@runtypelabs/cli 2.36.0 → 2.36.1

package/dist/index.js

@@ -19436,14 +19436,18 @@ var CANONICAL_LOG_COLUMN_SET = {
 };
 var CANONICAL_LOG_COLUMNS = Object.keys(CANONICAL_LOG_COLUMN_SET);
 
-// ../shared/dist/chunk-L2YZQMNS.mjs
+// ../shared/dist/chunk-NQDOLEMM.mjs
 var MODEL_FAMILY_PROVIDER_IDS = {
   "claude-fable-5": {
     "anthropic": "claude-fable-5",
     "vercel": "anthropic/claude-fable-5"
   },
   "claude-fable-5-1": {
-    "anthropic": "claude-fable-5-1"
+    "anthropic": "claude-fable-5-1",
+    "vercel": "anthropic/claude-fable-5.1"
+  },
+  "claude-fable-5.1": {
+    "vercel": "anthropic/claude-fable-5.1"
   },
   "claude-haiku-4-5": {
     "anthropic": "claude-haiku-4-5-20251001",
@@ -19638,6 +19642,10 @@ var MODEL_FAMILY_PROVIDER_IDS = {
     "google": "gemini-3.7-flash",
     "vercel": "google/gemini-3.7-flash"
   },
+  "gemini-3-8-flash": {
+    "google": "gemini-3.8-flash",
+    "vercel": "google/gemini-3.8-flash"
+  },
   "gemini-3-flash": {
     "google": "gemini-3-flash-preview"
   },
@@ -19689,6 +19697,10 @@ var MODEL_FAMILY_PROVIDER_IDS = {
     "google": "gemini-3.7-flash",
     "vercel": "google/gemini-3.7-flash"
   },
+  "gemini-3.8-flash": {
+    "google": "gemini-3.8-flash",
+    "vercel": "google/gemini-3.8-flash"
+  },
   "gemini-embedding-001": {
     "google": "gemini-embedding-001"
   },
@@ -20413,7 +20425,6 @@ var MODEL_FAMILY_PROVIDER_IDS = {
     "vercel": "alibaba/qwen3.5-plus"
   },
   "qwen3-6-27b": {
-    "mixlayer": "qwen/qwen3.6-27b",
     "vercel": "alibaba/qwen3.6-27b"
   },
   "qwen3-6-35b-a3b": {
@@ -20447,6 +20458,9 @@ var MODEL_FAMILY_PROVIDER_IDS = {
   "qwen3-8-max": {
     "vercel": "alibaba/qwen3.8-max"
   },
+  "qwen3-8-max-0902": {
+    "vercel": "alibaba/qwen3.8-max-0902"
+  },
   "qwen3-coder": {
     "vercel": "alibaba/qwen3-coder"
   },
@@ -20511,7 +20525,6 @@ var MODEL_FAMILY_PROVIDER_IDS = {
     "vercel": "alibaba/qwen3.5-plus"
   },
   "qwen3.6-27b": {
-    "mixlayer": "qwen/qwen3.6-27b",
     "vercel": "alibaba/qwen3.6-27b"
   },
   "qwen3.6-35b-a3b": {
@@ -20545,6 +20558,9 @@ var MODEL_FAMILY_PROVIDER_IDS = {
   "qwen3.8-max": {
     "vercel": "alibaba/qwen3.8-max"
   },
+  "qwen3.8-max-0902": {
+    "vercel": "alibaba/qwen3.8-max-0902"
+  },
   "rerank-v3-5": {
     "vercel": "cohere/rerank-v3.5"
   },
@@ -20650,6 +20666,14 @@ var DEEPSEEK_V4_FLASH_IDS = [
   "deepseek/deepseek-v4-flash-0731"
 ];
 var GLM_5_3_FLASH_IDS = ["glm-5.3-flash", "glm-5-3-flash", "zai/glm-5.3-flash"];
+var QWEN_3_8_27B_IDS = ["qwen3.8-27b", "qwen3-8-27b"];
+function mixlayerPrimary(ids) {
+  const config3 = {
+    providers: [{ provider: "mixlayer", weight: 100 }],
+    fallback: ["vercel"]
+  };
+  return Object.fromEntries(ids.map((id) => [id, config3]));
+}
 var PROVIDER_ROUTING_CONFIG = {
   "claude-sonnet-4-5": {
     providers: [
@@ -20877,7 +20901,8 @@ var PROVIDER_ROUTING_CONFIG = {
   },
   // WHY(docs/why/packages/shared/src.md#deepseek-v4-flash-mirrors-v4-pro-each-slash-form): DeepSeek V4 Flash mirrors V4 Pro: each slash form needs its own explicit Vercel fallback entry to escape the
   ...workersAiPrimary(DEEPSEEK_V4_FLASH_IDS),
-  // WHY(docs/why/packages/shared/src.md#qwen-3-8-27b-stays-off-workers-ai-its-262k-token): Qwen 3.8 27B stays off Workers AI: its 262K token cap is below Vercel's 1M, and fallback only engages when
+  // WHY(packages/shared/src/provider-routing.ts): Qwen 3.8 27B routes through Mixlayer because its 262K context is below Vercel's 1M.
+  ...mixlayerPrimary(QWEN_3_8_27B_IDS),
   // WHY(docs/why/packages/shared/src.md#vercel-renamed-its-xai-gateway-namespace-xai-to): Vercel renamed its xAI gateway namespace xai/ to spacexai/ on 2026-08-22; this hand-maintained table isn't
   ...workersAiPrimary(GLM_5_3_FLASH_IDS),
   "minimax-m2.7": {
@@ -21304,7 +21329,7 @@ var BuiltInToolIdPrefix = {
   PLATFORM: "platform"
 };
 
-// ../shared/dist/chunk-MICECRNL.mjs
+// ../shared/dist/chunk-UPJ4KSD4.mjs
 function sanitizeToolNameForAiSdk(name) {
   return name.replace(/[^a-zA-Z0-9_]/g, "_").toLowerCase().replace(/_+/g, "_").replace(/^_|_$/g, "");
 }
@@ -24188,15 +24213,8 @@ var GRANOLA_INTEGRATION = {
   ]
 };
 
-// ../shared/dist/chunk-IEAXJJXV.mjs
+// ../shared/dist/chunk-IMR6WLWN.mjs
 var DEFAULT_MODELS_FOR_NEW_ACCOUNTS = [
-  {
-    provider: "runtype",
-    modelId: "qwen/qwen3.6-27b",
-    isDefault: false,
-    strengths: "Fast, low-cost open-weight workhorse with reasoning, tool-use, and vision \u2014 good for high-volume extraction, classification, and simple transforms.",
-    bestFor: ["fast-cheap", "vision"]
-  },
   // WHY(docs/why/packages/shared/src.md#openai-models-route-via-openai-direct-with-verce): OpenAI models route via OpenAI direct with Vercel fallback; routed entries benefit from automatic provider
   {
     provider: "runtype",
@@ -24244,7 +24262,7 @@ var DEFAULT_MODELS_FOR_NEW_ACCOUNTS = [
   // WHY(docs/why/packages/shared/src.md#google-models-route-via-google-direct-with-verte): Google models route via Google direct, with Vertex/Vercel fallback.
   {
     provider: "runtype",
-    modelId: "gemini-3.7-flash",
+    modelId: "gemini-3.8-flash",
     isDefault: true,
     strengths: "Fast, low-cost, and natively multimodal with good reasoning, tool-use, and a 1M context \u2014 the account default and a strong choice for vision and high-throughput steps.",
     bestFor: ["default", "vision", "fast-cheap", "agentic-tools", "long-context"]
@@ -24257,14 +24275,6 @@ var DEFAULT_MODELS_FOR_NEW_ACCOUNTS = [
     strengths: "Specialised coding workhorse with strong tool-use, low cost, and a 1M context \u2014 ideal for code generation, debugging, and multi-file editing workflows.",
     bestFor: ["agentic-tools", "long-context"]
   },
-  // WHY(docs/why/packages/shared/src.md#nvidia-nemotron-ultra-reasoning-tool-use-flagshi): NVIDIA Nemotron Ultra: reasoning + tool-use flagship, routed via Vercel.
-  {
-    provider: "runtype",
-    modelId: "nemotron-3-ultra-550b-a55b",
-    isDefault: false,
-    strengths: "Large open-weight reasoning and tool-use flagship with a 1M context at a low price \u2014 strong analytical depth where an open model is preferred.",
-    bestFor: ["deep-reasoning", "agentic-tools", "long-context"]
-  },
   // WHY(docs/why/packages/shared/src.md#deepseek-v4-flash-routes-via-cloudflare-workers): DeepSeek V4 Flash routes via Cloudflare Workers AI (bare name resolves to
   {
     provider: "runtype",
@@ -24280,6 +24290,22 @@ var DEFAULT_MODELS_FOR_NEW_ACCOUNTS = [
     isDefault: false,
     strengths: "Agentic flagship \u2014 best-in-class at long, multi-round tool-call loops and coding/agent workflows, with a 1M context; well suited to tool-heavy agent steps.",
     bestFor: ["agentic-tools", "default", "deep-reasoning", "long-context"]
+  },
+  // WHY(packages/shared/src/provider-routing.ts): Z.ai GLM-5.3 Flash routes via Workers AI, with Vercel fallback.
+  {
+    provider: "runtype",
+    modelId: "glm-5.3-flash",
+    isDefault: false,
+    strengths: "Fast, low-cost model with strong reasoning, tool-use, and a 1M context \u2014 well suited to high-throughput agentic steps and long documents.",
+    bestFor: ["fast-cheap", "default", "agentic-tools", "long-context"]
+  },
+  // WHY(packages/shared/src/provider-routing.ts): Qwen 3.8 27B routes via Mixlayer, with Vercel fallback.
+  {
+    provider: "runtype",
+    modelId: "qwen3.8-27b",
+    isDefault: false,
+    strengths: "Fast open-weight model with strong reasoning, tool-use, and a 262K context \u2014 well suited to agentic tasks and long documents.",
+    bestFor: ["fast-cheap", "default", "deep-reasoning", "agentic-tools", "long-context"]
   }
 ];
 var DEFAULT_MODEL_ID = DEFAULT_MODELS_FOR_NEW_ACCOUNTS.find((m2) => m2.isDefault)?.modelId ?? "kimi-k2.6";
@@ -26504,7 +26530,8 @@ var AGENT_CONTENT_CONFIG_KEYS = {
   memory: true,
   sandbox: true,
   tenancyStrategy: true,
-  durability: true
+  durability: true,
+  state: true
 };
 var AGENT_CONTENT_CONFIG_KEY_LIST = Object.keys(AGENT_CONTENT_CONFIG_KEYS).sort();
 function stripTrailingChars(value, char) {
@@ -26825,6 +26852,28 @@ var MAX_MESSAGE_CONTENT_BYTES = 1024 * 1024;
 var MAX_MESSAGES_TOTAL_BYTES = 32 * 1024 * 1024;
 var encoder = new TextEncoder();
 var UNEVALUATED_VERDICT = Object.freeze({ status: "unevaluated" });
+var AGENT_STATE_DEFAULT_MERGE_STRATEGY = "merge";
+var agentPredictStateMappingSchema = external_exports.object({
+  /** Top-level state key the predicted argument projects into. */
+  stateKey: external_exports.string().min(1),
+  // WHY(docs/why/packages/shared/src.md#the-tool-may-be-one-the-agent-never-declares-the): The tool may be one the agent never declares: the canonical demo predicts from a CLIENT-declared frontend tool.
+  /** Name of the tool whose arguments are predicted from. */
+  tool: external_exports.string().min(1),
+  /** Argument of that tool carrying the predicted value. */
+  toolArgument: external_exports.string().min(1)
+});
+var agentStateChannelSchema = external_exports.object({
+  /** Initial state when a run starts with no client-supplied state. */
+  initial: external_exports.record(external_exports.string(), external_exports.unknown()).optional(),
+  /** Optional JSON Schema describing the state shape (documentation + client codegen). */
+  schema: external_exports.record(external_exports.string(), external_exports.unknown()).optional(),
+  // WHY(docs/why/packages/shared/src.md#how-client-supplied-state-merges-with-the-agents): How client-supplied state merges with the agent's own: `'replace'` (the client's state wins wholesale) or
+  /** How client-supplied state merges with the agent's own: `'replace'` (the client's state wins wholesale) or `'merge'` (shallow last-writer-wins per top-level key). */
+  mergeStrategy: external_exports.enum(["replace", "merge"]).default(AGENT_STATE_DEFAULT_MERGE_STRATEGY),
+  // WHY(docs/why/packages/shared/src.md#the-mapping-is-server-authored-because-only-the-a): The mapping is server-authored because only the agent knows which argument feeds which state key.
+  /** Tool arguments the client may optimistically project into state while they stream. */
+  predictState: external_exports.array(agentPredictStateMappingSchema).optional()
+});
 var tenancyAssuranceLevelSchema = external_exports.enum(["anonymous", "asserted", "verified"]);
 var tenancyStrategyPresetSchema = external_exports.enum([
   "internal",
@@ -27175,7 +27224,9 @@ var agentRuntimeConfigSchema = external_exports.object({
   // WHY(docs/why/packages/shared/src.md#tenancystrategy-is-optional-per-resource-isolati): tenancyStrategy is optional per-resource isolation opt-in; absent means platform-global default
   tenancyStrategy: tenancyStrategySchema.optional(),
   // WHY(docs/why/packages/shared/src.md#durable-turn-knobs-single-sourced-from-agentdura): Durable-turn knobs single-sourced from agentDurabilityConfigSchema; durability is the platform default
-  durability: agentDurabilityConfigSchema.optional()
+  durability: agentDurabilityConfigSchema.optional(),
+  // WHY(docs/why/packages/shared/src.md#the-same-schema-instance-the-export-plane-uses): The SAME instance the export plane uses, so the authoring and export planes cannot drift.
+  state: agentStateChannelSchema.optional()
 }).strict();
 var contextFallbackSchema = external_exports.union([
   external_exports.object({ type: external_exports.literal("retry"), delay: external_exports.number().optional() }),
@@ -30951,7 +31002,7 @@ var APPROVAL_UNATTENDED_SURFACES = [
   "a2a"
 ];
 var APPROVAL_PREFLIGHT_REJECTED_SURFACES = ["client-chat"];
-var APPROVAL_LEGACY_FOLD_SURFACES = APPROVAL_UNATTENDED_SURFACES.filter(
+var APPROVAL_REFUSING_SURFACES = APPROVAL_UNATTENDED_SURFACES.filter(
   (surface) => surface !== "client-chat"
 );
 function validatePromptApprovalSurfaceEnforcement(flowSteps, buckets) {
@@ -30963,7 +31014,7 @@ function validatePromptApprovalSurfaceEnforcement(flowSteps, buckets) {
       "warning",
       {
         code: APPROVAL_ENFORCED_ON_ATTENDED_SURFACES_ONLY_CODE,
-        message: `Step "${step.name}" arms a human approval gate (config.tools.approval.require), which is enforced only on attended surfaces that can present the prompt and resume the run: interactive dispatch and agent execute. Client chat rejects approval-gated execution because it has no secure continuation channel back to the visitor. On the other unattended surfaces (${APPROVAL_LEGACY_FOLD_SURFACES.join(", ")}) the flow runs on the legacy engine, which arms an approval gate only from the agent it dispatches and never from a flow step's own tools.approval, so these tools run without waiting for a decision. If they must not run unattended, keep this flow on attended surfaces, or move the approval requirement onto the agent an unattended surface dispatches so the gate travels with it.`,
+        message: `Step "${step.name}" arms a human approval gate (config.tools.approval.require), which is enforced only on attended surfaces that can present the prompt and resume the run: interactive dispatch and agent execute. Client chat rejects approval-gated execution because it has no secure continuation channel back to the visitor. The other unattended surfaces (${APPROVAL_REFUSING_SURFACES.join(", ")}) refuse the run (ineligible:surface-approval-pause): no engine answers the prompt without an attended client, so the flow does not run there at all. If it must run unattended, remove the approval requirement from this step.`,
         path: `flowSteps[${stepIndex}].config.tools.approval`,
         step: { index: stepIndex, name: step.name, type: step.type },
         details: {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@runtypelabs/cli",
-  "version": "2.36.0",
+  "version": "2.36.1",
   "description": "Command-line interface for Runtype AI platform",
   "type": "module",
   "main": "dist/index.js",
@@ -11,7 +11,7 @@
     "@modelcontextprotocol/client": "2.0.0",
     "@modelcontextprotocol/core": "2.0.0",
     "@runtypelabs/ink-components": "0.3.7",
-    "@runtypelabs/sdk": "9.6.0",
+    "@runtypelabs/sdk": "9.7.0",
     "@runtypelabs/terminal-animations": "0.2.4",
     "add-mcp": "2.0.0",
     "chalk": "^5.3.0",
@@ -33,7 +33,7 @@
   "devDependencies": {
     "@modelcontextprotocol/server": "2.0.0",
     "@runtypelabs/persona": "4.22.0",
-    "@runtypelabs/shared": "3.39.0",
+    "@runtypelabs/shared": "3.40.0",
     "@types/express": "^5.0.6",
     "@types/micromatch": "^4.0.9",
     "@types/node": "^25.3.3",