shortcutxl 0.3.76 → 0.3.78

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/model-selection.d.ts

@@ -1,9 +1,3 @@
-/**
- * Model + thinking-level resolution at session boot. Model order: explicit
- * boot model > saved session model (with valid api key) > first available.
- * Thinking order: explicit override > saved session thinking > default.
- * Reasoning is forced off for non-reasoning models.
- */
 import type { AgentModelRegistry } from '../contracts/agent-model.js';
 import type { AgentSessionContext } from '../contracts/agent-session-store.js';
 import type { ThinkingLevel } from '../core/core-types.js';

package/dist/embedded-agent/model-selection.js

@@ -1,4 +1,11 @@
-import { DEFAULT_THINKING_LEVEL, isValidThinkingLevel } from '../core/defaults.js';
+/**
+ * Model + thinking-level resolution at session boot. Model order: explicit
+ * boot model > saved session model (with valid api key) > first available.
+ * Thinking order: explicit override > saved session thinking > default.
+ * Reasoning is forced off for non-reasoning models.
+ */
+import { resolveModelReasoningLevel } from '../ai/models.js';
+import { isValidThinkingLevel } from '../core/defaults.js';
 function resolveSavedThinkingLevel(sessionContext, override) {
     if (override !== undefined) {
         return override;
@@ -7,7 +14,7 @@ function resolveSavedThinkingLevel(sessionContext, override) {
     if (isValidThinkingLevel(savedThinkingLevel)) {
         return savedThinkingLevel;
     }
-    return DEFAULT_THINKING_LEVEL;
+    return undefined;
 }
 async function resolveSavedModel(sessionContext, modelRegistry) {
     if (sessionContext.model == null) {
@@ -38,11 +45,11 @@ export async function selectInitialRuntimeModel(options) {
         const availableModels = await options.modelRegistry.getAvailable();
         model = availableModels[0];
     }
-    // Force reasoning off when the chosen model can't reason, so callers don't
-    // accidentally request thinking on a non-reasoning model.
-    if (!model || !model.reasoning) {
-        return { model, thinkingLevel: 'off' };
-    }
-    return { model, thinkingLevel };
+    return {
+        model,
+        thinkingLevel: model
+            ? resolveModelReasoningLevel(model, thinkingLevel)
+            : 'off'
+    };
 }
 //# sourceMappingURL=model-selection.js.map

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

@@ -1,17 +1,8 @@
 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, type ShortcutXLModelUsageStatusLogger } from './provider-native-request.js';
 import type { RunShortcutXLModelStreamRequestOptions } from './stream.js';
-export interface OpenAICompletionsGatewayAuth {
-    readonly headerName: 'Authorization' | 'x-api-key';
-    readonly token: string;
-}
-export interface OpenAICompletionsTransportOptions {
-    readonly auth: OpenAICompletionsGatewayAuth;
-    readonly baseUrl: string;
-    readonly fetchImpl?: typeof fetch;
-    readonly fetchOptions?: RequestInit;
-    readonly headers?: Readonly<Record<string, string | null>>;
+export interface OpenAICompletionsTransportOptions extends ProviderGatewayOptions {
     readonly qwenCompatibility?: {
         readonly disableThinkingForSuggestions?: boolean;
     };

package/dist/embedded-agent/openai-completions-transport.js

@@ -1,61 +1,8 @@
-import { convertMessages, streamSimpleOpenAICompletions } from '../ai/providers/openai-completions.js';
+import { buildOpenAICompletionsRequest, streamSimpleOpenAICompletions } from '../ai/providers/openai-completions.js';
 import { ShortcutXLModelCallTransportKind, ShortcutXLProviderUsageStatus } from '../contracts/model-call.js';
 import { toProviderUsageFromAi, toShortcutXLModelStreamEvent } from './anthropic-messages-transport.js';
-import { withV1Segment } from './gateway-base-url.js';
-import { shortcutProviderFastModeFromRequest, withProviderFastModePayload, withProviderFastModePayloadOverride } from './provider-fast-mode.js';
+import { applyGatewayProviderPayload, concreteProviderHeaders, createGatewayProviderModel, createGatewayProviderOptions } from './provider-native-request.js';
 import { normalizeTransportFetchError, throwInvalidResponse, throwTransportHttpError } from './transport-errors.js';
-function normalizeBaseUrl(baseUrl) {
-    return baseUrl.replace(/\/+$/, '');
-}
-const normalizeAiOpenAIBaseUrl = withV1Segment;
-function createAuthHeader(auth) {
-    if (auth.headerName === 'Authorization') {
-        return {
-            Authorization: auth.token.startsWith('Bearer ') ? auth.token : `Bearer ${auth.token}`
-        };
-    }
-    return { 'x-api-key': auth.token };
-}
-function buildGatewayAiModel(request, baseUrl) {
-    return {
-        api: 'openai-completions',
-        baseUrl: normalizeAiOpenAIBaseUrl(baseUrl),
-        compat: {
-            maxTokensField: 'max_completion_tokens',
-            supportsDeveloperRole: false,
-            supportsStore: false
-        },
-        contextWindow: request.model.contextWindow,
-        cost: { cacheRead: 0, cacheWrite: 0, input: 0, output: 0 },
-        id: request.model.id,
-        input: request.model.input,
-        maxTokens: request.model.maxTokens,
-        name: request.model.name,
-        provider: 'openai',
-        reasoning: request.model.reasoning
-    };
-}
-function buildGatewayAiStreamOptions(request, options) {
-    const fastMode = shortcutProviderFastModeFromRequest('openai', request);
-    return withProviderFastModePayloadOverride({
-        apiKey: options.auth.token,
-        cacheRetention: request.options?.cacheRetention,
-        fetch: options.fetchImpl,
-        fetchOptions: options.fetchOptions,
-        headers: {
-            ...request.options?.headers,
-            ...options.headers,
-            ...createAuthHeader(options.auth)
-        },
-        maxRetryDelayMs: request.options?.maxRetryDelayMs,
-        maxTokens: request.options?.maxTokens,
-        metadata: request.options?.metadata,
-        reasoning: request.options?.reasoning,
-        signal: options.signal,
-        temperature: request.options?.temperature,
-        transport: request.options?.transport
-    }, fastMode);
-}
 function toProviderUsage(usage) {
     if (!usage)
         return undefined;
@@ -69,64 +16,6 @@ function toProviderUsage(usage) {
         totalTokens: usage.total_tokens ?? input + output
     };
 }
-function buildInvokeRequestBody(request, options) {
-    const baseUrl = options.baseUrl;
-    const model = buildGatewayAiModel(request, baseUrl);
-    const toolChoice = request.options?.toolChoice;
-    const disableQwenThinking = request.options?.requestType === 'suggestions' &&
-        options.qwenCompatibility?.disableThinkingForSuggestions === true;
-    return withProviderFastModePayload({
-        ...(disableQwenThinking ? { enable_thinking: false } : {}),
-        max_completion_tokens: request.options?.maxTokens ?? 4096,
-        messages: convertMessages(model, request.context, {
-            maxTokensField: 'max_completion_tokens',
-            openRouterRouting: {},
-            requiresAssistantAfterToolResult: false,
-            requiresMistralToolIds: false,
-            requiresThinkingAsText: false,
-            requiresToolResultName: false,
-            supportsDeveloperRole: false,
-            supportsReasoningEffort: true,
-            supportsStore: false,
-            supportsStrictMode: true,
-            supportsUsageInStreaming: true,
-            thinkingFormat: 'openai',
-            vercelGatewayRouting: {}
-        }),
-        model: model.id,
-        stream: false,
-        ...(request.context.tools
-            ? {
-                tools: request.context.tools.map((tool) => ({
-                    function: {
-                        description: tool.description,
-                        name: tool.name,
-                        parameters: tool.parameters
-                    },
-                    type: 'function'
-                }))
-            }
-            : {}),
-        ...(typeof toolChoice === 'object'
-            ? {
-                tool_choice: {
-                    function: { name: toolChoice.name },
-                    type: 'function'
-                }
-            }
-            : {}),
-        ...(toolChoice === 'required' || toolChoice === 'any' ? { tool_choice: 'required' } : {}),
-        ...(toolChoice === 'auto' || toolChoice === 'none' ? { tool_choice: toolChoice } : {})
-    }, shortcutProviderFastModeFromRequest('openai', request));
-}
-function concreteHeaders(headers) {
-    const result = {};
-    for (const [key, value] of Object.entries(headers ?? {})) {
-        if (value !== null)
-            result[key] = value;
-    }
-    return result;
-}
 function parseInvokeResponse(data) {
     if (!data || typeof data !== 'object') {
         throwInvalidResponse('OpenAI Chat Completions invoke response must be a JSON object.');
@@ -148,8 +37,10 @@ function parseInvokeResponse(data) {
     };
 }
 export async function streamAiOpenAICompletionsModelRequest(request, options) {
-    const model = buildGatewayAiModel(request, options.baseUrl);
-    const stream = streamSimpleOpenAICompletions(model, request.context, buildGatewayAiStreamOptions(request, options));
+    const model = createGatewayProviderModel(request, 'openai-completions', options.baseUrl);
+    const stream = streamSimpleOpenAICompletions(model, request.context, createGatewayProviderOptions(request, 'openai-completions', options, {
+        signal: options.signal
+    }));
     let finalMessage;
     for await (const event of stream) {
         if (event.type === 'done')
@@ -173,14 +64,23 @@ export async function streamAiOpenAICompletionsModelRequest(request, options) {
 }
 export async function invokeOpenAICompletionsModel(request, options) {
     const fetchImpl = options.fetchImpl ?? fetch;
+    const model = createGatewayProviderModel(request, 'openai-completions', options.baseUrl);
+    const requestOptions = createGatewayProviderOptions(request, 'openai-completions', options, {
+        maxTokens: 4096,
+        signal: options.signal
+    });
+    const body = applyGatewayProviderPayload(request, 'openai-completions', buildOpenAICompletionsRequest(model, request.context, requestOptions, false));
+    if (request.options?.requestType === 'suggestions' &&
+        options.qwenCompatibility?.disableThinkingForSuggestions === true) {
+        body.enable_thinking = false;
+    }
     let response;
     try {
-        response = await fetchImpl(`${normalizeBaseUrl(options.baseUrl)}/v1/chat/completions`, {
-            body: JSON.stringify(buildInvokeRequestBody(request, options)),
-            headers: concreteHeaders({
+        response = await fetchImpl(`${model.baseUrl}/chat/completions`, {
+            body: JSON.stringify(body),
+            headers: concreteProviderHeaders({
                 'content-type': 'application/json',
-                ...options.headers,
-                ...createAuthHeader(options.auth)
+                ...requestOptions.headers
             }),
             method: 'POST',
             credentials: options.fetchOptions?.credentials,

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

@@ -1,17 +1,8 @@
 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, type ShortcutXLModelUsageStatusLogger } from './provider-native-request.js';
 import type { RunShortcutXLModelStreamRequestOptions } from './stream.js';
-export interface OpenAIResponsesGatewayAuth {
-    readonly headerName: 'Authorization' | 'x-api-key';
-    readonly token: string;
-}
-export interface OpenAIResponsesTransportOptions {
-    readonly auth: OpenAIResponsesGatewayAuth;
-    readonly baseUrl: string;
-    readonly fetchImpl?: typeof fetch;
-    readonly fetchOptions?: RequestInit;
-    readonly headers?: Readonly<Record<string, string | null>>;
+export interface OpenAIResponsesTransportOptions extends ProviderGatewayOptions {
     readonly usageStatusLogger?: ShortcutXLModelUsageStatusLogger;
 }
 export declare function streamAiOpenAIResponsesModelRequest(request: ShortcutXLAiModelStreamRequest, options: OpenAIResponsesTransportOptions & {

package/dist/embedded-agent/openai-responses-transport.js

@@ -1,161 +1,8 @@
-import { convertResponsesTools } from '../ai/providers/openai-responses-shared.js';
-import { streamSimpleOpenAIResponses } from '../ai/providers/openai-responses.js';
+import { buildOpenAIResponsesRequest, streamSimpleOpenAIResponses } from '../ai/providers/openai-responses.js';
 import { ShortcutXLModelCallTransportKind, ShortcutXLProviderUsageStatus } from '../contracts/model-call.js';
-import { aiUserContentToOpenAIResponsesContent } from './ai-content.js';
 import { toProviderUsageFromAi, toShortcutXLModelStreamEvent } from './anthropic-messages-transport.js';
-import { withV1Segment } from './gateway-base-url.js';
-import { OPENAI_PRIORITY_SERVICE_TIER, shortcutProviderFastModeFromRequest, withProviderFastModePayload, withProviderFastModePayloadOverride } from './provider-fast-mode.js';
+import { applyGatewayProviderPayload, concreteProviderHeaders, createGatewayProviderModel, createGatewayProviderOptions } from './provider-native-request.js';
 import { normalizeTransportFetchError, throwInvalidResponse, throwTransportHttpError } from './transport-errors.js';
-function normalizeBaseUrl(baseUrl) {
-    return baseUrl.replace(/\/+$/, '');
-}
-const normalizeAiOpenAIResponsesBaseUrl = withV1Segment;
-function createAuthHeader(auth) {
-    if (auth.headerName === 'Authorization') {
-        return {
-            Authorization: auth.token.startsWith('Bearer ') ? auth.token : `Bearer ${auth.token}`
-        };
-    }
-    return { 'x-api-key': auth.token };
-}
-function buildGatewayAiModel(request, baseUrl) {
-    return {
-        api: 'openai-responses',
-        baseUrl: normalizeAiOpenAIResponsesBaseUrl(baseUrl),
-        contextWindow: request.model.contextWindow,
-        cost: { cacheRead: 0, cacheWrite: 0, input: 0, output: 0 },
-        id: request.model.id,
-        input: request.model.input,
-        maxTokens: request.model.maxTokens,
-        name: request.model.name,
-        provider: 'openai',
-        reasoning: request.model.reasoning
-    };
-}
-function buildInvokeAiModel(request, baseUrl) {
-    return {
-        api: 'openai-responses',
-        baseUrl: normalizeAiOpenAIResponsesBaseUrl(baseUrl),
-        contextWindow: request.model.contextWindow,
-        cost: { cacheRead: 0, cacheWrite: 0, input: 0, output: 0 },
-        id: request.model.id,
-        input: ['text', 'image'],
-        maxTokens: request.options?.maxTokens ?? request.model.maxTokens,
-        name: request.model.name,
-        provider: 'openai',
-        reasoning: request.model.reasoning
-    };
-}
-function buildGatewayAiStreamOptions(request, options) {
-    const fastMode = shortcutProviderFastModeFromRequest('openai', request);
-    return withProviderFastModePayloadOverride({
-        apiKey: options.auth.token,
-        cacheRetention: request.options?.cacheRetention,
-        fetch: options.fetchImpl,
-        fetchOptions: options.fetchOptions,
-        headers: {
-            ...request.options?.headers,
-            ...options.headers,
-            ...createAuthHeader(options.auth)
-        },
-        maxRetryDelayMs: request.options?.maxRetryDelayMs,
-        maxTokens: request.options?.maxTokens,
-        metadata: request.options?.metadata,
-        reasoning: request.options?.reasoning,
-        ...(fastMode === 'openai' ? { serviceTier: OPENAI_PRIORITY_SERVICE_TIER } : {}),
-        signal: options.signal,
-        temperature: request.options?.temperature,
-        transport: request.options?.transport
-    }, fastMode);
-}
-// Local invoke converter because upstream owned AI's Responses converter only accepts
-// text/image content. The shared owned AI content helper owns the extra input_file case.
-function convertResponsesInvokeMessages(request) {
-    const input = [];
-    if (request.context.systemPrompt) {
-        input.push({
-            content: request.context.systemPrompt,
-            role: request.model.reasoning ? 'developer' : 'system'
-        });
-    }
-    for (const message of request.context.messages) {
-        if (message.role === 'user') {
-            const content = aiUserContentToOpenAIResponsesContent(message.content);
-            if (content.length > 0)
-                input.push({ content, role: 'user' });
-            continue;
-        }
-        if (message.role === 'assistant') {
-            for (const block of message.content) {
-                if (block.type === 'text') {
-                    input.push({
-                        content: [{ annotations: [], text: block.text, type: 'output_text' }],
-                        role: 'assistant',
-                        status: 'completed',
-                        type: 'message'
-                    });
-                }
-                else if (block.type === 'toolCall') {
-                    const [callId, itemId] = block.id.split('|');
-                    input.push({
-                        arguments: JSON.stringify(block.arguments),
-                        call_id: callId,
-                        ...(itemId ? { id: itemId.startsWith('fc') ? itemId : `fc_${itemId}` } : {}),
-                        name: block.name,
-                        type: 'function_call'
-                    });
-                }
-            }
-            continue;
-        }
-        const output = message.content
-            .filter((block) => block.type === 'text')
-            .map((block) => block.text)
-            .join('\n');
-        const attachments = message.content.filter((block) => block.type === 'image' || block.type === 'document');
-        const [callId] = message.toolCallId.split('|');
-        input.push({
-            call_id: callId,
-            output: output || '(see attached file)',
-            type: 'function_call_output'
-        });
-        if (attachments.length > 0) {
-            input.push({
-                content: aiUserContentToOpenAIResponsesContent([
-                    { text: 'Attached file(s) from tool result:', type: 'text' },
-                    ...attachments
-                ]),
-                role: 'user'
-            });
-        }
-    }
-    return input;
-}
-function buildInvokeRequestBody(request, baseUrl) {
-    const model = buildInvokeAiModel(request, baseUrl);
-    return withProviderFastModePayload({
-        input: convertResponsesInvokeMessages(request),
-        max_output_tokens: request.options?.maxTokens ?? 4096,
-        model: model.id,
-        stream: false,
-        // Browser-direct enterprise gateways must not persist helper conversation
-        // state by default; keep this explicit while Responses invoke is direct.
-        store: false,
-        ...(request.context.tools
-            ? {
-                tools: convertResponsesTools(request.context.tools, { strict: false })
-            }
-            : {}),
-        ...(typeof request.options?.toolChoice === 'object'
-            ? {
-                tool_choice: {
-                    name: request.options.toolChoice.name,
-                    type: 'function'
-                }
-            }
-            : {})
-    }, shortcutProviderFastModeFromRequest('openai', request));
-}
 function toProviderUsage(usage) {
     if (!usage)
         return undefined;
@@ -169,14 +16,6 @@ function toProviderUsage(usage) {
         totalTokens: usage.total_tokens ?? input + output
     };
 }
-function concreteHeaders(headers) {
-    const result = {};
-    for (const [key, value] of Object.entries(headers ?? {})) {
-        if (value !== null)
-            result[key] = value;
-    }
-    return result;
-}
 function parseInvokeResponse(data) {
     if (!data || typeof data !== 'object') {
         throwInvalidResponse('OpenAI Responses invoke response must be a JSON object.');
@@ -213,8 +52,10 @@ function parseInvokeResponse(data) {
     };
 }
 export async function streamAiOpenAIResponsesModelRequest(request, options) {
-    const model = buildGatewayAiModel(request, options.baseUrl);
-    const stream = streamSimpleOpenAIResponses(model, request.context, buildGatewayAiStreamOptions(request, options));
+    const model = createGatewayProviderModel(request, 'openai-responses', options.baseUrl);
+    const stream = streamSimpleOpenAIResponses(model, request.context, createGatewayProviderOptions(request, 'openai-responses', options, {
+        signal: options.signal
+    }));
     let finalMessage;
     for await (const event of stream) {
         if (event.type === 'done')
@@ -238,14 +79,18 @@ export async function streamAiOpenAIResponsesModelRequest(request, options) {
 }
 export async function invokeOpenAIResponsesModel(request, options) {
     const fetchImpl = options.fetchImpl ?? fetch;
+    const model = createGatewayProviderModel(request, 'openai-responses', options.baseUrl);
+    const requestOptions = createGatewayProviderOptions(request, 'openai-responses', options, {
+        maxTokens: 4096,
+        signal: options.signal
+    });
     let response;
     try {
-        response = await fetchImpl(`${normalizeBaseUrl(options.baseUrl)}/v1/responses`, {
-            body: JSON.stringify(buildInvokeRequestBody(request, options.baseUrl)),
-            headers: concreteHeaders({
+        response = await fetchImpl(`${model.baseUrl}/responses`, {
+            body: JSON.stringify(applyGatewayProviderPayload(request, 'openai-responses', buildOpenAIResponsesRequest(model, request.context, requestOptions, false))),
+            headers: concreteProviderHeaders({
                 'content-type': 'application/json',
-                ...options.headers,
-                ...createAuthHeader(options.auth)
+                ...requestOptions.headers
             }),
             method: 'POST',
             credentials: options.fetchOptions?.credentials,

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

@@ -1,5 +1,5 @@
 import { AgentExecuteCodeMode } from '../../contracts/agent-execute-code.js';
-import { AGENT_PHILOSOPHY, COMMUNICATION, DATA_FORMATTING, TASK_EXECUTION, WRITING_CODE } from '../../core/prompts/agent-guidelines.js';
+import { COMMUNICATION, DATA_FORMATTING, TASK_EXECUTION, WRITING_CODE } from '../../core/prompts/agent-guidelines.js';
 import { buildShortcutXLApiReferenceSection } from '../../utils/api-reference.js';
 import { TOOLS_BY_MODE_NAMES } from '../host-tools/mode-host-tools.js';
 import { buildActionModeMainSection } from './modes/action.js';
@@ -29,7 +29,6 @@ function buildModeSections(mode, platform) {
         case AgentExecuteCodeMode.Action:
             return [
                 buildActionModeMainSection(platformIdentity, currentDate),
-                AGENT_PHILOSOPHY,
                 TASK_EXECUTION,
                 COMMUNICATION,
                 SANDBOX_FILESYSTEM,

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

@@ -0,0 +1,29 @@
+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;
+}
+export interface ProviderGatewayOptions {
+    readonly auth: ProviderGatewayAuth;
+    readonly baseUrl: string;
+    readonly fetchImpl?: typeof fetch;
+    readonly fetchOptions?: RequestInit;
+    readonly headers?: Readonly<Record<string, string | null>>;
+}
+export declare function createGatewayProviderModel<TApi extends GatewayTransport>(request: ProviderRequest, transport: TApi, baseUrl: string): Model<TApi>;
+export declare function concreteProviderHeaders(headers: Readonly<Record<string, string | null>> | undefined): Record<string, string>;
+export declare function createGatewayProviderOptions(request: ProviderRequest, transport: GatewayTransport, options: ProviderGatewayOptions, defaults?: {
+    readonly maxTokens?: number;
+    readonly signal?: AbortSignal;
+}): SimpleStreamOptions;
+export declare function applyGatewayProviderPayload<T extends Record<string, unknown>>(request: ProviderRequest, transport: GatewayTransport, payload: T): T;
+export {};
+//# sourceMappingURL=provider-native-request.d.ts.map