fix(llm): per-model protocol bindings for OpenCode Zen/Go presets

TelegramSearchBot.LLM.Test/Service/AI/LLM/AnthropicTransportSessionTests.cs

@@ -0,0 +1,29 @@
+using System.Linq;
+using System.Net.Http;
+using TelegramSearchBot.Service.AI.LLM.Transports;
+using Xunit;
+
+namespace TelegramSearchBot.LLM.Test.Service.AI.LLM {
+    public class AnthropicTransportSessionTests {
+        [Fact]
+        public void CreateSessionHttpClient_AppliesStableSessionForOpenCodeEndpoints() {
+            using var client = AnthropicMessagesTransport.CreateSessionHttpClient("https://opencode.ai/zen/go/v1", 42);
+
+            Assert.Equal("tsb-42", client.DefaultRequestHeaders.GetValues("x-opencode-session").Single());
+        }
+
+        [Fact]
+        public void CreateSessionHttpClient_UsesGlobalSessionWithoutChatId() {
+            using var client = AnthropicMessagesTransport.CreateSessionHttpClient("https://opencode.ai/zen/v1");
+
+            Assert.Equal("tsb-global", client.DefaultRequestHeaders.GetValues("x-opencode-session").Single());
+        }
+
+        [Fact]
+        public void CreateSessionHttpClient_LeavesOtherEndpointsUntouched() {
+            using var client = AnthropicMessagesTransport.CreateSessionHttpClient("https://api.anthropic.com/v1", 42);
+
+            Assert.False(client.DefaultRequestHeaders.Contains("x-opencode-session"));
+        }
+    }
+}

TelegramSearchBot.LLM.Test/Service/AI/LLM/LlmProviderCatalogTests.cs

@@ -0,0 +1,100 @@
+using System;
+using System.Linq;
+using TelegramSearchBot.Model.AI;
+using TelegramSearchBot.Service.AI.LLM;
+using Xunit;
+
+namespace TelegramSearchBot.LLM.Test.Service.AI.LLM {
+    public class LlmProviderCatalogTests {
+        [Theory]
+        [InlineData("claude-sonnet-5", "anthropic")]
+        [InlineData("Claude-Opus-5", "anthropic")]
+        [InlineData("gpt-5.6-luna", "responses")]
+        [InlineData("grok-4.6", "responses")]
+        [InlineData("gemini-3.8-flash", "google")]
+        public void OpenCodeZen_RulesMapModelsToProtocols(string modelName, string expectedBindingId) {
+            var zen = LlmProviderCatalog.FindById("opencode-zen")!;
+
+            Assert.Equal(expectedBindingId, zen.ModelBindingRules.ResolveBindingId(modelName));
+        }
+
+        [Theory]
+        [InlineData("glm-5.3", null)]
+        [InlineData("kimi-k3", null)]
+        [InlineData("deepseek-v4-pro", null)]
+        public void OpenCodeZen_UnmatchedModelsKeepDefaultBinding(string modelName, string? expectedBindingId) {
+            var zen = LlmProviderCatalog.FindById("opencode-zen")!;
+
+            Assert.Equal(expectedBindingId, zen.ModelBindingRules.ResolveBindingId(modelName));
+        }
+
+        [Theory]
+        [InlineData("grok-4.6", "responses")]
+        [InlineData("gpt-5.6-luna", "responses")]
+        [InlineData("glm-5.3", null)]
+        public void OpenCodeGo_RulesMapResponsesModelsOnly(string modelName, string? expectedBindingId) {
+            var go = LlmProviderCatalog.FindById("opencode-go")!;
+
+            Assert.Equal(expectedBindingId, go.ModelBindingRules.ResolveBindingId(modelName));
+        }
+
+        [Fact]
+        public void OpenCodeGo_SeedsOfficialModels_WithResponsesBindingForGrokAndGpt() {
+            var go = LlmProviderCatalog.FindById("opencode-go")!;
+
+            Assert.Contains("grok-4.6", go.DefaultModels);
+            Assert.Contains("minimax-m3", go.DefaultModels);
+            Assert.Contains("qwen3.7-max", go.DefaultModels);
+            Assert.Equal("responses", go.ModelBindingRules.ResolveBindingId("grok-4.6"));
+            Assert.Equal("responses", go.ModelBindingRules.ResolveBindingId("gpt-5.6-luna"));
+            Assert.Null(go.ModelBindingRules.ResolveBindingId("glm-5.3"));
+            Assert.True(go.CatalogIsEntitlement);
+        }
+
+        [Fact]
+        public void OpenCodeZen_IsNotEntitlement_AndSeedsNoModels() {
+            var zen = LlmProviderCatalog.FindById("opencode-zen")!;
+
+            Assert.False(zen.CatalogIsEntitlement);
+            Assert.Empty(zen.DefaultModels);
+            Assert.Equal("https://opencode.ai/zen/v1", zen.DefaultGateway);
+            Assert.Equal(3, zen.Bindings!.Count);
+        }
+
+        [Theory]
+        [InlineData("https://opencode.ai/zen/go/v1", "opencode-go")]
+        [InlineData("https://opencode.ai/zen/go", "opencode-go")]
+        [InlineData("https://opencode.ai/zen/v1", "opencode-zen")]
+        [InlineData("https://opencode.ai/zen", "opencode-zen")]
+        [InlineData("https://api.openai.com/v1", null)]
+        [InlineData("not a url", null)]
+        public void FindForEndpoint_ClassifiesOpenCodeGateways(string endpoint, string? expectedPresetId) {
+            Assert.Equal(expectedPresetId, LlmProviderCatalog.FindForEndpoint(endpoint)?.Id);
+        }
+
+        [Theory]
+        [InlineData("https://opencode.ai/zen", "/v1", "https://opencode.ai/zen/v1")]
+        [InlineData("https://opencode.ai/zen/v1", "/v1", "https://opencode.ai/zen/v1")]
+        [InlineData("https://opencode.ai/zen/v1/", "/v1", "https://opencode.ai/zen/v1")]
+        [InlineData("https://example.com", null, "https://example.com")]
+        public void BuildBindingEndpoint_AppendsSuffixAtMostOnce(string gateway, string? suffix, string expected) {
+            Assert.Equal(expected, LlmProviderCatalog.BuildBindingEndpoint(gateway, suffix));
+        }
+
+        [Fact]
+        public void EveryPreset_HasResolvableRuleBindings() {
+            foreach (var preset in LlmProviderCatalog.Presets) {
+                if (preset.ModelBindingRules == null) {
+                    continue;
+                }
+
+                var bindingIds = preset.Bindings?.Select(b => b.Id).ToHashSet(StringComparer.OrdinalIgnoreCase)
+                                 ?? new System.Collections.Generic.HashSet<string>();
+                foreach (var rule in preset.ModelBindingRules) {
+                    // 规则引用的 binding 必须存在于预设声明中（默认 binding 由渠道提供，不在此列）。
+                    Assert.Contains(rule.BindingId, bindingIds);
+                }
+            }
+        }
+    }
+}

TelegramSearchBot.LLM/Service/AI/LLM/LlmProviderCatalog.cs

@@ -10,10 +10,47 @@
         string DisplayName,
         LLMProvider Provider,
         /// <summary>Null = gateway must be entered during creation (e.g. user-owned OpenCode Go gateway).</summary>
         string? DefaultGateway,
         string[] DefaultModels,
         bool RequiresApiKey,
-        string? Notes = null);
+        string? Notes = null,
+        /// <summary>Extra protocol bindings created alongside the channel default (multi-protocol gateways).</summary>
+        IReadOnlyList<LlmPresetBinding>? Bindings = null,
+        /// <summary>Model-name prefix → binding id; unmatched models stay on the channel default binding.</summary>
+        IReadOnlyList<LlmModelBindingRule>? ModelBindingRules = null,
+        /// <summary>
+        /// True when the gateway catalog IS the entitlement (subscription, e.g. OpenCode Go): refresh may add
+        /// discovered models. False for pay-per-token catalogs (e.g. OpenCode Zen) where listing ≠ authorization.
+        /// </summary>
+        bool CatalogIsEntitlement = false);
+
+    /// <summary>Extra API binding a preset creates: same gateway, different wire protocol/auth.</summary>
+    public sealed record LlmPresetBinding(
+        string Id,
+        LlmProtocol Protocol,
+        LlmAuthProfile AuthProfile,
+        /// <summary>Path appended to the entered gateway; not duplicated when the gateway already ends with it.</summary>
+        string EndpointSuffix = "/v1");
+
+    /// <summary>Model-name (case-insensitive) prefix → binding id.</summary>
+    public sealed record LlmModelBindingRule(string BindingId, params string[] Prefixes);
+
+    public static class LlmModelBindingRules {
+        /// <summary>First matching rule wins; null = keep the channel default binding.</summary>
+        public static string? ResolveBindingId(this IReadOnlyList<LlmModelBindingRule>? rules, string? modelName) {
+            if (rules == null || string.IsNullOrWhiteSpace(modelName)) {
+                return null;
+            }
+
+            foreach (var rule in rules) {
+                if (rule.Prefixes.Any(prefix => modelName.StartsWith(prefix, StringComparison.OrdinalIgnoreCase))) {
+                    return rule.BindingId;
+                }
+            }
+
+            return null;
+        }
+    }
 
     /// <summary>
     /// Code-defined provider catalog (pi-style): common providers ship preconfigured so the bot
@@ -69,23 +106,84 @@
                 "https://api.moonshot.cn/v1",
                 new[] { "kimi-k2-0711-preview", "moonshot-v1-128k" },
                 RequiresApiKey: true),
+            // OpenCode 网关不是单协议服务商：Zen 按模型分属 Anthropic/OpenAI/Google 协议，
+            // Go 只有 Responses + Chat Completions。预设建渠道时按 ModelBindingRules 拆 binding。
             new LlmProviderPreset(
-                "opencode-zen", "OpenCode Zen (官方订阅目录)", LLMProvider.Anthropic,
-                "https://opencode.ai/zen",
+                "opencode-zen", "OpenCode Zen (官方订阅目录)", LLMProvider.OpenAI,
+                "https://opencode.ai/zen/v1",
                 Array.Empty<string>(),
                 RequiresApiKey: true,
-                Notes: "Anthropic 兼容；订阅 token 作为 API Key；目录模型不自动创建，授权模型请用 添加模型 手工维护。"),
+                Notes: "多协议网关（claude-*→Anthropic、gpt-*/grok-*→Responses、gemini-*→Google、其余→Chat Completions）；按量计费，目录不自动创建，授权模型请用 `添加模型`。",
+                Bindings: new[] {
+                    new LlmPresetBinding("anthropic", LlmProtocol.AnthropicMessages, LlmAuthProfile.AnthropicApiKey),
+                    new LlmPresetBinding("responses", LlmProtocol.OpenAIResponses, LlmAuthProfile.Bearer),
+                    new LlmPresetBinding("google", LlmProtocol.Gemini, LlmAuthProfile.Bearer)
+                },
+                ModelBindingRules: new[] {
+                    new LlmModelBindingRule("anthropic", "claude-"),
+                    new LlmModelBindingRule("responses", "gpt-", "grok-", "o3", "o4"),
+                    new LlmModelBindingRule("google", "gemini-")
+                }),
             new LlmProviderPreset(
-                "opencode-go", "OpenCode Go (自建网关)", LLMProvider.Anthropic,
-                null,
-                Array.Empty<string>(),
+                "opencode-go", "OpenCode Go (订阅网关)", LLMProvider.OpenAI,
+                "https://opencode.ai/zen/go/v1",
+                new[] {
+                    "grok-4.6", "gpt-5.6-luna",
+                    "glm-5.3-flash", "glm-5.3", "glm-5.2", "glm-5.1",
+                    "kimi-k3", "kimi-k2.7-code", "kimi-k2.6",
+                    "longcat-2.0",
+                    "deepseek-v4.1-flash", "deepseek-v4-pro", "deepseek-v4-flash", "deepseek-v4-flash-vision-exp",
+                    "minimax-m3", "minimax-m2.7",
+                    "mimo-v2.5", "mimo-v2.5-pro",
+                    "qwen3.8-max", "qwen3.8-flash", "qwen3.7-max", "qwen3.7-plus", "qwen3.6-plus",
+                    "muse-spark-1.3-contributor", "muse-spark-1.2-contributor",
+                    "hy4-preview", "hy3"
+                },
                 RequiresApiKey: true,
-                Notes: "创建时需输入你的 OpenCode Go 网关地址；Anthropic 兼容，自动携带 x-opencode-session 头；模型请用 添加模型 维护。"),
+                Notes: "OpenAI 兼容订阅网关（/responses + /chat/completions），自动携带 x-opencode-session；订阅覆盖目录模型，刷新会同步目录。自建网关请用 `新建渠道`。",
+                Bindings: new[] {
+                    new LlmPresetBinding("responses", LlmProtocol.OpenAIResponses, LlmAuthProfile.Bearer)
+                },
+                ModelBindingRules: new[] {
+                    new LlmModelBindingRule("responses", "grok-", "gpt-")
+                },
+                CatalogIsEntitlement: true),
         };
 
         public static LlmProviderPreset? FindById(string id) =>
             Presets.FirstOrDefault(p => p.Id.Equals(id, StringComparison.OrdinalIgnoreCase));
 
+        /// <summary>
+        /// 按渠道端点识别 OpenCode 预设（opencode.ai/zen/go/* → Go，其余 /zen/* → Zen）。
+        /// 用于刷新时决定目录策略与模型→协议规则。
+        /// </summary>
+        public static LlmProviderPreset? FindForEndpoint(string? endpoint) {
+            if (string.IsNullOrWhiteSpace(endpoint) ||
+                !Uri.TryCreate(endpoint, UriKind.Absolute, out var uri) ||
+                !string.Equals(uri.Host, "opencode.ai", StringComparison.OrdinalIgnoreCase)) {
+                return null;
+            }
+
+            var path = uri.AbsolutePath;
+            if (path.StartsWith("/zen/go", StringComparison.OrdinalIgnoreCase)) {
+                return FindById("opencode-go");
+            }
+            if (path.Equals("/zen", StringComparison.OrdinalIgnoreCase) ||
+                path.StartsWith("/zen/", StringComparison.OrdinalIgnoreCase)) {
+                return FindById("opencode-zen");
+            }
+            return null;
+        }
+
+        /// <summary>把用户输入的网关补成带协议路径后缀的 binding 端点（后缀已存在时不重复追加）。</summary>
+        public static string BuildBindingEndpoint(string? gateway, string? suffix) {
+            var trimmed = (gateway ?? string.Empty).TrimEnd('/');
+            if (string.IsNullOrEmpty(suffix)) {
+                return trimmed;
+            }
+            return trimmed.EndsWith(suffix, StringComparison.OrdinalIgnoreCase) ? trimmed : trimmed + suffix;
+        }
+
         /// <summary>Human-readable numbered list for bot selection.</summary>
         public static string FormatMenu() {
             var sb = new System.Text.StringBuilder("请选择要创建的渠道预设：\n");
@@ -99,6 +197,7 @@
                     sb.AppendLine($"   备注: {p.Notes}");
                 }
             }
+
             sb.Append("\n发送编号选择；发送 取消 退出。");
             return sb.ToString();
         }

TelegramSearchBot.LLM/Service/AI/LLM/LlmProviderRegistry.cs

@@ -91,7 +91,7 @@ public virtual async Task<LlmTransportBundle> GetTransport(LLMChannel channel, L
                         promptCachingEnabled, supportsVision, logger, httpClientFactory);
                 case LlmProtocol.AnthropicMessages:
                     return Transports.AnthropicMessagesTransport.Create(channel, binding, modelName, systemPrompt,
-                        nativeTools, promptCachingEnabled, logger);
+                        nativeTools, promptCachingEnabled, logger, chatId);
                 case LlmProtocol.Gemini:
                     return Transports.GeminiTransport.Create(channel, binding, modelName, supportsVision, logger, httpClientFactory);
                 case LlmProtocol.Ollama:

TelegramSearchBot.LLM/Service/AI/LLM/Transports/AnthropicMessagesTransport.cs

@@ -25,7 +25,7 @@ public sealed class AnthropicMessagesTransport : ILlmTransport {
         /// construction glue formerly in AnthropicModelApi.CreateClient.
         /// </summary>
         public static LlmTransportBundle Create(LLMChannel channel, LLMApiBinding binding, string modelName, string systemPrompt,
-            bool nativeTools, bool promptCachingEnabled, ILogger logger) {
+            bool nativeTools, bool promptCachingEnabled, ILogger logger, long chatId = 0) {
             var apiKey = LlmBindingSupport.ResolveApiKey(channel, binding);
             var options = new Anthropic.Core.ClientOptions {
                 ApiKey = apiKey,
@@ -40,6 +40,11 @@ public static LlmTransportBundle Create(LLMChannel channel, LLMApiBinding bindin
                 }
                 options.BaseUrl = trimmed;
             }
+
+            // OpenCode Go 要求每个请求携带稳定的 x-opencode-session（否则 400 MissingSessionID）；
+            // 非 opencode 端点时 Apply 自会跳过，其它渠道无感。
+            options.HttpClient = CreateSessionHttpClient(endpoint, chatId);
+
             var transport = new AnthropicMessagesTransport(logger, new AnthropicClient(options), systemPrompt, modelName, channel,
                 nativeTools, promptCachingEnabled);
             var config = new LlmTransportConfig {
@@ -54,6 +59,16 @@ public static LlmTransportBundle Create(LLMChannel channel, LLMApiBinding bindin
             return new LlmTransportBundle(transport, config);
         }
 
+        /// <summary>
+        /// 构造带 OpenCode 会话头的 HttpClient（opencode 端点生效，其它端点原样）。
+        /// 单独抽出以便测试验证 Go 的 x-opencode-session 行为。
+        /// </summary>
+        internal static HttpClient CreateSessionHttpClient(string? endpoint, long chatId = 0) {
+            var httpClient = new HttpClient();
+            OpencodeSessionHeaders.Apply(httpClient, endpoint, chatId > 0 ? $"tsb-{chatId}" : OpencodeSessionHeaders.GlobalSessionId);
+            return httpClient;
+        }
+
         private readonly ILogger _logger;
         private readonly AnthropicClient _client;
         private readonly string _systemPrompt;

TelegramSearchBot.Test/Manage/EditLLMConfTest.cs

@@ -508,5 +508,50 @@ public async Task ExecuteAsync_SetMaxImageRetryCount() {
             Assert.NotNull(config);
             Assert.Equal("500", config.Value);
         }
+
+        [Fact]
+        public async Task ExecuteAsync_PresetOpenCodeGo_CreatesResponseBindingAndAssignsModels() {
+            // Arrange
+            long chatId = 123;
+            var stateKey = $"llmconf:{chatId}:state";
+            var dataKey = $"llmconf:{chatId}:data";
+            var goIndex = LlmProviderCatalog.Presets.ToList().FindIndex(p => p.Id == "opencode-go") + 1;
+
+            _dbMock.SetupSequence(d => d.StringGetAsync(stateKey, It.IsAny<CommandFlags>()))
+                .ReturnsAsync(RedisValue.Null)
+                .ReturnsAsync("awaiting_preset_selection")
+                .ReturnsAsync("awaiting_preset_apikey");
+            _dbMock.SetupSequence(d => d.StringGetAsync(dataKey, It.IsAny<CommandFlags>()))
+                .ReturnsAsync("opencode-go");
+
+            helperMock.Setup(h => h.AddChannel(It.IsAny<string>(), It.IsAny<string>(), It.IsAny<string>(), It.IsAny<LLMProvider>(), It.IsAny<int>(), It.IsAny<int>()))
+                .ReturnsAsync(7);
+            helperMock.Setup(h => h.AddModelWithChannel(It.IsAny<int>(), It.IsAny<List<string>>()))
+                .ReturnsAsync(true);
+            helperMock.Setup(h => h.EnsureBinding(It.IsAny<int>(), It.IsAny<string>(), It.IsAny<LlmProtocol>(), It.IsAny<LlmAuthProfile>()))
+                .ReturnsAsync(100);
+            helperMock.Setup(h => h.AssignModelBinding(It.IsAny<int>(), It.IsAny<string>(), It.IsAny<int>()))
+                .ReturnsAsync(true);
+
+            // Act
+            var result1 = await _service.ExecuteAsync("预制渠道", chatId);
+            Assert.True(result1.Item1);
+            Assert.Contains("OpenCode Go", result1.Item2);
+
+            var result2 = await _service.ExecuteAsync(goIndex.ToString(), chatId);
+            Assert.True(result2.Item1);
+            Assert.Contains("API Key", result2.Item2);
+
+            var result3 = await _service.ExecuteAsync("go-token", chatId);
+
+            // Assert
+            Assert.True(result3.Item1);
+            Assert.Contains("渠道创建成功", result3.Item2);
+            helperMock.Verify(h => h.EnsureBinding(7, "https://opencode.ai/zen/go/v1", LlmProtocol.OpenAIResponses, LlmAuthProfile.Bearer), Times.Once);
+            helperMock.Verify(h => h.AssignModelBinding(7, "grok-4.6", 100), Times.Once);
+            helperMock.Verify(h => h.AssignModelBinding(7, "gpt-5.6-luna", 100), Times.Once);
+            helperMock.Verify(h => h.AssignModelBinding(7, "glm-5.3", It.IsAny<int>()), Times.Never);
+            helperMock.Verify(h => h.AssignModelBinding(7, It.IsAny<string>(), It.IsAny<int>()), Times.Exactly(2));
+        }
     }
 }