountSchedulerScoreSnapshot(
 	}
 	if weights.Reset > 0 {
 		for _, candidate := range candidates {
-			end := candidate.account.SessionWindowEnd
-			if end == nil || !now.Before(*end) {
+			end, ok := openAISchedulingResetWindowEnd(candidate.account, now)
+			if !ok {
 				continue
 			}
 			remaining := end.Sub(now).Seconds()
@@ -2763,7 +2763,7 @@ func buildOpenAIAccountSchedulerScoreSnapshot(
 		ttftFactor := 0.5
 		resetFactor := 0.0
 		if weights.Reset > 0 && hasResetSample {
-			if end := candidate.account.SessionWindowEnd; end != nil && now.Before(*end) {
+			if end, ok := openAISchedulingResetWindowEnd(candidate.account, now); ok {
 				if maxResetRemaining > minResetRemaining {
 					resetFactor = 1 - clamp01((end.Sub(now).Seconds()-minResetRemaining)/(maxResetRemaining-minResetRemaining))
 				} else {
@@ -2952,22 +2952,89 @@ func openAIQuotaHeadroomFactor(account *Account, now time.Time) float64 {
 	if account == nil || len(account.Extra) == 0 || openAIQuotaHeadroomSnapshotStale(account.Extra, now) {
 		return openAIQuotaHeadroomNeutralFactor
 	}
-	primaryUsedPercent, ok := resolveAccountExtraNumber(account.Extra, "codex_primary_used_percent", "codex_7d_used_percent")
-	if !ok || openAIQuotaWindowResetAny(account.Extra, now, "primary", "7d") {
+	window5h, window7d := openAICanonicalQuotaWindows(account.Extra, now)
+	if !window7d.hasUsed || window7d.reset {
 		return openAIQuotaHeadroomNeutralFactor
 	}

-	factor := 1 - clamp01(primaryUsedPercent/100)
-	if secondaryUsedPercent, ok := resolveAccountExtraNumber(account.Extra, "codex_secondary_used_percent", "codex_5h_used_percent"); ok &&
-		!openAIQuotaWindowResetAny(account.Extra, now, "secondary", "5h") {
-		secondaryRemaining := 1 - clamp01(secondaryUsedPercent/100)
-		if secondaryRemaining < openAIQuotaHeadroomSecondaryLowRemain {
+	factor := 1 - clamp01(window7d.usedPercent/100)
+	if window5h.hasUsed && !window5h.reset {
+		remaining := 1 - clamp01(window5h.usedPercent/100)
+		if remaining < openAIQuotaHeadroomSecondaryLowRemain {
 			factor *= openAIQuotaHeadroomNeutralFactor
 		}
 	}
 	return factor
 }

+type openAICanonicalQuotaWindow struct {
+	usedPercent float64
+	hasUsed     bool
+	reset       bool
+}
+
+// 规范字段优先；历史原始字段复用写入端 Normalize 的窗口分类，不能固定把 primary 当作 7d。
+func openAICanonicalQuotaWindows(extra map[string]any, now time.Time) (window5h, window7d openAICanonicalQuotaWindow) {
+	if used, ok := resolveAccountExtraNumber(extra, "codex_5h_used_percent"); ok {
+		window5h = openAICanonicalQuotaWindow{usedPercent: used, hasUsed: true, reset: openAIQuotaWindowReset(extra, "5h", now)}
+	}
+	if used, ok := resolveAccountExtraNumber(extra, "codex_7d_used_percent"); ok {
+		window7d = openAICanonicalQuotaWindow{usedPercent: used, hasUsed: true, reset: openAIQuotaWindowReset(extra, "7d", now)}
+	}
+	if window5h.hasUsed && window7d.hasUsed {
+		return window5h, window7d
+	}
+
+	snapshot := &OpenAICodexUsageSnapshot{}
+	if used, ok := resolveAccountExtraNumber(extra, "codex_primary_used_percent"); ok {
+		snapshot.PrimaryUsedPercent = &used
+	}
+	if used, ok := resolveAccountExtraNumber(extra, "codex_secondary_used_percent"); ok {
+		snapshot.SecondaryUsedPercent = &used
+	}
+	if minutes := parseExtraInt(extra["codex_primary_window_minutes"]); minutes > 0 {
+		snapshot.PrimaryWindowMinutes = &minutes
+	}
+	if minutes := parseExtraInt(extra["codex_secondary_window_minutes"]); minutes > 0 {
+		snapshot.SecondaryWindowMinutes = &minutes
+	}
+	normalized := snapshot.Normalize()
+	if normalized == nil {
+		return window5h, window7d
+	}
+	fromRaw := func(used *float64) openAICanonicalQuotaWindow {
+		if used == nil {
+			return openAICanonicalQuotaWindow{}
+		}
+		// Normalize 保留原始字段指针，据此配对同一个窗口的用量和重置时间。
+		window := "secondary"
+		if used == snapshot.PrimaryUsedPercent {
+			window = "primary"
+		}
+		return openAICanonicalQuotaWindow{usedPercent: *used, hasUsed: true, reset: openAIQuotaWindowReset(extra, window, now)}
+	}
+	if !window5h.hasUsed {
+		window5h = fromRaw(normalized.Used5hPercent)
+	}
+	if !window7d.hasUsed {
+		window7d = fromRaw(normalized.Used7dPercent)
+	}
+	return window5h, window7d
+}
+
+func openAISchedulingResetWindowEnd(account *Account, now time.Time) (time.Time, bool) {
+	if account == nil {
+		return time.Time{}, false
+	}
+	if end, ok := openAICodexWindowResetAt(account.Extra, "5h"); ok && now.Before(end) {
+		return end, true
+	}
+	if end := account.SessionWindowEnd; end != nil && now.Before(*end) {
+		return *end, true
+	}
+	return time.Time{}, false
+}
+
 func openAIQuotaHeadroomSnapshotStale(extra map[string]any, now time.Time) bool {
 	updatedRaw, ok := extra["codex_usage_updated_at"]
 	if !ok {
@@ -2980,15 +3047,6 @@ func openAIQuotaHeadroomSnapshotStale(extra map[string]any, now time.Time) bool
 	return now.Sub(updatedAt) >= openAIQuotaHeadroomSnapshotStaleAfter
 }

-func openAIQuotaWindowResetAny(extra map[string]any, now time.Time, windows ...string) bool {
-	for _, window := range windows {
-		if openAIQuotaWindowReset(extra, window, now) {
-			return true
-		}
-	}
-	return false
-}
-
 func clamp01(value float64) float64 {
 	switch {
 	case value < 0:
@@ -0,0 +1,141 @@
+package service
+
+import (
+	"context"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/require"
+)
+
+func TestOpenAIQuotaHeadroomFactorUsesCanonicalWindows(t *testing.T) {
+	now := time.Now()
+	shortMins, longMins := 300, 10080
+	shortReset, longReset := 3600, 86400
+	shortUsed, longUsed := 96.0, 25.0
+	for _, primaryIs5h := range []bool{true, false} {
+		name := "primary_is_7d"
+		snapshot := &OpenAICodexUsageSnapshot{
+			PrimaryUsedPercent: &longUsed, PrimaryWindowMinutes: &longMins, PrimaryResetAfterSeconds: &longReset,
+			SecondaryUsedPercent: &shortUsed, SecondaryWindowMinutes: &shortMins, SecondaryResetAfterSeconds: &shortReset,
+		}
+		if primaryIs5h {
+			name = "primary_is_5h"
+			snapshot = &OpenAICodexUsageSnapshot{
+				PrimaryUsedPercent: &shortUsed, PrimaryWindowMinutes: &shortMins, PrimaryResetAfterSeconds: &shortReset,
+				SecondaryUsedPercent: &longUsed, SecondaryWindowMinutes: &longMins, SecondaryResetAfterSeconds: &longReset,
+			}
+		}
+		t.Run(name, func(t *testing.T) {
+			for _, rawOnly := range []bool{false, true} {
+				fields := "canonical_and_raw"
+				if rawOnly {
+					fields = "raw_only"
+				}
+				t.Run(fields, func(t *testing.T) {
+					extra := buildCodexUsageExtraUpdates(snapshot, now)
+					if rawOnly {
+						for key := range extra {
+							if strings.HasPrefix(key, "codex_5h_") || strings.HasPrefix(key, "codex_7d_") {
+								delete(extra, key)
+							}
+						}
+					}
+					// 周余量 75%，5h 余量低于 10% 时沿用现有的减半规则。
+					require.InDelta(t, 0.375, openAIQuotaHeadroomFactor(&Account{Extra: extra}, now), 1e-9)
+				})
+			}
+		})
+	}
+}
+
+func TestOpenAIQuotaHeadroomFactorCanonicalPrecedence(t *testing.T) {
+	now := time.Now()
+	extra := map[string]any{
+		"codex_primary_used_percent":   40.0,
+		"codex_secondary_used_percent": 20.0,
+		"codex_usage_updated_at":       now.Format(time.RFC3339),
+	}
+	// 没有窗口长度的历史数据仍沿用 primary=7d、secondary=5h。
+	require.InDelta(t, 0.6, openAIQuotaHeadroomFactor(&Account{Extra: extra}, now), 1e-9)
+	extra["codex_7d_used_percent"] = 80.0
+	extra["codex_5h_used_percent"] = 10.0
+	require.InDelta(t, 0.2, openAIQuotaHeadroomFactor(&Account{Extra: extra}, now), 1e-9)
+
+	// 规范窗口有效时，不应被另一个原始窗口的过期时间误判为已重置。
+	extra["codex_7d_reset_at"] = now.Add(time.Hour).Format(time.RFC3339)
+	extra["codex_primary_reset_at"] = now.Add(-time.Minute).Format(time.RFC3339)
+	require.InDelta(t, 0.2, openAIQuotaHeadroomFactor(&Account{Extra: extra}, now), 1e-9)
+	extra["codex_7d_reset_at"] = now.Add(-time.Minute).Format(time.RFC3339)
+	require.Equal(t, openAIQuotaHeadroomNeutralFactor, openAIQuotaHeadroomFactor(&Account{Extra: extra}, now))
+}
+
+func TestOpenAIQuotaHeadroomFactorRawWindowReset(t *testing.T) {
+	now := time.Now()
+	extra := map[string]any{
+		"codex_primary_used_percent":   96.0,
+		"codex_primary_window_minutes": 300,
+		"codex_primary_reset_at":       now.Add(-time.Minute).Format(time.RFC3339),
+		"codex_secondary_used_percent": 25.0,
+		"codex_secondary_reset_at":     now.Add(time.Hour).Format(time.RFC3339),
+		"codex_usage_updated_at":       now.Add(-time.Hour).Format(time.RFC3339),
+	}
+	require.InDelta(t, 0.75, openAIQuotaHeadroomFactor(&Account{Extra: extra}, now), 1e-9)
+	extra["codex_secondary_reset_at"] = now.Add(-time.Minute).Format(time.RFC3339)
+	require.Equal(t, openAIQuotaHeadroomNeutralFactor, openAIQuotaHeadroomFactor(&Account{Extra: extra}, now))
+}
+
+func TestOpenAISchedulerCanonicalResetScoresMatchSnapshot(t *testing.T) {
+	now := time.Now()
+	scheduler := openAIResetTestScheduler(1)
+	accounts := []*Account{
+		{ID: 1, Extra: map[string]any{"codex_5h_reset_at": now.Add(10 * time.Minute).Format(time.RFC3339)}},
+		{ID: 2, Extra: map[string]any{"codex_5h_reset_at": now.Add(4 * time.Hour).Format(time.RFC3339)}},
+	}
+	plan := scheduler.buildOpenAIAccountLoadPlan(context.Background(), OpenAIAccountScheduleRequest{}, accounts, nil)
+	scores := openAIPlanScores(plan)
+	require.InDelta(t, 1, scores[1]-scores[2], 1e-9)
+
+	snapshots := buildOpenAIAccountSchedulerScoreSnapshot(accounts, nil, scheduler.service.openAIWSSchedulerWeights(), false, defaultOpenAIOAuthSchedulingRateMultiplier)
+	for _, account := range accounts {
+		require.InDelta(t, scores[account.ID], snapshots[account.ID].BaseScore, 1e-9)
+	}
+}
+
+func TestOpenAISchedulingResetWindowEnd(t *testing.T) {
+	now := time.Date(2026, 9, 1, 12, 0, 0, 0, time.UTC)
+	updated := now.Add(-30 * time.Minute)
+	account := &Account{Extra: map[string]any{
+		"codex_usage_updated_at":       updated.Format(time.RFC3339),
+		"codex_5h_reset_after_seconds": 3600,
+	}}
+	end, ok := openAISchedulingResetWindowEnd(account, now)
+	require.True(t, ok)
+	require.Equal(t, updated.Add(time.Hour), end)
+
+	account.Extra["codex_5h_reset_at"] = now.Add(10 * time.Minute).Format(time.RFC3339)
+	end, ok = openAISchedulingResetWindowEnd(account, now)
+	require.True(t, ok)
+	require.Equal(t, now.Add(10*time.Minute), end)
+
+	delete(account.Extra, "codex_5h_reset_at")
+	account.Extra["codex_5h_reset_after_seconds"] = 60
+	_, ok = openAISchedulingResetWindowEnd(account, now)
+	require.False(t, ok)
+	account.Extra["codex_5h_reset_after_seconds"] = 3600
+	delete(account.Extra, "codex_usage_updated_at")
+	_, ok = openAISchedulingResetWindowEnd(account, now)
+	require.False(t, ok, "缺少采样时间时不能把相对倒计时锚定到当前时间")
+	account.Extra["codex_usage_updated_at"] = "invalid"
+	_, ok = openAISchedulingResetWindowEnd(account, now)
+	require.False(t, ok)
+
+	sessionEnd := now.Add(2 * time.Hour)
+	account.SessionWindowEnd = &sessionEnd
+	end, ok = openAISchedulingResetWindowEnd(account, now)
+	require.True(t, ok)
+	require.Equal(t, sessionEnd, end)
+	_, ok = openAISchedulingResetWindowEnd(nil, now)
+	require.False(t, ok)
+}
@@ -0,0 +1,82 @@
+package service
+
+import (
+	"encoding/json"
+	"errors"
+	"net/url"
+	"strings"
+)
+
+func requiresSystemChatRole(account *Account, targetURL string) bool {
+	if account == nil || account.Type != AccountTypeAPIKey {
+		return false
+	}
+	switch account.Platform {
+	case PlatformDeepseek, PlatformKimi, PlatformZhipu:
+		return true
+	}
+	// OpenAI-compatible accounts can point at the same strict providers. Match
+	// the selected destination's exact hostname, never the requested model ID.
+	u, err := url.Parse(targetURL)
+	if err != nil {
+		return false
+	}
+	switch strings.ToLower(u.Hostname()) {
+	case "api.deepseek.com", "api.kimi.com", "api.moonshot.cn", "api.moonshot.ai", "open.bigmodel.cn", "api.z.ai":
+		return true
+	}
+	return false
+}
+
+// normalizeStrictChatDeveloperRoles adapts native Chat requests only after an
+// account and destination have been selected. Unlike the Responses bridge, it
+// does not reorder/merge messages or downgrade instructions to user messages.
+// RawMessage preserves unknown fields and numeric precision. The input remains
+// untouched so a retry on another account sees the original developer roles.
+func normalizeStrictChatDeveloperRoles(account *Account, targetURL string, body []byte) ([]byte, error) {
+	if !requiresSystemChatRole(account, targetURL) {
+		return body, nil
+	}
+	var root map[string]json.RawMessage
+	if json.Unmarshal(body, &root) != nil || root == nil {
+		return nil, errors.New("chat role normalization: invalid request object")
+	}
+	raw, exists := root["messages"]
+	if !exists {
+		return body, nil
+	}
+	var messages []json.RawMessage
+	if json.Unmarshal(raw, &messages) != nil {
+		return nil, errors.New("chat role normalization: invalid messages array")
+	}
+	changed := false
+	for i, raw := range messages {
+		var message map[string]json.RawMessage
+		if json.Unmarshal(raw, &message) != nil || message == nil {
+			return nil, errors.New("chat role normalization: invalid message object")
+		}
+		var role string
+		if json.Unmarshal(message["role"], &role) != nil {
+			return nil, errors.New("chat role normalization: invalid message role")
+		}
+		if role != "developer" {
+			continue
+		}
+		message["role"] = json.RawMessage(`"system"`)
+		updated, err := json.Marshal(message)
+		if err != nil {
+			return nil, errors.New("chat role normalization: cannot encode message")
+		}
+		messages[i] = updated
+		changed = true
+	}
+	if !changed {
+		return body, nil
+	}
+	updated, err := json.Marshal(messages)
+	if err != nil {
+		return nil, errors.New("chat role normalization: cannot encode messages")
+	}
+	root["messages"] = updated
+	return json.Marshal(root)
+}
@@ -0,0 +1,185 @@
+//go:build unit
+
+package service
+
+import (
+	"bytes"
+	"context"
+	"fmt"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/openai_compat"
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/require"
+	"github.com/tidwall/gjson"
+)
+
+// Only the upstream transport is stubbed. Requests go through the actual Chat
+// dispatcher, model mapping, request construction and response/usage handling.
+type chatRoleContractUpstream struct {
+	httpUpstreamRecorder
+	strict bool
+}
+
+func (u *chatRoleContractUpstream) Do(req *http.Request, proxyURL string, accountID int64, concurrency int) (*http.Response, error) {
+	body, err := io.ReadAll(req.Body)
+	if err != nil {
+		return nil, err
+	}
+	req.Body = io.NopCloser(bytes.NewReader(body))
+	status, contentType := http.StatusOK, "application/json"
+	result := `{"id":"chatcmpl_roles","object":"chat.completion","choices":[{"index":0,"message":{"role":"assistant","content":"pong"},"finish_reason":"stop"}],"usage":{"prompt_tokens":9,"completion_tokens":1,"total_tokens":10}}`
+	if u.strict {
+		for _, message := range gjson.GetBytes(body, "messages").Array() {
+			if message.Get("role").String() == "developer" {
+				status = http.StatusBadRequest
+				result = "{\"error\":{\"message\":\"unknown variant `developer`\",\"type\":\"invalid_request_error\"}}"
+			}
+		}
+	}
+	if status == http.StatusOK && gjson.GetBytes(body, "stream").Bool() {
+		contentType = "text/event-stream"
+		result = "data: {\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"pong\"},\"finish_reason\":\"stop\"}]}\n\ndata: {\"choices\":[],\"usage\":{\"prompt_tokens\":9,\"completion_tokens\":1,\"total_tokens\":10}}\n\ndata: [DONE]\n\n"
+	}
+	u.resp = &http.Response{StatusCode: status, Header: http.Header{"Content-Type": {contentType}}, Body: io.NopCloser(strings.NewReader(result))}
+	return u.httpUpstreamRecorder.Do(req, proxyURL, accountID, concurrency)
+}
+
+func TestForwardAsChatCompletions_StrictDeveloperRole(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	for _, stream := range []bool{false, true} {
+		for _, target := range []struct{ platform, model, base string }{
+			{PlatformDeepseek, "deepseek-flash", "https://api.deepseek.com"},
+			{PlatformKimi, "k3", "https://api.kimi.com/coding"},
+			{PlatformZhipu, "glm-5.3", "https://open.bigmodel.cn/api/paas/v4"},
+			{PlatformOpenAI, "deepseek-flash", "https://api.deepseek.com"},
+			{PlatformOpenAI, "k3", "https://api.kimi.com/coding"},
+			{PlatformOpenAI, "kimi-k3", "https://api.moonshot.ai"},
+			{PlatformOpenAI, "glm-5.3", "https://open.bigmodel.cn/api/paas/v4"},
+		} {
+			t.Run(fmt.Sprintf("%s/%s/%s/stream=%t", target.platform, target.model, target.base, stream), func(t *testing.T) {
+				body := []byte(fmt.Sprintf(`{"model":"example-alias","messages":[{"role":"developer","content":"Keep these instructions."},{"role":"user","content":"ping"}],"stream":%t,"max_tokens":10}`, stream))
+				original := bytes.Clone(body)
+				upstream := &chatRoleContractUpstream{strict: true}
+				svc := &OpenAIGatewayService{cfg: rawChatCompletionsTestConfig(), httpUpstream: upstream}
+				account := rawChatCompletionsTestAccount()
+				account.Platform = target.platform
+				account.Credentials["base_url"] = target.base
+				account.Credentials["api_protocol"] = APIProtocolChatCompletions
+				account.Credentials["model_mapping"] = map[string]any{"example-alias": target.model}
+				account.Extra = map[string]any{openai_compat.ExtraKeyResponsesSupported: false}
+
+				recorder := httptest.NewRecorder()
+				c, _ := gin.CreateTestContext(recorder)
+				c.Request = httptest.NewRequest(http.MethodPost, "/v1/chat/completions", bytes.NewReader(body))
+				result, err := svc.ForwardAsChatCompletions(context.Background(), c, account, body, "", "")
+				require.NoError(t, err)
+				require.NotNil(t, result)
+				require.Equal(t, 9, result.Usage.InputTokens)
+				require.Equal(t, 1, result.Usage.OutputTokens)
+				require.Equal(t, target.model, gjson.GetBytes(upstream.lastBody, "model").String())
+				require.Equal(t, "system", gjson.GetBytes(upstream.lastBody, "messages.0.role").String())
+				require.Equal(t, "Keep these instructions.", gjson.GetBytes(upstream.lastBody, "messages.0.content").String())
+				require.Equal(t, original, body, "per-account adaptation must not mutate the retry body")
+				require.Equal(t, http.StatusOK, recorder.Code)
+				require.Contains(t, recorder.Body.String(), "pong")
+				if stream {
+					require.Contains(t, recorder.Body.String(), "data: [DONE]")
+				}
+
+				// A subsequent attempt with the same body on a compatible account
+				// must still see developer, not the previous account's system role.
+				upstream.strict = false
+				account.Platform = PlatformOpenAI
+				account.Credentials["base_url"] = "https://compatible.example.test/v1"
+				recorder = httptest.NewRecorder()
+				c, _ = gin.CreateTestContext(recorder)
+				c.Request = httptest.NewRequest(http.MethodPost, "/v1/chat/completions", bytes.NewReader(body))
+				_, err = svc.ForwardAsChatCompletions(context.Background(), c, account, body, "", "")
+				require.NoError(t, err)
+				require.Equal(t, "developer", gjson.GetBytes(upstream.lastBody, "messages.0.role").String())
+			})
+		}
+	}
+}
+
+func TestNormalizeStrictChatDeveloperRoles_PreservesConversation(t *testing.T) {
+	body := []byte(`{"model":"alias","messages":[{"role":"system","content":"first"},{"role":"developer","content":[{"type":"text","text":"developer is just text","cache_control":{"type":"ephemeral"}}],"name":"rules"},{"role":"user","content":"ping"},{"role":"assistant","content":null,"tool_calls":[{"id":"call_1","type":"function","function":{"name":"echo","arguments":"{\"role\":\"developer\"}"}}]},{"role":"tool","tool_call_id":"call_1","content":"ok"},{"role":"developer","content":"last"}],"metadata":{"large":9007199254740993,"role":"developer"},"stream":true,"reasoning_effort":"high"}`)
+	original := bytes.Clone(body)
+	want := strings.ReplaceAll(string(body), `"role":"developer","content"`, `"role":"system","content"`)
+	for _, platform := range []string{PlatformDeepseek, PlatformKimi, PlatformZhipu} {
+		t.Run(platform, func(t *testing.T) {
+			account := &Account{Platform: platform, Type: AccountTypeAPIKey}
+			got, err := normalizeStrictChatDeveloperRoles(account, "https://relay.example.test/v1/chat/completions", body)
+			require.NoError(t, err)
+			require.JSONEq(t, want, string(got))
+			require.Contains(t, string(got), "9007199254740993")
+			require.Equal(t, original, body)
+			again, err := normalizeStrictChatDeveloperRoles(account, "https://relay.example.test/v1/chat/completions", got)
+			require.NoError(t, err)
+			require.Equal(t, got, again)
+		})
+	}
+}
+
+func TestNormalizeStrictChatDeveloperRoles_ExactOfficialHosts(t *testing.T) {
+	body := []byte(`{"messages":[{"role":"developer","content":"rules"}]}`)
+	account := &Account{Platform: PlatformOpenAI, Type: AccountTypeAPIKey}
+	for _, host := range []string{"api.deepseek.com", "api.kimi.com", "api.moonshot.cn", "api.moonshot.ai", "open.bigmodel.cn", "api.z.ai", "API.DEEPSEEK.COM:443"} {
+		t.Run(host, func(t *testing.T) {
+			got, err := normalizeStrictChatDeveloperRoles(account, "https://"+host+"/v1/chat/completions", body)
+			require.NoError(t, err)
+			require.JSONEq(t, `{"messages":[{"role":"system","content":"rules"}]}`, string(got))
+		})
+	}
+}
+
+func TestNormalizeStrictChatDeveloperRoles_OtherTargetsUntouched(t *testing.T) {
+	// Neither model names nor host-like strings in the path/userinfo may enable
+	// normalization on an otherwise compatible provider.
+	body := []byte("{ \"model\": \"deepseek-flash\", \"messages\": [{\"role\":\"developer\",\"content\":\"rules\"}] }")
+	account := &Account{Platform: PlatformOpenAI, Type: AccountTypeAPIKey}
+	for _, target := range []string{
+		"https://api.openai.com/v1/chat/completions",
+		"https://relay.example.test/v1/chat/completions",
+		"https://api.deepseek.com.example.test/v1/chat/completions",
+		"https://example.test/api.kimi.com/chat/completions",
+		"https://api.deepseek.com@example.test/v1/chat/completions",
+		"://invalid-url",
+	} {
+		got, err := normalizeStrictChatDeveloperRoles(account, target, body)
+		require.NoError(t, err)
+		require.Equal(t, body, got)
+	}
+	for _, account := range []*Account{nil, {Platform: PlatformKimi, Type: AccountTypeOAuth}, {Platform: PlatformOpenAI, Type: AccountTypeOAuth}} {
+		got, err := normalizeStrictChatDeveloperRoles(account, "https://api.kimi.com/coding/v1/chat/completions", body)
+		require.NoError(t, err)
+		require.Equal(t, body, got)
+	}
+}
+
+func TestNormalizeStrictChatDeveloperRoles_CompatibleBodyUntouched(t *testing.T) {
+	account := &Account{Platform: PlatformDeepseek, Type: AccountTypeAPIKey}
+	for _, body := range []string{
+		"{ \"messages\": [{\"role\":\"system\",\"content\":\"rules\"}] }",
+		`{"messages":[]}`,
+		`{"input":[{"role":"developer","content":"Responses is not Chat"}]}`,
+	} {
+		got, err := normalizeStrictChatDeveloperRoles(account, "https://api.deepseek.com/v1/chat/completions", []byte(body))
+		require.NoError(t, err)
+		require.Equal(t, body, string(got))
+	}
+}
+
+func TestNormalizeStrictChatDeveloperRoles_MalformedBodyDoesNotLeak(t *testing.T) {
+	account := &Account{Platform: PlatformDeepseek, Type: AccountTypeAPIKey}
+	for _, body := range []string{`{"secret":`, `null`, `[]`, `{"messages":"secret"}`, `{"messages":["secret"]}`, `{"messages":[null]}`, `{"messages":[{"content":"secret"}]}`, `{"messages":[{"role":12,"content":"secret"}]}`} {
+		_, err := normalizeStrictChatDeveloperRoles(account, "https://api.deepseek.com/v1/chat/completions", []byte(body))
+		require.Error(t, err)
+		require.NotContains(t, err.Error(), "secret")
+	}
+}
@@ -2473,13 +2473,9 @@ func validateCodexModelsManifestEnvelope(body []byte) error {
 		return errors.New("missing top-level models array")
 	}
 	models = bytes.TrimSpace(models)
-	var entries []json.RawMessage
 	if len(models) == 0 || models[0] != '[' {
 		return errors.New("top-level models field is not an array")
 	}
-	if err := json.Unmarshal(models, &entries); err != nil {
-		return fmt.Errorf("decode top-level models array: %w", err)
-	}
 	return nil
 }

@@ -2562,6 +2562,11 @@ func TestFetchCodexModelsManifestRejectsInvalidEnvelope(t *testing.T) {
 	}
 }

+func TestValidateCodexModelsManifestEnvelopeRequiresExactModelsKey(t *testing.T) {
+	require.NoError(t, validateCodexModelsManifestEnvelope([]byte(`{"models":[]}`)))
+	require.Error(t, validateCodexModelsManifestEnvelope([]byte(`{"Models":[]}`)))
+}
+
 func TestFetchCodexModelsManifestAPIKeyDoesNotCacheInvalidEnvelope(t *testing.T) {
 	var calls atomic.Int32
 	upstream := &codexModelsHTTPUpstreamStub{do: func(_ *http.Request, _ string, _ int64, _ int) (*http.Response, error) {
@@ -1228,7 +1228,15 @@ func normalizeOpenAIModelForUpstream(account *Account, model string) string {
 	if account == nil || account.UsesOpenAICodexProtocol() {
 		return normalizeCodexModel(model)
 	}
-	return strings.TrimSpace(model)
+	model = strings.TrimSpace(model)
+	if account.Platform == PlatformDeepseek {
+		// DeepSeek 走 OpenAI 兼容入口，不经 Anthropic 入站的 [1m] 后缀归一
+		// （parseGatewayRequestCurrentBody 仅处理 PlatformAnthropic 协议）。
+		// 官方 Claude Code 接入要求 ANTHROPIC_MODEL=deepseek-flash[1m] 写法，
+		// 出站前剥离泄漏的客户端上下文后缀，转发规范名给上游。
+		return normalizeClaudeCodeLongContextModel(model)
+	}
+	return model
 }

 func SupportsVerbosity(model string) bool {
@@ -114,7 +114,8 @@ func writeOpenAICompactSSEFailureMessage(c *gin.Context, statusCode int, errType
 	}
 	MarkOpsStreamError(c, errType, message, statusCode)
 	payload, err := json.Marshal(map[string]any{
-		"type": "response.failed",
+		"type":            "response.failed",
+		"sequence_number": 0,
 		"response": map[string]any{
 			"id":     "resp_" + strings.ReplaceAll(uuid.NewString(), "-", ""),
 			"object": "response",
@@ -143,7 +144,8 @@ func writeOpenAICompactSSEFailureMessage(c *gin.Context, statusCode int, errType
 // 携带完整 response 对象。Codex 的 SSE 解析只从 output_item.done 收集 item，
 // 并要求 response.completed 的 response.id 必填、usage（若存在）必须携带
 // input_tokens/output_tokens/total_tokens 整数字段，否则整条 completed 事件
-// 解析失败，故此处做兜底修补。
+// 解析失败，故此处做兜底修补。每帧还要带单调的 sequence_number：grok-build
+// 把它当必填，缺了整轮反序列化失败。
 func buildOpenAICompactSSEPayload(finalResponse []byte) ([]byte, bool) {
 	if len(finalResponse) == 0 || !gjson.ValidBytes(finalResponse) {
 		return nil, false
@@ -176,12 +178,19 @@ func buildOpenAICompactSSEPayload(finalResponse []byte) ([]byte, bool) {

 	var buf bytes.Buffer
 	outputIndex := 0
-	appendEvent := func(eventType string, data []byte) {
+	sequenceNumber := 0
+	appendEvent := func(eventType string, data []byte) bool {
+		numbered, err := sjson.SetBytes(data, "sequence_number", sequenceNumber)
+		if err != nil {
+			return false
+		}
+		sequenceNumber++
 		_, _ = buf.WriteString("event: ")
 		_, _ = buf.WriteString(eventType)
 		_, _ = buf.WriteString("\ndata: ")
-		_, _ = buf.Write(data)
+		_, _ = buf.Write(numbered)
 		_, _ = buf.WriteString("\n\n")
+		return true
 	}
 	for _, item := range gjson.GetBytes(response, "output").Array() {
 		if !item.IsObject() {
@@ -195,7 +204,9 @@ func buildOpenAICompactSSEPayload(finalResponse []byte) ([]byte, bool) {
 		if err != nil {
 			return nil, false
 		}
-		appendEvent("response.output_item.done", event)
+		if !appendEvent("response.output_item.done", event) {
+			return nil, false
+		}
 		outputIndex++
 	}

@@ -203,7 +214,9 @@ func buildOpenAICompactSSEPayload(finalResponse []byte) ([]byte, bool) {
 	if err != nil {
 		return nil, false
 	}
-	appendEvent("response.completed", completed)
+	if !appendEvent("response.completed", completed) {
+		return nil, false
+	}
 	return buf.Bytes(), true
 }

@@ -73,6 +73,8 @@ func TestBuildOpenAICompactSSEPayload_EmitsItemsAndCompleted(t *testing.T) {
 	require.Equal(t, "response.output_item.done", events[0][0])
 	first := events[0][1]
 	require.Equal(t, "response.output_item.done", gjson.Get(first, "type").String())
+	require.True(t, gjson.Get(first, "sequence_number").Exists(), "compact 合成帧必须带 sequence_number")
+	require.Equal(t, int64(0), gjson.Get(first, "sequence_number").Int())
 	require.Equal(t, int64(0), gjson.Get(first, "output_index").Int())
 	require.Equal(t, "compaction", gjson.Get(first, "item.type").String())
 	require.Equal(t, "cmp_1", gjson.Get(first, "item.id").String())
@@ -81,12 +83,14 @@ func TestBuildOpenAICompactSSEPayload_EmitsItemsAndCompleted(t *testing.T) {
 	require.True(t, gjson.Get(first, "item.opaque.kept").Bool(), "item 原始字段必须逐字节保留")

 	require.Equal(t, "response.output_item.done", events[1][0])
+	require.Equal(t, int64(1), gjson.Get(events[1][1], "sequence_number").Int())
 	require.Equal(t, int64(1), gjson.Get(events[1][1], "output_index").Int())
 	require.Equal(t, "message", gjson.Get(events[1][1], "item.type").String())

 	require.Equal(t, "response.completed", events[2][0])
 	completed := events[2][1]
 	require.Equal(t, "response.completed", gjson.Get(completed, "type").String())
+	require.Equal(t, int64(2), gjson.Get(completed, "sequence_number").Int())
 	require.Equal(t, "resp_compact_1", gjson.Get(completed, "response.id").String())
 	require.Equal(t, int64(13), gjson.Get(completed, "response.usage.total_tokens").Int())
 	require.Len(t, gjson.Get(completed, "response.output").Array(), 2)
@@ -31,8 +31,9 @@ func TestWriteOpenAICompactSSEFailureMessage_CarriesCreatedAt(t *testing.T) {
 	require.True(t, found, "SSE 帧必须带 data 行: %q", body)

 	var event struct {
-		Type     string `json:"type"`
-		Response struct {
+		Type           string `json:"type"`
+		SequenceNumber *int   `json:"sequence_number"`
+		Response       struct {
 			ID        string `json:"id"`
 			Object    string `json:"object"`
 			CreatedAt int64  `json:"created_at"`
@@ -42,6 +43,8 @@ func TestWriteOpenAICompactSSEFailureMessage_CarriesCreatedAt(t *testing.T) {
 	require.NoError(t, json.Unmarshal([]byte(strings.TrimSpace(payload)), &event))

 	require.Equal(t, "response.failed", event.Type)
+	require.NotNil(t, event.SequenceNumber, "response.failed 必须带 sequence_number，否则 grok-build 读不出这帧")
+	require.GreaterOrEqual(t, *event.SequenceNumber, 0)
 	require.Equal(t, "response", event.Response.Object)
 	require.Equal(t, "failed", event.Response.Status)
 	require.Greater(t, event.Response.CreatedAt, int64(0),
@@ -182,6 +182,11 @@ func (s *OpenAIGatewayService) sendCCUpstreamRequest(
 	userAgent string,
 	grokCacheIdentity string,
 ) (*http.Response, error) {
+	// DeepSeek thinking mode 要求历史 assistant 回传 reasoning_content。
+	// Responses→CC 回退在加密-only / 缺 reasoning item 且缓存未命中时会漏掉该
+	// 字段，上游 400 "The `reasoning_content` in the thinking mode must be
+	// passed back to the API"。在共用出站点补空格占位，真实明文不覆盖。
+	body = ensureDeepSeekChatReasoningPlaceholders(account, body)
 	upstreamCtx, releaseUpstreamCtx := detachUpstreamContext(ctx)
 	upstreamReq, err := http.NewRequestWithContext(upstreamCtx, http.MethodPost, targetURL, bytes.NewReader(body))
 	releaseUpstreamCtx()
@@ -178,6 +178,11 @@ func (s *OpenAIGatewayService) forwardAsRawChatCompletions(
 	if err != nil {
 		return nil, err
 	}
+	upstreamBody, err = normalizeStrictChatDeveloperRoles(account, targetURL, upstreamBody)
+	if err != nil {
+		writeChatCompletionsError(c, http.StatusBadRequest, "invalid_request_error", err.Error())
+		return nil, err
+	}
 	SetActualOpenAIUpstreamEndpoint(c, grokChatRawEndpoint)
 	customUA := account.GetOpenAIUserAgent()
 	if customUA == "" && account.IsGrokOAuth() {
@@ -0,0 +1,193 @@
+package service
+
+import (
+	"bytes"
+	"context"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/openai_compat"
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/require"
+	"github.com/tidwall/gjson"
+)
+
+func openaiPlatformDeepSeekAccount() *Account {
+	return &Account{
+		ID:       18,
+		Name:     "openai-deepseek-apikey",
+		Platform: PlatformOpenAI,
+		Type:     AccountTypeAPIKey,
+		Extra: map[string]any{
+			openai_compat.ExtraKeyResponsesMode: string(openai_compat.ResponsesSupportModeForceChatCompletions),
+		},
+		Credentials: map[string]any{
+			"api_key":  "sk-test",
+			"base_url": DefaultDeepseekBaseURL,
+		},
+		Status:      StatusActive,
+		Schedulable: true,
+	}
+}
+
+func deepSeekChatFallbackTestConfig() *config.Config {
+	return &config.Config{
+		Security: config.SecurityConfig{
+			URLAllowlist: config.URLAllowlistConfig{
+				Enabled:           false,
+				AllowInsecureHTTP: true,
+			},
+		},
+	}
+}
+
+func TestEnsureDeepSeekChatReasoningPlaceholders(t *testing.T) {
+	deepSeekAccount := openaiPlatformDeepSeekAccount()
+	nativeDeepSeek := &Account{Platform: PlatformDeepseek, Type: AccountTypeAPIKey}
+	otherOpenAI := &Account{
+		Platform:    PlatformOpenAI,
+		Type:        AccountTypeAPIKey,
+		Credentials: map[string]any{"api_key": "sk-test", "base_url": "http://upstream.example"},
+	}
+
+	missing := []byte(`{"model":"deepseek-chat","messages":[{"role":"user","content":"hi"},{"role":"assistant","content":"","tool_calls":[{"id":"call_1","type":"function","function":{"name":"exec","arguments":"{}"}}]},{"role":"tool","tool_call_id":"call_1","content":"ok"}]}`)
+	withPlain := []byte(`{"model":"deepseek-chat","messages":[{"role":"assistant","reasoning_content":"real thinking","content":"ok"}]}`)
+
+	t.Run("openai_platform_deepseek_fills_empty_assistant", func(t *testing.T) {
+		got := ensureDeepSeekChatReasoningPlaceholders(deepSeekAccount, missing)
+		require.Equal(t, deepSeekChatReasoningPlaceholderText, gjson.GetBytes(got, "messages.1.reasoning_content").String())
+		require.False(t, gjson.GetBytes(got, "messages.0.reasoning_content").Exists())
+		require.False(t, gjson.GetBytes(got, "messages.2.reasoning_content").Exists())
+	})
+
+	t.Run("native_deepseek_platform_fills_empty_assistant", func(t *testing.T) {
+		got := ensureDeepSeekChatReasoningPlaceholders(nativeDeepSeek, missing)
+		require.Equal(t, deepSeekChatReasoningPlaceholderText, gjson.GetBytes(got, "messages.1.reasoning_content").String())
+	})
+
+	t.Run("does_not_overwrite_plaintext", func(t *testing.T) {
+		got := ensureDeepSeekChatReasoningPlaceholders(deepSeekAccount, withPlain)
+		require.Equal(t, "real thinking", gjson.GetBytes(got, "messages.0.reasoning_content").String())
+		require.Equal(t, string(withPlain), string(got), "已有明文时必须原样返回")
+	})
+
+	t.Run("non_deepseek_upstream_unchanged", func(t *testing.T) {
+		got := ensureDeepSeekChatReasoningPlaceholders(otherOpenAI, missing)
+		require.Equal(t, string(missing), string(got))
+		require.False(t, gjson.GetBytes(got, "messages.1.reasoning_content").Exists())
+	})
+
+	t.Run("nil_account_unchanged", func(t *testing.T) {
+		got := ensureDeepSeekChatReasoningPlaceholders(nil, missing)
+		require.Equal(t, string(missing), string(got))
+	})
+}
+
+type reasoningHitCache struct {
+	stubGatewayCache
+	values map[string]string
+}
+
+func (c *reasoningHitCache) GetReasoningContent(_ context.Context, itemID string) (string, error) {
+	if v, ok := c.values[itemID]; ok {
+		return v, nil
+	}
+	return "", ErrReasoningContentNotFound
+}
+
+func deepSeekChatHistoryWithEncryptedReasoning() []byte {
+	return []byte(`{
+		"model":"gpt-5.6-sol",
+		"stream":false,
+		"input":[
+			{"type":"reasoning","id":"item_enc1","summary":[],"encrypted_content":"opaque"},
+			{"type":"function_call","call_id":"call_1","name":"exec","arguments":"{}"},
+			{"type":"function_call_output","call_id":"call_1","output":"ok"},
+			{"type":"message","role":"user","content":[{"type":"input_text","text":"go on"}]}
+		]
+	}`)
+}
+
+func newDeepSeekChatFallbackContext(t *testing.T, body []byte) *gin.Context {
+	t.Helper()
+	gin.SetMode(gin.TestMode)
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/responses", bytes.NewReader(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+	return c
+}
+
+func newOKChatCompletionsUpstream(requestID, body string) *httpUpstreamRecorder {
+	return &httpUpstreamRecorder{resp: &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"application/json"}, "x-request-id": []string{requestID}},
+		Body:       io.NopCloser(strings.NewReader(body)),
+	}}
+}
+
+const deepSeekChatFallbackOKBody = `{"id":"chatcmpl_ph","object":"chat.completion","model":"deepseek-chat","choices":[{"index":0,"message":{"role":"assistant","content":"ok"},"finish_reason":"stop"}],"usage":{"prompt_tokens":3,"completion_tokens":2,"total_tokens":5}}`
+
+func TestForwardResponses_DeepSeekChatFallbackInjectsReasoningPlaceholderOnCacheMiss(t *testing.T) {
+	body := deepSeekChatHistoryWithEncryptedReasoning()
+	c := newDeepSeekChatFallbackContext(t, body)
+	upstream := newOKChatCompletionsUpstream("rid_ds_rc_placeholder", deepSeekChatFallbackOKBody)
+	svc := &OpenAIGatewayService{
+		cfg:          deepSeekChatFallbackTestConfig(),
+		httpUpstream: upstream,
+	}
+
+	result, err := svc.Forward(context.Background(), c, openaiPlatformDeepSeekAccount(), body)
+	require.NoError(t, err)
+	require.NotNil(t, result)
+	require.Contains(t, upstream.lastReq.URL.String(), "/chat/completions")
+	require.Equal(t, deepSeekChatReasoningPlaceholderText, gjson.GetBytes(upstream.lastBody, "messages.0.reasoning_content").String())
+	require.Equal(t, "call_1", gjson.GetBytes(upstream.lastBody, "messages.0.tool_calls.0.id").String())
+}
+
+func TestForwardResponses_DeepSeekChatFallbackKeepsCachedReasoningContent(t *testing.T) {
+	body := deepSeekChatHistoryWithEncryptedReasoning()
+	c := newDeepSeekChatFallbackContext(t, body)
+	upstream := newOKChatCompletionsUpstream("rid_ds_rc_cached", deepSeekChatFallbackOKBody)
+	svc := &OpenAIGatewayService{
+		cfg:          deepSeekChatFallbackTestConfig(),
+		httpUpstream: upstream,
+		cache:        &reasoningHitCache{values: map[string]string{"item_enc1": "cached thinking"}},
+	}
+
+	result, err := svc.Forward(context.Background(), c, openaiPlatformDeepSeekAccount(), body)
+	require.NoError(t, err)
+	require.NotNil(t, result)
+	require.Equal(t, "cached thinking", gjson.GetBytes(upstream.lastBody, "messages.0.reasoning_content").String())
+}
+
+func TestForwardResponses_NonDeepSeekChatFallbackDoesNotInjectReasoningPlaceholder(t *testing.T) {
+	body := deepSeekChatHistoryWithEncryptedReasoning()
+	c := newDeepSeekChatFallbackContext(t, body)
+	upstream := newOKChatCompletionsUpstream("rid_other_rc", deepSeekChatFallbackOKBody)
+	account := &Account{
+		ID:       99,
+		Platform: PlatformOpenAI,
+		Type:     AccountTypeAPIKey,
+		Extra: map[string]any{
+			openai_compat.ExtraKeyResponsesMode: string(openai_compat.ResponsesSupportModeForceChatCompletions),
+		},
+		Credentials: map[string]any{
+			"api_key":  "sk-test",
+			"base_url": "http://upstream.example",
+		},
+	}
+	svc := &OpenAIGatewayService{
+		cfg:          deepSeekChatFallbackTestConfig(),
+		httpUpstream: upstream,
+	}
+
+	result, err := svc.Forward(context.Background(), c, account, body)
+	require.NoError(t, err)
+	require.NotNil(t, result)
+	require.False(t, gjson.GetBytes(upstream.lastBody, "messages.0.reasoning_content").Exists())
+}
@@ -10,6 +10,7 @@ import (
 	"strings"
 	"unicode/utf8"

+	"github.com/Wei-Shaw/sub2api/internal/pkg/apicompat"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/ctxkey"
 	"github.com/Wei-Shaw/sub2api/internal/util/urlvalidator"
 	"github.com/gin-gonic/gin"
@@ -141,7 +142,25 @@ func normalizeDeepSeekResponsesRequestBody(account *Account, body []byte) []byte
 	if stripped, err := sjson.DeleteBytes(normalized, "previous_response_id"); err == nil {
 		normalized = stripped
 	}
-	return normalized
+
+	var requestBody map[string]any
+	if err := decodeOpenAIJSONUseNumber(normalized, &requestBody); err != nil {
+		return normalized
+	}
+	input, exists := requestBody["input"]
+	if !exists {
+		return normalized
+	}
+	liftedInput, changed := apicompat.LiftResponsesToolOutputMedia(input)
+	if !changed {
+		return normalized
+	}
+	requestBody["input"] = liftedInput
+	rebuilt, err := marshalOpenAIUpstreamJSON(requestBody)
+	if err != nil {
+		return normalized
+	}
+	return rebuilt
 }

 func trimOpenAIEncryptedReasoningItems(reqBody map[string]any) bool {
@@ -1464,12 +1464,23 @@ func (s *OpenAIGatewayService) bindHTTPResponseAccount(ctx context.Context, c *g
 	if store == nil {
 		return
 	}
+	// The client may close the stream immediately after receiving its terminal
+	// event, canceling the request context before these durable affinity writes
+	// run. Preserve request values, but give all Redis writes one bounded budget
+	// independent of the downstream connection lifecycle.
+	bindBaseCtx := context.Background()
+	if ctx != nil {
+		bindBaseCtx = context.WithoutCancel(ctx)
+	}
+	bindCtx, cancel := context.WithTimeout(bindBaseCtx, openAIWSStateStoreRedisTimeout)
+	defer cancel()
+
 	groupID := getOpenAIGroupIDFromContext(c)
 	ttl := s.openAIWSResponseStickyTTL()
-	logOpenAIWSBindResponseAccountWarn(groupID, account.ID, responseID, store.BindResponseAccount(ctx, groupID, responseID, account.ID, ttl))
+	logOpenAIWSBindResponseAccountWarn(groupID, account.ID, responseID, store.BindResponseAccount(bindCtx, groupID, responseID, account.ID, ttl))
 	if rawOwner, ok := c.Get(openAIHTTPResponseOwnerContextKey); ok {
 		if owner, ok := rawOwner.(openAIHTTPResponseOwner); ok && owner.userID > 0 && owner.apiKeyID > 0 {
-			if err := s.BindOpenAIHTTPResponseOwner(ctx, groupID, responseID, owner.userID, owner.apiKeyID); err != nil {
+			if err := s.BindOpenAIHTTPResponseOwner(bindCtx, groupID, responseID, owner.userID, owner.apiKeyID); err != nil {
 				logger.L().Warn(
 					"openai.http_bind_response_owner_failed",
 					zap.Int64("group_id", groupID),
@@ -7,6 +7,8 @@ import (
 	"errors"
 	"fmt"
 	"net/http"
+	"net/url"
+	"strconv"
 	"strings"
 	"time"

@@ -14,6 +16,8 @@ import (
 	"github.com/Wei-Shaw/sub2api/internal/pkg/logger"
 	"github.com/Wei-Shaw/sub2api/internal/util/responseheaders"
 	"github.com/gin-gonic/gin"
+	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
 	"go.uber.org/zap"
 )

@@ -399,3 +403,67 @@ func (s *OpenAIGatewayService) setReasoningContent(itemID, content string) {
 		)
 	}
 }
+
+// deepSeekChatReasoningPlaceholderText 是 Chat Completions 侧 thinking-mode
+// 占位明文。必须是单个空格：DeepSeek 拒绝空串，非空即可通过；LiteLLM 同样注入
+// 单个空格。
+const deepSeekChatReasoningPlaceholderText = " "
+
+// targetsDeepSeekAPIHost 报告该账号实际上游是否是 DeepSeek 官方 API。
+// platform=deepseek 直接命中；platform=openai 但 base_url 指向
+// api.deepseek.com 的映射账号同样命中（Codex 把 GPT 模型名映射到 DeepSeek
+// 时的典型接入方式）。
+func targetsDeepSeekAPIHost(account *Account) bool {
+	if account == nil {
+		return false
+	}
+	if account.Platform == PlatformDeepseek {
+		return true
+	}
+	u, err := url.Parse(strings.TrimSpace(account.GetOpenAIBaseURL()))
+	if err != nil {
+		return false
+	}
+	ds, err := url.Parse(DefaultDeepseekBaseURL)
+	if err != nil {
+		return false
+	}
+	return strings.EqualFold(u.Hostname(), ds.Hostname())
+}
+
+// ensureDeepSeekChatReasoningPlaceholders 给缺 reasoning_content 的 assistant
+// 消息补单个空格占位。DeepSeek thinking mode 要求历史里每条产生过思维的
+// assistant 消息都回传该字段，否则 400
+// "The `reasoning_content` in the thinking mode must be passed back to the API"。
+//
+// 桥接会从 summary / 缓存回注真实明文；这里只填仍为空的缺口，不覆盖已有内容。
+// 非 DeepSeek 上游原样返回（字节不变）。
+func ensureDeepSeekChatReasoningPlaceholders(account *Account, body []byte) []byte {
+	if !targetsDeepSeekAPIHost(account) {
+		return body
+	}
+	messages := gjson.GetBytes(body, "messages")
+	if !messages.IsArray() {
+		return body
+	}
+	updated := body
+	changed := false
+	for i, msg := range messages.Array() {
+		if strings.TrimSpace(msg.Get("role").String()) != "assistant" {
+			continue
+		}
+		if msg.Get("reasoning_content").String() != "" {
+			continue
+		}
+		next, err := sjson.SetBytes(updated, "messages."+strconv.Itoa(i)+".reasoning_content", deepSeekChatReasoningPlaceholderText)
+		if err != nil {
+			return body
+		}
+		updated = next
+		changed = true
+	}
+	if !changed {
+		return body
+	}
+	return updated
+}
@@ -710,26 +710,29 @@ func openAICodexSnapshotStaleForPause(extra map[string]any, now time.Time) bool
 // timestamp and falls back to codex_<window>_reset_after_seconds anchored at
 // codex_usage_updated_at, mirroring AccountUsageService's window-progress logic.
 func openAIQuotaWindowReset(extra map[string]any, window string, now time.Time) bool {
+	resetAt, ok := openAICodexWindowResetAt(extra, window)
+	return ok && !now.Before(resetAt)
+}
+
+// 绝对时间优先；相对倒计时必须锚定快照采样时间，不能随每次评分向后滑动。
+func openAICodexWindowResetAt(extra map[string]any, window string) (time.Time, bool) {
 	if len(extra) == 0 {
-		return false
+		return time.Time{}, false
 	}
 	if resetAtRaw, ok := extra["codex_"+window+"_reset_at"]; ok {
 		if resetAt, err := parseTime(fmt.Sprint(resetAtRaw)); err == nil {
-			return !now.Before(resetAt)
+			return resetAt, true
 		}
 	}
 	resetAfter := parseExtraInt(extra["codex_"+window+"_reset_after_seconds"])
 	if resetAfter <= 0 {
-		return false
+		return time.Time{}, false
 	}
-	base := now
-	if updatedRaw, ok := extra["codex_usage_updated_at"]; ok {
-		if updatedAt, err := parseTime(fmt.Sprint(updatedRaw)); err == nil {
-			base = updatedAt
-		}
+	updatedAt, err := parseTime(fmt.Sprint(extra["codex_usage_updated_at"]))
+	if err != nil {
+		return time.Time{}, false
 	}
-	resetAt := base.Add(time.Duration(resetAfter) * time.Second)
-	return !now.Before(resetAt)
+	return updatedAt.Add(time.Duration(resetAfter) * time.Second), true
 }

 func readOpenAIQuotaUsedPercent(extra map[string]any, window string) float64 {
@@ -589,6 +589,37 @@ func TestOpenAIGatewayService_BindHTTPResponseAccount(t *testing.T) {
 	require.False(t, owned)
 }

+func TestOpenAIGatewayService_BindHTTPResponseAccount_DetachesCanceledRequestContext(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/openai/v1/responses", nil)
+	groupID := int64(4201)
+	c.Set("api_key", &APIKey{ID: 501, GroupID: &groupID})
+	SetOpenAIHTTPResponseOwner(c, 601, 501)
+
+	cache := &responseBindContextProbeCache{}
+	svc := &OpenAIGatewayService{cache: cache}
+	account := &Account{ID: 37001, Platform: PlatformOpenAI, Type: AccountTypeAPIKey}
+	requestCtx, cancelRequest := context.WithCancel(context.Background())
+	cancelRequest()
+	startedAt := time.Now()
+
+	svc.bindHTTPResponseAccount(requestCtx, c, account, "resp_http_canceled_001")
+
+	require.Len(t, cache.setContextErrors, 3)
+	for _, contextErr := range cache.setContextErrors {
+		require.NoError(t, contextErr, "response affinity writes must survive downstream cancellation")
+	}
+	require.Len(t, cache.setDeadlines, 3)
+	for _, deadline := range cache.setDeadlines {
+		require.True(t, deadline.After(startedAt))
+		require.LessOrEqual(t, deadline.Sub(startedAt), openAIWSStateStoreRedisTimeout+100*time.Millisecond)
+	}
+	require.Len(t, cache.sessionBindings, 3)
+	require.Contains(t, cache.sessionBindings, openAIWSResponseAccountCacheKey("resp_http_canceled_001"))
+}
+
 func TestOpenAIGatewayService_GenerateExplicitSessionHash_SkipsContentFallback(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 	svc := &OpenAIGatewayService{}
@@ -710,6 +741,23 @@ type stubGatewayCache struct {
 	deletedSessions map[string]int
 }

+type responseBindContextProbeCache struct {
+	stubGatewayCache
+	setContextErrors []error
+	setDeadlines     []time.Time
+}
+
+func (c *responseBindContextProbeCache) SetSessionAccountID(ctx context.Context, groupID int64, sessionHash string, accountID int64, ttl time.Duration) error {
+	c.setContextErrors = append(c.setContextErrors, ctx.Err())
+	if deadline, ok := ctx.Deadline(); ok {
+		c.setDeadlines = append(c.setDeadlines, deadline)
+	}
+	if err := ctx.Err(); err != nil {
+		return err
+	}
+	return c.stubGatewayCache.SetSessionAccountID(ctx, groupID, sessionHash, accountID, ttl)
+}
+
 func (c *stubGatewayCache) GetSessionAccountID(ctx context.Context, groupID int64, sessionHash string) (int64, error) {
 	if id, ok := c.sessionBindings[sessionHash]; ok {
 		return id, nil
@@ -82,6 +82,43 @@ func isOpenAIOAuthServableModel(requestedModel string) bool {
 	return true
 }

+// deepseekServableModels 列出 DeepSeek 平台账号在「未配置 model_mapping」时
+// 可服务的官方模型名（精确匹配，小写比较）。deepseek-flash 与 deepseek-v4-pro
+// 为官方现行名；deepseek-v4-flash / deepseek-v4-flash-vision-exp 为处于兼容
+// 路由期的旧名（上游仍接受并按 flash 价计费）；deepseek-v4-pro-0813 为与
+// 计费 pro 档口径一致的版本化名。
+var deepseekServableModels = []string{
+	"deepseek-flash",
+	"deepseek-v4-pro",
+	"deepseek-v4-flash",
+	"deepseek-v4-flash-vision-exp",
+	"deepseek-v4-pro-0813",
+}
+
+// isDeepseekServableModel 报告 DeepSeek 平台账号在「未配置 model_mapping」时
+// 可服务的模型名。官方现行模型为 deepseek-flash 与 deepseek-v4-pro；旧名
+// deepseek-v4-flash / deepseek-v4-flash-vision-exp 处于兼容路由期（上游仍接受并
+// 按 flash 价计费），deepseek-v4-pro-0813 为版本化名。其余一律拒绝：既避免把
+// 未知模型透传给上游（上游 404/400 会触发 per-model 冷却并掩盖真实配置错误），
+// 也避免上游把不认识的模型名静默兜底成 deepseek-flash。
+// Source: https://api-docs.deepseek.com/quick_start/pricing
+//
+//	https://api-docs.deepseek.com/guides/anthropic_api
+func isDeepseekServableModel(requestedModel string) bool {
+	// 官方 Claude Code 接入文档要求 ANTHROPIC_MODEL=deepseek-flash[1m]：
+	// [1m] 是客户端上下文选择器，先按同一规则归一化再比对白名单。
+	model := strings.ToLower(normalizeClaudeCodeLongContextModel(strings.TrimSpace(requestedModel)))
+	if model == "" {
+		return true // 空模型交由上层必填校验处理
+	}
+	for _, servable := range deepseekServableModels {
+		if model == servable {
+			return true
+		}
+	}
+	return false
+}
+
 // resolveOpenAICompactForwardModel determines the compact-only upstream model
 // for /responses/compact requests. It never affects normal /responses traffic.
 // When no compact-specific mapping matches, the input model is returned as-is.
@@ -429,6 +429,30 @@ func TestNormalizeOpenAIModelForUpstream(t *testing.T) {
 			model:   "gpt-4.1",
 			want:    "gpt-4.1",
 		},
+		{
+			name:    "deepseek strips claude code long context suffix",
+			account: &Account{Type: AccountTypeAPIKey, Platform: PlatformDeepseek},
+			model:   "deepseek-flash[1m]",
+			want:    "deepseek-flash",
+		},
+		{
+			name:    "deepseek strips duplicated long context suffix",
+			account: &Account{Type: AccountTypeAPIKey, Platform: PlatformDeepseek},
+			model:   "deepseek-flash[1M][1m]",
+			want:    "deepseek-flash",
+		},
+		{
+			name:    "deepseek preserves plain model",
+			account: &Account{Type: AccountTypeAPIKey, Platform: PlatformDeepseek},
+			model:   "deepseek-flash",
+			want:    "deepseek-flash",
+		},
+		{
+			name:    "non deepseek preserves long context suffix",
+			account: &Account{Type: AccountTypeAPIKey, Platform: PlatformOpenAI},
+			model:   "deepseek-flash[1m]",
+			want:    "deepseek-flash[1m]",
+		},
 	}

 	for _, tt := range tests {
@@ -0,0 +1,79 @@
+package service
+
+import (
+	"context"
+	"net/http"
+	"strings"
+)
+
+// OpenAIGatewayService implements service.PluginAccountDirectory for the OpenAI
+// OAuth outbound transport capability. The directory is intentionally scoped to
+// OpenAI OAuth-like, non-shadow accounts regardless of the requested filter, so a
+// plugin can never enumerate or resolve credentials outside that set. The host
+// additionally only wires this directory into plugins whose manifest declares the
+// matching capability (see PluginManager.buildHostServices).
+
+// ListPluginAccounts returns the ids of active OpenAI OAuth-like accounts.
+func (s *OpenAIGatewayService) ListPluginAccounts(ctx context.Context, platform, accountType string) ([]int64, error) {
+	if s == nil || s.accountRepo == nil {
+		return nil, nil
+	}
+	if p := strings.TrimSpace(platform); p != "" && p != PlatformOpenAI {
+		return nil, nil
+	}
+	if at := strings.TrimSpace(accountType); at != "" && at != AccountTypeOAuth {
+		// Setup-token style accounts are OAuth-like but not AccountTypeOAuth; the
+		// current binding only routes AccountTypeOAuth, so honour that filter.
+		return nil, nil
+	}
+	accounts, err := s.accountRepo.ListByPlatform(ctx, PlatformOpenAI)
+	if err != nil {
+		return nil, err
+	}
+	ids := make([]int64, 0, len(accounts))
+	for i := range accounts {
+		account := accounts[i]
+		if account.Status == StatusActive && account.IsOpenAIOAuthLike() && !account.IsShadow() && account.Type == AccountTypeOAuth {
+			ids = append(ids, account.ID)
+		}
+	}
+	return ids, nil
+}
+
+// ResolvePluginOutboundIdentity resolves the access token plus the outbound
+// identity headers and proxy the host would attach to a live request for the
+// account. It returns (nil, nil) for out-of-scope accounts or when no token can
+// be resolved.
+func (s *OpenAIGatewayService) ResolvePluginOutboundIdentity(ctx context.Context, accountID int64) (*PluginOutboundIdentity, error) {
+	if s == nil || s.accountRepo == nil || accountID <= 0 {
+		return nil, nil
+	}
+	account, err := s.accountRepo.GetByID(ctx, accountID)
+	if err != nil {
+		return nil, err
+	}
+	if account == nil || account.Type != AccountTypeOAuth || !account.IsOpenAIOAuthLike() || account.IsShadow() {
+		return nil, nil
+	}
+	token, _, err := s.GetAccessToken(ctx, account)
+	if err != nil {
+		return nil, err
+	}
+	if strings.TrimSpace(token) == "" {
+		return nil, nil
+	}
+	headers := http.Header{}
+	if err := resolveAndSetOpenAIChatGPTAccountHeaders(ctx, s.accountRepo, headers, account); err != nil {
+		return nil, err
+	}
+	ensureCodexIdentityHeaders(headers)
+	enforceCodexIdentityHeaders(headers)
+	return &PluginOutboundIdentity{
+		AccountID:   account.ID,
+		Platform:    account.Platform,
+		AccountType: account.Type,
+		ProxyURL:    resolveAccountProxyURL(account),
+		Token:       token,
+		Headers:     headers,
+	}, nil
+}
@@ -354,8 +354,9 @@ func buildOpenAIWSHTTPBridgeErrorEvent(statusCode int, message string) []byte {
 		message = "upstream request failed"
 	}
 	event := map[string]any{
-		"type":   "error",
-		"status": statusCode,
+		"type":            "error",
+		"sequence_number": 0,
+		"status":          statusCode,
 		"error": map[string]any{
 			"type":    "upstream_error",
 			"message": message,
@@ -363,7 +364,7 @@ func buildOpenAIWSHTTPBridgeErrorEvent(statusCode int, message string) []byte {
 	}
 	body, err := json.Marshal(event)
 	if err != nil {
-		return []byte(`{"type":"error","error":{"type":"upstream_error","message":"upstream request failed"}}`)
+		return []byte(`{"type":"error","sequence_number":0,"error":{"type":"upstream_error","message":"upstream request failed"}}`)
 	}
 	return body
 }
@@ -398,9 +399,9 @@ func buildOpenAIWSHTTPBridgeFailedEvent(responseID, model string, source []byte,
 	if model = strings.TrimSpace(model); model != "" {
 		response["model"] = model
 	}
-	body, err := json.Marshal(map[string]any{"type": "response.failed", "response": response})
+	body, err := json.Marshal(map[string]any{"type": "response.failed", "sequence_number": 0, "response": response})
 	if err != nil {
-		return []byte(`{"type":"response.failed","response":{"status":"failed","output":[],"error":{"code":"upstream_error","message":"Upstream response failed"}}}`)
+		return []byte(`{"type":"response.failed","sequence_number":0,"response":{"status":"failed","output":[],"error":{"code":"upstream_error","message":"Upstream response failed"}}}`)
 	}
 	return body
 }
@@ -0,0 +1,265 @@
+package service
+
+import (
+	"context"
+	"net/http"
+	"time"
+
+	pluginv1 "github.com/Wei-Shaw/sub2api/pkg/pluginapi/v1"
+	"google.golang.org/grpc/codes"
+	"google.golang.org/grpc/status"
+)
+
+// 通用宿主服务（HostService）的资源上限。这些限制与任何具体插件能力无关，
+// 只用于约束单个插件对共享存储的占用，属于防御性护栏而非业务策略。
+const (
+	// pluginKVMaxPluginKeyLen 与清单 id 的 maxLength 一致（manifest.schema.json），
+	// 确保任何可安装插件的 pluginKey 都能通过校验，不会被误判为不可用。
+	pluginKVMaxPluginKeyLen  = 160
+	pluginKVMaxNamespaceLen  = 128
+	pluginKVMaxKeyLen        = 256
+	pluginKVMaxValueBytes    = 256 * 1024
+	pluginKVMaxListLimit     = 1000
+	pluginKVDefaultListLimit = 100
+	pluginKVMaxTTL           = 90 * 24 * time.Hour
+)
+
+// PluginKVStore 是宿主向插件提供的通用命名空间键值存储端口。它对存储介质保持中立
+// （由 repository 层用 Redis 等实现），供任何插件持久化跨请求 / 跨副本 / 跨重启的
+// 状态。pluginKey 由宿主根据服务该连接的运行时注入，插件无法伪造，从而保证不同
+// 插件之间命名空间严格隔离。
+type PluginKVStore interface {
+	Get(ctx context.Context, pluginKey, namespace, key string) ([]byte, bool, error)
+	Set(ctx context.Context, pluginKey, namespace, key string, value []byte, ttl time.Duration) error
+	Delete(ctx context.Context, pluginKey, namespace, key string) error
+	List(ctx context.Context, pluginKey, namespace, keyPrefix string, limit int) ([]string, error)
+}
+
+// PluginOutboundIdentity 是宿主为某账号解析出的、可直接用于出站请求的身份材料：
+// 访问令牌、宿主会附加的出站请求头，以及账号代理。
+type PluginOutboundIdentity struct {
+	AccountID   int64
+	Platform    string
+	AccountType string
+	ProxyURL    string
+	Token       string
+	Headers     http.Header
+}
+
+// PluginAccountDirectory 让插件枚举其能力所覆盖的账号，并按需解析这些账号的出站身份，
+// 无需等待一条真实请求流经插件。这是一项敏感能力（会把账号凭据交给插件进程），因此
+// 宿主只对「其声明能力确实覆盖这些账号」的插件开放（见 PluginManager.buildHostServices）。
+// 实现方自身也必须把返回范围收敛到该能力对应的账号集合。
+type PluginAccountDirectory interface {
+	ListPluginAccounts(ctx context.Context, platform, accountType string) ([]int64, error)
+	ResolvePluginOutboundIdentity(ctx context.Context, accountID int64) (*PluginOutboundIdentity, error)
+}
+
+// pluginHostServiceServer 实现 pluginv1.HostServiceServer，是宿主经 go-plugin broker
+// 反向暴露给单个插件进程的服务端点。它绑定到具体插件的 pluginKey，因此每个运行时都有
+// 自己的实例；所有键值操作都被强制限定在该插件的命名空间内。
+type pluginHostServiceServer struct {
+	pluginv1.UnimplementedHostServiceServer
+	pluginKey string
+	store     PluginKVStore
+	directory PluginAccountDirectory
+}
+
+func newPluginHostServiceServer(pluginKey string, store PluginKVStore, directory PluginAccountDirectory) *pluginHostServiceServer {
+	return &pluginHostServiceServer{pluginKey: pluginKey, store: store, directory: directory}
+}
+
+func (s *pluginHostServiceServer) ready() bool {
+	return s != nil && s.store != nil && isValidPluginKVSegment(s.pluginKey, pluginKVMaxPluginKeyLen)
+}
+
+func (s *pluginHostServiceServer) KVGet(ctx context.Context, req *pluginv1.KVGetRequest) (*pluginv1.KVGetResponse, error) {
+	if !s.ready() {
+		return nil, status.Error(codes.Unavailable, "宿主键值存储不可用")
+	}
+	if req == nil {
+		return nil, status.Error(codes.InvalidArgument, "请求为空")
+	}
+	if err := validatePluginKVNamespace(req.Namespace); err != nil {
+		return nil, err
+	}
+	if err := validatePluginKVKey(req.Key); err != nil {
+		return nil, err
+	}
+	value, found, err := s.store.Get(ctx, s.pluginKey, req.Namespace, req.Key)
+	if err != nil {
+		return nil, status.Errorf(codes.Internal, "读取键值失败: %v", err)
+	}
+	if !found {
+		return &pluginv1.KVGetResponse{Found: false}, nil
+	}
+	return &pluginv1.KVGetResponse{Found: true, Value: value}, nil
+}
+
+func (s *pluginHostServiceServer) KVSet(ctx context.Context, req *pluginv1.KVSetRequest) (*pluginv1.KVSetResponse, error) {
+	if !s.ready() {
+		return nil, status.Error(codes.Unavailable, "宿主键值存储不可用")
+	}
+	if req == nil {
+		return nil, status.Error(codes.InvalidArgument, "请求为空")
+	}
+	if err := validatePluginKVNamespace(req.Namespace); err != nil {
+		return nil, err
+	}
+	if err := validatePluginKVKey(req.Key); err != nil {
+		return nil, err
+	}
+	if len(req.Value) > pluginKVMaxValueBytes {
+		return nil, status.Errorf(codes.InvalidArgument, "值超过 %d 字节上限", pluginKVMaxValueBytes)
+	}
+	ttl, err := pluginKVTTL(req.TtlSeconds)
+	if err != nil {
+		return nil, err
+	}
+	if err := s.store.Set(ctx, s.pluginKey, req.Namespace, req.Key, req.Value, ttl); err != nil {
+		return nil, status.Errorf(codes.Internal, "写入键值失败: %v", err)
+	}
+	return &pluginv1.KVSetResponse{}, nil
+}
+
+func (s *pluginHostServiceServer) KVDelete(ctx context.Context, req *pluginv1.KVDeleteRequest) (*pluginv1.KVDeleteResponse, error) {
+	if !s.ready() {
+		return nil, status.Error(codes.Unavailable, "宿主键值存储不可用")
+	}
+	if req == nil {
+		return nil, status.Error(codes.InvalidArgument, "请求为空")
+	}
+	if err := validatePluginKVNamespace(req.Namespace); err != nil {
+		return nil, err
+	}
+	if err := validatePluginKVKey(req.Key); err != nil {
+		return nil, err
+	}
+	if err := s.store.Delete(ctx, s.pluginKey, req.Namespace, req.Key); err != nil {
+		return nil, status.Errorf(codes.Internal, "删除键值失败: %v", err)
+	}
+	return &pluginv1.KVDeleteResponse{}, nil
+}
+
+func (s *pluginHostServiceServer) KVList(ctx context.Context, req *pluginv1.KVListRequest) (*pluginv1.KVListResponse, error) {
+	if !s.ready() {
+		return nil, status.Error(codes.Unavailable, "宿主键值存储不可用")
+	}
+	if req == nil {
+		return nil, status.Error(codes.InvalidArgument, "请求为空")
+	}
+	if err := validatePluginKVNamespace(req.Namespace); err != nil {
+		return nil, err
+	}
+	if req.KeyPrefix != "" {
+		if err := validatePluginKVKey(req.KeyPrefix); err != nil {
+			return nil, err
+		}
+	}
+	limit := int(req.Limit)
+	if limit <= 0 {
+		limit = pluginKVDefaultListLimit
+	}
+	if limit > pluginKVMaxListLimit {
+		limit = pluginKVMaxListLimit
+	}
+	keys, err := s.store.List(ctx, s.pluginKey, req.Namespace, req.KeyPrefix, limit)
+	if err != nil {
+		return nil, status.Errorf(codes.Internal, "列举键值失败: %v", err)
+	}
+	return &pluginv1.KVListResponse{Keys: keys}, nil
+}
+
+func (s *pluginHostServiceServer) ListAccounts(ctx context.Context, req *pluginv1.ListAccountsRequest) (*pluginv1.ListAccountsResponse, error) {
+	if s == nil || s.directory == nil {
+		return nil, status.Error(codes.Unavailable, "账号目录不可用")
+	}
+	if req == nil {
+		return nil, status.Error(codes.InvalidArgument, "请求为空")
+	}
+	ids, err := s.directory.ListPluginAccounts(ctx, req.Platform, req.AccountType)
+	if err != nil {
+		return nil, status.Errorf(codes.Internal, "列举账号失败: %v", err)
+	}
+	return &pluginv1.ListAccountsResponse{AccountIds: ids}, nil
+}
+
+func (s *pluginHostServiceServer) ResolveOutboundIdentity(ctx context.Context, req *pluginv1.ResolveOutboundIdentityRequest) (*pluginv1.ResolveOutboundIdentityResponse, error) {
+	if s == nil || s.directory == nil {
+		return nil, status.Error(codes.Unavailable, "账号目录不可用")
+	}
+	if req == nil || req.AccountId <= 0 {
+		return nil, status.Error(codes.InvalidArgument, "account_id 无效")
+	}
+	identity, err := s.directory.ResolvePluginOutboundIdentity(ctx, req.AccountId)
+	if err != nil {
+		return nil, status.Errorf(codes.Internal, "解析账号出站身份失败: %v", err)
+	}
+	if identity == nil {
+		return &pluginv1.ResolveOutboundIdentityResponse{Found: false}, nil
+	}
+	return &pluginv1.ResolveOutboundIdentityResponse{
+		Found:       true,
+		AccountId:   identity.AccountID,
+		Platform:    identity.Platform,
+		AccountType: identity.AccountType,
+		ProxyUrl:    identity.ProxyURL,
+		Token:       identity.Token,
+		Headers:     headersToPlugin(identity.Headers),
+	}, nil
+}
+
+func pluginKVTTL(seconds int64) (time.Duration, error) {
+	if seconds < 0 {
+		return 0, status.Error(codes.InvalidArgument, "ttl_seconds 不能为负")
+	}
+	if seconds == 0 {
+		return 0, nil
+	}
+	// 先在整数秒上比较上限，避免 seconds*time.Second 溢出 int64 后回绕成负值、
+	// 从而绕过上限检查把负 TTL 传给 Redis。
+	maxSeconds := int64(pluginKVMaxTTL / time.Second)
+	if seconds > maxSeconds {
+		return 0, status.Errorf(codes.InvalidArgument, "ttl_seconds 超过上限 %d", maxSeconds)
+	}
+	return time.Duration(seconds) * time.Second, nil
+}
+
+func validatePluginKVNamespace(namespace string) error {
+	if namespace == "" {
+		return status.Error(codes.InvalidArgument, "namespace 不能为空")
+	}
+	if !isValidPluginKVSegment(namespace, pluginKVMaxNamespaceLen) {
+		return status.Error(codes.InvalidArgument, "namespace 仅允许字母、数字、'.'、'_'、'-' 且长度受限")
+	}
+	return nil
+}
+
+func validatePluginKVKey(key string) error {
+	if key == "" {
+		return status.Error(codes.InvalidArgument, "key 不能为空")
+	}
+	if !isValidPluginKVSegment(key, pluginKVMaxKeyLen) {
+		return status.Error(codes.InvalidArgument, "key 仅允许字母、数字、'.'、'_'、'-' 且长度受限")
+	}
+	return nil
+}
+
+// isValidPluginKVSegment 限定命名段字符集为 [A-Za-z0-9._-]。这既避免 Redis glob
+// 元字符（*?[]）污染 SCAN，也避免 ':' 破坏内部键结构，从而保证命名空间隔离可靠。
+func isValidPluginKVSegment(value string, maxLen int) bool {
+	if value == "" || len(value) > maxLen {
+		return false
+	}
+	for _, r := range value {
+		switch {
+		case r >= 'a' && r <= 'z':
+		case r >= 'A' && r <= 'Z':
+		case r >= '0' && r <= '9':
+		case r == '.' || r == '_' || r == '-':
+		default:
+			return false
+		}
+	}
+	return true
+}
@@ -0,0 +1,130 @@
+package service
+
+import (
+	"context"
+	"sync"
+	"testing"
+	"time"
+
+	pluginv1 "github.com/Wei-Shaw/sub2api/pkg/pluginapi/v1"
+	hcplugin "github.com/hashicorp/go-plugin"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// brokerProbePlugin 是仅用于测试的插件传输实现：它实现 HostBrokerReceiver 以拿到
+// broker，并在 InitHostServices 中用宿主下发的 broker id 拨号回宿主的 HostService，
+// 走一次真实 gRPC 的 KVSet/KVGet 往返。
+type brokerProbePlugin struct {
+	pluginv1.UnimplementedTransportPluginServer
+	broker *hcplugin.GRPCBroker
+
+	mu       sync.Mutex
+	ready    bool
+	getFound bool
+	getValue []byte
+	failure  string
+}
+
+func (p *brokerProbePlugin) SetHostBroker(broker *hcplugin.GRPCBroker) { p.broker = broker }
+
+func (p *brokerProbePlugin) InitHostServices(ctx context.Context, req *pluginv1.InitHostServicesRequest) (*pluginv1.InitHostServicesResponse, error) {
+	if p.broker == nil {
+		return &pluginv1.InitHostServicesResponse{Ready: false, Message: "broker 未注入"}, nil
+	}
+	conn, err := p.broker.Dial(req.HostServiceId)
+	if err != nil {
+		p.record("dial: " + err.Error())
+		return &pluginv1.InitHostServicesResponse{Ready: false, Message: err.Error()}, nil
+	}
+	defer func() { _ = conn.Close() }()
+	client := pluginv1.NewHostServiceClient(conn)
+	if _, err := client.KVSet(ctx, &pluginv1.KVSetRequest{Namespace: "state", Key: "probe", Value: []byte("hello"), TtlSeconds: 60}); err != nil {
+		p.record("set: " + err.Error())
+		return &pluginv1.InitHostServicesResponse{Ready: false, Message: err.Error()}, nil
+	}
+	got, err := client.KVGet(ctx, &pluginv1.KVGetRequest{Namespace: "state", Key: "probe"})
+	if err != nil {
+		p.record("get: " + err.Error())
+		return &pluginv1.InitHostServicesResponse{Ready: false, Message: err.Error()}, nil
+	}
+	p.mu.Lock()
+	p.ready = true
+	p.getFound = got.Found
+	p.getValue = got.Value
+	p.mu.Unlock()
+	return &pluginv1.InitHostServicesResponse{Ready: true}, nil
+}
+
+func (p *brokerProbePlugin) record(failure string) {
+	p.mu.Lock()
+	p.failure = failure
+	p.mu.Unlock()
+}
+
+// noHostServicesPlugin 模拟老插件：不实现 InitHostServices（返回 Unimplemented），
+// 也不接收 broker。
+type noHostServicesPlugin struct {
+	pluginv1.UnimplementedTransportPluginServer
+}
+
+func dispenseTransportClient(t *testing.T, impl pluginv1.TransportPluginServer) *pluginv1.TransportClient {
+	t.Helper()
+	client, _ := hcplugin.TestPluginGRPCConn(t, false, map[string]hcplugin.Plugin{
+		pluginv1.TransportPluginName: &pluginv1.GRPCPlugin{Impl: impl},
+	})
+	t.Cleanup(func() { _ = client.Close() })
+	dispensed, err := client.Dispense(pluginv1.TransportPluginName)
+	require.NoError(t, err)
+	tc, ok := dispensed.(*pluginv1.TransportClient)
+	require.True(t, ok)
+	require.NotNil(t, tc.Broker)
+	return tc
+}
+
+// 端到端验证 broker 反向通道：宿主服务 AcceptAndServe + 插件 Dial 回来，经真实 gRPC
+// 完成 KV 往返，且写入落到宿主注入的 pluginKey 命名空间下。
+func TestOfferPluginHostServices_BrokerRoundtrip(t *testing.T) {
+	probe := &brokerProbePlugin{}
+	tc := dispenseTransportClient(t, probe)
+
+	store := newFakePluginKVStore()
+	hostServer := newPluginHostServiceServer("test.plugin", store, nil)
+	offerPluginHostServices(context.Background(), &PluginInstallation{PluginKey: "test.plugin"}, tc.TransportPluginClient, tc.Broker, hostServer, 5*time.Second)
+
+	probe.mu.Lock()
+	defer probe.mu.Unlock()
+	require.Empty(t, probe.failure)
+	require.True(t, probe.ready)
+	assert.True(t, probe.getFound)
+	assert.Equal(t, []byte("hello"), probe.getValue)
+
+	stored, found, err := store.Get(context.Background(), "test.plugin", "state", "probe")
+	require.NoError(t, err)
+	require.True(t, found)
+	assert.Equal(t, []byte("hello"), stored)
+}
+
+// 老插件不实现 InitHostServices 时，offerPluginHostServices 必须优雅降级、不 panic、
+// 不写入任何状态。
+func TestOfferPluginHostServices_UnimplementedIsGraceful(t *testing.T) {
+	tc := dispenseTransportClient(t, &noHostServicesPlugin{})
+
+	store := newFakePluginKVStore()
+	hostServer := newPluginHostServiceServer("test.plugin", store, nil)
+	require.NotPanics(t, func() {
+		offerPluginHostServices(context.Background(), &PluginInstallation{PluginKey: "test.plugin"}, tc.TransportPluginClient, tc.Broker, hostServer, 5*time.Second)
+	})
+
+	_, found, err := store.Get(context.Background(), "test.plugin", "state", "probe")
+	require.NoError(t, err)
+	assert.False(t, found)
+}
+
+// hostServices 为 nil（未配置键值存储）时不得触发任何 broker 交互。
+func TestOfferPluginHostServices_NilHostServicesNoop(t *testing.T) {
+	tc := dispenseTransportClient(t, &noHostServicesPlugin{})
+	require.NotPanics(t, func() {
+		offerPluginHostServices(context.Background(), &PluginInstallation{PluginKey: "test.plugin"}, tc.TransportPluginClient, tc.Broker, nil, 5*time.Second)
+	})
+}
@@ -0,0 +1,331 @@
+package service
+
+import (
+	"context"
+	"math"
+	"net/http"
+	"sort"
+	"strings"
+	"sync"
+	"testing"
+	"time"
+
+	pluginv1 "github.com/Wei-Shaw/sub2api/pkg/pluginapi/v1"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"google.golang.org/grpc/codes"
+	"google.golang.org/grpc/status"
+)
+
+// fakePluginKVStore 是内存版 PluginKVStore，按 (pluginKey, namespace, key) 三元组隔离，
+// 用于确定性地验证宿主服务端的校验与命名空间隔离逻辑。
+type fakePluginKVStore struct {
+	mu     sync.Mutex
+	values map[string][]byte
+	ttls   map[string]time.Duration
+}
+
+func newFakePluginKVStore() *fakePluginKVStore {
+	return &fakePluginKVStore{values: map[string][]byte{}, ttls: map[string]time.Duration{}}
+}
+
+func (f *fakePluginKVStore) compose(pluginKey, namespace, key string) string {
+	return pluginKey + "\x00" + namespace + "\x00" + key
+}
+
+func (f *fakePluginKVStore) Get(_ context.Context, pluginKey, namespace, key string) ([]byte, bool, error) {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	value, ok := f.values[f.compose(pluginKey, namespace, key)]
+	if !ok {
+		return nil, false, nil
+	}
+	return append([]byte(nil), value...), true, nil
+}
+
+func (f *fakePluginKVStore) Set(_ context.Context, pluginKey, namespace, key string, value []byte, ttl time.Duration) error {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	composed := f.compose(pluginKey, namespace, key)
+	f.values[composed] = append([]byte(nil), value...)
+	f.ttls[composed] = ttl
+	return nil
+}
+
+func (f *fakePluginKVStore) Delete(_ context.Context, pluginKey, namespace, key string) error {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	composed := f.compose(pluginKey, namespace, key)
+	delete(f.values, composed)
+	delete(f.ttls, composed)
+	return nil
+}
+
+func (f *fakePluginKVStore) List(_ context.Context, pluginKey, namespace, keyPrefix string, limit int) ([]string, error) {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	prefix := f.compose(pluginKey, namespace, keyPrefix)
+	scope := f.compose(pluginKey, namespace, "")
+	var keys []string
+	for composed := range f.values {
+		if strings.HasPrefix(composed, prefix) {
+			keys = append(keys, strings.TrimPrefix(composed, scope))
+		}
+	}
+	sort.Strings(keys)
+	if limit > 0 && len(keys) > limit {
+		keys = keys[:limit]
+	}
+	return keys, nil
+}
+
+func TestPluginHostServiceServer_SetGetDeleteRoundtrip(t *testing.T) {
+	store := newFakePluginKVStore()
+	server := newPluginHostServiceServer("local.example.plugin", store, nil)
+	ctx := context.Background()
+
+	_, err := server.KVSet(ctx, &pluginv1.KVSetRequest{Namespace: "state", Key: "alpha", Value: []byte("v1"), TtlSeconds: 60})
+	require.NoError(t, err)
+
+	got, err := server.KVGet(ctx, &pluginv1.KVGetRequest{Namespace: "state", Key: "alpha"})
+	require.NoError(t, err)
+	assert.True(t, got.Found)
+	assert.Equal(t, []byte("v1"), got.Value)
+	assert.Equal(t, time.Minute, store.ttls[store.compose("local.example.plugin", "state", "alpha")])
+
+	_, err = server.KVDelete(ctx, &pluginv1.KVDeleteRequest{Namespace: "state", Key: "alpha"})
+	require.NoError(t, err)
+
+	got, err = server.KVGet(ctx, &pluginv1.KVGetRequest{Namespace: "state", Key: "alpha"})
+	require.NoError(t, err)
+	assert.False(t, got.Found)
+}
+
+func TestPluginHostServiceServer_GetNotFound(t *testing.T) {
+	server := newPluginHostServiceServer("local.example.plugin", newFakePluginKVStore(), nil)
+	got, err := server.KVGet(context.Background(), &pluginv1.KVGetRequest{Namespace: "state", Key: "missing"})
+	require.NoError(t, err)
+	assert.False(t, got.Found)
+	assert.Nil(t, got.Value)
+}
+
+func TestPluginHostServiceServer_NamespaceIsolationAcrossPlugins(t *testing.T) {
+	store := newFakePluginKVStore()
+	first := newPluginHostServiceServer("local.plugin.one", store, nil)
+	second := newPluginHostServiceServer("local.plugin.two", store, nil)
+	ctx := context.Background()
+
+	_, err := first.KVSet(ctx, &pluginv1.KVSetRequest{Namespace: "state", Key: "shared", Value: []byte("first")})
+	require.NoError(t, err)
+
+	// 另一个插件即便使用完全相同的 namespace/key 也读不到，因为 pluginKey 由宿主注入。
+	got, err := second.KVGet(ctx, &pluginv1.KVGetRequest{Namespace: "state", Key: "shared"})
+	require.NoError(t, err)
+	assert.False(t, got.Found)
+
+	got, err = first.KVGet(ctx, &pluginv1.KVGetRequest{Namespace: "state", Key: "shared"})
+	require.NoError(t, err)
+	assert.True(t, got.Found)
+	assert.Equal(t, []byte("first"), got.Value)
+}
+
+func TestPluginHostServiceServer_ListPrefixAndLimit(t *testing.T) {
+	store := newFakePluginKVStore()
+	server := newPluginHostServiceServer("local.example.plugin", store, nil)
+	ctx := context.Background()
+	for _, key := range []string{"account-1", "account-2", "account-3", "other-1"} {
+		_, err := server.KVSet(ctx, &pluginv1.KVSetRequest{Namespace: "state", Key: key, Value: []byte("x")})
+		require.NoError(t, err)
+	}
+
+	resp, err := server.KVList(ctx, &pluginv1.KVListRequest{Namespace: "state", KeyPrefix: "account-"})
+	require.NoError(t, err)
+	assert.ElementsMatch(t, []string{"account-1", "account-2", "account-3"}, resp.Keys)
+
+	resp, err = server.KVList(ctx, &pluginv1.KVListRequest{Namespace: "state", KeyPrefix: "account-", Limit: 2})
+	require.NoError(t, err)
+	assert.Len(t, resp.Keys, 2)
+}
+
+func TestPluginHostServiceServer_Validation(t *testing.T) {
+	server := newPluginHostServiceServer("local.example.plugin", newFakePluginKVStore(), nil)
+	ctx := context.Background()
+	oversized := make([]byte, pluginKVMaxValueBytes+1)
+
+	cases := []struct {
+		name string
+		call func() error
+	}{
+		{"empty namespace", func() error {
+			_, err := server.KVGet(ctx, &pluginv1.KVGetRequest{Namespace: "", Key: "k"})
+			return err
+		}},
+		{"namespace with colon", func() error {
+			_, err := server.KVGet(ctx, &pluginv1.KVGetRequest{Namespace: "a:b", Key: "k"})
+			return err
+		}},
+		{"key with glob", func() error {
+			_, err := server.KVGet(ctx, &pluginv1.KVGetRequest{Namespace: "state", Key: "a*"})
+			return err
+		}},
+		{"empty key", func() error {
+			_, err := server.KVGet(ctx, &pluginv1.KVGetRequest{Namespace: "state", Key: ""})
+			return err
+		}},
+		{"oversized value", func() error {
+			_, err := server.KVSet(ctx, &pluginv1.KVSetRequest{Namespace: "state", Key: "k", Value: oversized})
+			return err
+		}},
+		{"negative ttl", func() error {
+			_, err := server.KVSet(ctx, &pluginv1.KVSetRequest{Namespace: "state", Key: "k", Value: []byte("v"), TtlSeconds: -1})
+			return err
+		}},
+		{"ttl over max", func() error {
+			_, err := server.KVSet(ctx, &pluginv1.KVSetRequest{Namespace: "state", Key: "k", Value: []byte("v"), TtlSeconds: int64(pluginKVMaxTTL/time.Second) + 1})
+			return err
+		}},
+		{"ttl overflow", func() error {
+			// 溢出用例：seconds*time.Second 会回绕成负值，必须仍判为超限而非放行。
+			_, err := server.KVSet(ctx, &pluginv1.KVSetRequest{Namespace: "state", Key: "k", Value: []byte("v"), TtlSeconds: math.MaxInt64})
+			return err
+		}},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			err := tc.call()
+			require.Error(t, err)
+			assert.Equal(t, codes.InvalidArgument, status.Code(err))
+		})
+	}
+}
+
+func TestPluginHostServiceServer_UnavailableWhenNoStore(t *testing.T) {
+	server := newPluginHostServiceServer("local.example.plugin", nil, nil)
+	_, err := server.KVGet(context.Background(), &pluginv1.KVGetRequest{Namespace: "state", Key: "k"})
+	require.Error(t, err)
+	assert.Equal(t, codes.Unavailable, status.Code(err))
+
+	// pluginKey 为空同样视为不可用，避免键前缀塌缩导致的跨插件泄漏。
+	blank := newPluginHostServiceServer("", newFakePluginKVStore(), nil)
+	_, err = blank.KVGet(context.Background(), &pluginv1.KVGetRequest{Namespace: "state", Key: "k"})
+	require.Error(t, err)
+	assert.Equal(t, codes.Unavailable, status.Code(err))
+
+	// pluginKey 含 ':' 会破坏内部键结构，必须判定为不可用而非放行。
+	tainted := newPluginHostServiceServer("bad:key", newFakePluginKVStore(), nil)
+	_, err = tainted.KVGet(context.Background(), &pluginv1.KVGetRequest{Namespace: "state", Key: "k"})
+	require.Error(t, err)
+	assert.Equal(t, codes.Unavailable, status.Code(err))
+}
+
+// pluginKey 的长度上限必须覆盖清单 id 的 maxLength（160），否则长 id 的合法插件会被
+// 静默剥夺宿主服务能力。
+func TestPluginHostServiceServer_PluginKeyLengthBoundary(t *testing.T) {
+	ctx := context.Background()
+	maxLen := newPluginHostServiceServer(strings.Repeat("a", pluginKVMaxPluginKeyLen), newFakePluginKVStore(), nil)
+	_, err := maxLen.KVSet(ctx, &pluginv1.KVSetRequest{Namespace: "state", Key: "k", Value: []byte("v")})
+	require.NoError(t, err)
+
+	tooLong := newPluginHostServiceServer(strings.Repeat("a", pluginKVMaxPluginKeyLen+1), newFakePluginKVStore(), nil)
+	_, err = tooLong.KVSet(ctx, &pluginv1.KVSetRequest{Namespace: "state", Key: "k", Value: []byte("v")})
+	require.Error(t, err)
+	assert.Equal(t, codes.Unavailable, status.Code(err))
+}
+
+type fakeAccountDirectory struct {
+	ids      []int64
+	identity *PluginOutboundIdentity
+	lastReq  int64
+}
+
+func (f *fakeAccountDirectory) ListPluginAccounts(_ context.Context, _, _ string) ([]int64, error) {
+	return f.ids, nil
+}
+
+func (f *fakeAccountDirectory) ResolvePluginOutboundIdentity(_ context.Context, accountID int64) (*PluginOutboundIdentity, error) {
+	f.lastReq = accountID
+	if f.identity == nil || f.identity.AccountID != accountID {
+		return nil, nil
+	}
+	return f.identity, nil
+}
+
+func TestPluginHostServiceServer_AccountDirectory(t *testing.T) {
+	ctx := context.Background()
+	dir := &fakeAccountDirectory{
+		ids: []int64{3, 7},
+		identity: &PluginOutboundIdentity{
+			AccountID: 7, Platform: "openai", AccountType: "oauth", ProxyURL: "http://p:1",
+			Token: "tok", Headers: http.Header{"Originator": {"codex-tui"}},
+		},
+	}
+	server := newPluginHostServiceServer("local.example.plugin", newFakePluginKVStore(), dir)
+
+	list, err := server.ListAccounts(ctx, &pluginv1.ListAccountsRequest{Platform: "openai", AccountType: "oauth"})
+	require.NoError(t, err)
+	assert.Equal(t, []int64{3, 7}, list.AccountIds)
+
+	resolved, err := server.ResolveOutboundIdentity(ctx, &pluginv1.ResolveOutboundIdentityRequest{AccountId: 7})
+	require.NoError(t, err)
+	require.True(t, resolved.Found)
+	assert.Equal(t, "tok", resolved.Token)
+	assert.Equal(t, "http://p:1", resolved.ProxyUrl)
+	assert.Equal(t, "codex-tui", resolved.Headers["Originator"].Values[0])
+
+	// unknown account → found=false, no error
+	miss, err := server.ResolveOutboundIdentity(ctx, &pluginv1.ResolveOutboundIdentityRequest{AccountId: 99})
+	require.NoError(t, err)
+	assert.False(t, miss.Found)
+
+	// invalid account id → InvalidArgument
+	_, err = server.ResolveOutboundIdentity(ctx, &pluginv1.ResolveOutboundIdentityRequest{AccountId: 0})
+	assert.Equal(t, codes.InvalidArgument, status.Code(err))
+}
+
+// 无目录时账号目录 RPC 必须返回 Unavailable（KV 仍可用），保证未授权插件拿不到凭据。
+func TestPluginHostServiceServer_DirectoryUnavailableWithoutDirectory(t *testing.T) {
+	server := newPluginHostServiceServer("local.example.plugin", newFakePluginKVStore(), nil)
+	_, err := server.ListAccounts(context.Background(), &pluginv1.ListAccountsRequest{})
+	assert.Equal(t, codes.Unavailable, status.Code(err))
+	_, err = server.ResolveOutboundIdentity(context.Background(), &pluginv1.ResolveOutboundIdentityRequest{AccountId: 1})
+	assert.Equal(t, codes.Unavailable, status.Code(err))
+	// KV 仍可用
+	_, kvErr := server.KVGet(context.Background(), &pluginv1.KVGetRequest{Namespace: "state", Key: "k"})
+	require.NoError(t, kvErr)
+}
+
+func TestPluginDeclaresOpenAIOAuthCapability(t *testing.T) {
+	match := PluginManifest{Capabilities: []PluginCapability{{ID: PluginCapabilityOpenAIOAuthOutbound, Platform: PlatformOpenAI, AccountType: AccountTypeOAuth}}}
+	if !pluginDeclaresOpenAIOAuthCapability(match) {
+		t.Fatal("matching capability must be recognized")
+	}
+	wrongType := PluginManifest{Capabilities: []PluginCapability{{ID: PluginCapabilityOpenAIOAuthOutbound, Platform: PlatformOpenAI, AccountType: "apikey"}}}
+	if pluginDeclaresOpenAIOAuthCapability(wrongType) {
+		t.Fatal("wrong account_type must not match")
+	}
+	empty := PluginManifest{}
+	if pluginDeclaresOpenAIOAuthCapability(empty) {
+		t.Fatal("no capability must not match")
+	}
+}
+
+// buildHostServices 只对声明了 OpenAI OAuth 能力的插件注入账号目录。
+func TestBuildHostServicesGatesDirectoryByCapability(t *testing.T) {
+	dir := &fakeAccountDirectory{}
+	m := &PluginManager{kvStore: newFakePluginKVStore(), accountDirectory: dir}
+
+	authorized := &PluginInstallation{PluginKey: "p.authorized", Manifest: PluginManifest{
+		Capabilities: []PluginCapability{{ID: PluginCapabilityOpenAIOAuthOutbound, Platform: PlatformOpenAI, AccountType: AccountTypeOAuth}},
+	}}
+	srv, ok := m.buildHostServices(authorized).(*pluginHostServiceServer)
+	require.True(t, ok)
+	require.NotNil(t, srv.directory)
+
+	unauthorized := &PluginInstallation{PluginKey: "p.other", Manifest: PluginManifest{
+		Capabilities: []PluginCapability{{ID: "some.other.capability", Platform: "x", AccountType: "y"}},
+	}}
+	srv2, ok := m.buildHostServices(unauthorized).(*pluginHostServiceServer)
+	require.True(t, ok)
+	require.Nil(t, srv2.directory, "unauthorized plugin must not receive the account directory")
+	require.NotNil(t, srv2.store, "but KV remains available")
+}
@@ -46,6 +46,11 @@ type PluginManager struct {
 	cfg       *config.Config
 	hostInfo  PluginHostInfo
 	installer *PluginPackageInstaller
+	// kvStore 为运行中的插件提供通用宿主键值存储；为 nil 时不向插件暴露宿主服务。
+	kvStore PluginKVStore
+	// accountDirectory 为声明了对应能力的插件提供账号目录与出站身份解析（敏感能力）；
+	// 通过 SetAccountDirectory 在启动装配阶段注入，为 nil 时插件拿不到该能力。
+	accountDirectory PluginAccountDirectory

 	operationMu        sync.Mutex
 	mu                 sync.Mutex
@@ -57,13 +62,14 @@ type PluginManager struct {
 	route              atomic.Pointer[pluginRoute]
 }

-func NewPluginManager(repo PluginRepository, encryptor SecretEncryptor, cfg *config.Config, hostInfo PluginHostInfo) *PluginManager {
+func NewPluginManager(repo PluginRepository, encryptor SecretEncryptor, cfg *config.Config, hostInfo PluginHostInfo, kvStore PluginKVStore) *PluginManager {
 	return &PluginManager{
 		repo:               repo,
 		encryptor:          encryptor,
 		cfg:                cfg,
 		hostInfo:           hostInfo,
 		installer:          NewPluginPackageInstaller(cfg, hostInfo),
+		kvStore:            kvStore,
 		runtimes:           make(map[int64]*pluginRuntime),
 		localInstallations: make(map[int64]*PluginInstallation),
 	}
@@ -813,6 +819,26 @@ func (m *PluginManager) Test(ctx context.Context, id int64) (*pluginv1.TestConfi
 	return runtime.api.TestConfig(testCtx, &pluginv1.TestConfigRequest{ConfigJson: configJSON})
 }

+// Status returns the plugin's passive runtime status (its Health response, including
+// any status_json blob) for the config UI. Unlike Test it never applies config,
+// starts a temporary runtime, or reaches upstream — a plugin that is not currently
+// running simply reports "not running" with no status blob. This makes it safe to
+// serve from a lightweight, ungated, read-only endpoint used for status polling.
+func (m *PluginManager) Status(ctx context.Context, id int64) (*pluginv1.HealthResponse, error) {
+	if _, err := m.repo.GetByID(ctx, id); err != nil {
+		return nil, err
+	}
+	m.mu.Lock()
+	runtime := m.runtimes[id]
+	m.mu.Unlock()
+	if runtime == nil {
+		return &pluginv1.HealthResponse{Healthy: false, Message: "插件未运行"}, nil
+	}
+	statusCtx, cancel := context.WithTimeout(ctx, 10*time.Second)
+	defer cancel()
+	return runtime.status(statusCtx)
+}
+
 type pluginUIAssetClaims struct {
 	Version  int   `json:"version"`
 	PluginID int64 `json:"plugin_id"`
@@ -1018,7 +1044,44 @@ func (m *PluginManager) newRuntime(ctx context.Context, installation *PluginInst
 		return nil, err
 	}
 	timeout := time.Duration(m.cfg.Plugins.StartTimeoutSeconds) * time.Second
-	return startPluginRuntime(ctx, installation, timeout, socketDir)
+	return startPluginRuntime(ctx, installation, timeout, socketDir, m.buildHostServices(installation))
+}
+
+// SetAccountDirectory 注入账号目录实现（敏感能力）。仅在启动装配阶段调用一次，
+// 早于 Start，因此运行期读取无需额外同步。
+func (m *PluginManager) SetAccountDirectory(directory PluginAccountDirectory) {
+	m.mu.Lock()
+	m.accountDirectory = directory
+	m.mu.Unlock()
+}
+
+// buildHostServices 为单个插件构造绑定其 pluginKey 的宿主服务端点。返回 nil（未配置
+// 键值存储或缺少 pluginKey）时，startPluginRuntime 不会向插件暴露任何宿主服务。
+// 账号目录（会向插件交付账号凭据）只对「清单声明了 OpenAI OAuth 出站能力」的插件开放，
+// 从而把凭据暴露面收敛到本就要处理这些账号的插件。
+func (m *PluginManager) buildHostServices(installation *PluginInstallation) pluginv1.HostServiceServer {
+	if m.kvStore == nil || installation == nil || strings.TrimSpace(installation.PluginKey) == "" {
+		return nil
+	}
+	var directory PluginAccountDirectory
+	if pluginDeclaresOpenAIOAuthCapability(installation.Manifest) {
+		m.mu.Lock()
+		directory = m.accountDirectory
+		m.mu.Unlock()
+	}
+	return newPluginHostServiceServer(installation.PluginKey, m.kvStore, directory)
+}
+
+// pluginDeclaresOpenAIOAuthCapability reports whether the (install-validated)
+// manifest declares the OpenAI OAuth outbound transport capability.
+func pluginDeclaresOpenAIOAuthCapability(manifest PluginManifest) bool {
+	for _, capability := range manifest.Capabilities {
+		if capability.ID == PluginCapabilityOpenAIOAuthOutbound &&
+			capability.Platform == PlatformOpenAI && capability.AccountType == AccountTypeOAuth {
+			return true
+		}
+	}
+	return false
 }

 func (m *PluginManager) removeRuntimeLocked(id int64) *pluginRuntime {
@@ -112,3 +112,24 @@ func TestStablePluginBucketIsDeterministicAndBounded(t *testing.T) {
 		assert.Less(t, first, uint64(100))
 	}
 }
+
+type statusStubRepository struct {
+	PluginRepository
+}
+
+func (r *statusStubRepository) GetByID(context.Context, int64) (*PluginInstallation, error) {
+	return &PluginInstallation{ID: 7}, nil
+}
+
+// Status is the read-only, ungated runtime-status channel: when the plugin is not
+// running it must report "not running" with no status blob and never error or start
+// a runtime (that side-effecting behaviour belongs to Test, not Status).
+func TestPluginManagerStatusReportsNotRunningWithoutRuntime(t *testing.T) {
+	manager := &PluginManager{repo: &statusStubRepository{}}
+	resp, err := manager.Status(context.Background(), 7)
+	require.NoError(t, err)
+	require.NotNil(t, resp)
+	assert.False(t, resp.Healthy)
+	assert.Equal(t, "插件未运行", resp.Message)
+	assert.Empty(t, resp.StatusJson)
+}
@@ -8,6 +8,7 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"log/slog"
 	"net/http"
 	"os/exec"
 	"strconv"
@@ -19,6 +20,9 @@ import (
 	pluginv1 "github.com/Wei-Shaw/sub2api/pkg/pluginapi/v1"
 	hclog "github.com/hashicorp/go-hclog"
 	hcplugin "github.com/hashicorp/go-plugin"
+	"google.golang.org/grpc"
+	"google.golang.org/grpc/codes"
+	"google.golang.org/grpc/status"
 )

 type pluginRuntime struct {
@@ -31,7 +35,7 @@ type pluginRuntime struct {
 	doneOnce     sync.Once
 }

-func startPluginRuntime(ctx context.Context, installation *PluginInstallation, startTimeout time.Duration, socketDir string) (*pluginRuntime, error) {
+func startPluginRuntime(ctx context.Context, installation *PluginInstallation, startTimeout time.Duration, socketDir string, hostServices pluginv1.HostServiceServer) (*pluginRuntime, error) {
 	if installation == nil {
 		return nil, errors.New("插件安装记录为空")
 	}
@@ -66,11 +70,12 @@ func startPluginRuntime(ctx context.Context, installation *PluginInstallation, s
 		client.Kill()
 		return nil, fmt.Errorf("获取插件传输能力: %w", err)
 	}
-	api, ok := dispensed.(pluginv1.TransportPluginClient)
-	if !ok {
+	transportClient, ok := dispensed.(*pluginv1.TransportClient)
+	if !ok || transportClient.TransportPluginClient == nil {
 		client.Kill()
 		return nil, errors.New("插件未实现传输 gRPC 客户端")
 	}
+	api := transportClient.TransportPluginClient
 	runtime := &pluginRuntime{
 		installation: installation,
 		client:       client,
@@ -97,9 +102,56 @@ func startPluginRuntime(ctx context.Context, installation *PluginInstallation, s
 		}
 		return nil, fmt.Errorf("插件不健康: %s", health.Message)
 	}
+	// 可选地把宿主服务（HostService）反向暴露给插件。这是叠加在传输契约之上的能力：
+	// 老插件不实现 InitHostServices（返回 Unimplemented），此处静默跳过，绝不阻断启动。
+	offerPluginHostServices(ctx, installation, api, transportClient.Broker, hostServices, startTimeout)
 	return runtime, nil
 }

+// offerPluginHostServices 在 go-plugin broker 上启动一个宿主服务实例，并通过
+// InitHostServices 把 broker 流 id 交给插件。服务生命周期与插件进程绑定：client.Kill()
+// 会关闭 broker，AcceptAndServe 随之 GracefulStop，无需手动清理。整个过程尽力而为，
+// 任何失败都只记录日志、不影响插件转发能力。
+func offerPluginHostServices(
+	ctx context.Context,
+	installation *PluginInstallation,
+	api pluginv1.TransportPluginClient,
+	broker *hcplugin.GRPCBroker,
+	hostServices pluginv1.HostServiceServer,
+	startTimeout time.Duration,
+) {
+	if broker == nil || hostServices == nil || api == nil {
+		return
+	}
+	brokerID := broker.NextId()
+	go broker.AcceptAndServe(brokerID, func(opts []grpc.ServerOption) *grpc.Server {
+		server := grpc.NewServer(opts...)
+		pluginv1.RegisterHostServiceServer(server, hostServices)
+		return server
+	})
+	initCtx, cancel := context.WithTimeout(ctx, startTimeout)
+	defer cancel()
+	resp, err := api.InitHostServices(initCtx, &pluginv1.InitHostServicesRequest{
+		HostServiceId:         brokerID,
+		HostServiceApiVersion: pluginv1.HostServiceAPIVersion,
+	})
+	pluginKey := ""
+	if installation != nil {
+		pluginKey = installation.PluginKey
+	}
+	if err != nil {
+		if status.Code(err) == codes.Unimplemented {
+			slog.Debug("plugin_host_services_unimplemented", "plugin", pluginKey)
+		} else {
+			slog.Warn("plugin_host_services_init_failed", "plugin", pluginKey, "error", err)
+		}
+		return
+	}
+	if resp != nil && !resp.Ready {
+		slog.Debug("plugin_host_services_declined", "plugin", pluginKey, "message", resp.Message)
+	}
+}
+
 func (r *pluginRuntime) validateAndApplyConfig(ctx context.Context, configJSON []byte) error {
 	_, err := r.validateAndApplyNormalizedConfig(ctx, configJSON)
 	return err
@@ -161,6 +213,23 @@ func (r *pluginRuntime) checkHealth(ctx context.Context) error {
 	return nil
 }

+// status returns the plugin's passive Health response, including any status_json
+// blob it exposes for the config UI. It performs no config apply and no upstream
+// call, so it is safe to serve from a lightweight, ungated status endpoint.
+func (r *pluginRuntime) status(ctx context.Context) (*pluginv1.HealthResponse, error) {
+	if r == nil || r.api == nil || r.client == nil || r.client.Exited() {
+		return nil, errors.New("插件进程已退出")
+	}
+	health, err := r.api.Health(ctx, &pluginv1.HealthRequest{})
+	if err != nil {
+		return nil, fmt.Errorf("插件状态查询失败: %w", err)
+	}
+	if health == nil {
+		return nil, errors.New("插件未返回状态")
+	}
+	return health, nil
+}
+
 func (r *pluginRuntime) beginRequest() bool {
 	if r == nil || r.draining.Load() {
 		return false
@@ -47,7 +47,7 @@ func TestPluginRuntimeIntegration(t *testing.T) {
 		assert.FileExists(t, filepath.Join(installation.InstallPath, filepath.FromSlash(relative)))
 	}

-	runtime, err := startPluginRuntime(context.Background(), installation, 10*time.Second, filepath.Join(root, "runtime"))
+	runtime, err := startPluginRuntime(context.Background(), installation, 10*time.Second, filepath.Join(root, "runtime"), nil)
 	require.NoError(t, err)
 	defer runtime.kill()
 	require.NoError(t, runtime.validateAndApplyConfig(context.Background(), []byte(`{
@@ -6,6 +6,8 @@ import (
 	"net/http"
 	"strings"
 	"time"
+
+	"github.com/tidwall/gjson"
 )

 // 国产供应商（kimi/zhipu/deepseek）的响应式冷却辅助。
@@ -35,22 +37,65 @@ func isCNProviderConcurrencyLimit403(account *Account, upstreamMsg string) bool
 		strings.TrimSpace(upstreamMsg) == kimiConcurrentRequestLimitMessage
 }

+// cnQuotaExhausted403ErrorType 是 Kimi Coding Plan 配额窗口耗尽的 403 错误类型。
+// 与并发限制 403（见 isCNProviderConcurrencyLimit403）不同，这是 5h/weekly
+// 窗口耗尽的限流信号：窗口到期后自动恢复，绝不能落入通用 403 升级计数
+// （连续 3 次永久 SetError），必须按 429 口径冷却到真实窗口重置点。
+const cnQuotaExhausted403ErrorType = "access_terminated_error"
+
+// cnQuotaExhaustedReasonPrefix 是配额耗尽 403 临时停调 reason 的稳定前缀。
+const cnQuotaExhaustedReasonPrefix = "cn_quota_exhausted"
+
+// isCNProviderQuotaExhausted403 识别 CN 供应商配额窗口耗尽的 403。
+// 先做廉价的文案子串匹配（usage limit / quota will reset），未命中再解析
+// 响应体匹配结构化错误类型（Kimi 的 access_terminated_error），以兼容同族
+// 供应商的文案变体。与并发限制文案互斥，两类信号不会误判。
+// 仅限 Coding Plan 账号：滚动窗口快照（含重置时间）只有 Coding Plan 账号才有，
+// 非 Coding Plan 账号的同类 403 语义不明（可能是需要升级套餐的硬上限），
+// 应回退到通用 403 升级逻辑而非无限临时冷却。
+func isCNProviderQuotaExhausted403(account *Account, responseBody []byte, upstreamMsg string) bool {
+	if account == nil || !account.IsCNProvider() || !account.IsCodingPlan() {
+		return false
+	}
+	msg := strings.ToLower(strings.TrimSpace(upstreamMsg))
+	if strings.Contains(msg, "usage limit") || strings.Contains(msg, "quota will reset") {
+		return true
+	}
+	return strings.EqualFold(strings.TrimSpace(gjson.GetBytes(responseBody, "error.type").String()), cnQuotaExhausted403ErrorType)
+}
+
+// handleCNProviderQuotaExhausted403 把 CN 供应商配额窗口耗尽的 403 按 429
+// 同口径处理：有配额快照时冷却到最早的未来窗口重置点（SetRateLimited，
+// 窗口过期后调度自动恢复）；快照缺失（额度探测尚未刷新）时兜底默认 403
+// 冷却时长的临时停调，避免过度停调到数天后的窗口。无论哪条路径都不写
+// 账号 error 状态。
+func (s *RateLimitService) handleCNProviderQuotaExhausted403(
+	ctx context.Context,
+	account *Account,
+	upstreamMsg string,
+) {
+	if s.cooldownCNProviderToQuotaSnapshotReset(ctx, account, cnQuotaExhaustedReasonPrefix, "cn_quota_exhausted_rate_limited") != nil {
+		return
+	}
+
+	// upstreamMsg 由调用方（HandleUpstreamError）从同一响应体提取，此处直接复用；
+	// 为空时仅存前缀，保证 reason 稳定可检索。
+	reason := cnQuotaExhaustedReasonPrefix
+	if msg := strings.TrimSpace(upstreamMsg); msg != "" {
+		reason += ": " + msg
+	}
+	until := time.Now().Add(time.Duration(openAI403CooldownMinutesDefault) * time.Minute)
+	s.setCNProviderTempUnschedulable(ctx, account, until, cnQuotaExhaustedReasonPrefix, reason, "cn_quota_exhausted_temp_unschedulable")
+}
+
 func (s *RateLimitService) handleCNProviderConcurrencyLimit403(
 	ctx context.Context,
 	account *Account,
 ) {
 	until := time.Now().Add(time.Duration(openAI403CooldownMinutesDefault) * time.Minute)
-	reason := cnConcurrencyLimitReasonPrefix + ": " + kimiConcurrentRequestLimitMessage
-	s.notifyAccountSchedulingBlocked(account, until, cnConcurrencyLimitReasonPrefix)
-	if err := s.accountRepo.SetTempUnschedulable(ctx, account.ID, until, reason); err != nil {
-		slog.Warn("cn_concurrency_limit_set_temp_unschedulable_failed", "account_id", account.ID, "error", err)
-		return
-	}
-	slog.Info("cn_provider_concurrency_limited",
-		"account_id", account.ID,
-		"platform", account.Platform,
-		"until", until.UTC(),
-	)
+	s.setCNProviderTempUnschedulable(ctx, account, until,
+		cnConcurrencyLimitReasonPrefix, cnConcurrencyLimitReasonPrefix+": "+kimiConcurrentRequestLimitMessage,
+		"cn_provider_concurrency_limited")
 }

 // cnBalanceLowReason 构造余额不足临时停调的 reason（带稳定前缀）。
@@ -151,6 +196,45 @@ func cnProviderQuotaSnapshotReset(account *Account, now time.Time) *time.Time {
 	return earliest
 }

+// cooldownCNProviderToQuotaSnapshotReset 把账号冷却到配额快照中最早的未来窗口
+// 重置点（SetRateLimited，窗口过期后调度自动恢复），供 429（OpenCodeGo /
+// Coding Plan）与配额耗尽 403 共用。仅在冷却成功持久化后返回非 nil；
+// 无快照或写库失败时返回 nil，由调用方落入各自的兜底逻辑（避免写库失败
+// 时账号完全失去持久冷却）。
+func (s *RateLimitService) cooldownCNProviderToQuotaSnapshotReset(ctx context.Context, account *Account, reason, logEvent string) *time.Time {
+	until := cnProviderQuotaSnapshotReset(account, time.Now())
+	if until == nil {
+		return nil
+	}
+	s.notifyAccountSchedulingBlocked(account, *until, reason)
+	if err := s.accountRepo.SetRateLimited(ctx, account.ID, *until); err != nil {
+		slog.Warn("rate_limit_set_failed", "account_id", account.ID, "error", err)
+		return nil
+	}
+	slog.Info(logEvent,
+		"account_id", account.ID,
+		"platform", account.Platform,
+		"reset_at", until.UTC(),
+	)
+	return until
+}
+
+// setCNProviderTempUnschedulable 写临时停调（含调度层通知与日志），供 CN 403
+// 各分支（并发限制、配额耗尽兜底）共用。告警 key 由 notifyReason 派生，保持
+// 各分支既有日志 key 稳定（<prefix>_set_temp_unschedulable_failed）。
+func (s *RateLimitService) setCNProviderTempUnschedulable(ctx context.Context, account *Account, until time.Time, notifyReason, storeReason, logEvent string) {
+	s.notifyAccountSchedulingBlocked(account, until, notifyReason)
+	if err := s.accountRepo.SetTempUnschedulable(ctx, account.ID, until, storeReason); err != nil {
+		slog.Warn(notifyReason+"_set_temp_unschedulable_failed", "account_id", account.ID, "error", err)
+		return
+	}
+	slog.Info(logEvent,
+		"account_id", account.ID,
+		"platform", account.Platform,
+		"until", until.UTC(),
+	)
+}
+
 // applyCNProviderReactive429 处理国产供应商的 429 响应。
 // 返回 true 表示已处理（调用方应 return），false 表示未命中、继续走默认 429 逻辑。
 func (s *RateLimitService) applyCNProviderReactive429(
@@ -160,17 +244,7 @@ func (s *RateLimitService) applyCNProviderReactive429(
 	responseBody []byte,
 ) bool {
 	if account.IsOpenCodeGo() {
-		if until := cnProviderQuotaSnapshotReset(account, time.Now()); until != nil {
-			s.notifyAccountSchedulingBlocked(account, *until, "429")
-			if err := s.accountRepo.SetRateLimited(ctx, account.ID, *until); err != nil {
-				slog.Warn("rate_limit_set_failed", "account_id", account.ID, "error", err)
-				return true
-			}
-			slog.Info("opencode_go_rate_limited",
-				"account_id", account.ID,
-				"platform", account.Platform,
-				"reset_at", *until,
-			)
+		if s.cooldownCNProviderToQuotaSnapshotReset(ctx, account, "429", "opencode_go_rate_limited") != nil {
 			return true
 		}
 		if resetAt := parseOpenAIRateLimitResetTime(responseBody); resetAt != nil {
@@ -199,20 +273,9 @@ func (s *RateLimitService) applyCNProviderReactive429(
 	}
 	// 2) Coding Plan 窗口耗尽：冷却到快照中最早的窗口重置点（见
 	// cnProviderQuotaSnapshotReset：429 多由 5h 窗口触发，取较早点避免过度停调）。
-	if account.IsCodingPlan() {
-		if until := cnProviderQuotaSnapshotReset(account, time.Now()); until != nil {
-			s.notifyAccountSchedulingBlocked(account, *until, "429")
-			if err := s.accountRepo.SetRateLimited(ctx, account.ID, *until); err != nil {
-				slog.Warn("rate_limit_set_failed", "account_id", account.ID, "error", err)
-				return true
-			}
-			slog.Info("cn_coding_plan_rate_limited",
-				"account_id", account.ID,
-				"platform", account.Platform,
-				"reset_at", *until,
-			)
-			return true
-		}
+	if account.IsCodingPlan() &&
+		s.cooldownCNProviderToQuotaSnapshotReset(ctx, account, "429", "cn_coding_plan_rate_limited") != nil {
+		return true
 	}
 	return false
 }
@@ -992,6 +992,14 @@ func (s *RateLimitService) handle403(ctx context.Context, account *Account, upst
 		s.handleCNProviderConcurrencyLimit403(ctx, account)
 		return true
 	}
+	// Kimi 等 CN 供应商把 Coding Plan 配额窗口耗尽打成 403
+	// （error.type=access_terminated_error），这是窗口到期后自动恢复的限流
+	// 信号而非封禁：按 429 口径冷却到真实窗口重置点，避免落入下方通用 403
+	// 升级计数后被永久 SetError。
+	if isCNProviderQuotaExhausted403(account, responseBody, upstreamMsg) {
+		s.handleCNProviderQuotaExhausted403(ctx, account, upstreamMsg)
+		return true
+	}
 	// 国产供应商与 openai 同口径:HTML 403(CDN/代理拦截页)不构成账号失效证据,
 	// 且 403 在 failover 状态集里会被逐账号重放——直接 SetError 会让一个坏请求/
 	// 一层坏代理连环永久禁用整组账号。走 HTML 豁免 + N 次累计 + 临时冷却。
@@ -17,6 +17,7 @@ type rateLimitAccountRepoStub struct {
 	mockAccountRepoForGemini
 	setErrorCalls          int
 	tempCalls              int
+	rateLimitedCalls       int
 	updateCredentialsCalls int
 	updateExtraCalls       int
 	lastCredentials        map[string]any
@@ -25,6 +26,8 @@ type rateLimitAccountRepoStub struct {
 	lastTempReason         string
 	lastErrorID            int64
 	lastTempID             int64
+	lastRateLimitedID      int64
+	lastRateLimitedAt      time.Time
 	tempErr                error
 }

@@ -42,6 +45,13 @@ func (r *rateLimitAccountRepoStub) SetTempUnschedulable(ctx context.Context, id
 	return r.tempErr
 }

+func (r *rateLimitAccountRepoStub) SetRateLimited(ctx context.Context, id int64, resetAt time.Time) error {
+	r.rateLimitedCalls++
+	r.lastRateLimitedID = id
+	r.lastRateLimitedAt = resetAt
+	return nil
+}
+
 func (r *rateLimitAccountRepoStub) UpdateCredentials(ctx context.Context, id int64, credentials map[string]any) error {
 	r.updateCredentialsCalls++
 	r.lastCredentials = shallowCopyMap(credentials)
@@ -0,0 +1,171 @@
+//go:build unit
+
+package service
+
+import (
+	"context"
+	"fmt"
+	"net/http"
+	"testing"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/stretchr/testify/require"
+)
+
+// newCNQuotaExhaustedTestAccount 构造一个 Kimi Coding Plan 账号：
+// account_mode=coding 使 IsCodingPlan() 为真，Extra 带 5h/weekly 窗口快照。
+func newCNQuotaExhaustedTestAccount(resetIn time.Duration) *Account {
+	extra := map[string]any{}
+	if resetIn > 0 {
+		extra[cnExtraKey(PlatformKimi, cnExtraSuffix5hUsed)] = 100
+		extra[cnExtraKey(PlatformKimi, cnExtraSuffix5hReset)] = time.Now().Add(resetIn).UTC().Format(time.RFC3339)
+		extra[cnExtraKey(PlatformKimi, cnExtraSuffixWeeklyReset)] = time.Now().Add(7 * 24 * time.Hour).UTC().Format(time.RFC3339)
+	}
+	return &Account{
+		ID:          401,
+		Platform:    PlatformKimi,
+		Type:        AccountTypeAPIKey,
+		Credentials: map[string]any{"account_mode": AccountModeCoding},
+		Extra:       extra,
+	}
+}
+
+const kimiWeeklyQuotaExhaustedBody = `{"error":{"message":"You've reached your weekly (7-day) usage limit. Your quota will reset when the current 7-day window ends. To continue now, purchase extra usage or upgrade your plan: https://www.kimi.com/membership/subscription?tab=quota","type":"access_terminated_error"}}`
+
+// runKimi403 以 403 + 给定响应体驱动 HandleUpstreamError，返回 shouldDisable。
+func runKimi403(service *RateLimitService, account *Account, body string) bool {
+	return service.HandleUpstreamError(
+		context.Background(),
+		account,
+		http.StatusForbidden,
+		http.Header{},
+		[]byte(body),
+	)
+}
+
+func TestHandleUpstreamError_KimiQuotaExhausted403RateLimitedToWindowReset(t *testing.T) {
+	repo := &rateLimitAccountRepoStub{}
+	blocker := &runtimeBlockRecorder{}
+	service := NewRateLimitService(repo, nil, &config.Config{}, nil, nil)
+	service.SetAccountRuntimeBlocker(blocker)
+	account := newCNQuotaExhaustedTestAccount(2 * time.Hour)
+
+	shouldDisable := runKimi403(service, account, kimiWeeklyQuotaExhaustedBody)
+
+	require.True(t, shouldDisable)
+	require.Equal(t, 0, repo.setErrorCalls, "quota exhaustion must never SetError")
+	require.Equal(t, 0, repo.tempCalls)
+	require.Equal(t, 1, repo.rateLimitedCalls)
+	require.Equal(t, account.ID, repo.lastRateLimitedID)
+	require.True(t, repo.lastRateLimitedAt.After(time.Now()), "cooldown should target the future window reset")
+	require.True(t, repo.lastRateLimitedAt.Before(time.Now().Add(3*time.Hour)), "should cool down to the nearer 5h reset, not weekly")
+	require.Len(t, blocker.accounts, 1)
+	require.Equal(t, cnQuotaExhaustedReasonPrefix, blocker.reasons[0])
+}
+
+func TestHandleUpstreamError_KimiQuotaExhausted403WithoutSnapshotFallsBackToTemp(t *testing.T) {
+	repo := &rateLimitAccountRepoStub{}
+	service := NewRateLimitService(repo, nil, &config.Config{}, nil, nil)
+	account := newCNQuotaExhaustedTestAccount(0)
+
+	shouldDisable := runKimi403(service, account, kimiWeeklyQuotaExhaustedBody)
+
+	require.True(t, shouldDisable)
+	require.Equal(t, 0, repo.setErrorCalls)
+	require.Equal(t, 0, repo.rateLimitedCalls)
+	require.Equal(t, 1, repo.tempCalls)
+	require.Contains(t, repo.lastTempReason, cnQuotaExhaustedReasonPrefix)
+	require.Greater(t, len(repo.lastTempReason), len(cnQuotaExhaustedReasonPrefix))
+}
+
+func TestHandleUpstreamError_KimiQuotaExhausted403MatchedByMessageOnly(t *testing.T) {
+	repo := &rateLimitAccountRepoStub{}
+	service := NewRateLimitService(repo, nil, &config.Config{}, nil, nil)
+	account := newCNQuotaExhaustedTestAccount(2 * time.Hour)
+	// 无 error.type 字段，仅靠文案兜底匹配。
+	body := `{"error":{"message":"You've reached your weekly (7-day) usage limit. Your quota will reset when the current 7-day window ends."}}`
+
+	shouldDisable := runKimi403(service, account, body)
+
+	require.True(t, shouldDisable)
+	require.Equal(t, 0, repo.setErrorCalls)
+	require.Equal(t, 1, repo.rateLimitedCalls)
+}
+
+func TestHandleUpstreamError_KimiConcurrencyLimit403StillUsesConcurrencyPath(t *testing.T) {
+	repo := &rateLimitAccountRepoStub{}
+	service := NewRateLimitService(repo, nil, &config.Config{}, nil, nil)
+	account := newCNQuotaExhaustedTestAccount(2 * time.Hour)
+	body := fmt.Sprintf(`{"error":{"message":%q,"type":"access_terminated_error"}}`, kimiConcurrentRequestLimitMessage)
+
+	shouldDisable := runKimi403(service, account, body)
+
+	require.True(t, shouldDisable)
+	require.Equal(t, 0, repo.setErrorCalls)
+	require.Equal(t, 0, repo.rateLimitedCalls, "concurrency-limit message must keep its dedicated path")
+	require.Equal(t, 1, repo.tempCalls)
+	require.Contains(t, repo.lastTempReason, cnConcurrencyLimitReasonPrefix)
+}
+
+func TestHandleUpstreamError_KimiGeneric403StillEscalates(t *testing.T) {
+	repo := &rateLimitAccountRepoStub{}
+	counter := &openAI403CounterCacheStub{counts: []int64{3}}
+	service := NewRateLimitService(repo, nil, &config.Config{}, nil, nil)
+	service.SetOpenAI403CounterCache(counter)
+	account := newCNQuotaExhaustedTestAccount(2 * time.Hour)
+
+	shouldDisable := runKimi403(service, account, `{"error":{"message":"policy rejected this request"}}`)
+
+	require.True(t, shouldDisable)
+	require.Equal(t, 1, repo.setErrorCalls, "non-quota 403 keeps the generic escalation path")
+	require.Equal(t, 0, repo.rateLimitedCalls)
+}
+
+func TestHandleUpstreamError_KimiNonCodingPlanQuotaMessageUsesGeneric403(t *testing.T) {
+	repo := &rateLimitAccountRepoStub{}
+	counter := &openAI403CounterCacheStub{counts: []int64{3}}
+	service := NewRateLimitService(repo, nil, &config.Config{}, nil, nil)
+	service.SetOpenAI403CounterCache(counter)
+	account := newCNQuotaExhaustedTestAccount(2 * time.Hour)
+	account.Credentials = map[string]any{"account_mode": AccountModePayG}
+
+	shouldDisable := runKimi403(service, account, kimiWeeklyQuotaExhaustedBody)
+
+	require.True(t, shouldDisable)
+	require.Equal(t, 1, repo.setErrorCalls, "non-coding-plan account must keep the generic 403 escalation")
+	require.Equal(t, 0, repo.rateLimitedCalls)
+	require.Equal(t, 0, repo.tempCalls)
+}
+
+func TestHandleUpstreamError_NonCNAccessTerminated403KeepsGenericPath(t *testing.T) {
+	repo := &rateLimitAccountRepoStub{}
+	counter := &openAI403CounterCacheStub{counts: []int64{1}}
+	service := NewRateLimitService(repo, nil, &config.Config{}, nil, nil)
+	service.SetOpenAI403CounterCache(counter)
+	account := &Account{
+		ID:       402,
+		Platform: PlatformOpenAI,
+		Type:     AccountTypeOAuth,
+	}
+
+	shouldDisable := runKimi403(service, account, kimiWeeklyQuotaExhaustedBody)
+
+	require.True(t, shouldDisable)
+	require.Equal(t, 0, repo.rateLimitedCalls, "non-CN platform must not enter the CN quota path")
+	require.Equal(t, 0, repo.setErrorCalls)
+	require.Equal(t, 1, repo.tempCalls)
+	require.Contains(t, repo.lastTempReason, "(1/3)")
+}
+
+func TestIsCNProviderQuotaExhausted403_Classification(t *testing.T) {
+	kimi := newCNQuotaExhaustedTestAccount(0)
+	require.True(t, isCNProviderQuotaExhausted403(kimi, []byte(kimiWeeklyQuotaExhaustedBody), ""))
+	require.True(t, isCNProviderQuotaExhausted403(kimi, []byte(`{"error":{"message":"You've reached your weekly (7-day) usage limit."}}`), "You've reached your weekly (7-day) usage limit."))
+	require.False(t, isCNProviderQuotaExhausted403(kimi, []byte(`{"error":{"message":"policy rejected"}}`), "policy rejected"))
+	require.False(t, isCNProviderQuotaExhausted403(kimi, nil, ""))
+
+	openai := &Account{ID: 1, Platform: PlatformOpenAI, Type: AccountTypeOAuth}
+	require.False(t, isCNProviderQuotaExhausted403(openai, []byte(kimiWeeklyQuotaExhaustedBody), ""))
+	require.False(t, isCNProviderQuotaExhausted403(nil, []byte(kimiWeeklyQuotaExhaustedBody), ""))
+}
@@ -742,3 +742,8 @@ func (s *RedeemService) reduceOrCancelSubscription(ctx context.Context, userID,

 	return nil
 }
+
+// GetUserHistoryPaginated returns all redemption types for the authenticated user.
+func (s *RedeemService) GetUserHistoryPaginated(ctx context.Context, userID int64, params pagination.PaginationParams) ([]RedeemCode, *pagination.PaginationResult, error) {
+	return s.redeemRepo.ListByUserPaginated(ctx, userID, params, "")
+}
@@ -0,0 +1,155 @@
+package service
+
+import (
+	"bytes"
+	"context"
+	"fmt"
+	"net/http"
+	"strings"
+	"time"
+
+	"github.com/gin-gonic/gin"
+	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
+)
+
+const (
+	SeedanceEndpointCreate           GrokMediaEndpoint        = "seedance_create"
+	SeedanceEndpointStatus           GrokMediaEndpoint        = "seedance_status"
+	SeedanceEndpointDelete           GrokMediaEndpoint        = "seedance_delete"
+	OpenAIEndpointCapabilitySeedance OpenAIEndpointCapability = "seedance"
+)
+
+func (e GrokMediaEndpoint) IsSeedance() bool {
+	return e == SeedanceEndpointCreate || e == SeedanceEndpointStatus || e == SeedanceEndpointDelete
+}
+
+// SeedanceTaskKey isolates ownership and billing keys from other video providers.
+func SeedanceTaskKey(id string) string { return "seedance:" + strings.TrimSpace(id) }
+
+func ParseSeedanceRequest(body []byte) (GrokMediaRequestInfo, error) {
+	var info GrokMediaRequestInfo
+	if !gjson.ValidBytes(body) || !gjson.ParseBytes(body).IsObject() {
+		return info, fmt.Errorf("request body must be a JSON object")
+	}
+	model := gjson.GetBytes(body, "model")
+	if model.Type != gjson.String || strings.TrimSpace(model.String()) == "" {
+		return info, fmt.Errorf("model is required")
+	}
+	content := gjson.GetBytes(body, "content")
+	if !content.IsArray() || len(content.Array()) == 0 {
+		return info, fmt.Errorf("content must be a non-empty array")
+	}
+	info.Model = strings.TrimSpace(model.String())
+	var texts []string
+	for _, item := range content.Array() {
+		switch item.Get("type").String() {
+		case "text":
+			texts = append(texts, item.Get("text").String())
+		case "image_url":
+			info.InputImageURLs = append(info.InputImageURLs, item.Get("image_url.url").String())
+		}
+	}
+	info.Prompt = strings.Join(texts, "\n")
+	return info, nil
+}
+
+func buildSeedanceURL(base string, endpoint GrokMediaEndpoint, taskID string) (string, error) {
+	base = strings.TrimRight(base, "/")
+	// Accept an origin, a proxy prefix, or the full Ark API base.
+	if !strings.HasSuffix(base, "/api/v3") && !strings.HasSuffix(base, "/v3") {
+		base += "/api/v3"
+	}
+	base += "/contents/generations/tasks"
+	if endpoint != SeedanceEndpointCreate {
+		if err := validateUpstreamPathSegment("Seedance task ID", taskID); err != nil || strings.TrimSpace(taskID) == "" {
+			return "", fmt.Errorf("invalid Seedance task ID")
+		}
+		base += "/" + taskID
+	}
+	return base, nil
+}
+
+// ForwardSeedance preserves the Ark protocol, including multimodal content and
+// future fields. Only model is rewritten using the account's configured mapping.
+func (s *OpenAIGatewayService) ForwardSeedance(ctx context.Context, c *gin.Context, account *Account, endpoint GrokMediaEndpoint, taskID string, body []byte) (*OpenAIForwardResult, error) {
+	if !account.SupportsOpenAIEndpointCapability(OpenAIEndpointCapabilitySeedance) || !endpoint.IsSeedance() {
+		return nil, fmt.Errorf("seedance requires an OpenAI API key account with a custom base URL")
+	}
+	base, err := s.validateUpstreamBaseURL(account.GetCredential("base_url"))
+	if err != nil {
+		return nil, err
+	}
+	target, err := buildSeedanceURL(base, endpoint, strings.TrimPrefix(taskID, "seedance:"))
+	if err != nil {
+		return nil, err
+	}
+	model, upstreamModel := "", ""
+	method := http.MethodGet
+	switch endpoint {
+	case SeedanceEndpointCreate:
+		info, parseErr := ParseSeedanceRequest(body)
+		if parseErr != nil {
+			return nil, parseErr
+		}
+		model = info.Model
+		upstreamModel = account.GetMappedModel(model)
+		body, err = sjson.SetBytes(body, "model", upstreamModel)
+		if err != nil {
+			return nil, err
+		}
+		method = http.MethodPost
+	case SeedanceEndpointDelete:
+		method = http.MethodDelete
+	}
+	token := strings.TrimSpace(account.GetCredential("api_key"))
+	if token == "" {
+		return nil, fmt.Errorf("seedance account missing api_key")
+	}
+	started := time.Now()
+	req, err := http.NewRequestWithContext(ctx, method, target, bytes.NewReader(body))
+	if err != nil {
+		return nil, err
+	}
+	req.Header.Set("Authorization", "Bearer "+token)
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Accept", "application/json")
+	account.ApplyHeaderOverrides(req.Header)
+	proxy := ""
+	if account.ProxyID != nil && account.Proxy != nil {
+		proxy = account.Proxy.URL()
+	}
+	resp, err := s.httpUpstream.Do(req, proxy, account.ID, account.Concurrency)
+	SetOpsLatencyMs(c, OpsUpstreamLatencyMsKey, time.Since(started).Milliseconds())
+	if err != nil {
+		return nil, err
+	}
+	defer func() { _ = resp.Body.Close() }()
+	responseBody, err := ReadUpstreamResponseBody(resp.Body, s.cfg, c, openAITooLargeError)
+	if err != nil {
+		return nil, err
+	}
+	// Do not retry ambiguous asynchronous creates: the upstream may already have
+	// accepted a billable job. Preserve native error codes and response bodies.
+	if resp.StatusCode >= 300 {
+		writeGrokMediaResponse(c, resp, responseBody, s.responseHeaderFilter)
+		return nil, fmt.Errorf("seedance upstream status %d", resp.StatusCode)
+	}
+	result := &OpenAIForwardResult{Model: model, BillingModel: model, UpstreamModel: upstreamModel, Duration: time.Since(started), ResponseHeaders: resp.Header.Clone()}
+	if endpoint == SeedanceEndpointCreate {
+		id := strings.TrimSpace(gjson.GetBytes(responseBody, "id").String())
+		if id == "" {
+			return nil, fmt.Errorf("seedance create response missing task ID")
+		}
+		result.ResponseID = SeedanceTaskKey(id)
+	}
+	if endpoint == SeedanceEndpointStatus {
+		result.ResponseID = taskID
+		result.UpstreamModel = gjson.GetBytes(responseBody, "model").String()
+		if gjson.GetBytes(responseBody, "status").String() == "succeeded" {
+			result.Usage.OutputTokens = max(0, int(gjson.GetBytes(responseBody, "usage.completion_tokens").Int()))
+		}
+	}
+	writeGrokMediaResponse(c, resp, responseBody, s.responseHeaderFilter)
+	return result, nil
+}
@@ -0,0 +1,112 @@
+//go:build unit
+
+package service
+
+import (
+	"context"
+	"io"
+	"net/http"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+	"github.com/tidwall/gjson"
+)
+
+func seedanceTestAccount() *Account {
+	return &Account{ID: 9, Platform: PlatformOpenAI, Type: AccountTypeAPIKey, Credentials: map[string]any{
+		"api_key": "ark-secret", "base_url": "https://ark.cn-beijing.volces.com/api/v3",
+		"openai_capabilities": []string{"seedance"},
+		"model_mapping":       map[string]any{"video": "ep-seedance"},
+	}}
+}
+
+func TestSeedanceNativeForwarding(t *testing.T) {
+	body := []byte(`{"model":"video","content":[{"type":"text","text":"waves"},{"type":"image_url","image_url":{"url":"https://example.com/first.png"},"role":"first_frame"},{"type":"audio_url","audio_url":{"url":"https://example.com/audio.mp3"}}],"duration":-1,"generate_audio":true,"future_field":{"keep":true}}`)
+	upstream := &grokMediaContentUpstreamStub{response: grokMediaContentStatusResponse(`{"id":"task-1"}`)}
+	svc := &OpenAIGatewayService{httpUpstream: upstream}
+	c, w := grokMediaContentTestContext(http.MethodPost, "/api/v3/contents/generations/tasks", nil)
+	result, err := svc.ForwardSeedance(context.Background(), c, seedanceTestAccount(), SeedanceEndpointCreate, "", body)
+	require.NoError(t, err)
+	require.JSONEq(t, `{"id":"task-1"}`, w.Body.String())
+	require.Equal(t, "seedance:task-1", result.ResponseID)
+	require.Zero(t, result.Usage.OutputTokens)
+	require.Equal(t, "video", result.BillingModel)
+	require.Equal(t, "ep-seedance", result.UpstreamModel)
+	require.Equal(t, "https://ark.cn-beijing.volces.com/api/v3/contents/generations/tasks", upstream.request.URL.String())
+	require.Equal(t, "Bearer ark-secret", upstream.request.Header.Get("Authorization"))
+	forwarded, err := io.ReadAll(upstream.request.Body)
+	require.NoError(t, err)
+	require.Equal(t, "ep-seedance", gjson.GetBytes(forwarded, "model").String())
+	for _, field := range []string{"content", "duration", "generate_audio", "future_field"} {
+		require.Equal(t, gjson.GetBytes(body, field).Raw, gjson.GetBytes(forwarded, field).Raw)
+	}
+}
+
+func TestSeedanceStatusAndDelete(t *testing.T) {
+	for _, status := range []string{"queued", "running", "failed", "cancelled", "expired", "succeeded"} {
+		t.Run(status, func(t *testing.T) {
+			body := `{"id":"task-1","status":"` + status + `","model":"ep-seedance","content":{"video_url":"https://cdn.example/video.mp4"},"usage":{"completion_tokens":12345}}`
+			upstream := &grokMediaContentUpstreamStub{response: grokMediaContentStatusResponse(body)}
+			svc := &OpenAIGatewayService{httpUpstream: upstream}
+			c, w := grokMediaContentTestContext(http.MethodGet, "/api/v3/contents/generations/tasks/task-1", nil)
+			result, err := svc.ForwardSeedance(context.Background(), c, seedanceTestAccount(), SeedanceEndpointStatus, "seedance:task-1", nil)
+			require.NoError(t, err)
+			require.JSONEq(t, body, w.Body.String())
+			require.Equal(t, "/api/v3/contents/generations/tasks/task-1", upstream.request.URL.Path)
+			if status == "succeeded" {
+				require.Equal(t, 12345, result.Usage.OutputTokens)
+			} else {
+				require.Zero(t, result.Usage.OutputTokens)
+			}
+			require.Zero(t, result.VideoCount, "must use token billing, not Grok seconds")
+		})
+	}
+	upstream := &grokMediaContentUpstreamStub{response: grokMediaContentStatusResponse("")}
+	upstream.response.StatusCode = http.StatusNoContent
+	svc := &OpenAIGatewayService{httpUpstream: upstream}
+	c, w := grokMediaContentTestContext(http.MethodDelete, "/api/v3/contents/generations/tasks/task-1", nil)
+	_, err := svc.ForwardSeedance(context.Background(), c, seedanceTestAccount(), SeedanceEndpointDelete, "seedance:task-1", nil)
+	require.NoError(t, err)
+	require.Equal(t, http.MethodDelete, upstream.request.Method)
+	require.Equal(t, http.StatusNoContent, w.Code)
+}
+
+func TestSeedanceValidationAndCapability(t *testing.T) {
+	for _, body := range []string{`{`, `[]`, `{}`, `{"model":12,"content":[{}]}`, `{"model":"x","content":[]}`} {
+		_, err := ParseSeedanceRequest([]byte(body))
+		require.Error(t, err)
+	}
+	info, err := ParseSeedanceRequest([]byte(`{"model":"x","content":[{"type":"text","text":"first"},{"type":"text","text":"second"},{"type":"image_url","image_url":{"url":"https://example.com/a.png"}}]}`))
+	require.NoError(t, err)
+	require.Contains(t, string(info.ModerationBody()), "first")
+	require.Contains(t, string(info.ModerationBody()), "second")
+	require.Contains(t, string(info.ModerationBody()), "https://example.com/a.png")
+	for _, id := range []string{"", "..", "a/b", "a?b", "a#b", "%2e%2e"} {
+		_, err := buildSeedanceURL("https://example.com", SeedanceEndpointStatus, id)
+		require.Error(t, err, id)
+	}
+	for _, base := range []string{"https://example.com", "https://example.com/api/v3/", "https://example.com/v3"} {
+		url, err := buildSeedanceURL(base, SeedanceEndpointCreate, "")
+		require.NoError(t, err)
+		require.NotContains(t, url, "/v3/api/v3")
+	}
+	a := seedanceTestAccount()
+	require.True(t, a.SupportsOpenAIEndpointCapability(OpenAIEndpointCapabilitySeedance))
+	a.Type = AccountTypeOAuth
+	require.False(t, a.SupportsOpenAIEndpointCapability(OpenAIEndpointCapabilitySeedance))
+	a.Type = AccountTypeAPIKey
+	delete(a.Credentials, "openai_capabilities")
+	require.False(t, a.SupportsOpenAIEndpointCapability(OpenAIEndpointCapabilitySeedance))
+}
+
+func TestSeedancePreservesUpstreamErrorsWithoutRetry(t *testing.T) {
+	upstream := &grokMediaContentUpstreamStub{response: grokMediaContentStatusResponse(`{"error":{"code":"QuotaExceeded","message":"quota exhausted"}}`)}
+	upstream.response.StatusCode = 429
+	svc := &OpenAIGatewayService{httpUpstream: upstream}
+	c, w := grokMediaContentTestContext(http.MethodPost, "/api/v3/contents/generations/tasks", nil)
+	_, err := svc.ForwardSeedance(context.Background(), c, seedanceTestAccount(), SeedanceEndpointCreate, "", []byte(`{"model":"video","content":[{"type":"text","text":"waves"}]}`))
+	require.Error(t, err)
+	require.Equal(t, 429, w.Code)
+	require.Contains(t, w.Body.String(), "QuotaExceeded")
+	require.Len(t, upstream.requests, 1)
+}
@@ -51,7 +51,6 @@ func (r *tokenRefreshCandidateRepo) ListOAuthRefreshCandidatePage(_ context.Cont
 			}
 		}
 		if options.ActiveOnly && account.Status != StatusActive ||
-			!account.Schedulable ||
 			account.Type != AccountTypeOAuth ||
 			!platformAllowed ||
 			options.RequireRefreshToken && strings.TrimSpace(refreshToken) == "" ||
@@ -176,7 +175,7 @@ func TestTokenRefreshService_ProcessRefreshUsesOAuthRefreshCandidates(t *testing
 				Type:        AccountTypeOAuth,
 				Status:      StatusActive,
 				Schedulable: false,
-				Credentials: map[string]any{"refresh_token": "permanently-rejected-token"},
+				Credentials: map[string]any{"refresh_token": "paused-account-token"},
 			},
 		},
 	}
@@ -195,7 +194,9 @@ func TestTokenRefreshService_ProcessRefreshUsesOAuthRefreshCandidates(t *testing
 	svc.processRefresh()

 	require.Zero(t, repo.listActiveCalls, "TokenRefreshService should not use the broad active-account query")
-	require.ElementsMatch(t, []int64{1, 6}, repo.updatedCredentialIDs)
+	// Account 7 is paused (schedulable=false) but active: it must still be
+	// refreshed so its stored access_token does not silently expire.
+	require.ElementsMatch(t, []int64{1, 6, 7}, repo.updatedCredentialIDs)
 	require.Equal(t, 1, repo.clearTempCalls, "successful refresh should clear the OAuth 401 temp-unschedulable state")
 }

@@ -30,6 +30,21 @@
 - Sub2API 继续负责响应状态处理、SSE 解析、错误映射、用量统计、计费和下游输出。
 - 灰度比例以账号 ID 稳定分桶，未命中的 OAuth 账号继续使用原有内置路径。

+## 宿主服务（HostService）
+
+`HostService` 是宿主经 go-plugin broker 反向暴露给插件的通用能力层，独立于具体插件类型，供有状态插件使用。它通过 `TransportPlugin.InitHostServices` 在运行时协商：宿主启动后把一个 broker 流 id 交给插件，插件用它拨号回宿主并获得 `HostServiceClient`。
+
+- **可选且向后兼容**：未实现 `InitHostServices` 的旧插件返回 `Unimplemented`，宿主静默跳过，转发能力不受影响。宿主服务有独立的 `HostServiceAPIVersion`，新增能力不会改变传输契约版本，也不会使既有插件失效。
+- **随进程回收**：宿主服务实例与插件进程生命周期绑定，插件退出时 broker 关闭并自动 `GracefulStop`，无需插件手动清理。
+
+当前提供的通用设施：
+
+- **命名空间键值存储（KV）**：`KVGet` / `KVSet` / `KVDelete` / `KVList`，为插件持久化跨请求、跨副本、跨重启的状态。存储由 Redis 支撑，因此多实例部署天然共享同一份状态。
+  - **命名空间隔离**：宿主根据服务该连接的运行时注入插件身份（`pluginKey`），插件无法伪造，也无法读写其它插件的命名空间。
+  - **护栏**：`namespace` / `key` 仅允许 `[A-Za-z0-9._-]`；单值上限 256 KiB；`ttl_seconds` 为 0 表示不过期、正值有上限；`KVList` 返回条数有上限。
+
+新增宿主设施时，在 `HostService` 上追加 RPC 即可，无需改动传输契约或清单格式。
+
 ## 包结构

 ```text
@@ -61,15 +76,21 @@ ui/assets/...

 插件 UI 由包内静态文件实现，宿主使用只有 `allow-scripts` 权限的 sandbox iframe 加载。iframe 没有管理员 Token，也不能直接访问管理 API。宿主为每次打开配置页生成短时资源 URL 和独立 Bridge Token，并且同时校验消息来源窗口与 Token。

-UI 可以发送以下消息：
+UI 可以发送以下消息。消息按语义分层，鉴权与副作用一致对应（读操作免二次验证，写/主动测试需要二次验证），任何插件都可复用，不针对具体插件定制：

- `config.load`
- `config.save`
- `config.test`
- `ui.resize`
- `ui.notify`
+| 消息 | 语义 | 二次验证 | 映射的插件 RPC |
+| --- | --- | --- | --- |
+| `config.load` | 读取已保存配置 | 否 | 宿主数据库 |
+| `config.save` | 写入配置 | 是 | `ValidateConfig` + `ApplyConfig` |
+| `config.test` | 主动测试配置/连通性（可产生副作用） | 是 | `TestConfig` |
+| `plugin.status` | 读取运行时状态（无副作用） | 否 | `Health`（`status_json`） |
+| `ui.resize` / `ui.notify` | 仅 UI 交互 | — | — |

-每个请求消息带 `request_id`，宿主以 `<type>.result` 返回结果。配置整体使用 Sub2API 的密钥加密后存入数据库；运行中插件会先验证并应用新配置，数据库写入失败时恢复旧配置。
+每个请求消息带 `request_id`，宿主以 `<type>.result` 返回结果。
+
+- 配置整体使用 Sub2API 的密钥加密后存入数据库；运行中插件会先验证并应用新配置，数据库写入失败时恢复旧配置。
+- `config.test` 的结果由插件 UI 自行展示（内联或经 `ui.notify`），宿主不再对成功结果强制弹出提示，避免插件把它当作轻量状态轮询时刷屏。
+- `plugin.status` 是通用的**只读**状态通道：宿主经 `GET /admin/plugins/:id/status` 调用运行中插件的 `Health`，返回 `{healthy, message, status_json}`。`status_json` 是插件自定义的**不透明** JSON 快照（宿主不解析、不参与健康判定），插件必须以无副作用方式生成（不得应用配置、访问上游或触发探测），因此该端点只读、免二次验证。插件未运行时返回 `healthy=false` 且不含 `status_json`。这样带状态面板的插件无需滥用 `config.test` 即可展示实时状态。

 ## 协议源码

@@ -44,10 +44,13 @@ UI 到宿主：
 | `config.load` | 无 | `config` |
 | `config.save` | `config` 对象 | 规范化后的 `config` |
 | `config.test` | 无 | `result` |
+| `plugin.status` | 无 | `result`（`Health`：`{healthy, message, status_json}`） |
 | `ui.resize` | `height` | 无响应 |
 | `ui.notify` | `level`、`message` | 无响应 |

-`config.test` 在 v1 中测试已保存配置。UI 若要测试当前表单，应先调用 `config.save`。
+`config.test` 在 v1 中测试已保存配置（需二次验证，可产生副作用）。UI 若要测试当前表单，应先调用 `config.save`。
+
+`plugin.status` 是只读运行时状态通道：无副作用、免二次验证、不弹宿主提示，供状态面板轮询。它映射到插件 `Health`，`result.status_json` 是插件自定义的不透明 JSON 快照。带状态展示的插件应使用它，而不是把 `config.test` 当作状态轮询。

 ## 必须执行的校验

@@ -11,6 +11,28 @@ service TransportPlugin {
  rpc ApplyConfig(ApplyConfigRequest) returns (ApplyConfigResponse);
  rpc TestConfig(TestConfigRequest) returns (TestConfigResponse);
  rpc Forward(stream ForwardRequest) returns (stream ForwardResponse);
+  // InitHostServices hands the plugin a go-plugin broker stream id it can dial
+  // back to reach the host-provided HostService. It is an optional, generic
+  // lifecycle hook: plugins built against an older contract leave it
+  // Unimplemented and the host simply runs without offering host services.
+  rpc InitHostServices(InitHostServicesRequest) returns (InitHostServicesResponse);
+}
+
+// HostService is implemented by the host and dialed by the plugin over the
+// go-plugin broker id delivered via InitHostServices. It exposes generic,
+// capability-neutral facilities that any plugin may use; new facilities are
+// added here as new RPCs without changing the transport contract version.
+service HostService {
+  rpc KVGet(KVGetRequest) returns (KVGetResponse);
+  rpc KVSet(KVSetRequest) returns (KVSetResponse);
+  rpc KVDelete(KVDeleteRequest) returns (KVDeleteResponse);
+  rpc KVList(KVListRequest) returns (KVListResponse);
+  // Account directory: enumerate the accounts the plugin is bound to act on and
+  // resolve the outbound identity (credentials/headers/proxy) the host would
+  // attach to a live request. The host restricts both calls to accounts the
+  // requesting plugin's capability already covers.
+  rpc ListAccounts(ListAccountsRequest) returns (ListAccountsResponse);
+  rpc ResolveOutboundIdentity(ResolveOutboundIdentityRequest) returns (ResolveOutboundIdentityResponse);
 }

 message GetInfoRequest {}
@@ -28,6 +50,12 @@ message HealthRequest {}
 message HealthResponse {
  bool healthy = 1;
  string message = 2;
+  // status_json is an optional plugin-defined JSON blob describing current runtime
+  // state (e.g. a status summary for the config UI). It MUST be produced passively
+  // — reading it must not apply config, reach upstream, or cause side effects — so
+  // the host can expose it via a lightweight, ungated status endpoint. Ignored by
+  // the host's health gating.
+  string status_json = 3;
 }

 message ValidateConfigRequest {
@@ -57,6 +85,9 @@ message TestConfigResponse {
  bool success = 1;
  string message = 2;
  int64 latency_ms = 3;
+  // status_json is an optional plugin-defined JSON blob surfaced to the config UI
+  // alongside the test result (e.g. a runtime status summary). Ignored by the host.
+  string status_json = 4;
 }

 message HeaderValues {
@@ -115,3 +146,79 @@ message ForwardResponse {
    ForwardResponseError error = 4;
  }
 }
+
+message InitHostServicesRequest {
+  // host_service_id is the broker stream id the plugin dials to reach HostService.
+  uint32 host_service_id = 1;
+  // host_service_api_version lets the plugin reject a host it cannot talk to.
+  uint32 host_service_api_version = 2;
+}
+
+message InitHostServicesResponse {
+  // ready is true once the plugin has successfully dialed back the host service.
+  bool ready = 1;
+  string message = 2;
+}
+
+// KV messages back a namespaced, per-plugin persistent key-value store. The
+// host derives the owning plugin from the runtime that serves the connection,
+// so a plugin can never read or write another plugin's namespace.
+message KVGetRequest {
+  string namespace = 1;
+  string key = 2;
+}
+
+message KVGetResponse {
+  bool found = 1;
+  bytes value = 2;
+}
+
+message KVSetRequest {
+  string namespace = 1;
+  string key = 2;
+  bytes value = 3;
+  // ttl_seconds > 0 sets an expiry; 0 stores the entry without expiry.
+  int64 ttl_seconds = 4;
+}
+
+message KVSetResponse {}
+
+message KVDeleteRequest {
+  string namespace = 1;
+  string key = 2;
+}
+
+message KVDeleteResponse {}
+
+message KVListRequest {
+  string namespace = 1;
+  string key_prefix = 2;
+  int32 limit = 3;
+}
+
+message KVListResponse {
+  repeated string keys = 1;
+}
+
+message ListAccountsRequest {
+  string platform = 1;
+  string account_type = 2;
+}
+
+message ListAccountsResponse {
+  repeated int64 account_ids = 1;
+}
+
+message ResolveOutboundIdentityRequest {
+  int64 account_id = 1;
+}
+
+message ResolveOutboundIdentityResponse {
+  bool found = 1;
+  int64 account_id = 2;
+  string platform = 3;
+  string account_type = 4;
+  string proxy_url = 5;
+  string token = 6;
+  map<string, HeaderValues> headers = 7;
+}
@@ -19,12 +19,13 @@ import (
 const _ = grpc.SupportPackageIsVersion9

 const (
-	TransportPlugin_GetInfo_FullMethodName        = "/sub2api.plugin.v1.TransportPlugin/GetInfo"
-	TransportPlugin_Health_FullMethodName         = "/sub2api.plugin.v1.TransportPlugin/Health"
-	TransportPlugin_ValidateConfig_FullMethodName = "/sub2api.plugin.v1.TransportPlugin/ValidateConfig"
-	TransportPlugin_ApplyConfig_FullMethodName    = "/sub2api.plugin.v1.TransportPlugin/ApplyConfig"
-	TransportPlugin_TestConfig_FullMethodName     = "/sub2api.plugin.v1.TransportPlugin/TestConfig"
-	TransportPlugin_Forward_FullMethodName        = "/sub2api.plugin.v1.TransportPlugin/Forward"
+	TransportPlugin_GetInfo_FullMethodName          = "/sub2api.plugin.v1.TransportPlugin/GetInfo"
+	TransportPlugin_Health_FullMethodName           = "/sub2api.plugin.v1.TransportPlugin/Health"
+	TransportPlugin_ValidateConfig_FullMethodName   = "/sub2api.plugin.v1.TransportPlugin/ValidateConfig"
+	TransportPlugin_ApplyConfig_FullMethodName      = "/sub2api.plugin.v1.TransportPlugin/ApplyConfig"
+	TransportPlugin_TestConfig_FullMethodName       = "/sub2api.plugin.v1.TransportPlugin/TestConfig"
+	TransportPlugin_Forward_FullMethodName          = "/sub2api.plugin.v1.TransportPlugin/Forward"
+	TransportPlugin_InitHostServices_FullMethodName = "/sub2api.plugin.v1.TransportPlugin/InitHostServices"
 )

 // TransportPluginClient is the client API for TransportPlugin service.
@@ -37,6 +38,11 @@ type TransportPluginClient interface {
 	ApplyConfig(ctx context.Context, in *ApplyConfigRequest, opts ...grpc.CallOption) (*ApplyConfigResponse, error)
 	TestConfig(ctx context.Context, in *TestConfigRequest, opts ...grpc.CallOption) (*TestConfigResponse, error)
 	Forward(ctx context.Context, opts ...grpc.CallOption) (grpc.BidiStreamingClient[ForwardRequest, ForwardResponse], error)
+	// InitHostServices hands the plugin a go-plugin broker stream id it can dial
+	// back to reach the host-provided HostService. It is an optional, generic
+	// lifecycle hook: plugins built against an older contract leave it
+	// Unimplemented and the host simply runs without offering host services.
+	InitHostServices(ctx context.Context, in *InitHostServicesRequest, opts ...grpc.CallOption) (*InitHostServicesResponse, error)
 }

 type transportPluginClient struct {
@@ -110,6 +116,16 @@ func (c *transportPluginClient) Forward(ctx context.Context, opts ...grpc.CallOp
 // This type alias is provided for backwards compatibility with existing code that references the prior non-generic stream type by name.
 type TransportPlugin_ForwardClient = grpc.BidiStreamingClient[ForwardRequest, ForwardResponse]

+func (c *transportPluginClient) InitHostServices(ctx context.Context, in *InitHostServicesRequest, opts ...grpc.CallOption) (*InitHostServicesResponse, error) {
+	cOpts := append([]grpc.CallOption{grpc.StaticMethod()}, opts...)
+	out := new(InitHostServicesResponse)
+	err := c.cc.Invoke(ctx, TransportPlugin_InitHostServices_FullMethodName, in, out, cOpts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
 // TransportPluginServer is the server API for TransportPlugin service.
 // All implementations must embed UnimplementedTransportPluginServer
 // for forward compatibility.
@@ -120,6 +136,11 @@ type TransportPluginServer interface {
 	ApplyConfig(context.Context, *ApplyConfigRequest) (*ApplyConfigResponse, error)
 	TestConfig(context.Context, *TestConfigRequest) (*TestConfigResponse, error)
 	Forward(grpc.BidiStreamingServer[ForwardRequest, ForwardResponse]) error
+	// InitHostServices hands the plugin a go-plugin broker stream id it can dial
+	// back to reach the host-provided HostService. It is an optional, generic
+	// lifecycle hook: plugins built against an older contract leave it
+	// Unimplemented and the host simply runs without offering host services.
+	InitHostServices(context.Context, *InitHostServicesRequest) (*InitHostServicesResponse, error)
 	mustEmbedUnimplementedTransportPluginServer()
 }

@@ -148,6 +169,9 @@ func (UnimplementedTransportPluginServer) TestConfig(context.Context, *TestConfi
 func (UnimplementedTransportPluginServer) Forward(grpc.BidiStreamingServer[ForwardRequest, ForwardResponse]) error {
 	return status.Error(codes.Unimplemented, "method Forward not implemented")
 }
+func (UnimplementedTransportPluginServer) InitHostServices(context.Context, *InitHostServicesRequest) (*InitHostServicesResponse, error) {
+	return nil, status.Error(codes.Unimplemented, "method InitHostServices not implemented")
+}
 func (UnimplementedTransportPluginServer) mustEmbedUnimplementedTransportPluginServer() {}
 func (UnimplementedTransportPluginServer) testEmbeddedByValue()                         {}

@@ -266,6 +290,24 @@ func _TransportPlugin_Forward_Handler(srv interface{}, stream grpc.ServerStream)
 // This type alias is provided for backwards compatibility with existing code that references the prior non-generic stream type by name.
 type TransportPlugin_ForwardServer = grpc.BidiStreamingServer[ForwardRequest, ForwardResponse]

+func _TransportPlugin_InitHostServices_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(InitHostServicesRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(TransportPluginServer).InitHostServices(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: TransportPlugin_InitHostServices_FullMethodName,
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(TransportPluginServer).InitHostServices(ctx, req.(*InitHostServicesRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
 // TransportPlugin_ServiceDesc is the grpc.ServiceDesc for TransportPlugin service.
 // It's only intended for direct use with grpc.RegisterService,
 // and not to be introspected or modified (even as a copy)
@@ -293,6 +335,10 @@ var TransportPlugin_ServiceDesc = grpc.ServiceDesc{
 			MethodName: "TestConfig",
 			Handler:    _TransportPlugin_TestConfig_Handler,
 		},
+		{
+			MethodName: "InitHostServices",
+			Handler:    _TransportPlugin_InitHostServices_Handler,
+		},
 	},
 	Streams: []grpc.StreamDesc{
 		{
@@ -304,3 +350,313 @@ var TransportPlugin_ServiceDesc = grpc.ServiceDesc{
 	},
 	Metadata: "plugin.proto",
 }
+
+const (
+	HostService_KVGet_FullMethodName                   = "/sub2api.plugin.v1.HostService/KVGet"
+	HostService_KVSet_FullMethodName                   = "/sub2api.plugin.v1.HostService/KVSet"
+	HostService_KVDelete_FullMethodName                = "/sub2api.plugin.v1.HostService/KVDelete"
+	HostService_KVList_FullMethodName                  = "/sub2api.plugin.v1.HostService/KVList"
+	HostService_ListAccounts_FullMethodName            = "/sub2api.plugin.v1.HostService/ListAccounts"
+	HostService_ResolveOutboundIdentity_FullMethodName = "/sub2api.plugin.v1.HostService/ResolveOutboundIdentity"
+)
+
+// HostServiceClient is the client API for HostService service.
+//
+// For semantics around ctx use and closing/ending streaming RPCs, please refer to https://pkg.go.dev/google.golang.org/grpc/?tab=doc#ClientConn.NewStream.
+//
+// HostService is implemented by the host and dialed by the plugin over the
+// go-plugin broker id delivered via InitHostServices. It exposes generic,
+// capability-neutral facilities that any plugin may use; new facilities are
+// added here as new RPCs without changing the transport contract version.
+type HostServiceClient interface {
+	KVGet(ctx context.Context, in *KVGetRequest, opts ...grpc.CallOption) (*KVGetResponse, error)
+	KVSet(ctx context.Context, in *KVSetRequest, opts ...grpc.CallOption) (*KVSetResponse, error)
+	KVDelete(ctx context.Context, in *KVDeleteRequest, opts ...grpc.CallOption) (*KVDeleteResponse, error)
+	KVList(ctx context.Context, in *KVListRequest, opts ...grpc.CallOption) (*KVListResponse, error)
+	// Account directory: enumerate the accounts the plugin is bound to act on and
+	// resolve the outbound identity (credentials/headers/proxy) the host would
+	// attach to a live request. The host restricts both calls to accounts the
+	// requesting plugin's capability already covers.
+	ListAccounts(ctx context.Context, in *ListAccountsRequest, opts ...grpc.CallOption) (*ListAccountsResponse, error)
+	ResolveOutboundIdentity(ctx context.Context, in *ResolveOutboundIdentityRequest, opts ...grpc.CallOption) (*ResolveOutboundIdentityResponse, error)
+}
+
+type hostServiceClient struct {
+	cc grpc.ClientConnInterface
+}
+
+func NewHostServiceClient(cc grpc.ClientConnInterface) HostServiceClient {
+	return &hostServiceClient{cc}
+}
+
+func (c *hostServiceClient) KVGet(ctx context.Context, in *KVGetRequest, opts ...grpc.CallOption) (*KVGetResponse, error) {
+	cOpts := append([]grpc.CallOption{grpc.StaticMethod()}, opts...)
+	out := new(KVGetResponse)
+	err := c.cc.Invoke(ctx, HostService_KVGet_FullMethodName, in, out, cOpts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *hostServiceClient) KVSet(ctx context.Context, in *KVSetRequest, opts ...grpc.CallOption) (*KVSetResponse, error) {
+	cOpts := append([]grpc.CallOption{grpc.StaticMethod()}, opts...)
+	out := new(KVSetResponse)
+	err := c.cc.Invoke(ctx, HostService_KVSet_FullMethodName, in, out, cOpts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *hostServiceClient) KVDelete(ctx context.Context, in *KVDeleteRequest, opts ...grpc.CallOption) (*KVDeleteResponse, error) {
+	cOpts := append([]grpc.CallOption{grpc.StaticMethod()}, opts...)
+	out := new(KVDeleteResponse)
+	err := c.cc.Invoke(ctx, HostService_KVDelete_FullMethodName, in, out, cOpts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *hostServiceClient) KVList(ctx context.Context, in *KVListRequest, opts ...grpc.CallOption) (*KVListResponse, error) {
+	cOpts := append([]grpc.CallOption{grpc.StaticMethod()}, opts...)
+	out := new(KVListResponse)
+	err := c.cc.Invoke(ctx, HostService_KVList_FullMethodName, in, out, cOpts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *hostServiceClient) ListAccounts(ctx context.Context, in *ListAccountsRequest, opts ...grpc.CallOption) (*ListAccountsResponse, error) {
+	cOpts := append([]grpc.CallOption{grpc.StaticMethod()}, opts...)
+	out := new(ListAccountsResponse)
+	err := c.cc.Invoke(ctx, HostService_ListAccounts_FullMethodName, in, out, cOpts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *hostServiceClient) ResolveOutboundIdentity(ctx context.Context, in *ResolveOutboundIdentityRequest, opts ...grpc.CallOption) (*ResolveOutboundIdentityResponse, error) {
+	cOpts := append([]grpc.CallOption{grpc.StaticMethod()}, opts...)
+	out := new(ResolveOutboundIdentityResponse)
+	err := c.cc.Invoke(ctx, HostService_ResolveOutboundIdentity_FullMethodName, in, out, cOpts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+// HostServiceServer is the server API for HostService service.
+// All implementations must embed UnimplementedHostServiceServer
+// for forward compatibility.
+//
+// HostService is implemented by the host and dialed by the plugin over the
+// go-plugin broker id delivered via InitHostServices. It exposes generic,
+// capability-neutral facilities that any plugin may use; new facilities are
+// added here as new RPCs without changing the transport contract version.
+type HostServiceServer interface {
+	KVGet(context.Context, *KVGetRequest) (*KVGetResponse, error)
+	KVSet(context.Context, *KVSetRequest) (*KVSetResponse, error)
+	KVDelete(context.Context, *KVDeleteRequest) (*KVDeleteResponse, error)
+	KVList(context.Context, *KVListRequest) (*KVListResponse, error)
+	// Account directory: enumerate the accounts the plugin is bound to act on and
+	// resolve the outbound identity (credentials/headers/proxy) the host would
+	// attach to a live request. The host restricts both calls to accounts the
+	// requesting plugin's capability already covers.
+	ListAccounts(context.Context, *ListAccountsRequest) (*ListAccountsResponse, error)
+	ResolveOutboundIdentity(context.Context, *ResolveOutboundIdentityRequest) (*ResolveOutboundIdentityResponse, error)
+	mustEmbedUnimplementedHostServiceServer()
+}
+
+// UnimplementedHostServiceServer must be embedded to have
+// forward compatible implementations.
+//
+// NOTE: this should be embedded by value instead of pointer to avoid a nil
+// pointer dereference when methods are called.
+type UnimplementedHostServiceServer struct{}
+
+func (UnimplementedHostServiceServer) KVGet(context.Context, *KVGetRequest) (*KVGetResponse, error) {
+	return nil, status.Error(codes.Unimplemented, "method KVGet not implemented")
+}
+func (UnimplementedHostServiceServer) KVSet(context.Context, *KVSetRequest) (*KVSetResponse, error) {
+	return nil, status.Error(codes.Unimplemented, "method KVSet not implemented")
+}
+func (UnimplementedHostServiceServer) KVDelete(context.Context, *KVDeleteRequest) (*KVDeleteResponse, error) {
+	return nil, status.Error(codes.Unimplemented, "method KVDelete not implemented")
+}
+func (UnimplementedHostServiceServer) KVList(context.Context, *KVListRequest) (*KVListResponse, error) {
+	return nil, status.Error(codes.Unimplemented, "method KVList not implemented")
+}
+func (UnimplementedHostServiceServer) ListAccounts(context.Context, *ListAccountsRequest) (*ListAccountsResponse, error) {
+	return nil, status.Error(codes.Unimplemented, "method ListAccounts not implemented")
+}
+func (UnimplementedHostServiceServer) ResolveOutboundIdentity(context.Context, *ResolveOutboundIdentityRequest) (*ResolveOutboundIdentityResponse, error) {
+	return nil, status.Error(codes.Unimplemented, "method ResolveOutboundIdentity not implemented")
+}
+func (UnimplementedHostServiceServer) mustEmbedUnimplementedHostServiceServer() {}
+func (UnimplementedHostServiceServer) testEmbeddedByValue()                     {}
+
+// UnsafeHostServiceServer may be embedded to opt out of forward compatibility for this service.
+// Use of this interface is not recommended, as added methods to HostServiceServer will
+// result in compilation errors.
+type UnsafeHostServiceServer interface {
+	mustEmbedUnimplementedHostServiceServer()
+}
+
+func RegisterHostServiceServer(s grpc.ServiceRegistrar, srv HostServiceServer) {
+	// If the following call panics, it indicates UnimplementedHostServiceServer was
+	// embedded by pointer and is nil.  This will cause panics if an
+	// unimplemented method is ever invoked, so we test this at initialization
+	// time to prevent it from happening at runtime later due to I/O.
+	if t, ok := srv.(interface{ testEmbeddedByValue() }); ok {
+		t.testEmbeddedByValue()
+	}
+	s.RegisterService(&HostService_ServiceDesc, srv)
+}
+
+func _HostService_KVGet_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(KVGetRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(HostServiceServer).KVGet(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: HostService_KVGet_FullMethodName,
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(HostServiceServer).KVGet(ctx, req.(*KVGetRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _HostService_KVSet_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(KVSetRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(HostServiceServer).KVSet(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: HostService_KVSet_FullMethodName,
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(HostServiceServer).KVSet(ctx, req.(*KVSetRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _HostService_KVDelete_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(KVDeleteRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(HostServiceServer).KVDelete(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: HostService_KVDelete_FullMethodName,
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(HostServiceServer).KVDelete(ctx, req.(*KVDeleteRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _HostService_KVList_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(KVListRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(HostServiceServer).KVList(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: HostService_KVList_FullMethodName,
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(HostServiceServer).KVList(ctx, req.(*KVListRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _HostService_ListAccounts_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(ListAccountsRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(HostServiceServer).ListAccounts(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: HostService_ListAccounts_FullMethodName,
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(HostServiceServer).ListAccounts(ctx, req.(*ListAccountsRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _HostService_ResolveOutboundIdentity_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(ResolveOutboundIdentityRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(HostServiceServer).ResolveOutboundIdentity(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: HostService_ResolveOutboundIdentity_FullMethodName,
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(HostServiceServer).ResolveOutboundIdentity(ctx, req.(*ResolveOutboundIdentityRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+// HostService_ServiceDesc is the grpc.ServiceDesc for HostService service.
+// It's only intended for direct use with grpc.RegisterService,
+// and not to be introspected or modified (even as a copy)
+var HostService_ServiceDesc = grpc.ServiceDesc{
+	ServiceName: "sub2api.plugin.v1.HostService",
+	HandlerType: (*HostServiceServer)(nil),
+	Methods: []grpc.MethodDesc{
+		{
+			MethodName: "KVGet",
+			Handler:    _HostService_KVGet_Handler,
+		},
+		{
+			MethodName: "KVSet",
+			Handler:    _HostService_KVSet_Handler,
+		},
+		{
+			MethodName: "KVDelete",
+			Handler:    _HostService_KVDelete_Handler,
+		},
+		{
+			MethodName: "KVList",
+			Handler:    _HostService_KVList_Handler,
+		},
+		{
+			MethodName: "ListAccounts",
+			Handler:    _HostService_ListAccounts_Handler,
+		},
+		{
+			MethodName: "ResolveOutboundIdentity",
+			Handler:    _HostService_ResolveOutboundIdentity_Handler,
+		},
+	},
+	Streams:  []grpc.StreamDesc{},
+	Metadata: "plugin.proto",
+}
--- a/Show More
+++ b/Show More
@@ -1 +1 @@
 .2.4
 .2.5