diff --git a/src/resources/extensions/sf/escalation.ts b/src/resources/extensions/sf/escalation.ts
index 63ddd03fc..9e198343e 100644
--- a/src/resources/extensions/sf/escalation.ts
+++ b/src/resources/extensions/sf/escalation.ts
@@ -142,6 +142,38 @@ export function writeEscalationArtifact(
 		}),
 	);
 
+	// continueWithDefault=true: the agent already proceeded with its
+	// recommendation. The artifact is the audit trail, but the choice +
+	// rationale should ALSO land in the memory store so future tasks see
+	// it via getRelevantMemoriesRanked — otherwise the rationale lives
+	// only in the JSON artifact and never reaches downstream prompts.
+	// resolveEscalation handles this for the continueWithDefault=false
+	// path; we do the symmetric write here. Best-effort.
+	if (artifact.continueWithDefault) {
+		try {
+			const recommendedOption = artifact.options.find(
+				(o) => o.id === artifact.recommendation,
+			);
+			const memoryContent = formatEscalationMemoryContent(
+				artifact,
+				recommendedOption,
+				`auto-applied default: ${artifact.recommendationRationale}`,
+			);
+			createMemory({
+				category: "architecture",
+				content: memoryContent,
+				confidence: 0.85,
+				source_unit_type: "execute-task",
+				source_unit_id: artifact.taskId,
+			});
+		} catch (memoryErr) {
+			logWarning(
+				"tool",
+				`escalation: continueWithDefault memory write failed: ${(memoryErr as Error).message}`,
+			);
+		}
+	}
+
 	return path;
 }