fix(sandbox): enforce E2B output limits on success

Fail closed even when the E2B SDK resolves a command after reporting a callback overflow.

Author: Bill Leoutsakos

apps/sim/lib/execution/remote-sandbox/conformance.test.ts

@@ -138,7 +138,11 @@ describe.each(PROVIDERS)('sandbox conformance [%s]', (provider) => {
       mockE2BCommandsRun.mockImplementation(
         async (_command: string, options: { onStdout: (chunk: string) => void }) => {
           options.onStdout('12345')
-          options.onStdout('6')
+          try {
+            options.onStdout('6')
+          } catch {
+            // E2B may report callback errors separately and still resolve the command.
+          }
           return { stdout: '123456', stderr: '', exitCode: 0 }
         }
       )

apps/sim/lib/execution/remote-sandbox/e2b.ts

@@ -83,6 +83,15 @@ class E2BSandboxHandle implements SandboxHandle {
         onStdout: limiter.stdout,
         onStderr: limiter.stderr,
       })
+      if (limiter.exceeded()) {
+        await Promise.resolve(this.sandbox.kill()).catch(() => {})
+        return {
+          stdout: '',
+          stderr: 'Sandbox command output limit exceeded',
+          exitCode: 137,
+          outputLimitExceeded: true,
+        }
+      }
       return { stdout: result.stdout, stderr: result.stderr, exitCode: result.exitCode }
     } catch (error) {
       if (limiter.exceeded()) {