diff --git a/src/bub/builtin/tools.py b/src/bub/builtin/tools.py
index 31af4bc3..9c821573 100644
--- a/src/bub/builtin/tools.py
+++ b/src/bub/builtin/tools.py
@@ -23,6 +23,14 @@
 DEFAULT_REQUEST_TIMEOUT_SECONDS = 10
 
 
+def _raise_for_failed_shell(returncode: int | None, output: str) -> None:
+    if returncode in (None, 0):
+        return
+
+    body = output.strip() or "(no output)"
+    raise RuntimeError(f"command exited with code {returncode}\noutput:\n{body}")
+
+
 def _get_agent(context: ToolContext) -> Agent:
     if "_runtime_agent" not in context.state:
         raise RuntimeError("no runtime agent found in tool context")
@@ -76,10 +84,11 @@ async def bash(
         return f"started: {shell.shell_id}"
     try:
         async with asyncio.timeout(timeout_seconds):
-            await shell_manager.wait_closed(shell.shell_id)
+            shell = await shell_manager.wait_closed(shell.shell_id)
     except TimeoutError:
         await shell_manager.terminate(shell.shell_id)
         return f"command timed out after {timeout_seconds} seconds and was terminated"
+    _raise_for_failed_shell(shell.returncode, shell.output)
     return shell.output.strip() or "(no output)"
 
 
diff --git a/tests/test_builtin_tools.py b/tests/test_builtin_tools.py
index 3431492c..36c7e41e 100644
--- a/tests/test_builtin_tools.py
+++ b/tests/test_builtin_tools.py
@@ -6,6 +6,8 @@
 
 import pytest
 from republic import ToolContext
+from republic.core.errors import ErrorKind
+from republic.tools.executor import ToolExecutor
 
 from bub.builtin.tools import bash, bash_output, kill_bash
 
@@ -25,6 +27,31 @@ async def test_bash_returns_stdout_for_foreground_command(tmp_path) -> None:
     assert result == "hello"
 
 
+@pytest.mark.asyncio
+async def test_bash_non_zero_exit_is_returned_as_tool_error(tmp_path) -> None:
+    command = _python_shell("import sys; print('boom'); sys.exit(7)")
+    executor = ToolExecutor()
+    tool_call = {
+        "type": "function",
+        "function": {
+            "name": bash.name,
+            "arguments": {"cmd": command},
+        },
+    }
+
+    result = await executor.execute_async([tool_call], tools=[bash], context=_tool_context(tmp_path))
+
+    assert result.error is not None
+    assert result.error.kind is ErrorKind.TOOL
+    assert len(result.tool_results) == 1
+    tool_result = result.tool_results[0]
+    assert tool_result["kind"] == "tool"
+    assert tool_result["message"] == "Tool 'bash' execution failed."
+    error_detail = tool_result["details"]["error"]
+    assert "command exited with code 7" in error_detail
+    assert "boom" in error_detail
+
+
 @pytest.mark.asyncio
 async def test_background_bash_exposes_output_via_bash_output(tmp_path) -> None:
     command = _python_shell(