@skybridge/test 0.0.0-dev.c19b437 → 0.0.0-dev.c2c0f98

package/README.md

@@ -4,19 +4,20 @@ Evals for [Skybridge](https://skybridge.tech) MCP apps: they answer whether a
 prompt reaches the right tool with the right arguments, which no unit test can
 check.
 
-**Beta.** The API is still settling, so it publishes under the `beta`
-dist-tag and may change between releases.
-
-Add the `evals` option to the Skybridge Vite plugin, then write scenarios with
-vitest's own `it` and `expect`.
+Add the `evals` option to the Skybridge Vite plugin so `expect.chat` exists,
+then write scenarios with vitest's own `it` and `expect`. Pass the app itself
+to `start`: it is served in-process, dialing `app.fetchHandler` directly, and
+the tool names and argument shapes come from the app value, so no type
+parameter is needed.
 
 ```ts
+import { anthropic } from "@ai-sdk/anthropic";
 import { expect, it } from "vitest";
 import { start } from "@skybridge/test";
-import type { AppType } from "../src/server.js";
+import { app } from "../src/server.js";
 
 it("finds products by category", async () => {
-  const chat = await start<AppType>();
+  const chat = await start({ app, model: anthropic("claude-sonnet-4-5") });
   await chat.send("I am looking for ski goggles");
 
   expect.chat(chat).toHaveCalledToolWith("search-products", {
@@ -25,68 +26,70 @@ it("finds products by category", async () => {
 });
 ```
 
-## Choosing a model
-
-`model` is a `provider/model-id` string. Only the provider you name is
-imported, so install just that one as a dev dependency, and the key is read
-from its own environment variable.
+There is no child process and no server to boot, so the evals run anywhere
+vitest does. The caveat is that they bypass the Express layer, so nothing
+verifies a bearer token; pass `authInfo` to claim an identity.
 
-| prefix | package | key |
-| --- | --- | --- |
-| `anthropic` | `@ai-sdk/anthropic` | `ANTHROPIC_API_KEY` |
-| `openai` | `@ai-sdk/openai` | `OPENAI_API_KEY` |
-| `mistral` | `@ai-sdk/mistral` | `MISTRAL_API_KEY` |
-| `google` | `@ai-sdk/google` | `GOOGLE_GENERATIVE_AI_API_KEY` |
+## Authenticated evals
 
 ```ts
-skybridge({
-  evals: {
-    model: "anthropic/claude-sonnet-4-5",
-    project: { cwd: ".", command: ["node", "dist/server.js"] },
+const chat = await start({
+  app,
+  model: anthropic("claude-sonnet-4-5"),
+  authInfo: {
+    token: "eval-token",
+    clientId: "eval-client",
+    scopes: ["checkout"],
+    extra: { sub: "user_123", email: "ada@example.com" },
   },
 });
 ```
 
-`command` has to start the server process directly, since wrappers such as
-`pnpm run` do not forward the channel the runner uses to learn the port.
+`authInfo` injects a claimed identity: the app's own scheme and scope
+enforcement runs for real, in-band challenges included, and only token
+signature verification is skipped. Omit it to eval the anonymous path, which is
+how you assert that a protected tool challenges instead of running.
 
-Scenarios wait on a real model, so raise vitest's `testTimeout` well above its
-default: `test: { testTimeout: 120_000 }`.
+## Choosing a model
+
+`start` takes any AI SDK model instance, so the provider and its credentials
+are the project's own. Install the provider package you want alongside `ai`.
 
-Any other prefix is passed to the AI SDK untouched, so a string such as
-`openai/gpt-5` also resolves through the Vercel AI Gateway if you have
-`AI_GATEWAY_API_KEY` set and no provider package installed.
+```ts
+import { anthropic } from "@ai-sdk/anthropic";
+import { start } from "@skybridge/test";
 
-## Models the prefixes cannot reach
+const chat = await start({ app, model: anthropic("claude-sonnet-4-5") });
+```
 
-For a local runtime, a company gateway, or a model that needs middleware,
-build the model yourself and register it from a vitest setup file. It takes
-precedence over `model`, which can then be left out.
+A local runtime or a company gateway works the same way, through the AI SDK's
+own `createOpenAICompatible` or `customProvider`.
 
 ```ts
-// eval-model.ts
 import { createOpenAICompatible } from "@ai-sdk/openai-compatible";
-import { defineEvalModel } from "@skybridge/test";
 
-defineEvalModel(
-  createOpenAICompatible({ baseURL: "http://localhost:11434/v1" })("llama3.1"),
-);
+const ollama = createOpenAICompatible({
+  name: "ollama",
+  baseURL: "http://localhost:11434/v1",
+});
+const chat = await start({ app, model: ollama("llama3.1") });
 ```
 
+The plugin's `evals` option also carries the defaults every conversation starts
+from, so a project can set them once instead of per scenario.
+
 ```ts
-// vitest.config.ts
-export default defineConfig({
-  plugins: [skybridge({ evals: { project: { ... } } })],
-  test: { setupFiles: ["./eval-model.ts"] },
-});
+skybridge({ evals: { temperature: 0, maxSteps: 6 } });
 ```
 
-Tool calls the server rejected do not count towards a positive assertion, so a
-scenario cannot pass on a call that errored. To assert an error on purpose, use
-`toHaveFailedToolCall`.
+## Matchers
+
+Assertions on what the assistant said work the same way, though model prose
+varies between runs, so keep them loose.
 
 ```ts
-expect.chat(chat).toHaveFailedToolCall("create-checkout");
+expect.chat(chat).toHaveSaid("in stock");
+expect.chat(chat).not.toHaveSaid("refund");
 ```
 
 See the [documentation](https://docs.skybridge.tech) for setup and the full

package/dist/chat.d.ts

@@ -1,4 +1,4 @@
-import type { ToolCall, ToolDefinition } from "@skybridge/vite-plugin/evals";
+import type { ToolCall } from "@skybridge/vite-plugin/evals";
 import { type LanguageModel } from "ai";
 interface HostConfig {
     model: LanguageModel;
@@ -6,13 +6,22 @@ interface HostConfig {
     systemPrompt?: string;
     maxSteps?: number;
 }
+/**
+ * A `fetch` replacement the transport dials instead of the network. Mirrors the
+ * MCP client's own `fetch` option so an in-process app can serve the session.
+ */
+export type TransportFetch = (url: string | URL, init?: RequestInit) => Promise<Response>;
 /**
  * One conversation against one MCP session. A scenario asserts on
  * {@link Chat.toolCalls}, the sequence of calls the model made.
  */
 export declare class Chat<App = unknown> {
+    /** Type-only anchor so `expect.chat` infers `App` from the conversation. */
+    readonly $app: App;
     /** Every call the model made, in order, as it crossed the wire. */
     readonly toolCalls: ToolCall<App>[];
+    /** What the assistant said, one entry per turn, intermediate ones included. */
+    readonly assistantTurns: string[];
     private readonly messages;
     private readonly failures;
     private readonly client;
@@ -20,11 +29,9 @@ export declare class Chat<App = unknown> {
     private readonly host;
     private tools;
     private constructor();
-    static open<App>(serverUrl: string, host: HostConfig): Promise<Chat<App>>;
-    /** The tool definitions the model was looking at, for failure reports. */
-    get toolDefinitions(): ToolDefinition[];
-    /** Takes a turn and returns the model's answer. */
-    send(prompt: string): Promise<string>;
+    static open<App>(host: HostConfig, fetchImpl: TransportFetch): Promise<Chat<App>>;
+    /** Takes a turn. Assert on the result with `expect.chat`. */
+    send(prompt: string): Promise<void>;
     close(): Promise<void>;
     private failureFor;
     private toolSet;

package/dist/chat.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"chat.d.ts","sourceRoot":"","sources":["../src/chat.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,QAAQ,EAAE,cAAc,EAAE,MAAM,8BAA8B,CAAC;AAC7E,OAAO,EAIL,KAAK,aAAa,EAGnB,MAAM,IAAI,CAAC;AAEZ,UAAU,UAAU;IAClB,KAAK,EAAE,aAAa,CAAC;IACrB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAID;;;GAGG;AACH,qBAAa,IAAI,CAAC,GAAG,GAAG,OAAO;IAC7B,mEAAmE;IACnE,QAAQ,CAAC,SAAS,EAAE,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAM;IAEzC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAsB;IAC/C,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAA6B;IACtD,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAS;IAChC,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAgC;IAC1D,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAa;IAClC,OAAO,CAAC,KAAK,CAAwB;IAErC,OAAO;WAUM,IAAI,CAAC,GAAG,EACnB,SAAS,EAAE,MAAM,EACjB,IAAI,EAAE,UAAU,GACf,OAAO,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAerB,0EAA0E;IAC1E,IAAI,eAAe,IAAI,cAAc,EAAE,CAEtC;IAED,mDAAmD;IAC7C,IAAI,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;IA4BrC,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC;IAI5B,OAAO,CAAC,UAAU;IAalB,OAAO,CAAC,OAAO;CAmChB"}
+{"version":3,"file":"chat.d.ts","sourceRoot":"","sources":["../src/chat.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,8BAA8B,CAAC;AAC7D,OAAO,EAIL,KAAK,aAAa,EAGnB,MAAM,IAAI,CAAC;AAEZ,UAAU,UAAU;IAClB,KAAK,EAAE,aAAa,CAAC;IACrB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAMD;;;GAGG;AACH,MAAM,MAAM,cAAc,GAAG,CAC3B,GAAG,EAAE,MAAM,GAAG,GAAG,EACjB,IAAI,CAAC,EAAE,WAAW,KACf,OAAO,CAAC,QAAQ,CAAC,CAAC;AAEvB;;;GAGG;AACH,qBAAa,IAAI,CAAC,GAAG,GAAG,OAAO;IAC7B,4EAA4E;IAC5E,SAAiB,IAAI,EAAE,GAAG,CAAC;IAE3B,mEAAmE;IACnE,QAAQ,CAAC,SAAS,EAAE,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAM;IAEzC,+EAA+E;IAC/E,QAAQ,CAAC,cAAc,EAAE,MAAM,EAAE,CAAM;IAEvC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAsB;IAC/C,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAA6B;IACtD,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAS;IAChC,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAgC;IAC1D,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAa;IAClC,OAAO,CAAC,KAAK,CAAc;IAE3B,OAAO;WAUM,IAAI,CAAC,GAAG,EACnB,IAAI,EAAE,UAAU,EAChB,SAAS,EAAE,cAAc,GACxB,OAAO,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAkBrB,6DAA6D;IACvD,IAAI,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC;IAiCnC,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC;IAI5B,OAAO,CAAC,UAAU;IAalB,OAAO,CAAC,OAAO;CAqChB"}

package/dist/chat.js

@@ -1,7 +1,7 @@
-import { Client } from "@modelcontextprotocol/sdk/client/index.js";
-import { StreamableHTTPClientTransport } from "@modelcontextprotocol/sdk/client/streamableHttp.js";
+import { Client, StreamableHTTPClientTransport, } from "@modelcontextprotocol/client";
 import { dynamicTool, generateText, jsonSchema, stepCountIs, } from "ai";
 const DEFAULT_MAX_STEPS = 8;
+const IN_PROCESS_URL = "http://in-process.skybridge.test/mcp";
 /**
  * One conversation against one MCP session. A scenario asserts on
  * {@link Chat.toolCalls}, the sequence of calls the model made.
@@ -9,6 +9,8 @@ const DEFAULT_MAX_STEPS = 8;
 export class Chat {
     /** Every call the model made, in order, as it crossed the wire. */
     toolCalls = [];
+    /** What the assistant said, one entry per turn, intermediate ones included. */
+    assistantTurns = [];
     messages = [];
     failures = new Map();
     client;
@@ -20,9 +22,9 @@ export class Chat {
         this.transport = transport;
         this.host = host;
     }
-    static async open(serverUrl, host) {
+    static async open(host, fetchImpl) {
         const client = new Client({ name: "skybridge-eval", version: "0" });
-        const transport = new StreamableHTTPClientTransport(new URL(serverUrl));
+        const transport = new StreamableHTTPClientTransport(new URL(IN_PROCESS_URL), { fetch: fetchImpl });
         await client.connect(transport);
         const chat = new Chat(client, transport, host);
         try {
@@ -35,11 +37,7 @@ export class Chat {
         }
         return chat;
     }
-    /** The tool definitions the model was looking at, for failure reports. */
-    get toolDefinitions() {
-        return this.tools;
-    }
-    /** Takes a turn and returns the model's answer. */
+    /** Takes a turn. Assert on the result with `expect.chat`. */
     async send(prompt) {
         this.messages.push({ role: "user", content: prompt });
         const result = await generateText({
@@ -61,8 +59,12 @@ export class Chat {
                 });
             }
         }
+        for (const step of result.steps) {
+            if (step.text !== "") {
+                this.assistantTurns.push(step.text);
+            }
+        }
         this.messages.push(...result.response.messages);
-        return result.text;
     }
     async close() {
         await this.transport.close();

package/dist/chat.js.map

@@ -1 +1 @@
-{"version":3,"file":"chat.js","sourceRoot":"","sources":["../src/chat.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,MAAM,2CAA2C,CAAC;AACnE,OAAO,EAAE,6BAA6B,EAAE,MAAM,oDAAoD,CAAC;AAEnG,OAAO,EACL,WAAW,EACX,YAAY,EACZ,UAAU,EAGV,WAAW,GACZ,MAAM,IAAI,CAAC;AASZ,MAAM,iBAAiB,GAAG,CAAC,CAAC;AAE5B;;;GAGG;AACH,MAAM,OAAO,IAAI;IACf,mEAAmE;IAC1D,SAAS,GAAoB,EAAE,CAAC;IAExB,QAAQ,GAAmB,EAAE,CAAC;IAC9B,QAAQ,GAAG,IAAI,GAAG,EAAkB,CAAC;IACrC,MAAM,CAAS;IACf,SAAS,CAAgC;IACzC,IAAI,CAAa;IAC1B,KAAK,GAAqB,EAAE,CAAC;IAErC,YACE,MAAc,EACd,SAAwC,EACxC,IAAgB;QAEhB,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QACrB,IAAI,CAAC,SAAS,GAAG,SAAS,CAAC;QAC3B,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC;IACnB,CAAC;IAED,MAAM,CAAC,KAAK,CAAC,IAAI,CACf,SAAiB,EACjB,IAAgB;QAEhB,MAAM,MAAM,GAAG,IAAI,MAAM,CAAC,EAAE,IAAI,EAAE,gBAAgB,EAAE,OAAO,EAAE,GAAG,EAAE,CAAC,CAAC;QACpE,MAAM,SAAS,GAAG,IAAI,6BAA6B,CAAC,IAAI,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC;QACxE,MAAM,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;QAChC,MAAM,IAAI,GAAG,IAAI,IAAI,CAAM,MAAM,EAAE,SAAS,EAAE,IAAI,CAAC,CAAC;QACpD,IAAI,CAAC;YACH,MAAM,EAAE,KAAK,EAAE,GAAG,MAAM,MAAM,CAAC,SAAS,EAAE,CAAC;YAC3C,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACrB,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,SAAS,CAAC,KAAK,EAAE,CAAC;YACxB,MAAM,KAAK,CAAC;QACd,CAAC;QACD,OAAO,IAAI,CAAC;IACd,CAAC;IAED,0EAA0E;IAC1E,IAAI,eAAe;QACjB,OAAO,IAAI,CAAC,KAAK,CAAC;IACpB,CAAC;IAED,mDAAmD;IACnD,KAAK,CAAC,IAAI,CAAC,MAAc;QACvB,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;QAEtD,MAAM,MAAM,GAAG,MAAM,YAAY,CAAC;YAChC,KAAK,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK;YACtB,WAAW,EAAE,IAAI,CAAC,IAAI,CAAC,WAAW,IAAI,CAAC;YACvC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,YAAY,KAAK,SAAS;gBACtC,CAAC,CAAC,EAAE;gBACJ,CAAC,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,CAAC;YACvC,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,KAAK,EAAE,IAAI,CAAC,OAAO,EAAE;YACrB,QAAQ,EAAE,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,IAAI,iBAAiB,CAAC;SAC/D,CAAC,CAAC;QAEH,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,KAAK,EAAE,CAAC;YAChC,KAAK,MAAM,IAAI,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC;gBAClC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC;oBAClB,IAAI,EAAE,IAAI,CAAC,QAAQ;oBACnB,SAAS,EAAE,CAAC,IAAI,CAAC,KAAK,IAAI,EAAE,CAA4B;oBACxD,GAAG,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC;iBACR,CAAC,CAAC;YACtB,CAAC;QACH,CAAC;QAED,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;QAChD,OAAO,MAAM,CAAC,IAAI,CAAC;IACrB,CAAC;IAED,KAAK,CAAC,KAAK;QACT,MAAM,IAAI,CAAC,SAAS,CAAC,KAAK,EAAE,CAAC;IAC/B,CAAC;IAEO,UAAU,CAAC,IAA+C;QAGhE,MAAM,QAAQ,GAAG,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;QACpD,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;YAC3B,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC;QAC9B,CAAC;QACD,IAAI,IAAI,CAAC,OAAO,KAAK,IAAI,EAAE,CAAC;YAC1B,OAAO,EAAE,MAAM,EAAE,2CAA2C,EAAE,CAAC;QACjE,CAAC;QACD,OAAO,EAAE,CAAC;IACZ,CAAC;IAEO,OAAO;QACb,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,UAAU,EAAE,EAAE;YAC5C,MAAM,EAAE,OAAO,EAAE,QAAQ,EAAE,GAAG,UAAU,EAAE,GAAG,UAAU,CAAC,WAAW,CAAC;YACpE,OAAO;gBACL,UAAU,CAAC,IAAI;gBACf,WAAW,CAAC;oBACV,WAAW,EAAE,UAAU,CAAC,WAAW,IAAI,EAAE;oBACzC,WAAW,EAAE,UAAU,CAAC,UAAU,CAAC;oBACnC,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,EAAE,UAAU,EAAE,EAAE,EAAE;wBACvC,IAAI,MAA+C,CAAC;wBACpD,IAAI,CAAC;4BACH,MAAM,GAAG,MAAM,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC;gCAClC,IAAI,EAAE,UAAU,CAAC,IAAI;gCACrB,SAAS,EAAE,CAAC,KAAK,IAAI,EAAE,CAA4B;6BACpD,CAAC,CAAC;wBACL,CAAC;wBAAC,OAAO,KAAK,EAAE,CAAC;4BACf,IAAI,CAAC,QAAQ,CAAC,GAAG,CACf,UAAU,EACV,sCAAsC,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAC/F,CAAC;4BACF,MAAM,KAAK,CAAC;wBACd,CAAC;wBACD,IAAI,MAAM,CAAC,OAAO,KAAK,IAAI,EAAE,CAAC;4BAC5B,IAAI,CAAC,QAAQ,CAAC,GAAG,CACf,UAAU,EACV,+BAA+B,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,OAAO,CAAC,EAAE,CAChE,CAAC;wBACJ,CAAC;wBACD,OAAO,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC;oBAClC,CAAC;iBACF,CAAC;aACM,CAAC;QACb,CAAC,CAAC,CAAC;QACH,OAAO,MAAM,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC;IACrC,CAAC;CACF"}
+{"version":3,"file":"chat.js","sourceRoot":"","sources":["../src/chat.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,MAAM,EACN,6BAA6B,GAE9B,MAAM,8BAA8B,CAAC;AAEtC,OAAO,EACL,WAAW,EACX,YAAY,EACZ,UAAU,EAGV,WAAW,GACZ,MAAM,IAAI,CAAC;AASZ,MAAM,iBAAiB,GAAG,CAAC,CAAC;AAE5B,MAAM,cAAc,GAAG,sCAAsC,CAAC;AAW9D;;;GAGG;AACH,MAAM,OAAO,IAAI;IAIf,mEAAmE;IAC1D,SAAS,GAAoB,EAAE,CAAC;IAEzC,+EAA+E;IACtE,cAAc,GAAa,EAAE,CAAC;IAEtB,QAAQ,GAAmB,EAAE,CAAC;IAC9B,QAAQ,GAAG,IAAI,GAAG,EAAkB,CAAC;IACrC,MAAM,CAAS;IACf,SAAS,CAAgC;IACzC,IAAI,CAAa;IAC1B,KAAK,GAAW,EAAE,CAAC;IAE3B,YACE,MAAc,EACd,SAAwC,EACxC,IAAgB;QAEhB,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QACrB,IAAI,CAAC,SAAS,GAAG,SAAS,CAAC;QAC3B,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC;IACnB,CAAC;IAED,MAAM,CAAC,KAAK,CAAC,IAAI,CACf,IAAgB,EAChB,SAAyB;QAEzB,MAAM,MAAM,GAAG,IAAI,MAAM,CAAC,EAAE,IAAI,EAAE,gBAAgB,EAAE,OAAO,EAAE,GAAG,EAAE,CAAC,CAAC;QACpE,MAAM,SAAS,GAAG,IAAI,6BAA6B,CACjD,IAAI,GAAG,CAAC,cAAc,CAAC,EACvB,EAAE,KAAK,EAAE,SAAS,EAAE,CACrB,CAAC;QACF,MAAM,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;QAChC,MAAM,IAAI,GAAG,IAAI,IAAI,CAAM,MAAM,EAAE,SAAS,EAAE,IAAI,CAAC,CAAC;QACpD,IAAI,CAAC;YACH,MAAM,EAAE,KAAK,EAAE,GAAG,MAAM,MAAM,CAAC,SAAS,EAAE,CAAC;YAC3C,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACrB,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,SAAS,CAAC,KAAK,EAAE,CAAC;YACxB,MAAM,KAAK,CAAC;QACd,CAAC;QACD,OAAO,IAAI,CAAC;IACd,CAAC;IAED,6DAA6D;IAC7D,KAAK,CAAC,IAAI,CAAC,MAAc;QACvB,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;QAEtD,MAAM,MAAM,GAAG,MAAM,YAAY,CAAC;YAChC,KAAK,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK;YACtB,WAAW,EAAE,IAAI,CAAC,IAAI,CAAC,WAAW,IAAI,CAAC;YACvC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,YAAY,KAAK,SAAS;gBACtC,CAAC,CAAC,EAAE;gBACJ,CAAC,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,CAAC;YACvC,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,KAAK,EAAE,IAAI,CAAC,OAAO,EAAE;YACrB,QAAQ,EAAE,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,IAAI,iBAAiB,CAAC;SAC/D,CAAC,CAAC;QAEH,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,KAAK,EAAE,CAAC;YAChC,KAAK,MAAM,IAAI,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC;gBAClC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC;oBAClB,IAAI,EAAE,IAAI,CAAC,QAAQ;oBACnB,SAAS,EAAE,CAAC,IAAI,CAAC,KAAK,IAAI,EAAE,CAA4B;oBACxD,GAAG,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC;iBACR,CAAC,CAAC;YACtB,CAAC;QACH,CAAC;QAED,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,KAAK,EAAE,CAAC;YAChC,IAAI,IAAI,CAAC,IAAI,KAAK,EAAE,EAAE,CAAC;gBACrB,IAAI,CAAC,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACtC,CAAC;QACH,CAAC;QAED,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IAClD,CAAC;IAED,KAAK,CAAC,KAAK;QACT,MAAM,IAAI,CAAC,SAAS,CAAC,KAAK,EAAE,CAAC;IAC/B,CAAC;IAEO,UAAU,CAAC,IAA+C;QAGhE,MAAM,QAAQ,GAAG,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;QACpD,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;YAC3B,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC;QAC9B,CAAC;QACD,IAAI,IAAI,CAAC,OAAO,KAAK,IAAI,EAAE,CAAC;YAC1B,OAAO,EAAE,MAAM,EAAE,2CAA2C,EAAE,CAAC;QACjE,CAAC;QACD,OAAO,EAAE,CAAC;IACZ,CAAC;IAEO,OAAO;QACb,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,UAAU,EAAE,EAAE;YAC5C,MAAM,EAAE,OAAO,EAAE,QAAQ,EAAE,GAAG,UAAU,EAAE,GAAG,UAAU,CAAC,WAAW,CAAC;YACpE,OAAO;gBACL,UAAU,CAAC,IAAI;gBACf,WAAW,CAAC;oBACV,WAAW,EAAE,UAAU,CAAC,WAAW,IAAI,EAAE;oBACzC,WAAW,EAAE,UAAU,CACrB,UAA8C,CAC/C;oBACD,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,EAAE,UAAU,EAAE,EAAE,EAAE;wBACvC,IAAI,MAA+C,CAAC;wBACpD,IAAI,CAAC;4BACH,MAAM,GAAG,MAAM,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC;gCAClC,IAAI,EAAE,UAAU,CAAC,IAAI;gCACrB,SAAS,EAAE,CAAC,KAAK,IAAI,EAAE,CAA4B;6BACpD,CAAC,CAAC;wBACL,CAAC;wBAAC,OAAO,KAAK,EAAE,CAAC;4BACf,IAAI,CAAC,QAAQ,CAAC,GAAG,CACf,UAAU,EACV,sCAAsC,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAC/F,CAAC;4BACF,MAAM,KAAK,CAAC;wBACd,CAAC;wBACD,IAAI,MAAM,CAAC,OAAO,KAAK,IAAI,EAAE,CAAC;4BAC5B,IAAI,CAAC,QAAQ,CAAC,GAAG,CACf,UAAU,EACV,+BAA+B,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,OAAO,CAAC,EAAE,CAChE,CAAC;wBACJ,CAAC;wBACD,OAAO,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC;oBAClC,CAAC;iBACF,CAAC;aACM,CAAC;QACb,CAAC,CAAC,CAAC;QACH,OAAO,MAAM,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC;IACrC,CAAC;CACF"}

package/dist/index.d.ts

@@ -1,5 +1,5 @@
 export type { ChatAssertion, ChatMatchers, } from "@skybridge/vite-plugin/evals";
 export type { Chat } from "./chat.js";
-export { defineEvalModel } from "./model.js";
+export type { EvalApp, EvalIdentity } from "./session-registry.js";
 export { start } from "./session-registry.js";
 //# sourceMappingURL=index.d.ts.map

package/dist/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,YAAY,EACV,aAAa,EACb,YAAY,GACb,MAAM,8BAA8B,CAAC;AACtC,YAAY,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACtC,OAAO,EAAE,eAAe,EAAE,MAAM,YAAY,CAAC;AAC7C,OAAO,EAAE,KAAK,EAAE,MAAM,uBAAuB,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,YAAY,EACV,aAAa,EACb,YAAY,GACb,MAAM,8BAA8B,CAAC;AACtC,YAAY,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACtC,YAAY,EAAE,OAAO,EAAE,YAAY,EAAE,MAAM,uBAAuB,CAAC;AACnE,OAAO,EAAE,KAAK,EAAE,MAAM,uBAAuB,CAAC"}

package/dist/index.js

@@ -1,3 +1,2 @@
-export { defineEvalModel } from "./model.js";
 export { start } from "./session-registry.js";
 //# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAKA,OAAO,EAAE,eAAe,EAAE,MAAM,YAAY,CAAC;AAC7C,OAAO,EAAE,KAAK,EAAE,MAAM,uBAAuB,CAAC"}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAMA,OAAO,EAAE,KAAK,EAAE,MAAM,uBAAuB,CAAC"}

package/dist/session-registry.d.ts

@@ -1,16 +1,59 @@
+import type { LanguageModel } from "ai";
 import { Chat } from "./chat.js";
 /**
- * Opens a fresh MCP session and conversation against the server the plugin
- * started. The session is closed when the current test finishes, so scenarios
- * never leak sessions and never write teardown, and tests that run
- * concurrently cannot close each other's sessions.
- *
- * The type parameter pins the assertions to the project's registry:
- * `start<AppType>()` returns a `Chat<AppType>`, and `expect.chat` infers the
- * tool names and argument shapes from it.
+ * The minimum a Skybridge app has to expose to be served in-process: the
+ * `$types` marker the assertions infer tool names from, and the fetch handler
+ * the session is dialed through. Structural on purpose, so this package needs
+ * no runtime dependency on `skybridge`.
+ */
+export interface EvalApp {
+    readonly $types: {
+        readonly tools: object;
+    };
+    readonly fetchHandler: {
+        fetch: (request: Request, options?: {
+            authInfo?: EvalIdentity;
+        }) => Promise<Response>;
+    };
+}
+/**
+ * The identity an in-process session claims, handed to the app as the
+ * request's `authInfo`. Mirrors the SDK's `AuthInfo` structurally; `extra`
+ * carries whatever claims the app's verifier would have produced.
  */
-export declare function start<App>(overrides?: {
+export interface EvalIdentity {
+    token: string;
+    clientId: string;
+    scopes: string[];
+    expiresAt?: number;
+    resource?: URL;
+    extra?: Record<string, unknown>;
+}
+interface StartOptions {
+    /** Any AI SDK model instance, built by the project. */
+    model: LanguageModel;
     systemPrompt?: string;
     temperature?: number;
+    maxSteps?: number;
+}
+/**
+ * Opens a fresh MCP session and conversation against the app under test. The
+ * app is served in-process, and the session is closed when the current test
+ * finishes, so scenarios never leak sessions and never write teardown, and
+ * tests that run concurrently cannot close each other's sessions.
+ *
+ * The assertions are inferred from the app value itself, so `expect.chat` gets
+ * the project's tool names and argument shapes with no type parameter. The
+ * session talks straight to `app.fetchHandler`, which never closes: the
+ * handler is memoized on the app and shared by every test.
+ *
+ * `authInfo` claims an identity for the session: the app's per-tool scheme and
+ * scope enforcement runs against it for real, only token verification is
+ * skipped. Omit it to exercise the anonymous path, challenges included.
+ */
+export declare function start<App extends EvalApp>(options: StartOptions & {
+    app: App;
+    authInfo?: EvalIdentity;
 }): Promise<Chat<App>>;
+export {};
 //# sourceMappingURL=session-registry.d.ts.map

package/dist/session-registry.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"session-registry.d.ts","sourceRoot":"","sources":["../src/session-registry.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAGjC;;;;;;;;;GASG;AACH,wBAAsB,KAAK,CAAC,GAAG,EAAE,SAAS,CAAC,EAAE;IAC3C,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB,GAAG,OAAO,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAgBrB"}
+{"version":3,"file":"session-registry.d.ts","sourceRoot":"","sources":["../src/session-registry.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AAExC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAEjC;;;;;GAKG;AACH,MAAM,WAAW,OAAO;IACtB,QAAQ,CAAC,MAAM,EAAE;QAAE,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC;IAC5C,QAAQ,CAAC,YAAY,EAAE;QACrB,KAAK,EAAE,CACL,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE;YAAE,QAAQ,CAAC,EAAE,YAAY,CAAA;SAAE,KAClC,OAAO,CAAC,QAAQ,CAAC,CAAC;KACxB,CAAC;CACH;AAED;;;;GAIG;AACH,MAAM,WAAW,YAAY;IAC3B,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,QAAQ,CAAC,EAAE,GAAG,CAAC;IACf,KAAK,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACjC;AAED,UAAU,YAAY;IACpB,uDAAuD;IACvD,KAAK,EAAE,aAAa,CAAC;IACrB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAUD;;;;;;;;;;;;;;GAcG;AACH,wBAAsB,KAAK,CAAC,GAAG,SAAS,OAAO,EAC7C,OAAO,EAAE,YAAY,GAAG;IAAE,GAAG,EAAE,GAAG,CAAC;IAAC,QAAQ,CAAC,EAAE,YAAY,CAAA;CAAE,GAC5D,OAAO,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAoBpB"}

package/dist/session-registry.js

@@ -1,28 +1,37 @@
 import { inject, onTestFinished } from "vitest";
 import { Chat } from "./chat.js";
-import { resolveModel } from "./model.js";
+function sharedDefaults() {
+    try {
+        return inject("skybridgeEvals");
+    }
+    catch {
+        return undefined;
+    }
+}
 /**
- * Opens a fresh MCP session and conversation against the server the plugin
- * started. The session is closed when the current test finishes, so scenarios
- * never leak sessions and never write teardown, and tests that run
- * concurrently cannot close each other's sessions.
+ * Opens a fresh MCP session and conversation against the app under test. The
+ * app is served in-process, and the session is closed when the current test
+ * finishes, so scenarios never leak sessions and never write teardown, and
+ * tests that run concurrently cannot close each other's sessions.
  *
- * The type parameter pins the assertions to the project's registry:
- * `start<AppType>()` returns a `Chat<AppType>`, and `expect.chat` infers the
- * tool names and argument shapes from it.
+ * The assertions are inferred from the app value itself, so `expect.chat` gets
+ * the project's tool names and argument shapes with no type parameter. The
+ * session talks straight to `app.fetchHandler`, which never closes: the
+ * handler is memoized on the app and shared by every test.
+ *
+ * `authInfo` claims an identity for the session: the app's per-tool scheme and
+ * scope enforcement runs against it for real, only token verification is
+ * skipped. Omit it to exercise the anonymous path, challenges included.
  */
-export async function start(overrides) {
-    const config = inject("skybridgeEvals");
-    const url = inject("skybridgeEvalsUrl");
-    if (config === undefined || url === undefined) {
-        throw new Error("No eval server is running. Add `skybridge({ evals: {...} })` to the vitest config.");
-    }
-    const chat = await Chat.open(url, {
-        model: await resolveModel(config.model),
-        temperature: overrides?.temperature ?? config.temperature,
-        systemPrompt: overrides?.systemPrompt ?? config.systemPrompt,
-        maxSteps: config.maxSteps,
-    });
+export async function start(options) {
+    const config = sharedDefaults();
+    const { app, authInfo } = options;
+    const chat = await Chat.open({
+        model: options.model,
+        temperature: options.temperature ?? config?.temperature,
+        systemPrompt: options.systemPrompt ?? config?.systemPrompt,
+        maxSteps: options.maxSteps ?? config?.maxSteps,
+    }, (url, init) => app.fetchHandler.fetch(new Request(url, init), authInfo === undefined ? undefined : { authInfo }));
     onTestFinished(() => chat.close());
     return chat;
 }

package/dist/session-registry.js.map

@@ -1 +1 @@
-{"version":3,"file":"session-registry.js","sourceRoot":"","sources":["../src/session-registry.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,cAAc,EAAE,MAAM,QAAQ,CAAC;AAChD,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAE1C;;;;;;;;;GASG;AACH,MAAM,CAAC,KAAK,UAAU,KAAK,CAAM,SAGhC;IACC,MAAM,MAAM,GAAG,MAAM,CAAC,gBAAgB,CAAC,CAAC;IACxC,MAAM,GAAG,GAAG,MAAM,CAAC,mBAAmB,CAAC,CAAC;IACxC,IAAI,MAAM,KAAK,SAAS,IAAI,GAAG,KAAK,SAAS,EAAE,CAAC;QAC9C,MAAM,IAAI,KAAK,CACb,oFAAoF,CACrF,CAAC;IACJ,CAAC;IACD,MAAM,IAAI,GAAG,MAAM,IAAI,CAAC,IAAI,CAAM,GAAG,EAAE;QACrC,KAAK,EAAE,MAAM,YAAY,CAAC,MAAM,CAAC,KAAK,CAAC;QACvC,WAAW,EAAE,SAAS,EAAE,WAAW,IAAI,MAAM,CAAC,WAAW;QACzD,YAAY,EAAE,SAAS,EAAE,YAAY,IAAI,MAAM,CAAC,YAAY;QAC5D,QAAQ,EAAE,MAAM,CAAC,QAAQ;KAC1B,CAAC,CAAC;IACH,cAAc,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;IACnC,OAAO,IAAI,CAAC;AACd,CAAC"}
+{"version":3,"file":"session-registry.js","sourceRoot":"","sources":["../src/session-registry.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,MAAM,EAAE,cAAc,EAAE,MAAM,QAAQ,CAAC;AAChD,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAwCjC,SAAS,cAAc;IACrB,IAAI,CAAC;QACH,OAAO,MAAM,CAAC,gBAAgB,CAAC,CAAC;IAClC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,SAAS,CAAC;IACnB,CAAC;AACH,CAAC;AAED;;;;;;;;;;;;;;GAcG;AACH,MAAM,CAAC,KAAK,UAAU,KAAK,CACzB,OAA6D;IAE7D,MAAM,MAAM,GAAG,cAAc,EAAE,CAAC;IAChC,MAAM,EAAE,GAAG,EAAE,QAAQ,EAAE,GAAG,OAAO,CAAC;IAElC,MAAM,IAAI,GAAG,MAAM,IAAI,CAAC,IAAI,CAC1B;QACE,KAAK,EAAE,OAAO,CAAC,KAAK;QACpB,WAAW,EAAE,OAAO,CAAC,WAAW,IAAI,MAAM,EAAE,WAAW;QACvD,YAAY,EAAE,OAAO,CAAC,YAAY,IAAI,MAAM,EAAE,YAAY;QAC1D,QAAQ,EAAE,OAAO,CAAC,QAAQ,IAAI,MAAM,EAAE,QAAQ;KAC/C,EACD,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,CACZ,GAAG,CAAC,YAAY,CAAC,KAAK,CACpB,IAAI,OAAO,CAAC,GAAG,EAAE,IAAI,CAAC,EACtB,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,CAClD,CACJ,CAAC;IAEF,cAAc,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;IACnC,OAAO,IAAI,CAAC;AACd,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@skybridge/test",
-  "version": "0.0.0-dev.c19b437",
+  "version": "0.0.0-dev.c2c0f98",
   "description": "Eval runner for Skybridge MCP apps: does a prompt reach the right tool with the right arguments",
   "license": "ISC",
   "type": "module",
@@ -12,58 +12,31 @@
     ".": {
       "types": "./dist/index.d.ts",
       "default": "./dist/index.js"
-    },
-    "./vitest-config": {
-      "types": "./dist/vitest-config.d.ts",
-      "default": "./dist/vitest-config.js"
     }
   },
   "files": [
     "dist"
   ],
   "dependencies": {
-    "@modelcontextprotocol/sdk": "^1.29.0",
-    "ai": "^5.0.0",
-    "@skybridge/vite-plugin": "0.0.0-dev.c19b437"
+    "@modelcontextprotocol/client": "^2.0.0",
+    "@skybridge/vite-plugin": "0.0.0-dev.c2c0f98"
   },
   "peerDependencies": {
-    "@ai-sdk/anthropic": "^2.0.0",
-    "@ai-sdk/google": "^2.0.0",
-    "@ai-sdk/mistral": "^2.0.0",
-    "@ai-sdk/openai": "^2.0.0",
+    "ai": "^5.0.0",
     "skybridge": "^2.0.0",
     "vitest": "^4.1.0"
   },
-  "peerDependenciesMeta": {
-    "@ai-sdk/anthropic": {
-      "optional": true
-    },
-    "@ai-sdk/google": {
-      "optional": true
-    },
-    "@ai-sdk/mistral": {
-      "optional": true
-    },
-    "@ai-sdk/openai": {
-      "optional": true
-    }
-  },
   "devDependencies": {
-    "@ai-sdk/anthropic": "^2.0.0",
-    "@ai-sdk/google": "^2.0.0",
-    "@ai-sdk/mistral": "^2.0.0",
-    "@ai-sdk/openai": "^2.0.0",
     "@types/node": "^24.13.3",
+    "ai": "^5.0.0",
     "typescript": "^5.9.3",
     "vitest": "^4.1.10",
-    "skybridge": "0.0.0-dev.c19b437"
+    "zod": "^4.4.3",
+    "skybridge": "0.0.0-dev.c2c0f98"
   },
   "engines": {
     "node": ">=22.23.1"
   },
-  "publishConfig": {
-    "tag": "beta"
-  },
   "scripts": {
     "build": "tsc -p tsconfig.json",
     "format": "biome check --write --error-on-warnings",

package/dist/model.d.ts

@@ -1,15 +0,0 @@
-import type { LanguageModel } from "ai";
-/**
- * Registers a model instance from a setup file, taking precedence over the
- * configured `model` string. The escape hatch for anything the provider
- * prefixes cannot express, such as a local runtime or a company LLM gateway.
- */
-export declare function defineEvalModel(model: LanguageModel): void;
-/**
- * Turns a `provider/model-id` string into a model instance, importing only the
- * provider named. Each provider package reads its own conventional API key
- * variable, so no key configuration is needed. An unrecognised prefix is
- * returned untouched for the AI SDK's default provider to resolve.
- */
-export declare function resolveModel(spec: string | undefined): Promise<LanguageModel>;
-//# sourceMappingURL=model.d.ts.map

package/dist/model.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"model.d.ts","sourceRoot":"","sources":["../src/model.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AAaxC;;;;GAIG;AACH,wBAAgB,eAAe,CAAC,KAAK,EAAE,aAAa,GAAG,IAAI,CAE1D;AAED;;;;;GAKG;AACH,wBAAsB,YAAY,CAChC,IAAI,EAAE,MAAM,GAAG,SAAS,GACvB,OAAO,CAAC,aAAa,CAAC,CAwBxB"}

package/dist/model.js

@@ -1,43 +0,0 @@
-const PROVIDERS = {
-    anthropic: () => import("@ai-sdk/anthropic").then((m) => m.anthropic),
-    google: () => import("@ai-sdk/google").then((m) => m.google),
-    mistral: () => import("@ai-sdk/mistral").then((m) => m.mistral),
-    openai: () => import("@ai-sdk/openai").then((m) => m.openai),
-};
-let registered;
-/**
- * Registers a model instance from a setup file, taking precedence over the
- * configured `model` string. The escape hatch for anything the provider
- * prefixes cannot express, such as a local runtime or a company LLM gateway.
- */
-export function defineEvalModel(model) {
-    registered = model;
-}
-/**
- * Turns a `provider/model-id` string into a model instance, importing only the
- * provider named. Each provider package reads its own conventional API key
- * variable, so no key configuration is needed. An unrecognised prefix is
- * returned untouched for the AI SDK's default provider to resolve.
- */
-export async function resolveModel(spec) {
-    if (registered !== undefined) {
-        return registered;
-    }
-    if (spec === undefined) {
-        throw new Error("No model configured. Set `model` on the evals plugin, or call defineEvalModel() from a setup file.");
-    }
-    const separator = spec.indexOf("/");
-    if (separator === -1) {
-        return spec;
-    }
-    const name = spec.slice(0, separator);
-    const load = PROVIDERS[name];
-    if (load === undefined) {
-        return spec;
-    }
-    const provider = await load().catch(() => {
-        throw new Error(`The model "${spec}" needs @ai-sdk/${name}@^2. Install it as a dev dependency.`);
-    });
-    return provider(spec.slice(separator + 1));
-}
-//# sourceMappingURL=model.js.map

package/dist/model.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"model.js","sourceRoot":"","sources":["../src/model.ts"],"names":[],"mappings":"AAIA,MAAM,SAAS,GAAmD;IAChE,SAAS,EAAE,GAAG,EAAE,CAAC,MAAM,CAAC,mBAAmB,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;IACrE,MAAM,EAAE,GAAG,EAAE,CAAC,MAAM,CAAC,gBAAgB,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC;IAC5D,OAAO,EAAE,GAAG,EAAE,CAAC,MAAM,CAAC,iBAAiB,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC;IAC/D,MAAM,EAAE,GAAG,EAAE,CAAC,MAAM,CAAC,gBAAgB,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC;CAC7D,CAAC;AAEF,IAAI,UAAqC,CAAC;AAE1C;;;;GAIG;AACH,MAAM,UAAU,eAAe,CAAC,KAAoB;IAClD,UAAU,GAAG,KAAK,CAAC;AACrB,CAAC;AAED;;;;;GAKG;AACH,MAAM,CAAC,KAAK,UAAU,YAAY,CAChC,IAAwB;IAExB,IAAI,UAAU,KAAK,SAAS,EAAE,CAAC;QAC7B,OAAO,UAAU,CAAC;IACpB,CAAC;IACD,IAAI,IAAI,KAAK,SAAS,EAAE,CAAC;QACvB,MAAM,IAAI,KAAK,CACb,oGAAoG,CACrG,CAAC;IACJ,CAAC;IACD,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IACpC,IAAI,SAAS,KAAK,CAAC,CAAC,EAAE,CAAC;QACrB,OAAO,IAAI,CAAC;IACd,CAAC;IACD,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC,CAAC;IACtC,MAAM,IAAI,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC;IAC7B,IAAI,IAAI,KAAK,SAAS,EAAE,CAAC;QACvB,OAAO,IAAI,CAAC;IACd,CAAC;IACD,MAAM,QAAQ,GAAG,MAAM,IAAI,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE;QACvC,MAAM,IAAI,KAAK,CACb,cAAc,IAAI,mBAAmB,IAAI,sCAAsC,CAChF,CAAC;IACJ,CAAC,CAAC,CAAC;IACH,OAAO,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC;AAC7C,CAAC"}