Fix InstructionsType decoding for reusable prompts (issue #187)

Sources/OpenAI/Public/ResponseModels/Response/ResponseModel.swift

@@ -54,10 +54,11 @@ public struct ResponseModel: Decodable {
     }
   }
 
-  /// Instructions type - can be a string or an array of strings
+  /// Instructions type - can be a string, an array of strings, or an array of messages (for reusable prompts)
   public enum InstructionsType: Decodable {
     case string(String)
     case array([String])
+    case messages([InputMessage])
 
     public init(from decoder: Decoder) throws {
       let container = try decoder.singleValueContainer()
@@ -66,10 +67,12 @@ public struct ResponseModel: Decodable {
         self = .string(stringValue)
       } else if let arrayValue = try? container.decode([String].self) {
         self = .array(arrayValue)
+      } else if let messagesValue = try? container.decode([InputMessage].self) {
+        self = .messages(messagesValue)
       } else {
         throw DecodingError.dataCorruptedError(
           in: container,
-          debugDescription: "Expected String or [String] for instructions")
+          debugDescription: "Expected String, [String], or [InputMessage] for instructions")
       }
     }
   }

Tests/OpenAITests/ResponseModelValidationTests.swift

@@ -194,6 +194,62 @@ final class ResponseModelValidationTests: XCTestCase {
     XCTAssertEqual(responseModel.usage?.outputTokens, 1035)
   }
 
+  // MARK: - InstructionsType Tests
+
+  func testInstructionsTypeStringDecoding() throws {
+    let decoder = JSONDecoder()
+    let responseModel = try decoder.decode(ResponseModel.self, from: instructionsStringJSON.data(using: .utf8)!)
+
+    XCTAssertNotNil(responseModel.instructions)
+    if case .string(let value) = responseModel.instructions {
+      XCTAssertEqual(value, "You are a helpful assistant.")
+    } else {
+      XCTFail("Expected string instructions type")
+    }
+  }
+
+  func testInstructionsTypeArrayOfStringsDecoding() throws {
+    let decoder = JSONDecoder()
+    let responseModel = try decoder.decode(ResponseModel.self, from: instructionsArrayOfStringsJSON.data(using: .utf8)!)
+
+    XCTAssertNotNil(responseModel.instructions)
+    if case .array(let values) = responseModel.instructions {
+      XCTAssertEqual(values.count, 2)
+      XCTAssertEqual(values[0], "Be helpful.")
+      XCTAssertEqual(values[1], "Be concise.")
+    } else {
+      XCTFail("Expected array of strings instructions type")
+    }
+  }
+
+  func testInstructionsTypeMessagesDecoding() throws {
+    // This tests the fix for issue #187 - reusable prompts return instructions as message objects
+    let decoder = JSONDecoder()
+    let responseModel = try decoder.decode(ResponseModel.self, from: instructionsMessagesJSON.data(using: .utf8)!)
+
+    XCTAssertNotNil(responseModel.instructions)
+    if case .messages(let messages) = responseModel.instructions {
+      XCTAssertEqual(messages.count, 2)
+      XCTAssertEqual(messages[0].role, "developer")
+      XCTAssertEqual(messages[0].type, "message")
+      XCTAssertEqual(messages[1].role, "assistant")
+
+      // Validate content of first message
+      if case .array(let contentItems) = messages[0].content {
+        XCTAssertEqual(contentItems.count, 1)
+        if case .text(let textContent) = contentItems[0] {
+          XCTAssertEqual(textContent.text, "You are a helpful assistant for {{customer_name}}.")
+        } else {
+          XCTFail("Expected text content item")
+        }
+      } else {
+        XCTFail("Expected array content in message")
+      }
+    } else {
+      XCTFail("Expected messages instructions type")
+    }
+  }
+
   // MARK: - Test Data
 
   private let textInputResponseJSON = """
@@ -679,4 +735,114 @@ final class ResponseModelValidationTests: XCTestCase {
       "metadata": {}
     }
     """
+
+  // MARK: - InstructionsType Test Data
+
+  private let instructionsStringJSON = """
+    {
+      "id": "resp_test_string_instructions",
+      "object": "response",
+      "created_at": 1741476542,
+      "status": "completed",
+      "error": null,
+      "incomplete_details": null,
+      "instructions": "You are a helpful assistant.",
+      "max_output_tokens": null,
+      "model": "gpt-4.1-2025-04-14",
+      "output": [],
+      "parallel_tool_calls": true,
+      "previous_response_id": null,
+      "reasoning": null,
+      "store": true,
+      "temperature": 1.0,
+      "text": null,
+      "tool_choice": "auto",
+      "tools": [],
+      "top_p": 1.0,
+      "truncation": "disabled",
+      "usage": {
+        "input_tokens": 10,
+        "output_tokens": 10,
+        "total_tokens": 20
+      },
+      "user": null,
+      "metadata": {}
+    }
+    """
+
+  private let instructionsArrayOfStringsJSON = """
+    {
+      "id": "resp_test_array_instructions",
+      "object": "response",
+      "created_at": 1741476542,
+      "status": "completed",
+      "error": null,
+      "incomplete_details": null,
+      "instructions": ["Be helpful.", "Be concise."],
+      "max_output_tokens": null,
+      "model": "gpt-4.1-2025-04-14",
+      "output": [],
+      "parallel_tool_calls": true,
+      "previous_response_id": null,
+      "reasoning": null,
+      "store": true,
+      "temperature": 1.0,
+      "text": null,
+      "tool_choice": "auto",
+      "tools": [],
+      "top_p": 1.0,
+      "truncation": "disabled",
+      "usage": {
+        "input_tokens": 10,
+        "output_tokens": 10,
+        "total_tokens": 20
+      },
+      "user": null,
+      "metadata": {}
+    }
+    """
+
+  /// This JSON represents the response format when using reusable prompts with variables (issue #187)
+  private let instructionsMessagesJSON = """
+    {
+      "id": "resp_test_messages_instructions",
+      "object": "response",
+      "created_at": 1741476542,
+      "status": "completed",
+      "error": null,
+      "incomplete_details": null,
+      "instructions": [
+        {
+          "type": "message",
+          "content": [{"type": "input_text", "text": "You are a helpful assistant for {{customer_name}}."}],
+          "role": "developer"
+        },
+        {
+          "type": "message",
+          "content": [{"type": "input_text", "text": ""}],
+          "role": "assistant"
+        }
+      ],
+      "max_output_tokens": null,
+      "model": "gpt-4.1-2025-04-14",
+      "output": [],
+      "parallel_tool_calls": true,
+      "previous_response_id": null,
+      "reasoning": null,
+      "store": true,
+      "temperature": 1.0,
+      "text": null,
+      "tool_choice": "auto",
+      "tools": [],
+      "top_p": 1.0,
+      "truncation": "disabled",
+      "usage": {
+        "input_tokens": 10,
+        "output_tokens": 10,
+        "total_tokens": 20
+      },
+      "user": null,
+      "metadata": {}
+    }
+    """
 }