@hue-run/sdk 0.7.0 → 0.8.0

package/dist/evals/runner.d.ts

@@ -50,7 +50,7 @@ interface RunnerOptions {
      * (for example a Hue-operated grading worker that owns the evaluator source) instead of
      * refusing the run. Their IDs are reported in `deferredScorerVersionIds`. */
     deferUnboundLocalScorers?: boolean;
-    /** Cases in flight at once, 1–16. Default 1. */
+    /** Cases in flight at once, 1–64. Default 1. */
     concurrency?: number;
     /** Deadline for JSON Schema scoring in its worker, 100–60000 ms. Default 2000. */
     schemaTimeoutMillis?: number;

package/dist/evals/runner.js

@@ -80,8 +80,8 @@ function settings(options) {
     if (typeof options.persistResultContent !== "boolean")
         throw new TypeError("Choose persistResultContent explicitly: true or false");
     const concurrency = options.concurrency ?? 1;
-    if (!Number.isInteger(concurrency) || concurrency < 1 || concurrency > 16)
-        throw new RangeError("concurrency must be 1–16");
+    if (!Number.isInteger(concurrency) || concurrency < 1 || concurrency > 64)
+        throw new RangeError("concurrency must be 1–64");
     const timeout = options.schemaTimeoutMillis ?? 2000;
     if (!Number.isInteger(timeout) || timeout < 100 || timeout > 60_000)
         throw new RangeError("schemaTimeoutMillis must be 100–60000");

package/dist/evals/simulation.d.ts

@@ -1,7 +1,7 @@
 import type { HueClient } from "../client.js";
 import { type EnvironmentClient } from "../environment/client.js";
 import type { EnvironmentTool } from "../environment/tools.js";
-import type { EnvironmentIdentity, PublishableEnvironmentDefinition } from "../environment/types.js";
+import type { EnvironmentIdentity, PublishableEnvironmentDefinition, WorldHandoff } from "../environment/types.js";
 import { type EvaluationClient } from "./client.js";
 import type { ActualAgentManifestInputV2, AttemptConnectionBundleV2, RequestedAttemptProviderV2 } from "./attempt.js";
 import { type RunnerReport } from "./runner.js";
@@ -109,10 +109,15 @@ export interface SimulationTargetContext {
     executionId: string;
     /** Stable world identity for adapter control operations such as coverage reporting. */
     environmentRunId: string;
-    /** Framework-neutral local callables backed by this attempt's isolated world. */
+    /** Framework-neutral local callables backed by this attempt's isolated world; empty for a
+     * gateway world, whose calls go to the provider mirrors in `world`. */
     tools: Record<string, EnvironmentTool>;
-    /** Short-lived capability for providers that execute MCP remotely. */
-    mcp: SimulationMcpCapability;
+    /** The mirror URLs, world token, environment carriers and MCP configuration of a gateway
+     * world; absent for a world created while the gateway was off. */
+    world?: WorldHandoff;
+    /** One MCP endpoint and bearer: the gateway world's first MCP mirror with the world token, or
+     * the deprecated execution-scoped `hue_sim_` capability of a legacy world. */
+    mcp?: SimulationMcpCapability;
     /** Credential-bearing provider connections for this callback only. Hue never
      * checkpoints, logs or adds this response to parity digests. */
     connectionBundle?: AttemptConnectionBundleV2;
@@ -149,7 +154,7 @@ export interface RunSimulationOptions {
     };
     /** Local scorer callbacks bound by their declared source digests. */
     localScorers?: LocalScorer[];
-    /** Cases in flight, 1–16; defaults to 1. */
+    /** Cases in flight, 1–64; defaults to 1. */
     concurrency?: number;
     /** JSON Schema worker deadline in milliseconds. */
     schemaTimeoutMillis?: number;
@@ -157,6 +162,10 @@ export interface RunSimulationOptions {
     maxSteps?: number;
     /** Per-world lease in seconds, 1–86400. */
     ttlSeconds?: number;
+    /** The agent revision under test, sent on world create for the world's fingerprint. */
+    agentRevision?: string;
+    /** Emit a one-time `DeprecationWarning` when the deployment serves a legacy world; on by default. */
+    deprecationWarnings?: boolean;
     /** Cooperative caller cancellation signal. */
     signal?: AbortSignal;
     /** Optional display name for the fresh experiment. */

package/dist/evals/simulation.js

@@ -378,10 +378,21 @@ export async function runSimulation(options) {
         baseUrl: options.client.baseUrl,
     });
     try {
-        const scenarioDigest = digest(definitionIdentity(definition));
+        // The agent revision is part of the attempt: a resume after it changed would otherwise
+        // finish the remaining cases under a different revision than the completed ones. A
+        // checkpoint written before revisions were tracked carries the definition digest alone and
+        // still matches its definition.
+        const scenarioDigest = digest({
+            definition: definitionIdentity(definition),
+            agentRevision: options.agentRevision ?? null,
+        });
         let attempt = await store.read("active-attempt");
-        if (attempt && attempt.stage !== "completed" && attempt.scenarioDigest !== scenarioDigest)
-            throw new Error("Recover the unfinished simulation before running a changed scenario");
+        const previousDigest = digest(definitionIdentity(definition));
+        if (attempt &&
+            attempt.stage !== "completed" &&
+            attempt.scenarioDigest !== scenarioDigest &&
+            attempt.scenarioDigest !== previousDigest)
+            throw new Error("Recover the unfinished simulation before running a changed scenario or agent revision");
         if (!attempt || attempt.stage === "completed") {
             attempt = {
                 scenarioDigest,
@@ -432,6 +443,8 @@ export async function runSimulation(options) {
                 requested,
                 maxSteps: options.maxSteps,
                 ttlSeconds: options.ttlSeconds,
+                agentRevision: options.agentRevision,
+                deprecationWarnings: options.deprecationWarnings,
                 signal: options.signal,
                 onProgress: (event) => options.onProgress?.({
                     ...event,
@@ -448,11 +461,16 @@ export async function runSimulation(options) {
                     executionId: targetContext.executionId,
                     environmentRunId: targetContext.environmentRunId,
                     tools: targetContext.tools,
-                    mcp: {
-                        url: targetContext.mcp.url,
-                        token: targetContext.mcp.token,
-                        expiresAt: targetContext.mcp.expiresAt,
-                    },
+                    ...(targetContext.world ? { world: structuredClone(targetContext.world) } : {}),
+                    ...(targetContext.mcp
+                        ? {
+                            mcp: {
+                                url: targetContext.mcp.url,
+                                token: targetContext.mcp.token,
+                                expiresAt: targetContext.mcp.expiresAt,
+                            },
+                        }
+                        : {}),
                     ...(targetContext.connectionBundle
                         ? { connectionBundle: structuredClone(targetContext.connectionBundle) }
                         : {}),

package/dist/version.d.ts

@@ -1,2 +1,2 @@
 /** Package version shared by the instrumentation scope and the export User-Agent. */
-export declare const sdkVersion = "0.7.0";
+export declare const sdkVersion = "0.8.0";

package/dist/version.js

@@ -1,3 +1,3 @@
 // Generated by scripts/write-version.mjs from package.json; do not edit by hand.
 /** Package version shared by the instrumentation scope and the export User-Agent. */
-export const sdkVersion = "0.7.0";
+export const sdkVersion = "0.8.0";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hue-run/sdk",
-  "version": "0.7.0",
+  "version": "0.8.0",
   "private": false,
   "license": "MIT",
   "publishConfig": {