@wix/evalforge-evaluator 0.199.0 → 0.201.0

package/build/index.mjs

@@ -7401,7 +7401,7 @@ import {
 import { createHash } from "crypto";
 import path from "path";
 import { spawn, execFileSync } from "child_process";
-var INSTALL_TIMEOUT_MS = 1e5;
+var INSTALL_TIMEOUT_MS = 12e4;
 var HEARTBEAT_INTERVAL_MS = 5e3;
 function reportRegistry(workDir, onProgress) {
   try {
@@ -11080,22 +11080,38 @@ async function executeWithAiSdk(context) {
       ...topLevelExtras,
       model,
       abortSignal: abortController.signal,
-      system: systemPrompt,
-      ...hasImages ? {
-        messages: [
+      // Anthropic prompt caching: the Vercel AI SDK can't attach cache_control
+      // to the top-level `system` string, so the system prompt is sent as a
+      // `system` message carrying an ephemeral cache breakpoint (Anthropic only).
+      // Anthropic renders tools -> system -> messages, so a breakpoint on the
+      // system block caches the tool definitions too — high value for MCP
+      // scenarios that share a large tool set + system prompt across runs.
+      messages: [
+        ...systemPrompt ? [
           {
-            role: "user",
-            content: [
-              { type: "text", text: scenario.triggerPrompt },
-              ...triggerPromptImages.map((img) => ({
-                type: "image",
-                image: img.base64,
-                mediaType: img.mediaType
-              }))
-            ]
+            role: "system",
+            content: systemPrompt,
+            ...isAnthropic ? {
+              providerOptions: {
+                anthropic: {
+                  cacheControl: { type: "ephemeral" }
+                }
+              }
+            } : {}
           }
-        ]
-      } : { prompt: scenario.triggerPrompt },
+        ] : [],
+        {
+          role: "user",
+          content: hasImages ? [
+            { type: "text", text: scenario.triggerPrompt },
+            ...triggerPromptImages.map((img) => ({
+              type: "image",
+              image: img.base64,
+              mediaType: img.mediaType
+            }))
+          ] : scenario.triggerPrompt
+        }
+      ],
       temperature: supportsThinking ? void 0 : cfg.temperature,
       topP: supportsThinking ? void 0 : cfg.topP,
       frequencyPenalty: cfg.frequencyPenalty,