@sanity/ailf 0.1.20 → 0.1.21

package/dist/pipeline/calculate-scores.js

@@ -324,13 +324,25 @@ function calculateScores(resultsPath, weights) {
 /**
  * Extracts agent behavior summary from a test result's metadata.
  * Returns null if the test was not run with the instrumented provider.
+ *
+ * Checks two locations because Promptfoo may flatten/merge the metadata
+ * object differently than how the provider originally set it:
+ *   1. metadata.agentBehaviorSummary (set directly by AgenticProvider)
+ *   2. metadata.agentBehavior.summary (nested in the full behavior log)
  */
 function extractAgentBehavior(test) {
     const { metadata } = test;
-    if (!metadata?.agentBehaviorSummary) {
-        return null;
+    if (metadata?.agentBehaviorSummary) {
+        return metadata.agentBehaviorSummary;
+    }
+    // Fallback: Promptfoo may drop the top-level agentBehaviorSummary
+    // field during serialization, but the data is nested inside the
+    // full agentBehavior log.
+    const behavior = metadata?.agentBehavior;
+    if (behavior?.summary) {
+        return behavior.summary;
     }
-    return metadata.agentBehaviorSummary;
+    return null;
 }
 /**
  * Extracts grader (assertion) cost from the raw Promptfoo results file.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sanity/ailf",
-  "version": "0.1.20",
+  "version": "0.1.21",
   "private": false,
   "publishConfig": {
     "access": "restricted"