Merge branch 'v0.9.0rc' into remove-sandbox-as-it-is-broken

apps/server/src/services/agent-service.ts

@@ -13,6 +13,8 @@ import {
   isAbortError,
   loadContextFiles,
   createLogger,
+  classifyError,
+  getUserFriendlyErrorMessage,
 } from '@automaker/utils';
 import { ProviderFactory } from '../providers/provider-factory.js';
 import { createChatOptions, validateWorkingDirectory } from '../lib/sdk-options.js';
@@ -171,6 +173,18 @@ export class AgentService {
       session.thinkingLevel = thinkingLevel;
     }
 
+    // Validate vision support before processing images
+    const effectiveModel = model || session.model;
+    if (imagePaths && imagePaths.length > 0 && effectiveModel) {
+      const supportsVision = ProviderFactory.modelSupportsVision(effectiveModel);
+      if (!supportsVision) {
+        throw new Error(
+          `This model (${effectiveModel}) does not support image input. ` +
+            `Please switch to a model that supports vision, or remove the images and try again.`
+        );
+      }
+    }
+
     // Read images and convert to base64
     const images: Message['images'] = [];
     if (imagePaths && imagePaths.length > 0) {
@@ -365,6 +379,53 @@ export class AgentService {
             content: responseText,
             toolUses,
           });
+        } else if (msg.type === 'error') {
+          // Some providers (like Codex CLI/SaaS or Cursor CLI) surface failures as
+          // streamed error messages instead of throwing. Handle these here so the
+          // Agent Runner UX matches the Claude/Cursor behavior without changing
+          // their provider implementations.
+          const rawErrorText =
+            (typeof msg.error === 'string' && msg.error.trim()) ||
+            'Unexpected error from provider during agent execution.';
+
+          const errorInfo = classifyError(new Error(rawErrorText));
+
+          // Keep the provider-supplied text intact (Codex already includes helpful tips),
+          // only add a small rate-limit hint when we can detect it.
+          const enhancedText = errorInfo.isRateLimit
+            ? `${rawErrorText}\n\nTip: It looks like you hit a rate limit. Try waiting a bit or reducing concurrent Agent Runner / Auto Mode tasks.`
+            : rawErrorText;
+
+          this.logger.error('Provider error during agent execution:', {
+            type: errorInfo.type,
+            message: errorInfo.message,
+          });
+
+          // Mark session as no longer running so the UI and queue stay in sync
+          session.isRunning = false;
+          session.abortController = null;
+
+          const errorMessage: Message = {
+            id: this.generateId(),
+            role: 'assistant',
+            content: `Error: ${enhancedText}`,
+            timestamp: new Date().toISOString(),
+            isError: true,
+          };
+
+          session.messages.push(errorMessage);
+          await this.saveSession(sessionId, session.messages);
+
+          this.emitAgentEvent(sessionId, {
+            type: 'error',
+            error: enhancedText,
+            message: errorMessage,
+          });
+
+          // Don't continue streaming after an error message
+          return {
+            success: false,
+          };
         }
       }
 

apps/server/src/services/auto-mode-service.ts

@@ -1991,6 +1991,18 @@ This helps parse your summary correctly in the output logs.`;
     const planningMode = options?.planningMode || 'skip';
     const previousContent = options?.previousContent;
 
+    // Validate vision support before processing images
+    const effectiveModel = model || 'claude-sonnet-4-20250514';
+    if (imagePaths && imagePaths.length > 0) {
+      const supportsVision = ProviderFactory.modelSupportsVision(effectiveModel);
+      if (!supportsVision) {
+        throw new Error(
+          `This model (${effectiveModel}) does not support image input. ` +
+            `Please switch to a model that supports vision (like Claude models), or remove the images and try again.`
+        );
+      }
+    }
+
     // Check if this planning mode can generate a spec/plan that needs approval
     // - spec and full always generate specs
     // - lite only generates approval-ready content when requirePlanApproval is true