enable dom agent

Author: miguelg719

.env.example

@@ -11,9 +11,3 @@ EXPERIMENTAL_EVAL_MODELS="gpt-4o,claude-3-5-sonnet-latest,o1-mini,o1-preview"
 EVAL_CATEGORIES="observe,act,combination,extract,experimental"
 AGENT_EVAL_MAX_STEPS=50
 STAGEHAND_API_URL="http://localhost:80"
-
-# Screenshot collector thresholds for determining when screenshots are too similar
-# SSIM (Structural Similarity Index) threshold: 0-1, higher = more similar (default: 0.85)
-SCREENSHOT_SSIM_THRESHOLD=0.85
-# MSE (Mean Squared Error) threshold: lower = more similar (default: 200)
-SCREENSHOT_MSE_THRESHOLD=200

evals/taskConfig.ts

@@ -109,7 +109,7 @@ const DEFAULT_AGENT_MODELS = process.env.EVAL_AGENT_MODELS
   : [
       "computer-use-preview-2025-03-11",
       "claude-sonnet-4-20250514",
-      // "anthropic/claude-sonnet-4-20250514",
+      "anthropic/claude-sonnet-4-20250514",
     ];
 
 /**

evals/tasks/agent/onlineMind2Web.ts

@@ -40,17 +40,11 @@ export const onlineMind2Web: EvalFunction = async ({
       timeout: 75_000,
     });
 
-    if (!(modelName in modelToAgentProviderMap)) {
-      return {
-        _success: false,
-        error: `Model ${modelName} is not supported for agent tasks. Supported models: ${Object.keys(modelToAgentProviderMap).join(", ")}`,
-        debugUrl,
-        sessionUrl,
-        logs: logger.getLogs(),
-      };
-    }
+    const provider =
+      modelName in modelToAgentProviderMap
+        ? modelToAgentProviderMap[modelName]
+        : undefined;
 
-    const provider = modelToAgentProviderMap[modelName];
     const agent = stagehand.agent({
       model: modelName,
       provider,
@@ -72,7 +66,7 @@ export const onlineMind2Web: EvalFunction = async ({
 
     screenshotCollector.start();
 
-    const maxSteps = Number(process.env.AGENT_EVAL_MAX_STEPS) || 50;
+    const maxSteps = Number(process.env.AGENT_EVAL_MAX_STEPS) || 5;
     const agentResult = await agent.execute({
       instruction: params.confirmed_task,
       maxSteps: maxSteps,