Dev/steven/update convo history examples

docs/quickstart.md

@@ -83,6 +83,45 @@ main();
 
 **That's it!** Your existing OpenAI code now includes automatic guardrail validation based on your pipeline configuration. The response object works exactly like the original OpenAI response with additional `guardrail_results` property.
 
+## Multi-Turn Conversations
+
+When maintaining conversation history across multiple turns, **only append messages after guardrails pass**. This prevents blocked input messages from polluting your conversation context.
+
+```typescript
+import { GuardrailsOpenAI, GuardrailTripwireTriggered } from '@openai/guardrails';
+
+const client = await GuardrailsOpenAI.create('./guardrails_config.json');
+const messages: Array<{ role: 'user' | 'assistant'; content: string }> = [];
+
+while (true) {
+  const userInput = await readUserInput(); // replace with your input routine
+
+  try {
+    // ✅ Pass user input inline (don't mutate messages first)
+    const response = await client.chat.completions.create({
+      model: 'gpt-4o',
+      messages: [...messages, { role: 'user', content: userInput }],
+    });
+
+    const responseContent = response.choices[0].message?.content ?? '';
+    console.log(`Assistant: ${responseContent}`);
+
+    // ✅ Only append AFTER guardrails pass
+    messages.push({ role: 'user', content: userInput });
+    messages.push({ role: 'assistant', content: responseContent });
+  } catch (error) {
+    if (error instanceof GuardrailTripwireTriggered) {
+      // ❌ Guardrail blocked - message NOT added to history
+      console.log('Message blocked by guardrails');
+      continue;
+    }
+    throw error;
+  }
+}
+```
+
+**Why this matters**: If you append the user message before the guardrail check, blocked messages remain in your conversation history and get sent on every subsequent turn, even though they violated your safety policies.
+
 ## Guardrail Execution Error Handling
 
 Guardrails supports two error handling modes for guardrail execution failures:

examples/basic/agents_sdk.ts

@@ -112,22 +112,26 @@ async function main(): Promise<void> {
 
         console.log('🤔 Processing...\n');
 
-        // Pass conversation history with the new user message
-        const result = await run(agent, thread.concat({ role: 'user', content: userInput }));
-
-        // Update thread with the complete history including newly generated items
+        // Pass conversation history with the new user message (without mutating thread yet)
+        const conversationWithUser = thread.concat({ role: 'user', content: userInput });
+
+        const result = await run(agent, conversationWithUser);
+
+        // Guardrails passed - now safe to update thread with complete history
         thread = result.history;
-        
+
         console.log(`Assistant: ${result.finalOutput}\n`);
       } catch (error: any) {
         // Handle guardrail tripwire exceptions
         const errorType = error?.constructor?.name;
 
         if (errorType === 'InputGuardrailTripwireTriggered' || error instanceof InputGuardrailTripwireTriggered) {
           console.log('🛑 Input guardrail triggered! Please try a different message.\n');
+          // Guardrail blocked - user message NOT added to history
           continue;
         } else if (errorType === 'OutputGuardrailTripwireTriggered' || error instanceof OutputGuardrailTripwireTriggered) {
           console.log('🛑 Output guardrail triggered! The response was blocked.\n');
+          // Guardrail blocked - assistant response NOT added to history
           continue;
         } else {
           console.error('❌ An error occurred:', error.message);

examples/basic/azure_example.ts

@@ -60,14 +60,20 @@ const PIPELINE_CONFIG = {
  * @param responseId Optional response ID for conversation tracking
  * @returns Promise resolving to a response ID
  */
+type ChatMessage = {
+  role: 'user' | 'assistant' | 'system';
+  content: string;
+};
+
 async function processInput(
   guardrailsClient: GuardrailsAzureOpenAI,
-  userInput: string
+  userInput: string,
+  messages: ChatMessage[]
 ): Promise<string> {
-  // Use the new GuardrailsAzureOpenAI - it handles all guardrail validation automatically
+  // Pass user input inline WITHOUT mutating messages first
   const response = await guardrailsClient.guardrails.chat.completions.create({
     model: process.env.AZURE_DEPLOYMENT!,
-    messages: [{ role: 'user', content: userInput }],
+    messages: [...messages, { role: 'user', content: userInput }],
   });
 
   console.log(`\nAssistant output: ${response.choices[0].message.content}`);
@@ -79,6 +85,13 @@ async function processInput(
     );
   }
 
+  // Guardrails passed - now safe to add to conversation history
+  messages.push({ role: 'user', content: userInput });
+  messages.push({
+    role: 'assistant',
+    content: response.choices[0].message.content ?? '',
+  });
+
   return response.id;
 }
 
@@ -126,6 +139,7 @@ async function main(): Promise<void> {
   });
 
   const rl = createReadlineInterface();
+  const messages: ChatMessage[] = [];
   // let responseId: string | undefined;
 
   // Handle graceful shutdown
@@ -150,14 +164,15 @@ async function main(): Promise<void> {
       }
 
       try {
-        await processInput(guardrailsClient, userInput);
+        await processInput(guardrailsClient, userInput, messages);
       } catch (error) {
         if (error instanceof GuardrailTripwireTriggered) {
           const stageName = error.guardrailResult.info?.stage_name || 'unknown';
           console.log(`\n🛑 Guardrail triggered in stage '${stageName}'!`);
           console.log('\n📋 Guardrail Result:');
           console.log(JSON.stringify(error.guardrailResult, null, 2));
           console.log('\nPlease rephrase your message to avoid triggering security checks.\n');
+          // Guardrail blocked - user message NOT added to history
         } else {
           console.error(`\n❌ Error: ${error instanceof Error ? error.message : String(error)}\n`);
         }

examples/basic/local_model.ts

@@ -34,25 +34,25 @@ const GEMMA3_PIPELINE_CONFIG = {
 async function processInput(
   guardrailsClient: GuardrailsOpenAI,
   userInput: string,
-  inputData: OpenAI.Chat.Completions.ChatCompletionMessageParam[]
+  conversation: OpenAI.Chat.Completions.ChatCompletionMessageParam[]
 ): Promise<void> {
   try {
-    // Use GuardrailsClient for chat completions with guardrails
+    // Pass user input inline WITHOUT mutating conversation history first
     const response = await guardrailsClient.guardrails.chat.completions.create({
-      messages: [...inputData, { role: 'user', content: userInput }],
+      messages: [...conversation, { role: 'user', content: userInput }],
       model: 'gemma3',
     });
 
     // Access response content using standard OpenAI API
-    const responseContent = response.choices[0].message.content;
+    const responseContent = response.choices[0].message.content ?? '';
     console.log(`\nAssistant output: ${responseContent}\n`);
 
-    // Add to conversation history
-    inputData.push({ role: 'user', content: userInput });
-    inputData.push({ role: 'assistant', content: responseContent || '' });
+    // Guardrails passed - now safe to add to conversation history
+    conversation.push({ role: 'user', content: userInput });
+    conversation.push({ role: 'assistant', content: responseContent });
   } catch (error) {
     if (error instanceof GuardrailTripwireTriggered) {
-      // Handle guardrail violations
+      // Guardrail blocked - user message NOT added to history
       throw error;
     }
     throw error;
@@ -69,7 +69,7 @@ async function main(): Promise<void> {
     apiKey: 'ollama',
   });
 
-  const inputData: OpenAI.Chat.Completions.ChatCompletionMessageParam[] = [];
+  const conversation: OpenAI.Chat.Completions.ChatCompletionMessageParam[] = [];
 
   try {
     // eslint-disable-next-line no-constant-condition
@@ -87,14 +87,15 @@ async function main(): Promise<void> {
           });
         });
 
-        await processInput(guardrailsClient, userInput, inputData);
+        await processInput(guardrailsClient, userInput, conversation);
       } catch (error) {
         if (error instanceof GuardrailTripwireTriggered) {
           const stageName = error.guardrailResult.info?.stage_name || 'unknown';
           const guardrailName = error.guardrailResult.info?.guardrail_name || 'unknown';
 
           console.log(`\n🛑 Guardrail '${guardrailName}' triggered in stage '${stageName}'!`);
           console.log('Guardrail Result:', error.guardrailResult);
+          // Guardrail blocked - conversation history unchanged
           continue;
         }
         throw error;

examples/basic/multiturn_with_prompt_injection_detection.ts

@@ -278,37 +278,38 @@ async function main(malicious: boolean = false): Promise<void> {
         continue;
       }
 
-      // Append user message as content parts
-      messages.push({
+      const userMessage = {
         role: 'user',
         content: [{ type: 'input_text', text: userInput }],
-      });
+      };
 
       // First call: ask the model (may request function_call)
       console.log(`🔄 Making initial API call...`);
 
       let response: GuardrailsResponse;
       let functionCalls: any[] = [];
+      let assistantOutputs: any[] = [];
 
       try {
         response = await client.guardrails.responses.create({
           model: 'gpt-4.1-nano',
           tools: tools,
-          input: messages,
+          input: messages.concat(userMessage),
         });
 
         printGuardrailResults('initial', response);
 
-        // Add the assistant response to conversation history
-        messages.push(...response.output);
+        assistantOutputs = response.output ?? [];
+
+        // Guardrails passed - now safe to add user message to conversation history
+        messages.push(userMessage);
 
         // Grab any function calls from the response
-        functionCalls = response.output.filter(
-          (item: any) => item.type === 'function_call'
-        );
+        functionCalls = assistantOutputs.filter((item: any) => item.type === 'function_call');
 
         // Handle the case where there are no function calls
         if (functionCalls.length === 0) {
+          messages.push(...assistantOutputs);
           console.log(`\n🤖 Assistant: ${response.output_text}`);
           continue;
         }
@@ -325,6 +326,7 @@ async function main(malicious: boolean = false): Promise<void> {
           );
           console.log(`Confidence: ${info.confidence || 'N/A'}`);
           console.log('='.repeat(50));
+          // Guardrail blocked - user message NOT added to history
           continue;
         } else {
           throw error;
@@ -333,6 +335,8 @@ async function main(malicious: boolean = false): Promise<void> {
 
       if (functionCalls && functionCalls.length > 0) {
         // Execute function calls and add results to conversation
+        const toolMessages: any[] = [];
+
         for (const fc of functionCalls) {
           const fname = fc.name;
           const fargs = JSON.parse(fc.arguments);
@@ -359,20 +363,20 @@ async function main(malicious: boolean = false): Promise<void> {
                 };
               }
 
-              messages.push({
+              toolMessages.push({
                 type: 'function_call_output',
                 call_id: fc.call_id,
                 output: JSON.stringify(result),
               });
             } catch (ex) {
-              messages.push({
+              toolMessages.push({
                 type: 'function_call_output',
                 call_id: fc.call_id,
                 output: JSON.stringify({ error: String(ex) }),
               });
             }
           } else {
-            messages.push({
+            toolMessages.push({
               type: 'function_call_output',
               call_id: fc.call_id,
               output: JSON.stringify({ error: `Unknown function: ${fname}` }),
@@ -386,13 +390,15 @@ async function main(malicious: boolean = false): Promise<void> {
           const response = await client.guardrails.responses.create({
             model: 'gpt-4.1-nano',
             tools: tools,
-            input: messages,
+            input: messages.concat(assistantOutputs, toolMessages),
           });
 
           printGuardrailResults('final', response);
           console.log(`\n🤖 Assistant: ${response.output_text}`);
 
-          // Add the final assistant response to conversation history
+          // Guardrails passed - now safe to add tool results and assistant responses to history
+          messages.push(...assistantOutputs);
+          messages.push(...toolMessages);
           messages.push(...response.output);
         } catch (error: any) {
           if (error instanceof GuardrailTripwireTriggered) {
@@ -408,6 +414,7 @@ async function main(malicious: boolean = false): Promise<void> {
             console.log(`Observation: ${info.observation || 'N/A'}`);
             console.log(`Confidence: ${info.confidence || 'N/A'}`);
             console.log('='.repeat(50));
+            // Guardrail blocked - tool results NOT added to history
             continue;
           } else {
             throw error;