fix(eve): keep agent router models serializable

Signed-off-by: Casey Gowrie <ctgowrie@gmail.com>

docs/tools/workflows.mdx

@@ -426,7 +426,7 @@ const target = await auto({
 });
 ```
 
-`message` and `agents` are required. `model`, `instructions`, and `abortSignal` are optional; `model` accepts an evaluation model instance or ID, and the helper defaults to `typesafe-ai/jev` and `"Which subagent should handle this task?"`. It trims descriptions, ignores empty entries, skips evaluation for a sole candidate, and returns the selected agent name.
+`message` and `agents` are required. `model`, `instructions`, and `abortSignal` are optional; `model` accepts a serializable evaluation model ID, and the helper defaults to `typesafe-ai/jev` and `"Which subagent should handle this task?"`. It trims descriptions, ignores empty entries, skips evaluation for a sole candidate, and returns the selected agent name.
 
 `ctx.agents` is a replay-stable metadata snapshot taken when the workflow starts. In a top-level root workflow, `ctx.agents.agent` always represents the root-copy target and carries the root's authored `description` when provided. The snapshot also includes declared agents hidden from the parent model with `tool: false` or `disableTool()`, but exposes no model definitions, credentials, or callbacks. A delegated root copy omits `agent`, so `agentRouter()` cannot select another root copy recursively. Invocation still checks each target's availability through `ctx.agent()`.
 
@@ -440,7 +440,7 @@ import { agentRouter } from "eve/tools/agent-router";
 export default agentRouter();
 ```
 
-Pass `model` or `instructions` to customize the provided router. Because factory options become durable workflow configuration, `model` must be a serializable model ID; use `auto()` directly when you need an evaluation model instance:
+Pass `model` or `instructions` to customize the provided router:
 
 ```ts title="agent/tools/agent.ts"
 import { agentRouter } from "eve/tools/agent-router";
@@ -451,6 +451,8 @@ export default agentRouter({
 });
 ```
 
+Both `auto()` and `agentRouter()` accept model IDs only because their configuration crosses durable workflow boundaries. To use a live evaluation model instance, write your own `"use step"` helper and create or import the model inside that step instead of passing it as an argument.
+
 Its input is `{ message: string, outputSchema?: object }`. `agentRouter()` ignores entries without a non-empty description. With two or more described targets, it sends the message and effective descriptions to JEV, then invokes the selected name through `ctx.agent()`. It invokes a sole target without evaluation and forwards an optional `outputSchema` unchanged. In a top-level root session, the candidates include the root-copy `agent`; delegated root copies and declared subagent contexts include only their declared targets.
 
 When an authored tool accepts a JSON Schema supplied by the model, represent that input with a permissive object such as `z.looseObject({})`. Avoid `z.record(z.string(), z.json())`: its generated JSON Schema uses `propertyNames`, which OpenAI does not support. Validate the supplied value at the point where your tool consumes it.

packages/eve/extension-contracts/reports/tool/v53.json

@@ -2,7 +2,7 @@
   "kind": "eve-extension-capability-contract",
   "capability": "tool",
   "epoch": 53,
-  "sha256": "cec3f516eccdd2bf4b08ec7bc75048b29cb50233e9effa983707b232b3282455",
+  "sha256": "f591e6389c338895fbb342d81d10507d1b3374fef6a5ce2dd90efe84e9fdd856",
   "exports": [
     "AgentRouterAutoOptions",
     "AgentRouterInput",

packages/eve/src/execution/tools/agent-router-workflow.ts

@@ -46,7 +46,7 @@ export async function auto({
   if (typeof instructions !== "string" || instructions.trim().length === 0) {
     throw new Error("agentRouter auto requires non-empty instructions when provided.");
   }
-  if (typeof model === "string" && model.trim().length === 0) {
+  if (typeof model !== "string" || model.trim().length === 0) {
     throw new Error("agentRouter auto requires a non-empty model ID when provided.");
   }
   const criteria = Object.fromEntries(

packages/eve/src/execution/tools/agent-router.ts

@@ -1,22 +1,11 @@
 import { z } from "#compiled/zod/index.js";
 import type { JsonObject } from "#shared/json.js";
-import type { evaluate } from "#ai/evaluate.js";
 
 export { auto, executeAgentRouterTool } from "#execution/tools/agent-router-workflow.js";
 
 export const AGENT_ROUTER_TOOL_DESCRIPTION =
   "Route a task to the best available subagent based on each subagent's declared description.";
 
-export interface AgentRouterAutoOptions {
-  readonly abortSignal?: AbortSignal;
-  readonly agents: Readonly<Record<string, string>>;
-  /** Instructions used to select an agent. */
-  readonly instructions?: string;
-  readonly message: string;
-  /** Evaluation model instance or ID. Defaults to TypeSafe Jev. */
-  readonly model?: Parameters<typeof evaluate>[0]["model"];
-}
-
 export interface AgentRouterOptions {
   /** Instructions used to select an agent. */
   readonly instructions?: string;
@@ -24,6 +13,12 @@ export interface AgentRouterOptions {
   readonly model?: string;
 }
 
+export interface AgentRouterAutoOptions extends AgentRouterOptions {
+  readonly abortSignal?: AbortSignal;
+  readonly agents: Readonly<Record<string, string>>;
+  readonly message: string;
+}
+
 export interface AgentRouterInput {
   readonly message: string;
   readonly outputSchema?: JsonObject;

packages/eve/src/tools/provided/agent-router.test.ts

@@ -184,6 +184,10 @@ describe("agentRouter", () => {
       error: "agentRouter auto requires a non-empty model ID when provided.",
       options: { agents: { researcher: "Research" }, message: "Research", model: "" },
     },
+    {
+      error: "agentRouter auto requires a non-empty model ID when provided.",
+      options: { agents: { researcher: "Research" }, message: "Research", model: {} },
+    },
     {
       error: 'agentRouter auto requires a string description for agent "researcher".',
       options: { agents: { researcher: 42 }, message: "Research" },