sunpeak 0.19.4 → 0.19.12

package/bin/lib/eval/model-registry.mjs

@@ -0,0 +1,73 @@
+/**
+ * Model registry — maps model ID strings to AI SDK provider instances.
+ *
+ * Provider packages are dynamically imported so users only need to install
+ * the providers they actually use.
+ */
+
+/**
+ * @typedef {{ modelId: string, providerPackage: string }} ModelMapping
+ */
+
+/**
+ * Detect which provider package a model ID belongs to.
+ * @param {string} modelId
+ * @returns {string} Provider package name
+ */
+function getProviderPackage(modelId) {
+  if (/^(gpt-|o[134]-|o[134]$|chatgpt-)/.test(modelId)) return '@ai-sdk/openai';
+  if (/^claude-/.test(modelId)) return '@ai-sdk/anthropic';
+  if (/^(gemini-|models\/gemini-)/.test(modelId)) return '@ai-sdk/google';
+  throw new Error(
+    `Unknown model: "${modelId}". Expected a recognized prefix (gpt-, claude-, gemini-, o1-, o3-, o4-).`
+  );
+}
+
+/**
+ * Resolve a model ID string to an AI SDK LanguageModel instance.
+ * @param {string} modelId - e.g., 'gpt-4o', 'claude-sonnet-4-20250514', 'gemini-2.0-flash'
+ * @returns {Promise<import('ai').LanguageModel>}
+ */
+export async function resolveModel(modelId) {
+  const pkg = getProviderPackage(modelId);
+
+  let provider;
+  try {
+    provider = await import(pkg);
+  } catch {
+    throw new Error(
+      `Provider package "${pkg}" is not installed. Install it to use ${modelId}:\n\n  pnpm add -D ${pkg}\n`
+    );
+  }
+
+  // Each provider package exports a default function or named function
+  // that creates model instances: openai('gpt-4o'), anthropic('claude-...'), google('gemini-...')
+  if (pkg === '@ai-sdk/openai') {
+    const { openai } = provider;
+    return openai(modelId);
+  }
+  if (pkg === '@ai-sdk/anthropic') {
+    const { anthropic } = provider;
+    return anthropic(modelId);
+  }
+  if (pkg === '@ai-sdk/google') {
+    const { google } = provider;
+    return google(modelId);
+  }
+
+  throw new Error(`No provider factory found for ${pkg}`);
+}
+
+/**
+ * Check that the `ai` core package is installed.
+ * @returns {Promise<void>}
+ */
+export async function checkAiSdkInstalled() {
+  try {
+    await import('ai');
+  } catch {
+    throw new Error(
+      'The "ai" package is not installed. Install it to use evals:\n\n  pnpm add -D ai\n'
+    );
+  }
+}

package/bin/lib/inspect/inspect-config.mjs

@@ -49,7 +49,7 @@ export function defineInspectConfig(options) {
   const serverArg = server.includes(' ') ? `"${server}"` : server;
   const command = [
     `SUNPEAK_SANDBOX_PORT=${sandboxPort}`,
-    'npx sunpeak inspect',
+    'sunpeak inspect',
     `--server ${serverArg}`,
     ...(simulationsDir ? [`--simulations ${simulationsDir}`] : []),
     `--port ${port}`,

package/bin/lib/sandbox-server.mjs

@@ -65,12 +65,15 @@ export async function startSandboxServer({ preferredPort = 24680 } = {}) {
     } else if (
       err.code === 'HPE_INVALID_METHOD' &&
       err.rawPacket instanceof Buffer &&
-      err.rawPacket[0] === 0x16
+      err.rawPacket[0] >= 0x14 &&
+      err.rawPacket[0] <= 0x18
     ) {
       console.error(
         'Received HTTPS request on sandbox server (port ' + port + '). ' +
         'If you\'re using ngrok, make sure the upstream is http:// (not https://). ' +
-        'Example: ngrok http 8000'
+        'Example: ngrok http 8000\n' +
+        'If this only happens in Safari, use Chrome for `sunpeak dev` instead. ' +
+        'Safari is not compatible with the dev server. Production deploys (`sunpeak start`) work in all browsers.'
       );
     } else {
       console.error('Sandbox server client error', err);

package/bin/lib/test/test-config.mjs

@@ -103,7 +103,7 @@ function buildInspectCommand({ server, port, sandboxPort, simulationsDir }) {
     }
   }
 
-  parts.push('npx sunpeak inspect');
+  parts.push('sunpeak inspect');
 
   if (server.url) {
     parts.push(`--server ${server.url}`);

package/bin/sunpeak.js

@@ -115,6 +115,7 @@ Testing (works with any MCP server):
     init                   Scaffold test infrastructure into a project
     --unit                 Run unit tests (vitest)
     --live                 Run live tests against real hosts
+    --eval                 Run evals against LLM models
 
 App framework (for sunpeak projects):
   sunpeak new [name] [resources]  Create a new project

package/dist/chatgpt/index.cjs

@@ -1,6 +1,6 @@
 Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
 const require_chunk = require("../chunk-9hOWP6kD.cjs");
-const require_inspector = require("../inspector-Cvq3yjNL.cjs");
+const require_inspector = require("../inspector-jY9O18z9.cjs");
 const require_inspector_url = require("../inspector-url-C3LTKgXt.cjs");
 const require_discovery = require("../discovery-Clu4uHp1.cjs");
 //#region src/chatgpt/index.ts

package/dist/chatgpt/index.js

@@ -1,5 +1,5 @@
 import { r as __exportAll } from "../chunk-D6g4UhsZ.js";
-import { _ as McpAppHost, d as ThemeProvider, f as useThemeContext, g as extractResourceCSP, h as IframeResource, n as resolveServerToolResult, t as Inspector, v as SCREEN_WIDTHS } from "../inspector-Bp9jrHIu.js";
+import { _ as McpAppHost, d as ThemeProvider, f as useThemeContext, g as extractResourceCSP, h as IframeResource, n as resolveServerToolResult, t as Inspector, v as SCREEN_WIDTHS } from "../inspector-D5DckQuU.js";
 import { t as createInspectorUrl } from "../inspector-url-CyQcuBI9.js";
 import { c as toPascalCase, i as findResourceKey, n as extractSimulationKey, r as findResourceDirs, s as getComponentName, t as extractResourceKey } from "../discovery-Cgoegt62.js";
 //#region src/chatgpt/index.ts

package/dist/claude/index.cjs

@@ -1,4 +1,4 @@
 Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
 require("../chunk-9hOWP6kD.cjs");
-const require_inspector = require("../inspector-Cvq3yjNL.cjs");
+const require_inspector = require("../inspector-jY9O18z9.cjs");
 exports.Inspector = require_inspector.Inspector;

package/dist/claude/index.js

@@ -1,2 +1,2 @@
-import { t as Inspector } from "../inspector-Bp9jrHIu.js";
+import { t as Inspector } from "../inspector-D5DckQuU.js";
 export { Inspector };

package/dist/host/chatgpt/index.cjs

@@ -1,6 +1,6 @@
 Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
 require("../../chunk-9hOWP6kD.cjs");
-const require_use_app = require("../../use-app-DHYiev3D.cjs");
+const require_use_app = require("../../use-app-CbsBEmwv.cjs");
 let react = require("react");
 //#region src/host/chatgpt/openai-types.ts
 /**

package/dist/host/chatgpt/index.js

@@ -1,4 +1,4 @@
-import { t as useApp } from "../../use-app-Ck5kR1Sf.js";
+import { t as useApp } from "../../use-app-Bfargfa3.js";
 import { useCallback } from "react";
 //#region src/host/chatgpt/openai-types.ts
 /**

package/dist/index.cjs

@@ -1,8 +1,8 @@
 Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
 const require_chunk = require("./chunk-9hOWP6kD.cjs");
 const require_protocol = require("./protocol-C8pFDmcy.cjs");
-const require_use_app = require("./use-app-DHYiev3D.cjs");
-const require_inspector = require("./inspector-Cvq3yjNL.cjs");
+const require_use_app = require("./use-app-CbsBEmwv.cjs");
+const require_inspector = require("./inspector-jY9O18z9.cjs");
 const require_host_index = require("./host/index.cjs");
 const require_inspector_index = require("./inspector/index.cjs");
 const require_chatgpt_index = require("./chatgpt/index.cjs");

package/dist/index.js

@@ -1,6 +1,6 @@
 import { C as EmbeddedResourceSchema, D as ImplementationSchema, J as RequestIdSchema, Y as ResourceLinkSchema, _ as ContentBlockSchema, ct as array, dt as number, et as ToolSchema, ft as object, gt as unknown, h as CallToolResultSchema, ht as union, lt as boolean, mt as string, pt as record, st as _undefined, ut as literal } from "./protocol-CRqiPTLT.js";
-import { $ as u, A as W, B as h, C as P, D as T, E as S, F as c, G as n, H as jQ, I as d, J as q, K as o, L as e, M as _, N as _Q, O as TQ, P as a, Q as t, R as g, S as OQ, T as R, U as k, V as j, W as l, X as r, Y as qQ, Z as s, _ as M, a as B, at as z, b as NQ, c as E, d as FQ, et as v, f as GQ, g as LQ, h as K, i as AQ, it as y, j as WQ, k as U, l as EQ, m as I, n as AppProvider, nt as wQ, o as BQ, ot as zQ, p as H, q as p, r as A, rt as x, s as CX, t as useApp, tt as w, u as F, v as MX, w as PQ, x as O, y as N, z as gQ } from "./use-app-Ck5kR1Sf.js";
-import { S as DEFAULT_STYLE_VARIABLES } from "./inspector-Bp9jrHIu.js";
+import { $ as u, A as W, B as h, C as P, D as T, E as S, F as c, G as n, H as jQ, I as d, J as q, K as o, L as e, M as _, N as _Q, O as TQ, P as a, Q as t, R as g, S as OQ, T as R, U as k, V as j, W as l, X as r, Y as qQ, Z as s, _ as M, a as B, at as z, b as NQ, c as E, d as FQ, et as v, f as GQ, g as LQ, h as K, i as AQ, it as y, j as WQ, k as U, l as EQ, m as I, n as AppProvider, nt as wQ, o as BQ, ot as zQ, p as H, q as p, r as A, rt as x, s as CX, t as useApp, tt as w, u as F, v as MX, w as PQ, x as O, y as N, z as gQ } from "./use-app-Bfargfa3.js";
+import { S as DEFAULT_STYLE_VARIABLES } from "./inspector-D5DckQuU.js";
 import { detectHost, isChatGPT, isClaude } from "./host/index.js";
 import { t as inspector_exports } from "./inspector/index.js";
 import { t as chatgpt_exports } from "./chatgpt/index.js";

package/dist/inspector/index.cjs

@@ -1,6 +1,6 @@
 Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
 const require_chunk = require("../chunk-9hOWP6kD.cjs");
-const require_inspector = require("../inspector-Cvq3yjNL.cjs");
+const require_inspector = require("../inspector-jY9O18z9.cjs");
 const require_inspector_url = require("../inspector-url-C3LTKgXt.cjs");
 const require_discovery = require("../discovery-Clu4uHp1.cjs");
 //#region src/inspector/index.ts

package/dist/inspector/index.js

@@ -1,5 +1,5 @@
 import { r as __exportAll } from "../chunk-D6g4UhsZ.js";
-import { _ as McpAppHost, a as SidebarControl, b as getRegisteredHosts, c as SidebarTextarea, d as ThemeProvider, f as useThemeContext, g as extractResourceCSP, h as IframeResource, i as SidebarCollapsibleControl, l as SidebarToggle, m as useInspectorState, n as resolveServerToolResult, o as SidebarInput, p as useMcpConnection, r as SidebarCheckbox, s as SidebarSelect, t as Inspector, u as SimpleSidebar, v as SCREEN_WIDTHS, x as registerHostShell, y as getHostShell } from "../inspector-Bp9jrHIu.js";
+import { _ as McpAppHost, a as SidebarControl, b as getRegisteredHosts, c as SidebarTextarea, d as ThemeProvider, f as useThemeContext, g as extractResourceCSP, h as IframeResource, i as SidebarCollapsibleControl, l as SidebarToggle, m as useInspectorState, n as resolveServerToolResult, o as SidebarInput, p as useMcpConnection, r as SidebarCheckbox, s as SidebarSelect, t as Inspector, u as SimpleSidebar, v as SCREEN_WIDTHS, x as registerHostShell, y as getHostShell } from "../inspector-D5DckQuU.js";
 import { t as createInspectorUrl } from "../inspector-url-CyQcuBI9.js";
 import { c as toPascalCase, i as findResourceKey, n as extractSimulationKey, r as findResourceDirs, s as getComponentName, t as extractResourceKey } from "../discovery-Cgoegt62.js";
 //#region src/inspector/index.ts

package/dist/{inspector-Bp9jrHIu.js → inspector-D5DckQuU.js}

@@ -3995,7 +3995,7 @@ function Inspector({ children, simulations: initialSimulations = {}, appName = "
 						onChange: setProdResources,
 						label: "Prod Resources",
 						tooltip: "Load resources from dist/ builds instead of HMR",
-						docsPath: "api-reference/cli/dev#prod-tools-and-prod-resources-flags"
+						docsPath: "app-framework/cli/dev#prod-tools-and-prod-resources-flags"
 					}),
 					hasTools && /* @__PURE__ */ jsxs("div", {
 						className: "grid grid-cols-2 gap-2",
@@ -4003,7 +4003,7 @@ function Inspector({ children, simulations: initialSimulations = {}, appName = "
 						children: [/* @__PURE__ */ jsx(SidebarControl, {
 							label: "Tool",
 							tooltip: "Tool to inspect",
-							docsPath: "api-reference/cli/dev",
+							docsPath: "app-framework/cli/dev",
 							"data-testid": "tool-selector",
 							children: /* @__PURE__ */ jsx(SidebarSelect, {
 								value: selectedToolName,
@@ -4017,7 +4017,7 @@ function Inspector({ children, simulations: initialSimulations = {}, appName = "
 							})
 						}), /* @__PURE__ */ jsx(SidebarControl, {
 							label: selectedToolInfo && selectedToolInfo.fixtureSimNames.length > 0 ? "Simulation" : /* @__PURE__ */ jsx("a", {
-								href: `${DOCS_BASE_URL}/api-reference/simulations/simulation`,
+								href: `${DOCS_BASE_URL}/testing/simulations`,
 								target: "_blank",
 								rel: "noopener noreferrer",
 								className: "no-underline transition-colors",
@@ -4031,7 +4031,7 @@ function Inspector({ children, simulations: initialSimulations = {}, appName = "
 								children: "Simulation"
 							}),
 							tooltip: selectedToolInfo && selectedToolInfo.fixtureSimNames.length > 0 ? "Test fixture with mock data" : "Create simulations for faster testing",
-							docsPath: "api-reference/simulations/simulation",
+							docsPath: "testing/simulations",
 							"data-testid": "simulation-selector",
 							children: /* @__PURE__ */ jsx(SidebarSelect, {
 								value: activeSimulationName ?? "__none__",
@@ -4051,7 +4051,7 @@ function Inspector({ children, simulations: initialSimulations = {}, appName = "
 						children: [registeredHosts.length > 1 && /* @__PURE__ */ jsx(SidebarControl, {
 							label: "Host",
 							tooltip: "Host runtime to simulate",
-							docsPath: "api-reference/functions/host-detection",
+							docsPath: "app-framework/functions/host-detection",
 							children: /* @__PURE__ */ jsx(SidebarSelect, {
 								value: state.activeHost,
 								onChange: (value) => state.setActiveHost(value),
@@ -4063,7 +4063,7 @@ function Inspector({ children, simulations: initialSimulations = {}, appName = "
 						}), /* @__PURE__ */ jsx(SidebarControl, {
 							label: "Width",
 							tooltip: "Chat width",
-							docsPath: "api-reference/simulations/inspector",
+							docsPath: "testing/inspector",
 							children: /* @__PURE__ */ jsx(SidebarSelect, {
 								value: state.screenWidth,
 								onChange: (value) => state.setScreenWidth(value),
@@ -4092,7 +4092,7 @@ function Inspector({ children, simulations: initialSimulations = {}, appName = "
 						label: "Host Context",
 						defaultCollapsed: false,
 						tooltip: "Host-provided environment",
-						docsPath: "api-reference/hooks/use-host-context",
+						docsPath: "app-framework/hooks/use-host-context",
 						children: /* @__PURE__ */ jsxs("div", {
 							className: "space-y-1",
 							children: [
@@ -4101,7 +4101,7 @@ function Inspector({ children, simulations: initialSimulations = {}, appName = "
 									children: [/* @__PURE__ */ jsx(SidebarControl, {
 										label: "Theme",
 										tooltip: "Host color theme",
-										docsPath: "api-reference/hooks/use-theme",
+										docsPath: "app-framework/hooks/use-theme",
 										children: /* @__PURE__ */ jsx(SidebarToggle, {
 											value: state.theme,
 											onChange: (value) => state.setTheme(value),
@@ -4116,7 +4116,7 @@ function Inspector({ children, simulations: initialSimulations = {}, appName = "
 									}), /* @__PURE__ */ jsx(SidebarControl, {
 										label: "Locale",
 										tooltip: "BCP 47 language tag",
-										docsPath: "api-reference/hooks/use-locale",
+										docsPath: "app-framework/hooks/use-locale",
 										children: /* @__PURE__ */ jsx(SidebarInput, {
 											applyOnBlur: true,
 											value: state.locale,
@@ -4128,7 +4128,7 @@ function Inspector({ children, simulations: initialSimulations = {}, appName = "
 								/* @__PURE__ */ jsx(SidebarControl, {
 									label: "Display Mode",
 									tooltip: "Host resource rendering paradigm",
-									docsPath: "api-reference/hooks/use-display-mode",
+									docsPath: "app-framework/hooks/use-display-mode",
 									children: /* @__PURE__ */ jsx(SidebarToggle, {
 										value: state.displayMode,
 										onChange: (value) => state.setDisplayMode(value),
@@ -4155,7 +4155,7 @@ function Inspector({ children, simulations: initialSimulations = {}, appName = "
 										children: /* @__PURE__ */ jsx(SidebarControl, {
 											label: "Platform",
 											tooltip: "End user device platform",
-											docsPath: "api-reference/hooks/use-platform",
+											docsPath: "app-framework/hooks/use-platform",
 											children: /* @__PURE__ */ jsx(SidebarSelect, {
 												value: state.platform,
 												onChange: (value) => {
@@ -4193,7 +4193,7 @@ function Inspector({ children, simulations: initialSimulations = {}, appName = "
 										children: /* @__PURE__ */ jsx(SidebarControl, {
 											label: "Capabilities",
 											tooltip: "End user device capabilities",
-											docsPath: "api-reference/hooks/use-device-capabilities",
+											docsPath: "app-framework/hooks/use-device-capabilities",
 											children: /* @__PURE__ */ jsxs("div", {
 												className: "flex gap-2",
 												children: [/* @__PURE__ */ jsx(SidebarCheckbox, {
@@ -4212,7 +4212,7 @@ function Inspector({ children, simulations: initialSimulations = {}, appName = "
 								/* @__PURE__ */ jsx(SidebarControl, {
 									label: "Time Zone",
 									tooltip: "End user IANA time zone",
-									docsPath: "api-reference/hooks/use-time-zone",
+									docsPath: "app-framework/hooks/use-time-zone",
 									children: /* @__PURE__ */ jsx(SidebarInput, {
 										applyOnBlur: true,
 										value: state.timeZone,
@@ -4223,7 +4223,7 @@ function Inspector({ children, simulations: initialSimulations = {}, appName = "
 								/* @__PURE__ */ jsx(SidebarControl, {
 									label: "Container Dimensions",
 									tooltip: "Host-enforced size constraints (px)",
-									docsPath: "api-reference/hooks/use-viewport",
+									docsPath: "app-framework/hooks/use-viewport",
 									children: /* @__PURE__ */ jsxs("div", {
 										className: "grid grid-cols-4 gap-1",
 										children: [
@@ -4273,7 +4273,7 @@ function Inspector({ children, simulations: initialSimulations = {}, appName = "
 								/* @__PURE__ */ jsx(SidebarControl, {
 									label: "Safe Area Insets",
 									tooltip: "Device safe area padding (px)",
-									docsPath: "api-reference/hooks/use-safe-area",
+									docsPath: "app-framework/hooks/use-safe-area",
 									children: /* @__PURE__ */ jsxs("div", {
 										className: "grid grid-cols-4 gap-1",
 										children: [
@@ -4355,7 +4355,7 @@ function Inspector({ children, simulations: initialSimulations = {}, appName = "
 						label: "App Context",
 						defaultCollapsed: true,
 						tooltip: "App-provided context shared with the model",
-						docsPath: "api-reference/hooks/use-app-state",
+						docsPath: "app-framework/hooks/use-app-state",
 						children: /* @__PURE__ */ jsx(SidebarTextarea, {
 							value: state.modelContextJson,
 							onChange: (json) => state.validateJSON(json, state.setModelContextJson, state.setModelContextError),
@@ -4371,7 +4371,7 @@ function Inspector({ children, simulations: initialSimulations = {}, appName = "
 						label: "Tool Input (JSON)",
 						defaultCollapsed: false,
 						tooltip: "Arguments passed to the tool",
-						docsPath: "api-reference/hooks/use-tool-data",
+						docsPath: "app-framework/hooks/use-tool-data",
 						children: /* @__PURE__ */ jsx(SidebarTextarea, {
 							value: state.toolInputJson,
 							onChange: (json) => state.validateJSON(json, state.setToolInputJson, state.setToolInputError),
@@ -4385,7 +4385,7 @@ function Inspector({ children, simulations: initialSimulations = {}, appName = "
 						label: "Tool Result (JSON)",
 						defaultCollapsed: false,
 						tooltip: "Structured content returned by the tool",
-						docsPath: "api-reference/hooks/use-tool-data",
+						docsPath: "app-framework/hooks/use-tool-data",
 						"data-testid": "tool-result-section",
 						children: /* @__PURE__ */ jsx(SidebarTextarea, {
 							value: state.toolResultJson,
@@ -4416,4 +4416,4 @@ function Inspector({ children, simulations: initialSimulations = {}, appName = "
 //#endregion
 export { DEFAULT_STYLE_VARIABLES as S, McpAppHost as _, SidebarControl as a, getRegisteredHosts as b, SidebarTextarea as c, ThemeProvider as d, useThemeContext as f, extractResourceCSP as g, IframeResource as h, SidebarCollapsibleControl as i, SidebarToggle as l, useInspectorState as m, resolveServerToolResult as n, SidebarInput as o, useMcpConnection as p, SidebarCheckbox as r, SidebarSelect as s, Inspector as t, SimpleSidebar as u, SCREEN_WIDTHS as v, registerHostShell as x, getHostShell as y };
 
-//# sourceMappingURL=inspector-Bp9jrHIu.js.map
+//# sourceMappingURL=inspector-D5DckQuU.js.map