diff --git a/src/adapter/openai-to-cli.ts b/src/adapter/openai-to-cli.ts
index c8ecaa1..a285c24 100644
--- a/src/adapter/openai-to-cli.ts
+++ b/src/adapter/openai-to-cli.ts
@@ -46,6 +46,28 @@ export function extractModel(model: string): ClaudeModel {
   return "opus";
 }
 
+/**
+ * Normalize message content to a plain string.
+ *
+ * The OpenAI chat completions API allows `content` to be either a string or
+ * an array of content parts (e.g. [{type: "text", text: "Hello"}]).  Clients
+ * such as OpenClaw always send the array form, so we need to handle both.
+ */
+function extractContent(content: string | Array<{type: string; text?: string}> | unknown): string {
+  if (typeof content === "string") return content;
+  if (Array.isArray(content)) {
+    return content
+      .map(part => {
+        if (typeof part === "string") return part;
+        if (part && part.type === "text" && typeof part.text === "string") return part.text;
+        return "";
+      })
+      .filter(Boolean)
+      .join("\n");
+  }
+  return String(content ?? "");
+}
+
 /**
  * Convert OpenAI messages array to a single prompt string for Claude CLI
  *
@@ -59,17 +81,17 @@ export function messagesToPrompt(messages: OpenAIChatRequest["messages"]): strin
     switch (msg.role) {
       case "system":
         // System messages become context instructions
-        parts.push(`<system>\n${msg.content}\n</system>\n`);
+        parts.push(`<system>\n${extractContent(msg.content)}\n</system>\n`);
         break;
 
       case "user":
         // User messages are the main prompt
-        parts.push(msg.content);
+        parts.push(extractContent(msg.content));
         break;
 
       case "assistant":
         // Previous assistant responses for context
-        parts.push(`<previous_response>\n${msg.content}\n</previous_response>\n`);
+        parts.push(`<previous_response>\n${extractContent(msg.content)}\n</previous_response>\n`);
         break;
     }
   }