@inferencesh/sdk 0.32.0 → 0.33.0

package/CHANGELOG.md

@@ -7,6 +7,22 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.33.0] - 2026-10-24
+
+### Breaking
+- `AuthResponse.otp_method` is gone: read `otp_methods`, the ordered `SecondFactor[]` (`totp`, `passkey`, `recovery`, `email`) a sign-in may be finished with.
+
+### Added
+- The agent reducer applies streamed reasoning as it arrives; before, only text streamed.
+- `AgentInfo.model_capabilities`: what the agent's model takes besides text (`image`, `file`, ...), from its core app version.
+- `isRunWaiting(run)`. A chat whose run waits on the agent's own wait call (`ChatStatusWaiting`) counts as an open turn, not as one needing a person.
+- Generated types: `ChatStatusWaiting`, `ToolInvocationStatusWaiting`, `ChatEventTypeRunFailed`, `ChatEventTypeRunEnded`, `ChatRunEvent`, `RunEndReason`, `SecondFactor`, `InfraRemote`, `ScopeContentWrite`, `ErrorCodeLastAuthenticator`, `InstancePriceHistoryPointDTO`, app capability and remote llm direct mode types.
+
+### Changed
+- Proxy endpoints: cancelling a queued message needs `conversations:write` only.
+
+## [0.32.0] - 2026-10-20
+
 ### Changed
 - `approveTool` and `rejectTool` resolve the call's interrupt (`POST /interrupts/{id}/resolve`), the same answer as every other pause (a wait, a hook gate, a held turn); they no longer post to `/tools/{id}/invoke|reject`. A call that is not awaiting approval rejects with "Tool call … is not awaiting approval".
 - `resolveInterrupt` (agent actions and `AgentsAPI`) takes an optional `reason`, sent with a deny as `data.reason`.

package/dist/agent/actions.js

@@ -392,13 +392,14 @@ export function createActions(ctx) {
                 const result = await api.sendMessage(client, agentConfig, chatId, trimmedText, files);
                 if (result) {
                     const { chatId: newChatId, userMessage } = result;
-                    // Add user message from POST response immediately
-                    if (userMessage)
+                    // An open SSE stream owns the message: it may already have moved it
+                    // on (a message that ends a wait is taken up before this response
+                    // returns). Polling refetches only when the chat's status changes,
+                    // so it shows the response, as does a chat with no stream yet.
+                    const manager = getStreamManager();
+                    if (userMessage && !(manager instanceof StreamableManager))
                         dispatch({ type: 'UPDATE_MESSAGE', payload: userMessage });
-                    // Start streaming if not already connected
-                    const streamManager = getStreamManager();
-                    if (newChatId && !streamManager) {
-                        // Either new chat or stream was stopped - restart streaming
+                    if (newChatId && !manager) {
                         if (!chatId) {
                             dispatch({ type: 'SET_CHAT_ID', payload: newChatId });
                             callbacks.onChatCreated?.(newChatId);

package/dist/agent/api.js

@@ -76,9 +76,11 @@ export async function fetchAgentInfo(client, agentRef) {
     try {
         const resp = await client.http.request('get', `/agents/${agentRef}`);
         const version = resp.data?.version;
+        const capabilities = version?.core_app?.app?.version?.metadata?.capabilities;
         return {
             description: version?.description,
             example_prompts: version?.example_prompts,
+            model_capabilities: Array.isArray(capabilities) ? capabilities : undefined,
         };
     }
     catch {

package/dist/agent/reducer.js

@@ -3,13 +3,15 @@
  *
  * Pure reducer for managing agent chat state.
  */
-import { ChatStatusBusy, ChatStatusAwaitingInput, ChatStatusIdle } from '../types.js';
-import { isMessageTerminal, isRunInterrupted, isRunWorking } from '../utils.js';
+import { ChatMessageContentTypeReasoning, ChatMessageContentTypeText, ChatStatusBusy, ChatStatusAwaitingInput, ChatStatusIdle, ChatStatusWaiting, } from '../types.js';
+import { isMessageTerminal, isRunInterrupted, isRunWaiting, isRunWorking } from '../utils.js';
 function deriveChatStatus(run) {
     if (!run)
         return ChatStatusIdle;
     if (isRunWorking(run.state))
         return ChatStatusBusy;
+    if (isRunWaiting(run))
+        return ChatStatusWaiting;
     if (isRunInterrupted(run.state))
         return ChatStatusAwaitingInput;
     return ChatStatusIdle;
@@ -60,26 +62,55 @@ function normalizeMessage(message, existing) {
     return { ...message, content: existing?.content ?? [] };
 }
 /**
- * Streamed text only grows until the message is done. Deltas build the text
- * locally, while a message update carries the server's saved copy, which can
- * lag behind them (often still empty). Taking that copy as-is blanks the text
- * until the next delta, so while the message is still generating, a text
- * that is an earlier version of what is shown keeps what is shown. Once the
- * message is terminal the server's copy is final and wins.
+ * The content blocks deltas stream, and the delta field each is built from.
+ * An optional block gets no block until something streams into it.
  */
-function keepStreamedText(updated, existing) {
+const streamedBlocks = [
+    { type: ChatMessageContentTypeText, field: 'response', optional: false },
+    { type: ChatMessageContentTypeReasoning, field: 'reasoning', optional: true },
+];
+/** content with the block of type set to text, added in front when missing. */
+function withBlockText(content, type, text) {
+    return content.some((c) => c.type === type)
+        ? content.map((c) => (c.type === type ? { ...c, text } : c))
+        : [{ type, text }, ...content];
+}
+const blockText = (message, type) => message.content.find((c) => c.type === type)?.text;
+/** copy is an earlier version of shown: shown already has all of it and more. */
+const isBehind = (copy, shown) => !!shown && shown.startsWith(copy ?? '');
+/**
+ * Streamed text and reasoning only grow until the message is done. Deltas
+ * build them locally, while a message update carries the server's saved
+ * copy, which can lag behind them (often still empty). While the message is
+ * generating, a server copy behind what is shown keeps what is shown. Once
+ * the message is terminal the server's copy is final and wins.
+ */
+function keepStreamedBlocks(updated, existing) {
     if (isMessageTerminal(updated.status))
         return updated;
-    const shown = existing.content.find((c) => c.type === 'text')?.text;
-    if (!shown)
-        return updated;
-    const incoming = updated.content.find((c) => c.type === 'text');
-    if (incoming?.text && !shown.startsWith(incoming.text))
-        return updated;
-    const content = incoming
-        ? updated.content.map((c) => (c === incoming ? { ...c, text: shown } : c))
-        : [{ type: 'text', text: shown }, ...updated.content];
-    return { ...updated, content };
+    let content = updated.content;
+    for (const { type } of streamedBlocks) {
+        const shown = blockText(existing, type);
+        if (isBehind(blockText(updated, type), shown))
+            content = withBlockText(content, type, shown);
+    }
+    return content === updated.content ? updated : { ...updated, content };
+}
+/**
+ * The message with what the deltas have built so far. Deltas from a stream
+ * joined mid-message start partway, behind what the message already shows,
+ * so the message keeps its copy until they catch up.
+ */
+function applyStreamedOutput(message, output) {
+    let content = message.content;
+    for (const { type, field, optional } of streamedBlocks) {
+        const streamed = output[field];
+        if (optional && !streamed)
+            continue;
+        if (!isBehind(streamed, blockText(message, type)))
+            content = withBlockText(content, type, streamed);
+    }
+    return content === message.content ? message : { ...message, content };
 }
 const normalizeMessages = (messages) => messages.map((m) => normalizeMessage(m));
 // =============================================================================
@@ -182,7 +213,7 @@ export function chatReducer(state, action) {
             const existingIndex = state.messages.findIndex((m) => m.id === message.id);
             if (existingIndex !== -1) {
                 const existing = state.messages[existingIndex];
-                const updated = keepStreamedText(action.partial ? normalizeMessage({ ...existing, ...message }, existing) : normalizeMessage(message), existing);
+                const updated = keepStreamedBlocks(action.partial ? normalizeMessage({ ...existing, ...message }, existing) : normalizeMessage(message), existing);
                 if (existing === updated)
                     return state;
                 const newMessages = [...state.messages];
@@ -207,16 +238,10 @@ export function chatReducer(state, action) {
             const target = msgs.find(m => m.id === messageId);
             if (!target)
                 return state;
-            const textBlock = target.content.find(c => c.type === 'text');
-            // Deltas from a stream joined mid-message start partway; the message
-            // update already showed more, so keep it until they catch up.
-            if (textBlock?.text && textBlock.text.startsWith(output.response ?? ''))
+            const next = applyStreamedOutput(target, output);
+            if (next === target)
                 return state;
-            const newContent = textBlock
-                ? target.content.map(c => c.type === 'text' ? { ...c, text: output.response } : c)
-                : [{ type: 'text', text: output.response }, ...target.content];
-            const newMessages = msgs.map(m => m.id === target.id ? { ...target, content: newContent } : m);
-            return { ...state, messages: newMessages };
+            return { ...state, messages: msgs.map(m => (m.id === target.id ? next : m)) };
         }
         case 'SET_CONNECTION_STATUS':
             return { ...state, connectionStatus: action.payload };

package/dist/agent/types.d.ts

@@ -89,7 +89,13 @@ export declare function isTemplateConfig(config: AgentOptions): config is Templa
 /**
  * Agent metadata fetched from the backend for template configs
  */
-export type AgentInfo = Partial<Pick<AgentVersionDTO, 'description' | 'example_prompts'>>;
+export type AgentInfo = Partial<Pick<AgentVersionDTO, 'description' | 'example_prompts'>> & {
+    /**
+     * What the agent's model takes besides text ("image", "file", ...), from
+     * its app version. Undefined when unknown; then assume everything.
+     */
+    model_capabilities?: string[];
+};
 /**
  * Status of the chat
  */

package/dist/index.d.ts

@@ -36,7 +36,7 @@ export type { ClientTool, ClientToolHandler, ClientToolResult } from './tool-bui
 export { lifecycleHook, learningHooks } from './hook-builder.js';
 export type { LifecycleHookBuilder } from './hook-builder.js';
 export { DeltaAccumulator, createLLMDeltaAccumulator, type DeltaEvent, type FieldTags, type FieldTagsRegistry } from './delta.js';
-export { parseStatus, isTerminalStatus, isChatBusy, isAwaitingHuman, isRunTerminal, isRunInterrupted, isRunSettled, isRunWorking, isMessageTerminal, pendingApprovals, heldTurn, pendingWait, } from './utils.js';
+export { parseStatus, isTerminalStatus, isChatBusy, isAwaitingHuman, isRunTerminal, isRunInterrupted, isRunWaiting, isRunSettled, isRunWorking, isMessageTerminal, pendingApprovals, heldTurn, pendingWait, } from './utils.js';
 export type { PendingApproval, HeldTurn, PendingWait } from './utils.js';
 export * from './types.js';
 export type { TaskDTO as Task } from './types.js';

package/dist/index.js

@@ -38,7 +38,7 @@ export { lifecycleHook, learningHooks } from './hook-builder.js';
 // Delta accumulator for streaming responses
 export { DeltaAccumulator, createLLMDeltaAccumulator } from './delta.js';
 // Status utilities (handle both int and string status values)
-export { parseStatus, isTerminalStatus, isChatBusy, isAwaitingHuman, isRunTerminal, isRunInterrupted, isRunSettled, isRunWorking, isMessageTerminal, pendingApprovals, heldTurn, pendingWait, } from './utils.js';
+export { parseStatus, isTerminalStatus, isChatBusy, isAwaitingHuman, isRunTerminal, isRunInterrupted, isRunWaiting, isRunSettled, isRunWorking, isMessageTerminal, pendingApprovals, heldTurn, pendingWait, } from './utils.js';
 // Types - includes TaskStatus constants and all DTOs
 export * from './types.js';
 // =============================================================================

package/dist/proxy/endpoints.json

@@ -140,7 +140,6 @@
             "path": "/chats/messages/{messageId}/cancel",
             "purpose": "cancel a message",
             "scopes": [
-                "conversations:read",
                 "conversations:write"
             ]
         },

package/dist/types.d.ts

@@ -608,7 +608,11 @@ export interface AuthResponse {
     session_id: string;
     is_new?: boolean;
     otp_required?: boolean;
-    otp_method?: string;
+    /**
+     * OTPMethods are the second factors the sign-in may be finished with,
+     * the default first (POST /auth/otp/verify with method).
+     */
+    otp_methods?: SecondFactor[];
     challenge_token?: string;
     redirect_to?: string;
     provider?: string;
@@ -911,7 +915,7 @@ export interface EstimateCostResponse {
 /**
  * Scope represents an API key permission scope string.
  */
-export type Scope = "*" | "agents" | "apps" | "conversations" | "files" | "datastores" | "templates" | "flows" | "projects" | "teams" | "billing" | "artifacts" | "agents:read" | "agents:write" | "agents:execute" | "apps:read" | "apps:write" | "apps:execute" | "conversations:read" | "conversations:write" | "files:read" | "files:write" | "datastores:read" | "datastores:write" | "flows:read" | "flows:write" | "flows:execute" | "projects:read" | "projects:write" | "teams:read" | "teams:write" | "billing:read" | "billing:write" | "secrets:read" | "secrets:write" | "secrets:reveal" | "credentials:read" | "credentials:write" | "engines:read" | "engines:write" | "remotes:read" | "remotes:write" | "apikeys:read" | "apikeys:write" | "knowledge:read" | "knowledge:write" | "artifacts:read" | "artifacts:write" | "user:read" | "user:write" | "settings:read" | "settings:write" | "approvals:write" | "admin:read" | "admin:write" | "store:pricing" | "store:submit";
+export type Scope = "*" | "agents" | "apps" | "conversations" | "files" | "datastores" | "templates" | "flows" | "projects" | "teams" | "billing" | "artifacts" | "agents:read" | "agents:write" | "agents:execute" | "apps:read" | "apps:write" | "apps:execute" | "conversations:read" | "conversations:write" | "files:read" | "files:write" | "datastores:read" | "datastores:write" | "flows:read" | "flows:write" | "flows:execute" | "projects:read" | "projects:write" | "teams:read" | "teams:write" | "billing:read" | "billing:write" | "secrets:read" | "secrets:write" | "secrets:reveal" | "credentials:read" | "credentials:write" | "engines:read" | "engines:write" | "remotes:read" | "remotes:write" | "apikeys:read" | "apikeys:write" | "knowledge:read" | "knowledge:write" | "artifacts:read" | "artifacts:write" | "user:read" | "user:write" | "settings:read" | "settings:write" | "approvals:write" | "admin:read" | "admin:write" | "store:pricing" | "store:submit" | "content:write";
 /**
  * API Key Scopes - hierarchical permission system.
  * Resource-level scopes (e.g., "agents") imply all action-level scopes (e.g., "agents:read").
@@ -1218,10 +1222,18 @@ export declare const ScopeStorePricing: Scope;
  * publisher's behalf. Held like ScopeStorePricing.
  */
 export declare const ScopeStoreSubmit: Scope;
+/**
+ * ScopeContentWrite writes the site's content (/content): blog and docs
+ * pages, menus and the platform media library. Held like
+ * ScopeStorePricing: only in the platform's workspace, by its members,
+ * their keys and runs, and its own keys. Reads of published content take
+ * no scope.
+ */
+export declare const ScopeContentWrite: Scope;
 /**
  * ScopeGroup identifies a category of scopes for UI grouping
  */
-export type ScopeGroup = "agents" | "apps" | "conversations" | "files" | "datastores" | "flows" | "projects" | "teams" | "billing" | "secrets" | "credentials" | "engines" | "remotes" | "apikeys" | "knowledge" | "artifacts" | "user" | "settings" | "approvals" | "admin" | "store";
+export type ScopeGroup = "agents" | "apps" | "conversations" | "files" | "datastores" | "flows" | "projects" | "teams" | "billing" | "secrets" | "credentials" | "engines" | "remotes" | "apikeys" | "knowledge" | "artifacts" | "user" | "settings" | "approvals" | "admin" | "store" | "content";
 export declare const ScopeGroupAgents: ScopeGroup;
 export declare const ScopeGroupApps: ScopeGroup;
 export declare const ScopeGroupConversations: ScopeGroup;
@@ -1243,6 +1255,7 @@ export declare const ScopeGroupSettings: ScopeGroup;
 export declare const ScopeGroupApprovals: ScopeGroup;
 export declare const ScopeGroupAdmin: ScopeGroup;
 export declare const ScopeGroupStore: ScopeGroup;
+export declare const ScopeGroupContent: ScopeGroup;
 /**
  * ScopeDefinition describes a single scope for UI rendering
  */
@@ -1398,9 +1411,6 @@ export interface AppPricing {
     description: string;
     description_rendered?: string;
 }
-/**
- * AppFunction represents a callable entry point within an app version.
- */
 export interface AppFunction {
     name: string;
     description?: string;
@@ -3062,7 +3072,7 @@ export interface DecideEntitlementRequestRequest {
  * and APIError.Code on the legacy envelope. Clients branch on these, so a
  * code is a named const here, where it generates into models and the SDKs.
  */
-export type ErrorCode = "invalid_request" | "validation_error" | "unauthorized" | "forbidden" | "not_found" | "conflict" | "name_conflict" | "already_exists" | "not_configured" | "method_not_allowed" | "rate_limited" | "internal_error" | "requirements_not_met" | "team_role_required" | "blocked_by_usage_policy" | "last_owner" | "account_deactivated" | "account_banned" | "person_required" | "otp_required" | "requires_sign_in" | "insufficient_scope" | "invalid_scope" | "impersonation_reason_required" | "mcp_auth_expired" | "admin_session_required" | "admin_scopes_not_approved" | "cli_session_required" | "admin_required" | "same_admin_required" | "browser_session_required" | "invalid_ttl" | "admin_authenticator_required" | "limit_exceeded" | "feature_not_available" | "entitlement_unavailable" | "payment_required" | "payment_method_required" | "form_closed" | "already_submitted" | "already_claimed" | "already_entitled" | "request_open" | "invalid_code" | "already_processed" | "expired" | "code_confirmation_required" | "app_retired" | "app_maintenance" | "invalid_transition" | "SESSION_NOT_FOUND" | "SESSION_EXPIRED" | "SESSION_ENDED" | "WORKER_LEASED" | "APP_MISMATCH" | "VERSION_MISMATCH" | "agents_disabled" | "remote_offline" | "remote_timeout" | "harness_not_drivable" | "harness_too_old";
+export type ErrorCode = "invalid_request" | "validation_error" | "unauthorized" | "forbidden" | "not_found" | "conflict" | "name_conflict" | "already_exists" | "not_configured" | "method_not_allowed" | "rate_limited" | "internal_error" | "requirements_not_met" | "team_role_required" | "blocked_by_usage_policy" | "last_owner" | "account_deactivated" | "account_banned" | "person_required" | "otp_required" | "requires_sign_in" | "insufficient_scope" | "invalid_scope" | "impersonation_reason_required" | "mcp_auth_expired" | "admin_session_required" | "admin_scopes_not_approved" | "cli_session_required" | "admin_required" | "same_admin_required" | "browser_session_required" | "invalid_ttl" | "admin_authenticator_required" | "last_authenticator" | "limit_exceeded" | "feature_not_available" | "entitlement_unavailable" | "payment_required" | "payment_method_required" | "form_closed" | "already_submitted" | "already_claimed" | "already_entitled" | "request_open" | "invalid_code" | "already_processed" | "expired" | "code_confirmation_required" | "app_retired" | "app_maintenance" | "invalid_transition" | "SESSION_NOT_FOUND" | "SESSION_EXPIRED" | "SESSION_ENDED" | "WORKER_LEASED" | "APP_MISMATCH" | "VERSION_MISMATCH" | "agents_disabled" | "remote_offline" | "remote_timeout" | "harness_not_drivable" | "harness_too_old";
 export declare const ErrorCodeInvalidRequest: ErrorCode;
 export declare const ErrorCodeValidationError: ErrorCode;
 export declare const ErrorCodeUnauthorized: ErrorCode;
@@ -3171,6 +3181,13 @@ export declare const ErrorCodeInvalidTTL: ErrorCode;
  * has not proved it gets otp_required instead: re-authenticate with it.
  */
 export declare const ErrorCodeAdminAuthenticatorRequired: ErrorCode;
+/**
+ * ErrorCodeLastAuthenticator (409): the change would leave two-step
+ * verification on with no method to prove it (removing the last
+ * authenticator app or passkey), or turn it on with none enrolled.
+ * Turn two-step verification off first, or enroll a method.
+ */
+export declare const ErrorCodeLastAuthenticator: ErrorCode;
 /**
  * Entitlements. LimitExceeded (402) and FeatureNotAvailable (403) carry
  * EntitlementErrorMeta. EntitlementUnavailable (500) means the plan could
@@ -3846,6 +3863,29 @@ export interface InstanceTypeBootTime {
     updated_at: string;
     sample_size: number;
 }
+/**
+ * InstancePriceHistoryPointDTO is one day's cheapest hourly prices, in cents,
+ * for a GPU type. List prices rarely move; the in-stock price does, as stock
+ * comes and goes. A price with no listing (or no stock) that day is omitted.
+ */
+export interface InstancePriceHistoryPointDTO {
+    day: string;
+    /**
+     * OnDemand and Spot are the cheapest listed prices, in stock or not.
+     */
+    on_demand?: number;
+    spot?: number;
+    /**
+     * OnDemandInStock and SpotInStock are the cheapest prices that had stock.
+     */
+    on_demand_in_stock?: number;
+    spot_in_stock?: number;
+    providers: number;
+    /**
+     * Regions is the number of provider regions with on-demand stock.
+     */
+    regions: number;
+}
 export interface InterruptDTO extends BaseModelDTO, PermissionModelDTO {
     run_id: string;
     chat_id: string;
@@ -5454,7 +5494,17 @@ export interface UserDTO extends BaseModelDTO {
     avatar_url: string;
     banned_at?: string;
     ban_note?: string;
+    /**
+     * TOTPEnabled mirrors TwoFactorEnabled for clients built before it
+     * (the Swift SDK decodes it as required).
+     * Deprecated: read TwoFactorEnabled. Kept until no Swift app build older
+     * than sdk-swift 0.23.0 is in use, since those fail to decode /me without it.
+     */
     totp_enabled: boolean;
+    /**
+     * TwoFactorEnabled is the account's two-step verification switch.
+     */
+    two_factor_enabled: boolean;
     metadata?: UserMetadataDTO;
 }
 /**
@@ -6248,10 +6298,15 @@ export declare const EntitlementRequestOpen: EntitlementRequestState;
 export declare const EntitlementRequestAccepted: EntitlementRequestState;
 export declare const EntitlementRequestDeclined: EntitlementRequestState;
 export declare const EntitlementRequestWithdrawn: EntitlementRequestState;
-export type ChatStatus = "busy" | "idle" | "awaiting_input" | "completed";
+export type ChatStatus = "busy" | "idle" | "awaiting_input" | "waiting" | "completed";
 export declare const ChatStatusBusy: ChatStatus;
 export declare const ChatStatusIdle: ChatStatus;
 export declare const ChatStatusAwaitingInput: ChatStatus;
+/**
+ * ChatStatusWaiting: the run is paused on the agent's own wait call. A
+ * message ends the wait early; nobody has to act on it.
+ */
+export declare const ChatStatusWaiting: ChatStatus;
 export declare const ChatStatusCompleted: ChatStatus;
 export type PlanStepStatus = "pending" | "in_progress" | "completed" | "cancelled";
 export declare const PlanStepStatusPending: PlanStepStatus;
@@ -6333,8 +6388,41 @@ export interface ChatMessageContent {
     tool_calls?: ToolCall[];
     event?: ChatEvent;
 }
-export type ChatEventType = "hook";
+export type ChatEventType = "hook" | "run_failed" | "run_ended";
 export declare const ChatEventTypeHook: ChatEventType;
+/**
+ * ChatEventTypeRunFailed marks the point a run ended in failure; the
+ * content's text is the run's error.
+ */
+export declare const ChatEventTypeRunFailed: ChatEventType;
+/**
+ * ChatEventTypeRunEnded marks the point a run completed or was stopped;
+ * the event's run says how, and the content's text, when set, why.
+ */
+export declare const ChatEventTypeRunEnded: ChatEventType;
+/**
+ * RunEndReason is why a run that completed ended the way it did, when its
+ * last turn did not end with a reply.
+ */
+export type RunEndReason = "output_limit" | "no_reply";
+/**
+ * RunEndReasonOutputLimit: the model's last turn hit its output limit
+ * (max_tokens) before it replied or called a tool.
+ */
+export declare const RunEndReasonOutputLimit: RunEndReason;
+/**
+ * RunEndReasonNoReply: the model ended its last turn with no reply and
+ * no tool call.
+ */
+export declare const RunEndReasonNoReply: RunEndReason;
+/**
+ * ChatRunEvent is how a run ended, on a run_ended event.
+ */
+export interface ChatRunEvent {
+    run_id: string;
+    state: AgentRunState;
+    reason?: RunEndReason;
+}
 /**
  * ChatEvent is the payload of an event-role message: system info shown in the
  * chat but never sent to the model.
@@ -6342,6 +6430,7 @@ export declare const ChatEventTypeHook: ChatEventType;
 export interface ChatEvent {
     type: ChatEventType;
     hook?: ChatHookEvent;
+    run?: ChatRunEvent;
 }
 /**
  * ChatHookEvent records one lifecycle hook handler run.
@@ -6732,8 +6821,8 @@ export declare const GraphEdgeTypeOutput: GraphEdgeType;
 export declare const InterruptResourceTurn: InterruptResourceType;
 /**
  * InterruptReasonWait is a tool call the agent made to wait: the run is
- * paused until the interrupt's deadline, when its own expiry trigger settles
- * it (allow), or until a person ends the wait sooner. Its resource is the
+ * paused until the interrupt's deadline, when the expiry sweep settles it
+ * (allow), or until a person ends the wait sooner. Its resource is the
  * wait call (InterruptResourceToolInvocation).
  */
 export declare const InterruptReasonWait: InterruptReason;
@@ -7058,6 +7147,33 @@ export type MCPServerAuthType = "oauth" | "api_key" | "none";
 export declare const MCPServerAuthOAuth: MCPServerAuthType;
 export declare const MCPServerAuthAPIKey: MCPServerAuthType;
 export declare const MCPServerAuthNone: MCPServerAuthType;
+/**
+ * SecondFactor is one way a person proves their second factor. It names
+ * what a session proved (AuthSession.SecondFactor), what a login challenge
+ * offers, the method a verify request picks, and the kind of an enrolled
+ * method (SecondFactorMethod.Kind: the enrollable ones).
+ */
+export type SecondFactor = "totp" | "passkey" | "recovery" | "email";
+/**
+ * SecondFactorTOTP: a code from an enrolled authenticator app.
+ */
+export declare const SecondFactorTOTP: SecondFactor;
+/**
+ * SecondFactorPasskey: a WebAuthn assertion from an enrolled passkey.
+ */
+export declare const SecondFactorPasskey: SecondFactor;
+/**
+ * SecondFactorRecovery: one of the account's single-use recovery codes.
+ * Not enrolled on its own: issued once an authenticator exists.
+ */
+export declare const SecondFactorRecovery: SecondFactor;
+/**
+ * SecondFactorEmail: a code emailed to the account. The weakest: the
+ * same mailbox a magic link lands in, so it is offered only to an
+ * account with two-step verification off (a platform admin's first
+ * sign-in, re-auth before they enroll).
+ */
+export declare const SecondFactorEmail: SecondFactor;
 /**
  * TeamInviteStatus represents the status of a team invitation
  */
@@ -7552,10 +7668,17 @@ export declare const TaskStatusRejected: TaskStatus;
 /**
  * Infra defines where a task should run.
  */
-export type Infra = "private" | "cloud" | "private_first";
+export type Infra = "private" | "cloud" | "private_first" | "remote";
 export declare const InfraPrivate: Infra;
 export declare const InfraCloud: Infra;
 export declare const InfraPrivateFirst: Infra;
+/**
+ * InfraRemote is a model call a harness on the caller's own remote
+ * answers (an agent whose model app is a harness model app). The
+ * scheduler never takes it: the remote model service sends it to the
+ * remote's daemon, whose frames end it.
+ */
+export declare const InfraRemote: Infra;
 /**
  * TaskLogType defines the type of task log.
  */
@@ -8070,11 +8193,12 @@ export declare const AgentEventSubagentCompleted: AgentEventType;
 /**
  * ToolInvocationStatus represents the execution status of a tool invocation
  */
-export type ToolInvocationStatus = "pending" | "in_progress" | "awaiting_input" | "awaiting_approval" | "completed" | "failed" | "cancelled";
+export type ToolInvocationStatus = "pending" | "in_progress" | "awaiting_input" | "awaiting_approval" | "waiting" | "completed" | "failed" | "cancelled";
 export declare const ToolInvocationStatusPending: ToolInvocationStatus;
 export declare const ToolInvocationStatusInProgress: ToolInvocationStatus;
 export declare const ToolInvocationStatusAwaitingInput: ToolInvocationStatus;
 export declare const ToolInvocationStatusAwaitingApproval: ToolInvocationStatus;
+export declare const ToolInvocationStatusWaiting: ToolInvocationStatus;
 export declare const ToolInvocationStatusCompleted: ToolInvocationStatus;
 export declare const ToolInvocationStatusFailed: ToolInvocationStatus;
 export declare const ToolInvocationStatusCancelled: ToolInvocationStatus;

package/dist/types.js

@@ -321,6 +321,14 @@ export const ScopeStorePricing = "store:pricing";
  * publisher's behalf. Held like ScopeStorePricing.
  */
 export const ScopeStoreSubmit = "store:submit";
+/**
+ * ScopeContentWrite writes the site's content (/content): blog and docs
+ * pages, menus and the platform media library. Held like
+ * ScopeStorePricing: only in the platform's workspace, by its members,
+ * their keys and runs, and its own keys. Reads of published content take
+ * no scope.
+ */
+export const ScopeContentWrite = "content:write";
 export const ScopeGroupAgents = "agents";
 export const ScopeGroupApps = "apps";
 export const ScopeGroupConversations = "conversations";
@@ -342,6 +350,7 @@ export const ScopeGroupSettings = "settings";
 export const ScopeGroupApprovals = "approvals";
 export const ScopeGroupAdmin = "admin";
 export const ScopeGroupStore = "store";
+export const ScopeGroupContent = "content";
 /**
  * AlwaysAllowScopeExact: this call exactly (each piece of the command,
  * the file, the domain).
@@ -485,6 +494,13 @@ export const ErrorCodeInvalidTTL = "invalid_ttl";
  * has not proved it gets otp_required instead: re-authenticate with it.
  */
 export const ErrorCodeAdminAuthenticatorRequired = "admin_authenticator_required";
+/**
+ * ErrorCodeLastAuthenticator (409): the change would leave two-step
+ * verification on with no method to prove it (removing the last
+ * authenticator app or passkey), or turn it on with none enrolled.
+ * Turn two-step verification off first, or enroll a method.
+ */
+export const ErrorCodeLastAuthenticator = "last_authenticator";
 /**
  * Entitlements. LimitExceeded (402) and FeatureNotAvailable (403) carry
  * EntitlementErrorMeta. EntitlementUnavailable (500) means the plan could
@@ -870,6 +886,11 @@ export const EntitlementRequestWithdrawn = "withdrawn";
 export const ChatStatusBusy = "busy";
 export const ChatStatusIdle = "idle";
 export const ChatStatusAwaitingInput = "awaiting_input";
+/**
+ * ChatStatusWaiting: the run is paused on the agent's own wait call. A
+ * message ends the wait early; nobody has to act on it.
+ */
+export const ChatStatusWaiting = "waiting";
 export const ChatStatusCompleted = "completed";
 export const PlanStepStatusPending = "pending";
 export const PlanStepStatusInProgress = "in_progress";
@@ -907,6 +928,26 @@ export const ChannelTypeDiscord = "discord";
 export const ChannelTypeTeams = "teams";
 export const ChannelTypeTelegram = "telegram";
 export const ChatEventTypeHook = "hook";
+/**
+ * ChatEventTypeRunFailed marks the point a run ended in failure; the
+ * content's text is the run's error.
+ */
+export const ChatEventTypeRunFailed = "run_failed";
+/**
+ * ChatEventTypeRunEnded marks the point a run completed or was stopped;
+ * the event's run says how, and the content's text, when set, why.
+ */
+export const ChatEventTypeRunEnded = "run_ended";
+/**
+ * RunEndReasonOutputLimit: the model's last turn hit its output limit
+ * (max_tokens) before it replied or called a tool.
+ */
+export const RunEndReasonOutputLimit = "output_limit";
+/**
+ * RunEndReasonNoReply: the model ended its last turn with no reply and
+ * no tool call.
+ */
+export const RunEndReasonNoReply = "no_reply";
 /**
  * DescriptionLimitListing bounds the description of an app, agent, flow,
  * MCP server or knowledge entry. A description is shown in listings and
@@ -994,8 +1035,8 @@ export const GraphEdgeTypeOutput = "output";
 export const InterruptResourceTurn = "turn";
 /**
  * InterruptReasonWait is a tool call the agent made to wait: the run is
- * paused until the interrupt's deadline, when its own expiry trigger settles
- * it (allow), or until a person ends the wait sooner. Its resource is the
+ * paused until the interrupt's deadline, when the expiry sweep settles it
+ * (allow), or until a person ends the wait sooner. Its resource is the
  * wait call (InterruptResourceToolInvocation).
  */
 export const InterruptReasonWait = "wait";
@@ -1111,6 +1152,26 @@ export const VideoRes4K = "4k";
 export const MCPServerAuthOAuth = "oauth";
 export const MCPServerAuthAPIKey = "api_key";
 export const MCPServerAuthNone = "none";
+/**
+ * SecondFactorTOTP: a code from an enrolled authenticator app.
+ */
+export const SecondFactorTOTP = "totp";
+/**
+ * SecondFactorPasskey: a WebAuthn assertion from an enrolled passkey.
+ */
+export const SecondFactorPasskey = "passkey";
+/**
+ * SecondFactorRecovery: one of the account's single-use recovery codes.
+ * Not enrolled on its own: issued once an authenticator exists.
+ */
+export const SecondFactorRecovery = "recovery";
+/**
+ * SecondFactorEmail: a code emailed to the account. The weakest: the
+ * same mailbox a magic link lands in, so it is offered only to an
+ * account with two-step verification off (a platform admin's first
+ * sign-in, re-auth before they enroll).
+ */
+export const SecondFactorEmail = "email";
 export const TeamInviteStatusPending = "pending";
 export const TeamInviteStatusAccepted = "accepted";
 export const TeamInviteStatusDeclined = "declined";
@@ -1483,6 +1544,13 @@ export const TaskStatusRejected = 13;
 export const InfraPrivate = "private";
 export const InfraCloud = "cloud";
 export const InfraPrivateFirst = "private_first";
+/**
+ * InfraRemote is a model call a harness on the caller's own remote
+ * answers (an agent whose model app is a harness model app). The
+ * scheduler never takes it: the remote model service sends it to the
+ * remote's daemon, whose frames end it.
+ */
+export const InfraRemote = "remote";
 export const TaskLogTypeBuild = 0;
 export const TaskLogTypeRun = 1;
 export const TaskLogTypeServe = 2;
@@ -1672,6 +1740,7 @@ export const ToolInvocationStatusPending = "pending";
 export const ToolInvocationStatusInProgress = "in_progress";
 export const ToolInvocationStatusAwaitingInput = "awaiting_input";
 export const ToolInvocationStatusAwaitingApproval = "awaiting_approval";
+export const ToolInvocationStatusWaiting = "waiting";
 export const ToolInvocationStatusCompleted = "completed";
 export const ToolInvocationStatusFailed = "failed";
 export const ToolInvocationStatusCancelled = "cancelled";

package/dist/utils.d.ts

@@ -4,7 +4,7 @@
  */
 import type { InterruptDTO } from './types.js';
 import { TaskStatus } from './types.js';
-import type { AgentRunState, ChatDTO, ChatMessageStatus } from './types.js';
+import type { AgentRunDTO, AgentRunState, ChatDTO, ChatMessageStatus } from './types.js';
 /**
  * Parse task status from int or string to TaskStatus number.
  * Handles both current int-based API and future string-based API.
@@ -18,12 +18,14 @@ export declare function isTerminalStatus(status: number | string | undefined | n
 export declare function isRunTerminal(state: AgentRunState | null | undefined): boolean;
 /** Parked on a human: a tool approval or input (input_required) or an authorization (auth_required). */
 export declare function isRunInterrupted(state: AgentRunState | null | undefined): boolean;
+/** Paused on the agent's own wait call: it resumes at its deadline or on a message, so nobody has to act. */
+export declare function isRunWaiting(run: Pick<AgentRunDTO, 'state' | 'interrupt_reason'> | null | undefined): boolean;
 /** The run produces no further events this turn: terminal or interrupted. */
 export declare function isRunSettled(state: AgentRunState | null | undefined): boolean;
 /** The run is executing: submitted or working. */
 export declare function isRunWorking(state: AgentRunState | null | undefined): boolean;
 export declare function isMessageTerminal(status: ChatMessageStatus | null | undefined): boolean;
-/** The turn is still open: the run is working or parked on a human (approval, input, authorization). */
+/** The turn is still open: the run is working, waiting, or parked on a human (approval, input, authorization). */
 export declare function isChatBusy(chat: ChatDTO | null | undefined): boolean;
 /** The chat is blocked on a human: nothing more arrives until someone approves, answers or authorizes. */
 export declare function isAwaitingHuman(chat: ChatDTO | null | undefined): boolean;

package/dist/utils.js

@@ -3,7 +3,7 @@
  * Provides future compatibility when API migrates from int to string status.
  */
 import { InterruptReasonToolApproval, InterruptReasonWait, InterruptResourceTurn, InterruptStatusPending } from './types.js';
-import { TaskStatusUnknown, TaskStatusReceived, TaskStatusQueued, TaskStatusDispatched, TaskStatusPreparing, TaskStatusServing, TaskStatusSettingUp, TaskStatusRunning, TaskStatusCancelling, TaskStatusUploading, TaskStatusCompleted, TaskStatusFailed, TaskStatusCancelled, TaskStatusRejected, AgentRunStateWorking, AgentRunStateSubmitted, AgentRunStateInputRequired, AgentRunStateAuthRequired, AgentRunStateCompleted, AgentRunStateFailed, AgentRunStateCanceled, AgentRunStateRejected, ChatMessageStatusReady, ChatMessageStatusFailed, ChatMessageStatusCancelled, ChatStatusBusy, ChatStatusAwaitingInput, } from './types.js';
+import { TaskStatusUnknown, TaskStatusReceived, TaskStatusQueued, TaskStatusDispatched, TaskStatusPreparing, TaskStatusServing, TaskStatusSettingUp, TaskStatusRunning, TaskStatusCancelling, TaskStatusUploading, TaskStatusCompleted, TaskStatusFailed, TaskStatusCancelled, TaskStatusRejected, AgentRunStateWorking, AgentRunStateSubmitted, AgentRunStateInputRequired, AgentRunStateAuthRequired, AgentRunStateCompleted, AgentRunStateFailed, AgentRunStateCanceled, AgentRunStateRejected, ChatMessageStatusReady, ChatMessageStatusFailed, ChatMessageStatusCancelled, ChatStatusBusy, ChatStatusAwaitingInput, ChatStatusWaiting, } from './types.js';
 /** Map string status names to TaskStatus values (for future string-based API) */
 const STATUS_STRING_MAP = {
     unknown: TaskStatusUnknown,
@@ -54,6 +54,10 @@ export function isRunTerminal(state) {
 export function isRunInterrupted(state) {
     return state === AgentRunStateInputRequired || state === AgentRunStateAuthRequired;
 }
+/** Paused on the agent's own wait call: it resumes at its deadline or on a message, so nobody has to act. */
+export function isRunWaiting(run) {
+    return !!run && isRunInterrupted(run.state) && run.interrupt_reason === InterruptReasonWait;
+}
 /** The run produces no further events this turn: terminal or interrupted. */
 export function isRunSettled(state) {
     return isRunTerminal(state) || isRunInterrupted(state);
@@ -65,19 +69,19 @@ export function isRunWorking(state) {
 export function isMessageTerminal(status) {
     return status === ChatMessageStatusReady || status === ChatMessageStatusFailed || status === ChatMessageStatusCancelled;
 }
-/** The turn is still open: the run is working or parked on a human (approval, input, authorization). */
+/** The turn is still open: the run is working, waiting, or parked on a human (approval, input, authorization). */
 export function isChatBusy(chat) {
     const run = chat?.active_run;
     if (run) {
         return isRunWorking(run.state) || isRunInterrupted(run.state);
     }
-    return chat?.status === ChatStatusBusy || chat?.status === ChatStatusAwaitingInput;
+    return chat?.status === ChatStatusBusy || chat?.status === ChatStatusAwaitingInput || chat?.status === ChatStatusWaiting;
 }
 /** The chat is blocked on a human: nothing more arrives until someone approves, answers or authorizes. */
 export function isAwaitingHuman(chat) {
     const run = chat?.active_run;
     if (run)
-        return isRunInterrupted(run.state);
+        return isRunInterrupted(run.state) && !isRunWaiting(run);
     return chat?.status === ChatStatusAwaitingInput;
 }
 /** An interrupt's meta, which arrives as an object or a JSON string. */

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@inferencesh/sdk",
-  "version": "0.32.0",
+  "version": "0.33.0",
   "description": "Official JavaScript/TypeScript SDK for inference.sh - Run AI models with a simple API",
   "type": "module",
   "main": "dist/index.js",