@bitkyc08/opencodex 2.7.1 → 2.7.2

package/src/providers/registry.ts

@@ -18,6 +18,9 @@ export interface ProviderRegistryEntry {
   adapter: string;
   baseUrl: string;
   authKind: ProviderAuthKind;
+  keyOptional?: boolean;
+  allowBaseUrlOverride?: boolean;
+  modelSuffixBracketStrip?: boolean;
   featured?: boolean;
   dashboardPreset?: boolean;
   note?: string;
@@ -55,7 +58,7 @@ export interface ProviderRegistryEntry {
 
 export type ProviderConfigSeed = Pick<
   OcxProviderConfig,
-  "adapter" | "baseUrl" | "authMode" | "defaultModel" | "models"
+  "adapter" | "baseUrl" | "authMode" | "keyOptional" | "modelSuffixBracketStrip" | "defaultModel" | "models"
   | "liveModels" | "contextWindow" | "modelContextWindows" | "modelInputModalities"
   | "reasoningEfforts" | "modelReasoningEfforts" | "reasoningEffortMap" | "modelReasoningEffortMap"
   | "noVisionModels" | "noReasoningModels" | "noTemperatureModels" | "noTopPModels" | "noPenaltyModels"
@@ -65,13 +68,25 @@ export type ProviderConfigSeed = Pick<
 
 // Shared between the OAuth (Claude account) and API-key Anthropic entries so both expose the
 // same static model seed.
-// 260709 refresh: claude-fable-5 added (official models overview); evidence in
-// devlog/model_update/260709_model_refresh/002_cursor_registry_drift.md.
+// 260710 context refresh: Tier-2 evidence in
+// devlog/_plan/260710_provider_hardening/001_research_frontier.md.
 const ANTHROPIC_MODELS = ["claude-fable-5", "claude-sonnet-5", "claude-opus-4-8", "claude-opus-4-7", "claude-opus-4-6", "claude-sonnet-4-6", "claude-haiku-4-5"];
-const ANTHROPIC_MODEL_CONTEXT_WINDOWS: Record<string, number> = { "claude-sonnet-5": 1_000_000, "claude-fable-5": 1_000_000 };
+const ANTHROPIC_MODEL_CONTEXT_WINDOWS: Record<string, number> = { "claude-sonnet-5": 1_000_000, "claude-fable-5": 1_000_000, "claude-opus-4-8": 1_000_000, "claude-haiku-4-5": 200_000 };
 
 const ZAI_GLM_52_MODELS = ["glm-5.2", "glm-5.2[1m]"];
 const ZAI_GLM_52_REASONING_EFFORTS = ["low", "medium", "high", "xhigh", "max"];
+// 260710 MiniMax models and context windows: Tier-2 evidence in
+// devlog/_plan/260710_provider_hardening/002_research_cn.md.
+const MINIMAX_MODELS = [
+  "MiniMax-M3",
+  "MiniMax-M2.7", "MiniMax-M2.7-highspeed",
+  "MiniMax-M2.5", "MiniMax-M2.5-highspeed",
+  "MiniMax-M2.1", "MiniMax-M2.1-highspeed",
+  "MiniMax-M2",
+];
+const MINIMAX_MODEL_CONTEXT_WINDOWS: Record<string, number> = Object.fromEntries(
+  MINIMAX_MODELS.map(id => [id, id === "MiniMax-M3" ? 1_000_000 : 204_800]),
+);
 const OPENAI_GPT56_MODELS = ["gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna"];
 const OPENAI_GPT56_CONTEXT_WINDOW = 372_000;
 const OPENAI_GPT56_CONTEXT_WINDOWS = {
@@ -80,10 +95,13 @@ const OPENAI_GPT56_CONTEXT_WINDOWS = {
   "gpt-5.6-luna": OPENAI_GPT56_CONTEXT_WINDOW,
 };
 const OPENROUTER_GPT56_MODELS = OPENAI_GPT56_MODELS.map(id => `openai/${id}`);
+// OpenRouter's live /endpoints routes report 1,050,000; keep this separate from the
+// unverified OpenAI API-key seed. Evidence: devlog/_plan/260710_provider_hardening/003_research_aggregators.md.
+const OPENROUTER_GPT56_CONTEXT_WINDOW = 1_050_000;
 const OPENROUTER_GPT56_CONTEXT_WINDOWS = {
-  "openai/gpt-5.6-sol": OPENAI_GPT56_CONTEXT_WINDOW,
-  "openai/gpt-5.6-terra": OPENAI_GPT56_CONTEXT_WINDOW,
-  "openai/gpt-5.6-luna": OPENAI_GPT56_CONTEXT_WINDOW,
+  "openai/gpt-5.6-sol": OPENROUTER_GPT56_CONTEXT_WINDOW,
+  "openai/gpt-5.6-terra": OPENROUTER_GPT56_CONTEXT_WINDOW,
+  "openai/gpt-5.6-luna": OPENROUTER_GPT56_CONTEXT_WINDOW,
 };
 
 /**
@@ -122,11 +140,22 @@ const DEEPSEEK_THINKING_REASONING_MAP: Record<string, string> = {
   xhigh: "max",
   max: "max",
 };
-const KIMI_THINKING_MODELS = ["kimi-k2.7-code", "kimi-k2.7-code-highspeed", "kimi-k2.6", "kimi-k2.5"];
-const KIMI_LOCKED_PARAMETER_MODELS = ["kimi-k2.7-code", "kimi-k2.7-code-highspeed", "kimi-k2.6", "kimi-k2.5"];
+// 260710 Kimi model aliases and context windows: Tier-2 evidence in
+// devlog/_plan/260710_provider_hardening/002_research_cn.md.
+const KIMI_API_MODELS = ["kimi-k2.7-code", "kimi-k2.7-code-highspeed", "kimi-k2.6", "kimi-k2.5"];
+const KIMI_CODING_MODELS = [...KIMI_API_MODELS, "kimi-for-coding"];
+const KIMI_THINKING_MODELS = KIMI_CODING_MODELS;
+const KIMI_LOCKED_PARAMETER_MODELS = KIMI_CODING_MODELS;
+const KIMI_AUTO_TOOL_CHOICE_ONLY_MODELS = ["kimi-k2.7-code", "kimi-k2.7-code-highspeed", "kimi-for-coding"];
+const KIMI_API_MODEL_CONTEXT_WINDOWS: Record<string, number> = Object.fromEntries(
+  KIMI_API_MODELS.map(id => [id, 262_144]),
+);
+const KIMI_CODING_MODEL_CONTEXT_WINDOWS: Record<string, number> = Object.fromEntries(
+  KIMI_CODING_MODELS.map(id => [id, 262_144]),
+);
 const NEURALWATT_REASONING_HISTORY_MODELS = [
-  "glm-5.2",
-  "moonshotai/Kimi-K2.5", "kimi-k2.6", "kimi-k2.7-code",
+  "glm-5.2", "glm-5.2-short",
+  "kimi-k2.6", "kimi-k2.7-code",
   "qwen3.5-397b", "qwen3.6-35b",
 ];
 const UMANS_MODELS = [
@@ -258,15 +287,16 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
     oauthId: "kimi",
     jawcodeBundle: "moonshot",
     note: "Log in with your Kimi account",
-    models: ["kimi-k2.7-code", "kimi-k2.7-code-highspeed", "kimi-k2.6", "kimi-k2.5"],
+    models: KIMI_CODING_MODELS,
     defaultModel: "kimi-k2.7-code",
+    modelContextWindows: KIMI_CODING_MODEL_CONTEXT_WINDOWS,
     // Kimi thinking is controlled by Kimi's `thinking` extension, not OpenAI `reasoning_effort`.
     noReasoningModels: KIMI_THINKING_MODELS,
     modelReasoningEfforts: Object.fromEntries(KIMI_THINKING_MODELS.map(id => [id, []])),
     noTemperatureModels: KIMI_LOCKED_PARAMETER_MODELS,
     noTopPModels: KIMI_LOCKED_PARAMETER_MODELS,
     noPenaltyModels: KIMI_LOCKED_PARAMETER_MODELS,
-    autoToolChoiceOnlyModels: ["kimi-k2.7-code", "kimi-k2.7-code-highspeed"],
+    autoToolChoiceOnlyModels: KIMI_AUTO_TOOL_CHOICE_ONLY_MODELS,
     preserveReasoningContentModels: KIMI_THINKING_MODELS,
   },
   {
@@ -351,9 +381,11 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
     authKind: "key",
     dashboardUrl: "https://portal.neuralwatt.com",
     defaultModel: "glm-5.2",
+    // 2026-07-10 live /v1/models: K2.5 rows were removed and GLM-5.2 short variants added.
+    // Evidence: devlog/_plan/260710_provider_hardening/003_research_aggregators.md and https://api.neuralwatt.com/v1/models.
     models: [
-      "glm-5.2", "glm-5.2-fast",
-      "moonshotai/Kimi-K2.5", "kimi-k2.5-fast", "kimi-k2.6", "kimi-k2.6-fast",
+      "glm-5.2", "glm-5.2-fast", "glm-5.2-short", "glm-5.2-short-fast",
+      "kimi-k2.6", "kimi-k2.6-fast",
       "kimi-k2.7-code",
       "qwen3.5-397b", "qwen3.5-397b-fast", "qwen3.6-35b", "qwen3.6-35b-fast",
     ],
@@ -361,8 +393,8 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
     modelReasoningEfforts: {
       "glm-5.2": ZAI_GLM_52_REASONING_EFFORTS,
       "glm-5.2-fast": [],
-      "moonshotai/Kimi-K2.5": [],
-      "kimi-k2.5-fast": [],
+      "glm-5.2-short": ZAI_GLM_52_REASONING_EFFORTS,
+      "glm-5.2-short-fast": [],
       "kimi-k2.6": [],
       "kimi-k2.6-fast": [],
       "kimi-k2.7-code": [],
@@ -374,8 +406,8 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
       "qwen3.6-35b-fast": [],
     },
     thinkingBudgetModels: THINKING_BUDGET_MODELS,
-    noReasoningModels: ["glm-5.2-fast", "kimi-k2.5-fast", "kimi-k2.6-fast", "qwen3.5-397b-fast", "qwen3.6-35b-fast"],
-    noVisionModels: ["glm-5.2", "glm-5.2-fast", "qwen3.5-397b", "qwen3.5-397b-fast"],
+    noReasoningModels: ["glm-5.2-fast", "glm-5.2-short-fast", "kimi-k2.6-fast", "qwen3.5-397b-fast", "qwen3.6-35b-fast"],
+    noVisionModels: ["glm-5.2", "glm-5.2-fast", "glm-5.2-short", "glm-5.2-short-fast", "qwen3.5-397b", "qwen3.5-397b-fast"],
     noTemperatureModels: ["kimi-k2.7-code"],
     noTopPModels: ["kimi-k2.7-code"],
     noPenaltyModels: ["kimi-k2.7-code"],
@@ -384,13 +416,26 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
   },
   { id: "openrouter", label: "OpenRouter", adapter: "openai-chat", baseUrl: "https://openrouter.ai/api/v1", authKind: "key", featured: true, dashboardUrl: "https://openrouter.ai/keys", jawcodeBundle: "openrouter", models: ["anthropic/claude-sonnet-5", ...OPENROUTER_GPT56_MODELS], modelContextWindows: { "anthropic/claude-sonnet-5": 1_000_000, ...OPENROUTER_GPT56_CONTEXT_WINDOWS } },
   { id: "groq", label: "Groq", adapter: "openai-chat", baseUrl: "https://api.groq.com/openai/v1", authKind: "key", featured: true, dashboardUrl: "https://console.groq.com/keys" },
-  { id: "google", label: "Google Gemini", adapter: "google", baseUrl: "https://generativelanguage.googleapis.com", authKind: "key", featured: true, dashboardUrl: "https://aistudio.google.com/apikey", defaultModel: "gemini-3-pro", jawcodeBundle: "google", extraMetadataAliases: ["gemini"] },
+  // 2026-07-10 Gemini API refresh: Tier-2 ai.google.dev evidence recorded in
+  // devlog/_plan/260710_provider_hardening/001_research_frontier.md.
+  {
+    id: "google", label: "Google Gemini", adapter: "google", baseUrl: "https://generativelanguage.googleapis.com", authKind: "key", featured: true,
+    dashboardUrl: "https://aistudio.google.com/apikey", defaultModel: "gemini-3.5-flash", models: ["gemini-3.5-flash", "gemini-3.1-pro-preview"],
+    modelContextWindows: { "gemini-3.5-flash": 1_000_000 },
+    modelReasoningEfforts: {
+      "gemini-3.5-flash": ["minimal", "low", "medium", "high"],
+      "gemini-3.1-pro-preview": ["low", "medium", "high"],
+    },
+    jawcodeBundle: "google", extraMetadataAliases: ["gemini"],
+  },
+  // 2026-07-10: defaultModel is frozen pending Vertex-specific Tier-2 evidence; Gemini API
+  // evidence from ai.google.dev does not establish Vertex publisher availability.
   { id: "google-vertex", label: "Google Vertex AI", adapter: "google", baseUrl: "https://aiplatform.googleapis.com", authKind: "key", dashboardUrl: "https://console.cloud.google.com/vertex-ai", defaultModel: "gemini-3-pro", googleMode: "vertex", jawcodeBundle: "google", extraMetadataAliases: ["gemini-vertex"] },
   { id: "google-antigravity", label: "Google Antigravity", adapter: "google", baseUrl: "https://daily-cloudcode-pa.googleapis.com", authKind: "oauth", dashboardUrl: "https://antigravity.google", models: ANTIGRAVITY_MODELS, defaultModel: "gemini-3.5-flash-low", modelContextWindows: ANTIGRAVITY_MODEL_CONTEXT_WINDOWS, googleMode: "cloud-code-assist", jawcodeBundle: "google", extraMetadataAliases: ["antigravity", "gemini-antigravity"] },
   { id: "azure-openai", label: "Azure OpenAI", adapter: "azure-openai", baseUrl: "https://{resource}.openai.azure.com/openai", authKind: "key", featured: true, dashboardUrl: "https://portal.azure.com" },
-  { id: "ollama", label: "Ollama (local)", adapter: "openai-chat", baseUrl: "http://localhost:11434/v1", authKind: "local", featured: true, note: "Local — key usually blank" },
-  { id: "vllm", label: "vLLM (local)", adapter: "openai-chat", baseUrl: "http://localhost:8000/v1", authKind: "local", featured: true, note: "Local — key usually blank" },
-  { id: "lm-studio", label: "LM Studio (local)", adapter: "openai-chat", baseUrl: "http://localhost:1234/v1", authKind: "local", featured: true, note: "Local — no key needed" },
+  { id: "ollama", label: "Ollama (local)", adapter: "openai-chat", baseUrl: "http://localhost:11434/v1", authKind: "local", allowBaseUrlOverride: true, featured: true, note: "Local — key usually blank" },
+  { id: "vllm", label: "vLLM (local)", adapter: "openai-chat", baseUrl: "http://localhost:8000/v1", authKind: "local", allowBaseUrlOverride: true, featured: true, note: "Local — key usually blank" },
+  { id: "lm-studio", label: "LM Studio (local)", adapter: "openai-chat", baseUrl: "http://localhost:1234/v1", authKind: "local", allowBaseUrlOverride: true, featured: true, note: "Local — no key needed" },
   {
     id: "deepseek",
     label: "DeepSeek",
@@ -398,8 +443,11 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
     adapter: "openai-chat",
     authKind: "key",
     dashboardUrl: "https://platform.deepseek.com/api_keys",
+    // deepseek-chat/deepseek-reasoner are upstream-deprecated at 2026-07-24 15:59 UTC;
+    // kept until then. Evidence: devlog/_plan/260710_provider_hardening/002_research_cn.md.
     models: ["deepseek-chat", "deepseek-reasoner", ...DEEPSEEK_THINKING_MODELS],
-    defaultModel: "deepseek-chat",
+    defaultModel: "deepseek-v4-flash",
+    modelContextWindows: { "deepseek-v4-flash": 1_000_000, "deepseek-v4-pro": 1_000_000 },
     /* [Decision Log]
     - 목적: DeepSeek V4 thinking mode multi-turn/tool-call requests must replay prior assistant reasoning_content.
     - 대안 분석: Globally preserve reasoning_content for all OpenAI-compatible models; preserve it for legacy deepseek-reasoner too; mark only V4 thinking models in registry metadata.
@@ -409,42 +457,67 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
     modelReasoningEffortMap: Object.fromEntries(DEEPSEEK_THINKING_MODELS.map(id => [id, DEEPSEEK_THINKING_REASONING_MAP])),
     preserveReasoningContentModels: DEEPSEEK_THINKING_MODELS,
   },
-  { id: "cerebras", label: "Cerebras", baseUrl: "https://api.cerebras.ai/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://cloud.cerebras.ai/platform/apikeys", defaultModel: "llama-3.3-70b" },
+  // llama-3.3-70b was deprecated by Cerebras on 2026-02-16. Evidence: devlog/_plan/260710_provider_hardening/003_research_aggregators.md.
+  { id: "cerebras", label: "Cerebras", baseUrl: "https://api.cerebras.ai/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://cloud.cerebras.ai/platform/apikeys", defaultModel: "gpt-oss-120b" },
+  // FREEZE 2026-07-10: exact serverless ids remain auth-gated/unverified. Evidence: devlog/_plan/260710_provider_hardening/003_research_aggregators.md.
   { id: "together", label: "Together", baseUrl: "https://api.together.xyz/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://api.together.xyz/settings/api-keys" },
   { id: "fireworks", label: "Fireworks", baseUrl: "https://api.fireworks.ai/inference/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://fireworks.ai/account/api-keys" },
-  { id: "firepass", label: "Fire Pass (Fireworks Kimi)", baseUrl: "https://api.fireworks.ai/inference/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://fireworks.ai/account/api-keys" },
+  {
+    id: "firepass", label: "Fire Pass (Fireworks Kimi)", baseUrl: "https://api.fireworks.ai/inference/v1", adapter: "openai-chat", authKind: "key",
+    dashboardUrl: "https://fireworks.ai/account/api-keys",
+    note: "Model data frozen pending Tier-2 entitlement proof",
+  },
   {
     id: "moonshot", label: "Moonshot (Kimi API)", baseUrl: "https://api.moonshot.ai/v1", adapter: "openai-chat", authKind: "key",
     dashboardUrl: "https://platform.moonshot.ai/console/api-keys", defaultModel: "kimi-k2.7-code", jawcodeBundle: "moonshot",
-    models: ["kimi-k2.7-code", "kimi-k2.7-code-highspeed", "kimi-k2.6", "kimi-k2.5"],
-    noReasoningModels: KIMI_THINKING_MODELS,
-    modelReasoningEfforts: Object.fromEntries(KIMI_THINKING_MODELS.map(id => [id, []])),
-    noTemperatureModels: KIMI_LOCKED_PARAMETER_MODELS,
-    noTopPModels: KIMI_LOCKED_PARAMETER_MODELS,
-    noPenaltyModels: KIMI_LOCKED_PARAMETER_MODELS,
+    models: KIMI_API_MODELS,
+    modelContextWindows: KIMI_API_MODEL_CONTEXT_WINDOWS,
+    noReasoningModels: KIMI_API_MODELS,
+    modelReasoningEfforts: Object.fromEntries(KIMI_API_MODELS.map(id => [id, []])),
+    noTemperatureModels: KIMI_API_MODELS,
+    noTopPModels: KIMI_API_MODELS,
+    noPenaltyModels: KIMI_API_MODELS,
     autoToolChoiceOnlyModels: ["kimi-k2.7-code", "kimi-k2.7-code-highspeed"],
-    preserveReasoningContentModels: KIMI_THINKING_MODELS,
+    preserveReasoningContentModels: KIMI_API_MODELS,
   },
   { id: "huggingface", label: "Hugging Face", baseUrl: "https://router.huggingface.co/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://huggingface.co/settings/tokens" },
   { id: "nvidia", label: "NVIDIA NIM", baseUrl: "https://integrate.api.nvidia.com/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://build.nvidia.com" },
   { id: "venice", label: "Venice", baseUrl: "https://api.venice.ai/api/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://venice.ai/settings/api" },
+  // 260710 GLM-5.2 context and path-specific ids: Tier-2 evidence in
+  // devlog/_plan/260710_provider_hardening/002_research_cn.md.
   {
     id: "zai", label: "Z.AI — GLM Coding Plan", baseUrl: "https://api.z.ai/api/coding/paas/v4", adapter: "openai-chat", authKind: "key",
     dashboardUrl: "https://z.ai/manage-apikey/apikey-list", defaultModel: "glm-5.2",
     note: "GLM-5.2 coding subscription",
     models: ["glm-5.2", "glm-5.2[1m]", "glm-5.1", "glm-5", "glm-4.6"],
+    modelContextWindows: { "glm-5.2": 1_000_000, "glm-5.2[1m]": 1_000_000 },
+    // Z.AI's OpenAI path returns 400 code 1211 for bracketed model ids.
+    modelSuffixBracketStrip: true,
     noVisionModels: ZAI_GLM_52_MODELS,
     modelReasoningEfforts: Object.fromEntries(ZAI_GLM_52_MODELS.map(id => [id, ZAI_GLM_52_REASONING_EFFORTS])),
     preserveReasoningContentModels: ZAI_GLM_52_MODELS,
   },
   { id: "nanogpt", label: "NanoGPT", baseUrl: "https://nano-gpt.com/api/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://nano-gpt.com/api" },
   { id: "synthetic", label: "Synthetic", baseUrl: "https://api.synthetic.new/openai/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://synthetic.new" },
+  // 2026-07-10: docs unverified; model data frozen. Evidence: devlog/_plan/260710_provider_hardening/002_research_cn.md.
   { id: "qwen-portal", label: "Qwen Portal", baseUrl: "https://portal.qwen.ai/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://portal.qwen.ai" },
+  // 2026-07-10: docs unverified; model data frozen. Evidence: devlog/_plan/260710_provider_hardening/002_research_cn.md.
   { id: "qianfan", label: "Qianfan (Baidu)", baseUrl: "https://qianfan.baidubce.com/v2", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://console.bce.baidu.com/iam/#/iam/apikey/list" },
+  // 2026-07-10: docs unverified; model data frozen. Evidence: devlog/_plan/260710_provider_hardening/002_research_cn.md.
   { id: "alibaba", label: "Alibaba Coding Plan", baseUrl: "https://coding-intl.dashscope.aliyuncs.com/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://dashscope.console.aliyun.com/apiKey" },
+  // NEEDS_HUMAN 2026-07-10: kept for config compatibility, but this is a dashboard URL,
+  // no /models endpoint is documented, and tools are silently ignored upstream per docs.parallel.ai.
+  // Evidence: devlog/_plan/260710_provider_hardening/003_research_aggregators.md.
   { id: "parallel", label: "Parallel", baseUrl: "https://platform.parallel.ai", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://platform.parallel.ai" },
+  // FREEZE 2026-07-10: model ids remain unverified. Evidence: devlog/_plan/260710_provider_hardening/003_research_aggregators.md.
   { id: "zenmux", label: "ZenMux", baseUrl: "https://zenmux.ai/api/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://zenmux.ai" },
-  { id: "litellm", label: "LiteLLM (self-hosted)", baseUrl: "http://localhost:4000/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://docs.litellm.ai/docs/proxy/quick_start" },
+  {
+    id: "litellm", label: "LiteLLM (self-hosted)", baseUrl: "http://localhost:4000/v1", adapter: "openai-chat", authKind: "key",
+    dashboardUrl: "https://docs.litellm.ai/docs/proxy/quick_start",
+    allowBaseUrlOverride: true,
+    // A self-hosted proxy may legitimately run without a master key.
+    keyOptional: true,
+  },
   {
     id: "ollama-cloud",
     label: "Ollama Cloud",
@@ -452,29 +525,43 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
     adapter: "openai-chat",
     authKind: "key",
     dashboardUrl: "https://ollama.com/settings/keys",
-    models: ["glm-5.2", "deepseek-v4-pro", "qwen3-coder", "gpt-oss:120b", "kimi-k2.6", "minimax-m3", "qwen3.5", "gemma4"],
+    // Live IDs verified 2026-07-10; qwen3-coder:480b retires 2026-07-15.
+    // Evidence: .codexclaw/evidence/260710_wp9_ollama_cloud_model_ids.md.
+    models: ["glm-5.2", "deepseek-v4-pro", "qwen3-coder:480b", "gpt-oss:120b", "kimi-k2.6", "minimax-m3", "qwen3.5:397b", "gemma4:31b"],
     defaultModel: "glm-5.2",
     noVisionModels: [
       "glm-5.2", "glm-5.1", "glm-5", "glm-4.7",
       "minimax-m2.7", "minimax-m2.5", "minimax-m2.1",
       "nemotron-3-ultra", "nemotron-3-super",
       "deepseek-v4-pro", "deepseek-v4-flash",
-      "gpt-oss", "qwen3-coder",
+      "gpt-oss", "qwen3-coder:480b",
     ],
   },
+  // FREEZE 2026-07-10: codestral-latest is unconfirmed behind auth. Evidence: devlog/_plan/260710_provider_hardening/003_research_aggregators.md.
   { id: "mistral", label: "Mistral", baseUrl: "https://api.mistral.ai/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://console.mistral.ai/api-keys", defaultModel: "codestral-latest" },
-  { id: "minimax", label: "MiniMax — Coding Plan", baseUrl: "https://api.minimax.io/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://platform.minimax.io", defaultModel: "MiniMax-M2.5", jawcodeBundle: "minimax", metadataModelIdNormalize: "case-insensitive", note: "Subscription Key or API Key" },
-  { id: "minimax-cn", label: "MiniMax — Coding Plan (CN)", baseUrl: "https://api.minimaxi.com/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://platform.minimaxi.com", defaultModel: "MiniMax-M2.5", jawcodeBundle: "minimax", metadataModelIdNormalize: "case-insensitive", note: "中国区 Subscription Key" },
+  {
+    id: "minimax", label: "MiniMax — Coding Plan", baseUrl: "https://api.minimax.io/v1", adapter: "openai-chat", authKind: "key",
+    dashboardUrl: "https://platform.minimax.io", defaultModel: "MiniMax-M3", models: MINIMAX_MODELS,
+    modelContextWindows: MINIMAX_MODEL_CONTEXT_WINDOWS,
+    jawcodeBundle: "minimax", metadataModelIdNormalize: "case-insensitive", note: "Subscription Key or API Key",
+  },
+  {
+    id: "minimax-cn", label: "MiniMax — Coding Plan (CN)", baseUrl: "https://api.minimaxi.com/v1", adapter: "openai-chat", authKind: "key",
+    dashboardUrl: "https://platform.minimaxi.com", defaultModel: "MiniMax-M3", models: MINIMAX_MODELS,
+    modelContextWindows: MINIMAX_MODEL_CONTEXT_WINDOWS,
+    jawcodeBundle: "minimax", metadataModelIdNormalize: "case-insensitive", note: "中国区 Subscription Key",
+  },
   {
     id: "kimi-code", label: "Kimi (coding)", baseUrl: "https://api.kimi.com/coding/v1", adapter: "openai-chat", authKind: "key",
     dashboardUrl: "https://platform.moonshot.cn/console/api-keys", defaultModel: "kimi-k2.7-code",
-    models: ["kimi-k2.7-code", "kimi-k2.7-code-highspeed", "kimi-k2.6", "kimi-k2.5"],
+    models: KIMI_CODING_MODELS,
+    modelContextWindows: KIMI_CODING_MODEL_CONTEXT_WINDOWS,
     noReasoningModels: KIMI_THINKING_MODELS,
     modelReasoningEfforts: Object.fromEntries(KIMI_THINKING_MODELS.map(id => [id, []])),
     noTemperatureModels: KIMI_LOCKED_PARAMETER_MODELS,
     noTopPModels: KIMI_LOCKED_PARAMETER_MODELS,
     noPenaltyModels: KIMI_LOCKED_PARAMETER_MODELS,
-    autoToolChoiceOnlyModels: ["kimi-k2.7-code", "kimi-k2.7-code-highspeed"],
+    autoToolChoiceOnlyModels: KIMI_AUTO_TOOL_CHOICE_ONLY_MODELS,
     preserveReasoningContentModels: KIMI_THINKING_MODELS,
   },
   { id: "opencode-zen", label: "opencode zen", baseUrl: "https://opencode.ai/zen/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://opencode.ai/auth" },
@@ -482,7 +569,9 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
   { id: "xiaomi", label: "Xiaomi MiMo", baseUrl: "https://api.xiaomimimo.com/anthropic", adapter: "anthropic", authKind: "key", dashboardUrl: "https://xiaomimimo.com", defaultModel: "mimo-v2.5-pro" },
   { id: "kilo", label: "Kilo", baseUrl: "https://api.kilo.ai/api/gateway", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://kilo.ai" },
   { id: "cloudflare-ai-gateway", label: "Cloudflare AI Gateway", baseUrl: "https://gateway.ai.cloudflare.com/v1/{account-id}/{gateway}/anthropic", adapter: "anthropic", authKind: "key", dashboardUrl: "https://dash.cloudflare.com/?to=/:account/ai/ai-gateway" },
+  // FREEZE 2026-07-10: /models is auth-gated, so ids remain unverified. Evidence: devlog/_plan/260710_provider_hardening/003_research_aggregators.md.
   { id: "github-copilot", label: "GitHub Copilot", baseUrl: "https://api.githubcopilot.com", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://github.com/settings/copilot" },
+  // FREEZE 2026-07-10: no public OpenAI-compatible endpoint is documented. Evidence: devlog/_plan/260710_provider_hardening/003_research_aggregators.md.
   { id: "gitlab-duo", label: "GitLab Duo", baseUrl: "https://cloud.gitlab.com/ai/v1/proxy/openai/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://gitlab.com/-/user_settings/personal_access_tokens" },
 ];
 

package/src/responses/parser.ts

@@ -250,6 +250,18 @@ export function parseRequest(body: unknown): OcxParsedRequest {
         continue;
       }
 
+      if (effectiveType === "additional_tools") {
+        // Codex Desktop responses_lite WS path: tools ride INSIDE input as an
+        // `additional_tools` item ({type, role, tools:[...]}) instead of body.tools.
+        // Same spec wire shapes (function/namespace/custom/tool_search) — collect and
+        // merge through the exact buildTools path so surface detection (collabSurface)
+        // and chat-model tool listing see them. The item itself never becomes a message;
+        // the native passthrough keeps it verbatim in _rawBody.
+        const at = item as { tools?: unknown[] };
+        if (Array.isArray(at.tools)) loadedToolSpecs.push(...at.tools);
+        continue;
+      }
+
       if (effectiveType === "compaction" || effectiveType === "compaction_summary" || effectiveType === "context_compaction") {
         // A stored summary from a previous compaction. Decode our ocx1 envelope into plain text so
         // the routed model keeps the compacted context; real OpenAI-encrypted blobs degrade to a note.

package/src/router.ts

@@ -97,11 +97,21 @@ function routedProviderConfig(providerName: string, provider: OcxProviderConfig)
   const preserveReasoningContentModels = mergeStringArray(registryEntry.preserveReasoningContentModels, provider.preserveReasoningContentModels);
   const thinkingToggleModels = mergeStringArray(registryEntry.thinkingToggleModels, provider.thinkingToggleModels);
   const thinkingBudgetModels = mergeStringArray(registryEntry.thinkingBudgetModels, provider.thinkingBudgetModels);
+  const registryBaseUrlIsTemplate = /\{[^}]*\}/.test(registryEntry.baseUrl);
+  const userBaseUrl = typeof provider.baseUrl === "string" ? provider.baseUrl.trim() : "";
+  const userBaseUrlIsResolved = userBaseUrl.length > 0 && !/\{[^}]*\}/.test(userBaseUrl);
+  if (registryEntry.allowBaseUrlOverride && !userBaseUrlIsResolved) {
+    throw new Error(`Invalid baseUrl for provider "${providerName}": expected a nonblank URL without unresolved placeholders`);
+  }
+  // Registry template URLs are presets; local/self-hosted entries opt in explicitly.
+  const baseUrl = (registryBaseUrlIsTemplate || registryEntry.allowBaseUrlOverride) && userBaseUrlIsResolved
+    ? userBaseUrl
+    : registryEntry.baseUrl;
 
   return {
     ...provider,
     adapter: registryEntry.adapter,
-    baseUrl: registryEntry.baseUrl,
+    baseUrl,
     authMode: canonicalAuthMode,
     apiKey: resolveEnvValue(provider.apiKey),
     // Backfill the Google wire mode + Vertex project/location from the registry when the user
@@ -113,6 +123,8 @@ function routedProviderConfig(providerName: string, provider: OcxProviderConfig)
     ...(provider.contextWindow === undefined && registryEntry.contextWindow !== undefined ? { contextWindow: registryEntry.contextWindow } : {}),
     ...(provider.reasoningEfforts === undefined && registryEntry.reasoningEfforts !== undefined ? { reasoningEfforts: registryEntry.reasoningEfforts } : {}),
     ...(provider.escapeBuiltinToolNames === undefined && registryEntry.escapeBuiltinToolNames !== undefined ? { escapeBuiltinToolNames: registryEntry.escapeBuiltinToolNames } : {}),
+    ...(provider.keyOptional === undefined && registryEntry.keyOptional !== undefined ? { keyOptional: registryEntry.keyOptional } : {}),
+    ...(provider.modelSuffixBracketStrip === undefined && registryEntry.modelSuffixBracketStrip !== undefined ? { modelSuffixBracketStrip: registryEntry.modelSuffixBracketStrip } : {}),
     // Scalar backfill: a persisted config created before the flag shipped inherits the registry
     // opt-in, while an explicit user `false` keeps overriding registry `true`.
     ...(provider.parallelToolCalls === undefined && registryEntry.parallelToolCalls !== undefined ? { parallelToolCalls: registryEntry.parallelToolCalls } : {}),

package/src/server/management-api.ts

@@ -535,12 +535,13 @@ export async function handleManagementAPI(req: Request, url: URL, config: OcxCon
     return jsonResponse({
       model: config.injectionModel ?? null,
       effort: config.injectionEffort ?? null,
+      prompt: config.injectionPrompt ?? null,
       efforts: CODEX_REASONING_LEVELS.map(l => l.effort),
       available: [...nativeModels, ...routedModels],
     });
   }
   if (url.pathname === "/api/injection-model" && req.method === "PUT") {
-    let body: { model?: unknown; effort?: unknown };
+    let body: { model?: unknown; effort?: unknown; prompt?: unknown };
     try { body = await req.json(); } catch { return jsonResponse({ error: "invalid JSON body" }, 400); }
     const { isCodexReasoningEffort } = await import("../reasoning-effort");
     const model = typeof body.model === "string" && body.model.length > 0 ? body.model : undefined;
@@ -559,8 +560,15 @@ export async function handleManagementAPI(req: Request, url: URL, config: OcxCon
     else delete config.injectionModel;
     if (effort) config.injectionEffort = effort;
     else delete config.injectionEffort;
+    // `prompt` key semantics mirror `effort`: absent -> unchanged; null/"" -> clear;
+    // non-empty string -> set (custom <multi_agent_mode> body, {{model}}/{{effort}}/{{roster}} placeholders).
+    if ("prompt" in body) {
+      if (typeof body.prompt === "string" && body.prompt.trim().length > 0) config.injectionPrompt = body.prompt;
+      else if (body.prompt === null || body.prompt === "") delete config.injectionPrompt;
+      else return jsonResponse({ error: "prompt must be a string or null" }, 400);
+    }
     saveConfig(config);
-    return jsonResponse({ ok: true, model: config.injectionModel ?? null, effort: config.injectionEffort ?? null });
+    return jsonResponse({ ok: true, model: config.injectionModel ?? null, effort: config.injectionEffort ?? null, prompt: config.injectionPrompt ?? null });
   }
 
   // Subagent model picker: which ≤5 routed models Codex's spawn_agent advertises (it shows the

package/src/server/request-log.ts

@@ -261,7 +261,10 @@ function captureUpstreamError(logCtx: RequestLogContext, text: string | null): v
       type?: unknown;
       error?: { message?: unknown };
       last_error?: { message?: unknown };
-      response?: { error?: { type?: unknown; code?: unknown; message?: unknown } };
+      response?: {
+        error?: { type?: unknown; code?: unknown; message?: unknown };
+        incomplete_details?: { reason?: unknown };
+      };
     };
     captureTerminalHttpStatus(logCtx, json);
     const message = json?.error?.message
@@ -269,12 +272,33 @@ function captureUpstreamError(logCtx: RequestLogContext, text: string | null): v
       ?? json?.response?.error?.message;
     if (typeof message === "string" && message.trim()) {
       logCtx.upstreamError = redactSecretString(message).slice(0, 500);
+      return;
+    }
+    // No human-readable error message: fall back to the structured incomplete reason emitted by
+    // the bridge on a stall-timeout or adapter EOF (response.incomplete). Maps the raw reason to a
+    // reader-facing label so a generic 502 in /api/logs explains WHY the turn ended, not just the
+    // mapped HTTP code.
+    const reason = json?.response?.incomplete_details?.reason;
+    if (typeof reason === "string" && reason.trim()) {
+      logCtx.upstreamError = redactSecretString(incompleteReasonLabel(reason.trim())).slice(0, 500);
     }
   } catch {
     /* not JSON; nothing to capture */
   }
 }
 
+/** Map a raw `incomplete_details.reason` (emitted by the bridge) to a reader-facing label. */
+function incompleteReasonLabel(reason: string): string {
+  switch (reason) {
+    case "upstream_stall_timeout":
+      return `Upstream stalled: no data for the stall-timeout window (${reason})`;
+    case "adapter_eof":
+      return `Upstream stream ended unexpectedly without a terminal event (${reason})`;
+    default:
+      return `Upstream incomplete: ${reason}`;
+  }
+}
+
 function captureTerminalHttpStatus(
   logCtx: RequestLogContext,
   json: {