shortcutxl 0.3.77 → 0.3.78

package/dist/constants.d.ts

@@ -25,8 +25,6 @@ export declare function getShortcutForkTraceDir(): string | undefined;
 export declare function getShortcutTuiRenderTraceSetting(): string | undefined;
 /** Optional TUI render trace full-text flag. */
 export declare function getShortcutTuiRenderTraceFullSetting(): string | undefined;
-/** WebSocket URL for the collab signaling server. */
-export declare const COLLAB_SERVER_URL: string;
 /** uv release version pinned to a known-good Astral release. */
 export declare const UV_VERSION = "0.7.2";
 /** Python version installed via uv on macOS. */

package/dist/constants.js

@@ -58,9 +58,6 @@ export function getShortcutTuiRenderTraceFullSetting() {
 }
 // -- Dev mode --------------------------------------------------------------
 // Dev mode is now a settings.json flag (settingsManager.getDevMode()), not an env var.
-// -- Collaboration ---------------------------------------------------------
-/** WebSocket URL for the collab signaling server. */
-export const COLLAB_SERVER_URL = process.env.COLLAB_SERVER_URL ?? 'ws://localhost:9400';
 // -- uv (Python toolchain for macOS) --------------------------------------
 /** uv release version pinned to a known-good Astral release. */
 export const UV_VERSION = '0.7.2';

package/dist/contracts/model-call.d.ts

@@ -4,7 +4,6 @@ export declare const ShortcutXLModelCallTransportKind: {
     readonly GoogleVertex: "google_vertex";
     readonly OpenAICompletions: "openai_completions";
     readonly OpenAIResponses: "openai_responses";
-    readonly ShortcutProxy: "shortcut_proxy";
 };
 export type ShortcutXLModelCallTransportKind = (typeof ShortcutXLModelCallTransportKind)[keyof typeof ShortcutXLModelCallTransportKind];
 /**

package/dist/contracts/model-call.js

@@ -3,8 +3,7 @@ export const ShortcutXLModelCallTransportKind = {
     AnthropicMessages: 'anthropic_messages',
     GoogleVertex: 'google_vertex',
     OpenAICompletions: 'openai_completions',
-    OpenAIResponses: 'openai_responses',
-    ShortcutProxy: 'shortcut_proxy'
+    OpenAIResponses: 'openai_responses'
 };
 /**
  * Request types emitted by private-gateway usage logging.

package/dist/contracts/wire.d.ts

@@ -15,6 +15,7 @@ export declare const WorkerMessageKind: {
     readonly ModelStreamRequest: "modelStreamRequest";
     readonly AgentCommandAck: "runtimeCommandAck";
     readonly AgentCommand: "runtimeCommand";
+    readonly AgentCommandReceived: "runtimeCommandReceived";
     readonly AgentEvent: "runtimeEvent";
 };
 export type WorkerMessageKind = (typeof WorkerMessageKind)[keyof typeof WorkerMessageKind];
@@ -30,6 +31,12 @@ export interface AgentCommandAckMessage {
     ok: boolean;
     error?: string;
 }
+export interface AgentCommandReceivedMessage {
+    kind: typeof WorkerMessageKind.AgentCommandReceived;
+    commandId: string;
+    receivedAt: number;
+    userMessageId: string;
+}
 export interface AgentEventMessage {
     kind: typeof WorkerMessageKind.AgentEvent;
     event: AgentEvent;
@@ -90,7 +97,7 @@ export interface ModelStreamCancelMessage {
     kind: typeof WorkerMessageKind.ModelStreamCancel;
     requestId: string;
 }
-export type WorkerMessage = AgentCommandAckMessage | AgentCommandMessage | AgentEventMessage | HostToolRequestMessage | HostToolResultMessage | ModelStreamCancelMessage | ModelStreamEventMessage | ModelStreamRequestMessage;
+export type WorkerMessage = AgentCommandAckMessage | AgentCommandMessage | AgentCommandReceivedMessage | AgentEventMessage | HostToolRequestMessage | HostToolResultMessage | ModelStreamCancelMessage | ModelStreamEventMessage | ModelStreamRequestMessage;
 /** Envelope-shape guard; tool-payload validation goes through the registry. */
 export declare function isWorkerMessage(value: unknown, registry: ToolRegistry): value is WorkerMessage;
 export declare function isShortcutXLAiModelInvokeRequest(value: unknown): value is ShortcutXLAiModelInvokeRequest;

package/dist/contracts/wire.js

@@ -12,6 +12,7 @@ export const WorkerMessageKind = {
     ModelStreamRequest: 'modelStreamRequest',
     AgentCommandAck: 'runtimeCommandAck',
     AgentCommand: 'runtimeCommand',
+    AgentCommandReceived: 'runtimeCommandReceived',
     AgentEvent: 'runtimeEvent'
 };
 /** Envelope-shape guard; tool-payload validation goes through the registry. */
@@ -30,6 +31,13 @@ export function isWorkerMessage(value, registry) {
                 typeof message.ok === 'boolean' &&
                 (message.error == null || typeof message.error === 'string'));
         }
+        case WorkerMessageKind.AgentCommandReceived: {
+            const message = candidate;
+            return (typeof message.commandId === 'string' &&
+                typeof message.receivedAt === 'number' &&
+                Number.isFinite(message.receivedAt) &&
+                typeof message.userMessageId === 'string');
+        }
         case WorkerMessageKind.AgentCommand:
             return (typeof candidate.commandId === 'string' &&
                 isAgentCommand(candidate.command) &&

package/dist/embedded-agent/ai-invoke.d.ts

@@ -1,5 +1,5 @@
 import type { Api, AssistantMessage, Model } from '../ai/types.js';
-import { type ShortcutXLModelInvokePayload, type ShortcutXLModelInvokeResponse } from '../contracts/model-call.js';
+import { type ShortcutXLModelInvokeResponse } from '../contracts/model-call.js';
 import type { ShortcutXLAiInvokeOptions, ShortcutXLAiModelInvokeRequest } from '../contracts/wire.js';
 import type { InvokeLlmParams } from './host-tools/invoke-llm/contract.js';
 export declare function createShortcutXLAiModelInvokeRequestFromParams(options: {
@@ -11,6 +11,5 @@ export declare function createShortcutXLAiModelInvokeRequestFromParams(options:
     readonly speed?: ShortcutXLAiInvokeOptions['speed'];
     readonly userMessageId?: string;
 }): ShortcutXLAiModelInvokeRequest;
-export declare function shortcutXLAiModelInvokeRequestToPayload(request: ShortcutXLAiModelInvokeRequest): ShortcutXLModelInvokePayload;
 export declare function assistantMessageToShortcutXLInvokeResponse(message: AssistantMessage | undefined): ShortcutXLModelInvokeResponse;
 //# sourceMappingURL=ai-invoke.d.ts.map

package/dist/embedded-agent/ai-invoke.js

@@ -1,5 +1,4 @@
 import { ShortcutXLProviderUsageStatus } from '../contracts/model-call.js';
-import { aiContentToShortcutProxyContent } from './ai-content.js';
 function toStructuredCloneSafe(value, seen = new WeakSet()) {
     if (value === null ||
         typeof value === 'string' ||
@@ -83,75 +82,6 @@ export function createShortcutXLAiModelInvokeRequestFromParams(options) {
         requestId: options.requestId
     });
 }
-function toolCallToShortcutProxyToolCall(toolCall) {
-    return {
-        arguments: JSON.stringify(toolCall.arguments ?? {}),
-        id: toolCall.id,
-        name: toolCall.name
-    };
-}
-export function shortcutXLAiModelInvokeRequestToPayload(request) {
-    const messages = [
-        ...(request.context.systemPrompt
-            ? [{ content: request.context.systemPrompt, role: 'system' }]
-            : [])
-    ];
-    for (const message of request.context.messages) {
-        if (message.role === 'user') {
-            messages.push({
-                content: aiContentToShortcutProxyContent(message.content, request.model.id),
-                role: 'human'
-            });
-            continue;
-        }
-        if (message.role === 'assistant') {
-            const content = typeof message.content === 'string'
-                ? message.content
-                : message.content
-                    .filter((block) => block.type === 'text')
-                    .map((block) => block.text)
-                    .join('');
-            const toolCalls = typeof message.content === 'string'
-                ? undefined
-                : message.content
-                    .filter((block) => block.type === 'toolCall')
-                    .map(toolCallToShortcutProxyToolCall);
-            messages.push({
-                content,
-                role: 'ai',
-                ...(toolCalls && toolCalls.length > 0 ? { toolCalls } : {})
-            });
-            continue;
-        }
-        if (message.role === 'toolResult') {
-            messages.push({
-                content: aiContentToShortcutProxyContent(message.content, request.model.id),
-                role: 'tool',
-                toolCallId: message.toolCallId
-            });
-        }
-    }
-    return {
-        effortLevel: request.options?.reasoning === 'low' ||
-            request.options?.reasoning === 'medium' ||
-            request.options?.reasoning === 'high'
-            ? request.options.reasoning
-            : undefined,
-        maxTokens: request.options?.maxTokens,
-        messages,
-        model: request.model.id,
-        speed: request.options?.speed,
-        thinkingType: request.options?.thinkingType,
-        toolChoice: typeof request.options?.toolChoice === 'object'
-            ? { name: request.options.toolChoice.name, type: 'tool' }
-            : undefined,
-        tools: request.context.tools?.map((tool) => ({
-            description: tool.description,
-            name: tool.name,
-            parameters: tool.parameters
-        }))
-    };
-}
 export function assistantMessageToShortcutXLInvokeResponse(message) {
     if (!message) {
         return { usageStatus: ShortcutXLProviderUsageStatus.Unavailable };

package/dist/embedded-agent/anthropic-messages-transport.d.ts

@@ -2,8 +2,7 @@ import type { AssistantMessageEvent, Usage } from '../ai/types.js';
 import { type ShortcutXLModelInvokeResponse, type ShortcutXLModelUsage } from '../contracts/model-call.js';
 import { type ShortcutXLModelStreamEvent } from '../contracts/model-stream.js';
 import type { ShortcutXLAiModelInvokeRequest, ShortcutXLAiModelStreamRequest } from '../contracts/wire.js';
-import type { ShortcutXLModelUsageStatusLogger } from './model-call-orchestrator.js';
-import { type ProviderGatewayOptions } from './provider-native-request.js';
+import { type ProviderGatewayOptions, type ShortcutXLModelUsageStatusLogger } from './provider-native-request.js';
 import type { RunShortcutXLModelStreamRequestOptions } from './stream.js';
 export interface AnthropicMessagesTransportOptions extends ProviderGatewayOptions {
     readonly usageStatusLogger?: ShortcutXLModelUsageStatusLogger;

package/dist/embedded-agent/google-vertex-transport.d.ts

@@ -1,7 +1,6 @@
 import { type ShortcutXLModelInvokeResponse } from '../contracts/model-call.js';
 import type { ShortcutXLAiModelInvokeRequest, ShortcutXLAiModelStreamRequest } from '../contracts/wire.js';
-import type { ShortcutXLModelUsageStatusLogger } from './model-call-orchestrator.js';
-import { type ProviderGatewayOptions } from './provider-native-request.js';
+import { type ProviderGatewayOptions, type ShortcutXLModelUsageStatusLogger } from './provider-native-request.js';
 import type { RunShortcutXLModelStreamRequestOptions } from './stream.js';
 export interface GoogleVertexTransportOptions extends ProviderGatewayOptions {
     readonly usageStatusLogger?: ShortcutXLModelUsageStatusLogger;

package/dist/embedded-agent/host-tools/invoke-llm/contract.d.ts

@@ -1,6 +1,6 @@
 /**
  * Bridge contract for `invoke_llm` — non-streaming LLM call dispatched to the
- * client-side `/api/llm/invoke` handler. Not a user-callable tool; the runtime
+ * client-side provider-native model worker. Not a user-callable tool; the runtime
  * uses this for internal one-shot LLM jobs (e.g. follow-up suggestions).
  */
 import type { AssistantMessage, ToolResultMessage, UserMessage } from '../../../ai/types.js';

package/dist/embedded-agent/host-tools/invoke-llm/contract.js

@@ -1,6 +1,6 @@
 /**
  * Bridge contract for `invoke_llm` — non-streaming LLM call dispatched to the
- * client-side `/api/llm/invoke` handler. Not a user-callable tool; the runtime
+ * client-side provider-native model worker. Not a user-callable tool; the runtime
  * uses this for internal one-shot LLM jobs (e.g. follow-up suggestions).
  */
 import { Type } from '@sinclair/typebox';

package/dist/embedded-agent/host-tools/shared/index.d.ts

@@ -1,5 +1,5 @@
 export { base64ToBytes, bytesToBase64, bytesToUtf8 } from './base64.js';
-export { LlmStreamError, readLlmStreamText, type LlmStreamRequest } from './llm-stream.js';
+export { LlmStreamError } from './llm-stream-error.js';
 export { MultimodalContentBlockFormat, buildAnthropicBlock, buildContentBlock, buildGeminiBlock } from './multimodal.js';
 export type { MultimodalBlock, MultimodalBlockInput, MultimodalBlockResult } from './multimodal.js';
 export { baseName, dirName, getFileDataDir, lowerExt, stem } from './sandbox-paths.js';

package/dist/embedded-agent/host-tools/shared/index.js

@@ -1,5 +1,5 @@
 export { base64ToBytes, bytesToBase64, bytesToUtf8 } from './base64.js';
-export { LlmStreamError, readLlmStreamText } from './llm-stream.js';
+export { LlmStreamError } from './llm-stream-error.js';
 export { MultimodalContentBlockFormat, buildAnthropicBlock, buildContentBlock, buildGeminiBlock } from './multimodal.js';
 export { baseName, dirName, getFileDataDir, lowerExt, stem } from './sandbox-paths.js';
 //# sourceMappingURL=index.js.map

package/dist/embedded-agent/host-tools/shared/llm-stream-error.d.ts

@@ -0,0 +1,7 @@
+import type { ShortcutXLModelStreamErrorDetails } from '../../../contracts/model-stream.js';
+/** Structured provider-native model-stream failure surfaced to helper tools. */
+export declare class LlmStreamError extends Error {
+    readonly details?: ShortcutXLModelStreamErrorDetails | undefined;
+    constructor(message: string, details?: ShortcutXLModelStreamErrorDetails | undefined);
+}
+//# sourceMappingURL=llm-stream-error.d.ts.map

package/dist/embedded-agent/host-tools/shared/llm-stream-error.js

@@ -0,0 +1,10 @@
+/** Structured provider-native model-stream failure surfaced to helper tools. */
+export class LlmStreamError extends Error {
+    details;
+    constructor(message, details) {
+        super(message);
+        this.details = details;
+        this.name = 'LlmStreamError';
+    }
+}
+//# sourceMappingURL=llm-stream-error.js.map

package/dist/embedded-agent/index.d.ts

@@ -17,20 +17,19 @@ import type { EmbeddedErrorRecoverySettings } from './settings.js';
 export { createAgentSnapshotFromAgentMessages } from '../core/agent-snapshot-builder.js';
 export { assertSessionInvariants, buildSessionContextMessageEntryIds, canonicalizeUserMessageEntries, parseSessionEntriesStrict, repairOrphanToolUses, serializeFileEntries } from '../core/session-schema.js';
 export type { CompactionEntry, FileEntry, LlmCallEntry, ModelChangeEntry, ParseSessionStrictResult, RepairOrphanToolUsesOptions, RepairOrphanToolUsesResult, SessionEntry, SessionHeader, SessionInvariantViolation, SessionMessageEntry, SessionParseError, SpeedChangeEntry, SystemPromptEntry, ThinkingLevelChangeEntry, ToolDefinitionsEntry, TurnEndEntry, TurnStartEntry } from '../core/session-schema.js';
-export { assistantMessageToShortcutXLInvokeResponse, createShortcutXLAiModelInvokeRequestFromParams, shortcutXLAiModelInvokeRequestToPayload } from './ai-invoke.js';
+export { assistantMessageToShortcutXLInvokeResponse, createShortcutXLAiModelInvokeRequestFromParams } from './ai-invoke.js';
 export { createShortcutXLAgentSettingsStore, createShortcutXLDefaultSessionContext } from './bootstrap.js';
 export { buildHostEmbeddedAgentOptions, type BuildHostEmbeddedAgentOptionsInput, type BuildHostEmbeddedAgentOptionsResult } from './host-runtime-options.js';
 export { detectEmbeddedReadOnlyViolations } from './host-tools/execute-code/index.js';
 export { getHostToolTimeoutMs } from './host-tools/timeouts.js';
-export { createShortcutProxyModelStreamTransport, createShortcutXLModelCallOrchestrator, runShortcutXLModelCallStreamRequest, type CreateShortcutXLModelCallOrchestratorOptions, type ShortcutXLModelCallOrchestrator, type ShortcutXLModelStreamTransport, type ShortcutXLModelUsageStatusLogger } from './model-call-orchestrator.js';
 export { createShortcutXLAgentModel, createShortcutXLAgentModelRegistry, ShortcutXLAgentModelId, ShortcutXLAgentProviderId, type CreateShortcutXLAgentModelOptions, type CreateShortcutXLAgentModelRegistryOptions } from './model-registry.js';
 export { buildShortcutXLAgentSystemPrompt, createShortcutXLAgentConfiguration, ShortcutXLPromptMode, type ShortcutXLPromptPlatform } from './prompt/build.js';
 export { createRunStatsAccumulator, type CreateRunStatsAccumulatorOptions, type RunStatsAccumulator, type RunStatsEntry, type RunStatsModelUsage, type RunStatsSnapshot, type RunStatsTokenUsageByModel } from './run-stats.js';
 export { EmbeddedSessionEntryBuilder, type AgentToolDefinitionSnapshot } from './session-entry-builder.js';
 export { createShortcutXLAgentSessionStore, type CreateShortcutXLAgentSessionStoreOptions, type ShortcutXLAgentSessionArtifacts, type ShortcutXLAgentSessionStore, type ShortcutXLSessionWriter } from './session-store.js';
 export { pushShortcutXLModelStreamEvent } from './stream-events.js';
-export { buildShortcutRuntimeRequestBody, createShortcutXLModelStream, createShortcutXLModelStreamRequest, runShortcutXLModelStreamRequest, type CreateShortcutXLModelStreamRequestOptions, type RunShortcutXLModelStreamRequestOptions, type ShortcutRuntimeRequestBody, type ShortcutXLModelStreamHostOptions } from './stream.js';
-export { createEmbeddedAgentPortClient, ShortcutXLModelStreamTraceKind, type CreateEmbeddedAgentPortClientOptions, type EmbeddedAgentPortClient, type ShortcutXLModelStreamBridgeMetrics, type ShortcutXLModelStreamRequestHandler, type ShortcutXLModelStreamTraceEntry, type ShortcutXLWorkerPort } from './worker-bridge/host-client.js';
+export { buildShortcutRuntimeRequestBody, createShortcutXLModelStreamRequest, type CreateShortcutXLModelStreamRequestOptions, type RunShortcutXLModelStreamRequestOptions, type ShortcutRuntimeRequestBody } from './stream.js';
+export { createEmbeddedAgentPortClient, ShortcutXLModelStreamTraceKind, type CreateEmbeddedAgentPortClientOptions, type EmbeddedAgentPortClient, type ShortcutXLInitialSubmitReceipt, type ShortcutXLModelStreamBridgeMetrics, type ShortcutXLModelStreamRequestHandler, type ShortcutXLModelStreamTraceEntry, type ShortcutXLWorkerPort } from './worker-bridge/host-client.js';
 export { AgentExecuteCodeModeValue as AgentExecuteCodeMode, ExecuteCodeSurfaceValue as ExecuteCodeSurface };
 /** Host-provided knobs the runtime reads at boot and per turn. */
 export interface EmbeddedAgentSettingsStore {

package/dist/embedded-agent/index.js

@@ -4,19 +4,18 @@ export { createAgentSnapshotFromAgentMessages } from '../core/agent-snapshot-bui
 export { 
 // Pure session helpers shared by the embedded host for upload and hydration.
 assertSessionInvariants, buildSessionContextMessageEntryIds, canonicalizeUserMessageEntries, parseSessionEntriesStrict, repairOrphanToolUses, serializeFileEntries } from '../core/session-schema.js';
-export { assistantMessageToShortcutXLInvokeResponse, createShortcutXLAiModelInvokeRequestFromParams, shortcutXLAiModelInvokeRequestToPayload } from './ai-invoke.js';
+export { assistantMessageToShortcutXLInvokeResponse, createShortcutXLAiModelInvokeRequestFromParams } from './ai-invoke.js';
 export { createShortcutXLAgentSettingsStore, createShortcutXLDefaultSessionContext } from './bootstrap.js';
 export { buildHostEmbeddedAgentOptions } from './host-runtime-options.js';
 export { detectEmbeddedReadOnlyViolations } from './host-tools/execute-code/index.js';
 export { getHostToolTimeoutMs } from './host-tools/timeouts.js';
-export { createShortcutProxyModelStreamTransport, createShortcutXLModelCallOrchestrator, runShortcutXLModelCallStreamRequest } from './model-call-orchestrator.js';
 export { createShortcutXLAgentModel, createShortcutXLAgentModelRegistry, ShortcutXLAgentModelId, ShortcutXLAgentProviderId } from './model-registry.js';
 export { buildShortcutXLAgentSystemPrompt, createShortcutXLAgentConfiguration, ShortcutXLPromptMode } from './prompt/build.js';
 export { createRunStatsAccumulator } from './run-stats.js';
 export { EmbeddedSessionEntryBuilder } from './session-entry-builder.js';
 export { createShortcutXLAgentSessionStore } from './session-store.js';
 export { pushShortcutXLModelStreamEvent } from './stream-events.js';
-export { buildShortcutRuntimeRequestBody, createShortcutXLModelStream, createShortcutXLModelStreamRequest, runShortcutXLModelStreamRequest } from './stream.js';
+export { buildShortcutRuntimeRequestBody, createShortcutXLModelStreamRequest } from './stream.js';
 export { createEmbeddedAgentPortClient, ShortcutXLModelStreamTraceKind } from './worker-bridge/host-client.js';
 export { AgentExecuteCodeModeValue as AgentExecuteCodeMode, ExecuteCodeSurfaceValue as ExecuteCodeSurface };
 export { EmbeddedAgentCreditGate } from './credit-gate.js';

package/dist/embedded-agent/openai-completions-transport.d.ts

@@ -1,7 +1,6 @@
 import { type ShortcutXLModelInvokeResponse } from '../contracts/model-call.js';
 import type { ShortcutXLAiModelInvokeRequest, ShortcutXLAiModelStreamRequest } from '../contracts/wire.js';
-import type { ShortcutXLModelUsageStatusLogger } from './model-call-orchestrator.js';
-import { type ProviderGatewayOptions } from './provider-native-request.js';
+import { type ProviderGatewayOptions, type ShortcutXLModelUsageStatusLogger } from './provider-native-request.js';
 import type { RunShortcutXLModelStreamRequestOptions } from './stream.js';
 export interface OpenAICompletionsTransportOptions extends ProviderGatewayOptions {
     readonly qwenCompatibility?: {

package/dist/embedded-agent/openai-responses-transport.d.ts

@@ -1,7 +1,6 @@
 import { type ShortcutXLModelInvokeResponse } from '../contracts/model-call.js';
 import type { ShortcutXLAiModelInvokeRequest, ShortcutXLAiModelStreamRequest } from '../contracts/wire.js';
-import type { ShortcutXLModelUsageStatusLogger } from './model-call-orchestrator.js';
-import { type ProviderGatewayOptions } from './provider-native-request.js';
+import { type ProviderGatewayOptions, type ShortcutXLModelUsageStatusLogger } from './provider-native-request.js';
 import type { RunShortcutXLModelStreamRequestOptions } from './stream.js';
 export interface OpenAIResponsesTransportOptions extends ProviderGatewayOptions {
     readonly usageStatusLogger?: ShortcutXLModelUsageStatusLogger;

package/dist/embedded-agent/prompt/sections.d.ts

@@ -6,7 +6,7 @@
  * Sandbox layout reference. Subagents that touch the sandbox (general,
  * document_reader) append this to their system prompt.
  */
-export declare const SANDBOX_FILESYSTEM = "\n======================\n## SANDBOX FILESYSTEM\n======================\n\nYour sandbox has these top-level directories \u2014 they are separate mounts:\n\n- `/workspace/` \u2014 Active workspace. Primary working directory. Uploaded files and generated output belong here for the selected workspace scope.\n- `/skills/` \u2014 Skills directory (completely separate from /workspace):\n  - `/skills/default/` \u2014 Built-in skills baked into the Modal image (read-only).\n  - `/skills/user/` \u2014 Personal skills mounted read-write.\n  - `/skills/team/your-team-id/` \u2014 Team-shared skills mounted according to your team permissions.\n- `/memories/` \u2014 Persistent memory (MEMORY.md, daily notes, traces)\n- `/tmp/` \u2014 Ephemeral local SSD scratch space. Use only when an operation needs normal local-disk behavior; cleared between sessions.\n\n**Storage notes:** `/skills/default` is baked into the Modal image. `/workspace`, `/skills/user`, `/skills/team`, and `/memories` are cloud-backed mounts. Cloud-backed mounts persist, but do not support normal local-disk behavior such as append operations (`>>`, `tee -a`) or random-write/seek-heavy file creation. `/tmp` is local SSD: it supports normal local file operations, but is ephemeral and cleared between sessions. Do not leave raw, intermediate, or final artifacts only in `/tmp`; organize durable work under `/workspace`.\n\nNEVER create skills under `workspace`. Skills MUST go in `skills`.\nNEVER save user files under `skills`. Generated files belong in `workspace`.\n\nUsers can browse, edit, preview, and download files via the **Workspace Files** panel in the UI. The panel shows the active workspace scope selected for the conversation.\n- Save outputs to `/workspace/`\n- When you have generated files relevant to the user's requests, tell users the relative path where files are saved and mention they can find files in the Workspace Files panel\n";
+export declare const SANDBOX_FILESYSTEM = "\n======================\n## SANDBOX FILESYSTEM\n======================\n\nYour sandbox has these top-level directories \u2014 they are separate mounts:\n\n- `/workspace/` \u2014 Active workspace. Primary working directory. Uploaded files and generated output belong here for the selected workspace scope.\n- `/skills/` \u2014 Skills directory (completely separate from /workspace):\n  - `/skills/default/` \u2014 Built-in skills baked into the Modal image (read-only).\n  - `/skills/user/` \u2014 Personal skills mounted read-write.\n  - `/skills/team/your-team-id/` \u2014 Team-shared skills mounted according to your team permissions.\n- `/memories/` \u2014 Persistent memory (MEMORY.md, daily notes, traces)\n- `/tmp/` \u2014 Ephemeral local SSD scratch space. Use only when an operation needs normal local-disk behavior; cleared between sessions.\n\n**Storage notes:** `/skills/default` is baked into the Modal image. `/workspace`, `/skills/user`, `/skills/team`, and `/memories` are cloud-backed mounts. Cloud-backed mounts persist, but do not support normal local-disk behavior such as append operations (`>>`, `tee -a`) or random-write/seek-heavy file creation. `/tmp` is local SSD: it supports normal local file operations, but is ephemeral and cleared between sessions. Do not leave raw, intermediate, or final artifacts only in `/tmp`; organize durable work under `/workspace`.\n\nDo not read files that are concurrently being created in the same turn. Wait for creation to finish.\n\nNEVER create skills under `workspace`. Skills MUST go in `skills`.\nNEVER save user files under `skills`. Generated files belong in `workspace`.\n\nUsers can browse, edit, preview, and download files via the **Workspace Files** panel in the UI. The panel shows the active workspace scope selected for the conversation.\n- Save outputs to `/workspace/`\n- When you have generated files relevant to the user's requests, tell users the relative path where files are saved and mention they can find files in the Workspace Files panel\n";
 /**
  * File-layout reference for the document_reader subagent. Mirrors
  * `backend/agent/constants/file_reading.py:FILE_STRUCTURE` verbatim, with

package/dist/embedded-agent/prompt/sections.js

@@ -24,6 +24,8 @@ Your sandbox has these top-level directories — they are separate mounts:
 
 **Storage notes:** \`/skills/default\` is baked into the Modal image. \`/workspace\`, \`/skills/user\`, \`/skills/team\`, and \`/memories\` are cloud-backed mounts. Cloud-backed mounts persist, but do not support normal local-disk behavior such as append operations (\`>>\`, \`tee -a\`) or random-write/seek-heavy file creation. \`/tmp\` is local SSD: it supports normal local file operations, but is ephemeral and cleared between sessions. Do not leave raw, intermediate, or final artifacts only in \`/tmp\`; organize durable work under \`/workspace\`.
 
+Do not read files that are concurrently being created in the same turn. Wait for creation to finish.
+
 NEVER create skills under \`workspace\`. Skills MUST go in \`skills\`.
 NEVER save user files under \`skills\`. Generated files belong in \`workspace\`.
 

package/dist/embedded-agent/provider-native-request.d.ts

@@ -1,7 +1,12 @@
 import type { Model, SimpleStreamOptions } from '../ai/types.js';
+import type { ShortcutXLProviderUsageEvent, ShortcutXLUsageLogResult } from '../contracts/model-call.js';
 import type { ShortcutXLAiModelInvokeRequest, ShortcutXLAiModelStreamRequest } from '../contracts/wire.js';
 import { type GatewayTransport } from './gateway-base-url.js';
 type ProviderRequest = ShortcutXLAiModelStreamRequest | ShortcutXLAiModelInvokeRequest;
+export interface ShortcutXLModelUsageStatusLogger {
+    recordProviderUsage(event: ShortcutXLProviderUsageEvent): void | Promise<void>;
+    recordUsageReportResult(result: ShortcutXLUsageLogResult): void | Promise<void>;
+}
 export interface ProviderGatewayAuth {
     readonly headerName: 'Authorization' | 'x-api-key';
     readonly token: string;

package/dist/embedded-agent/provider-native-request.js

@@ -38,8 +38,9 @@ export function createGatewayProviderOptions(request, transport, options, defaul
         ...authHeaders(options.auth, transport)
     }, fastMode);
     const thinkingType = request.options && 'thinkingType' in request.options ? request.options.thinkingType : undefined;
-    const reasoning = request.options?.reasoning ??
-        (thinkingType === 'on' ? 'high' : thinkingType === 'off' ? 'off' : undefined);
+    const reasoning = thinkingType === 'off'
+        ? 'off'
+        : (request.options?.reasoning ?? (thinkingType === 'on' ? 'high' : undefined));
     return withProviderFastModePayloadOverride({
         apiKey: options.auth.token,
         cacheRetention: request.options?.cacheRetention,

package/dist/embedded-agent/stream.d.ts

@@ -1,4 +1,5 @@
-import { type Api, type Context, type Model, type ReasoningLevel, type SimpleStreamOptions } from '../ai/types.js';
+/** Provider-neutral request tracing helpers shared by embedded transports. */
+import type { Api, Context, Model, ReasoningLevel, SimpleStreamOptions } from '../ai/types.js';
 import { type ShortcutXLModelStreamEvent, type ShortcutXLModelStreamFetchTrace, type ShortcutXLModelStreamMessage, type ShortcutXLModelStreamRequest } from '../contracts/model-stream.js';
 type ShortcutRuntimeMessage = ShortcutXLModelStreamMessage;
 export declare function toShortcutEffortLevel(effort: ReasoningLevel | undefined): ShortcutXLModelStreamRequest['effortLevel'];
@@ -18,12 +19,6 @@ export interface ShortcutRuntimeRequestBody {
         }[];
     }[];
     model: string;
-    user_message_id?: string;
-    /**
-     * Selected speed mode. The proxy's `LLMRequest.speed` field validates
-     * `'fast' | 'standard'`. Server-side gating ensures fast mode only
-     * applies to supported models.
-     */
     speed?: 'fast' | 'standard';
     thinking_type?: 'adaptive' | 'off';
     tools?: {
@@ -31,97 +26,22 @@ export interface ShortcutRuntimeRequestBody {
         name: string;
         parameters: Record<string, unknown>;
     }[];
+    user_message_id?: string;
 }
-interface ShortcutXLModelStreamWorkspaceContext {
-    readonly workspaceId?: string;
-    readonly workspaceType?: 'file' | 'project' | 'global';
-}
-/**
- * Owned AI `thinkingLevel` ↔ proxy's `effort_level`. Exported so the
- * run-stats accumulator can stamp the same effort label on the run-stats
- * entry that the proxy actually used. Keeping a single source of truth
- * prevents drift between what the LLM saw and what the dropdown shows.
- */
-/** Host plumbing the embedded runtime needs to reach the stream endpoint. */
-export interface ShortcutXLModelStreamHostOptions {
-    readonly baseUrl?: string;
-    readonly clientPlatform: string;
-    readonly fetchImpl?: typeof fetch;
-    /** Override only the legacy `/api/llm/stream` fetch URL. Defaults to `baseUrl`. */
-    readonly streamBaseUrl?: string;
-    readonly getAccessToken?: () => Promise<string | undefined>;
-    /**
-     * Refreshes host-owned credentials after a 401 from the stream proxy. Workers
-     * cannot refresh auth directly, so they call back into the host through this
-     * hook and retry the stream request once.
-     */
-    readonly refreshAccessToken?: () => Promise<string | undefined>;
-    /**
-     * Resolves the active conversation thread id at request time. Sent as the
-     * `X-Session-Id` header — the backend `/api/llm/stream` proxy uses it for
-     * `langfuse_session_id`, `thread_id` on token-event webhooks, and
-     * trace grouping. Without this, every call is bucketed under the
-     * synthetic fallback `llm-proxy-{user_id}` and per-thread reconciliation
-     * breaks.
-     */
-    readonly getSessionId?: () => string | undefined;
-    readonly getWorkspaceContext?: () => ShortcutXLModelStreamWorkspaceContext | undefined;
-    /**
-     * Observability hook fired after the runtime has normalized context/options
-     * into the exact JSON payload shape sent to `/api/llm/stream`.
-     * Credentials and transport headers are intentionally excluded.
-     */
-    readonly onRequest?: (request: ShortcutXLModelStreamRequest, body: ShortcutRuntimeRequestBody) => void;
-    /**
-     * Diagnostic hook fired when a stream fetch fails. Distinguishes failure
-     * stages that aren't visible in the generic `recovery_started` telemetry:
-     *   - `before_response`: `fetch()` rejected before any `Response` (DNS,
-     *     CORS preflight, network unreachable, captive portal). The
-     *     opaque "Failed to fetch" / "Load failed" case.
-     *   - `non_ok_response`: backend returned a response with status outside 2xx.
-     *   - `response_body`: the response started but the SSE read failed or
-     *     idle-timed out mid-stream.
-     * Headers/body intentionally excluded — only diagnostic metadata is sent.
-     */
+/** Provider-native transports use these shared stream callbacks. */
+export interface RunShortcutXLModelStreamRequestOptions {
+    onEvent(event: ShortcutXLModelStreamEvent): void;
     readonly onFetchTrace?: (event: ShortcutXLModelStreamFetchTrace) => void;
-    /** Retry budget for retryable stream transport failures. Defaults to 2 retries. */
-    readonly maxStreamRetries?: number;
-    readonly streamRetryBaseDelayMs?: number;
-    readonly streamRetryMaxDelayMs?: number;
-    /**
-     * Ceiling for server-supplied `Retry-After` delays. Distinct from
-     * `streamRetryMaxDelayMs` (which only bounds client-computed backoff).
-     */
-    readonly streamRetryServerMaxDelayMs?: number;
+    readonly signal?: AbortSignal;
 }
-/** Inputs needed to assemble the wire-shape stream request. */
 export interface CreateShortcutXLModelStreamRequestOptions {
     readonly context: Context;
     readonly model: Model<Api>;
     readonly options?: SimpleStreamOptions;
     readonly requestId: string;
 }
-/** Per-call options layered on top of host options for `runShortcutXL...`. */
-export interface RunShortcutXLModelStreamRequestOptions extends ShortcutXLModelStreamHostOptions {
-    readonly apiKey?: string;
-    onEvent(event: ShortcutXLModelStreamEvent): void;
-    readonly signal?: AbortSignal;
-    readonly streamIdleTimeoutMs?: number;
-}
-/** Convert owned AI `Context` + model into the ShortcutXL stream request shape. */
+/** Convert owned AI context into the stable request-trace shape. */
 export declare function createShortcutXLModelStreamRequest(options: CreateShortcutXLModelStreamRequestOptions): ShortcutXLModelStreamRequest;
 export declare function buildShortcutRuntimeRequestBody(request: ShortcutXLModelStreamRequest): ShortcutRuntimeRequestBody;
-/**
- * Drive one stream request to completion. Emits Start/Delta/.../Done|Error
- * events through `onEvent`. Always emits a terminal Error event on abort or
- * failure so consumers never need to time out.
- */
-export declare function runShortcutXLModelStreamRequest(request: ShortcutXLModelStreamRequest, options: RunShortcutXLModelStreamRequestOptions): Promise<void>;
-/**
- * Build a owned AI-compatible `streamFn` bound to the given host. The returned
- * function fires off one request per call and pushes events into a fresh
- * assistant-message event stream.
- */
-export declare function createShortcutXLModelStream(hostOptions: ShortcutXLModelStreamHostOptions): (model: Model<Api>, context: Context, options?: SimpleStreamOptions) => import("../ai/event-stream.js").AssistantMessageEventStream;
 export {};
 //# sourceMappingURL=stream.d.ts.map