diff --git a/cmdstanpy/model.py b/cmdstanpy/model.py
index d1d96f48..d9f4e9fc 100644
--- a/cmdstanpy/model.py
+++ b/cmdstanpy/model.py
@@ -2027,7 +2027,14 @@ def diagnose(
             Gradients are evaluated in the unconstrained space.
         """
 
-        with temp_single_json(data) as _data, temp_single_json(inits) as _inits:
+        with (
+            temp_single_json(data) as _data,
+            temp_single_json(inits) as _inits,
+            # the output is read below and is not needed after this call.
+            tempfile.TemporaryDirectory(
+                prefix=self.name, dir=_TMPDIR
+            ) as output_dir,
+        ):
             cmd = [
                 str(self.exe_file),
                 "diagnose",
@@ -2036,14 +2043,12 @@ def diagnose(
             if epsilon is not None:
                 cmd.append(f"epsilon={epsilon}")
             if error is not None:
-                cmd.append(f"epsilon={error}")
+                cmd.append(f"error={error}")
             if _data is not None:
                 cmd += ["data", f"file={_data}"]
             if _inits is not None:
                 cmd.append(f"init={_inits}")
 
-            output_dir = tempfile.mkdtemp(prefix=self.name, dir=_TMPDIR)
-
             output = os.path.join(output_dir, "output.csv")
             cmd += ["output", f"file={output}"]
             if sig_figs is not None:
diff --git a/test/test_model.py b/test/test_model.py
index d725eed0..a6af29b5 100644
--- a/test/test_model.py
+++ b/test/test_model.py
@@ -11,6 +11,7 @@
 import numpy as np
 import pytest
 
+from cmdstanpy import _TMPDIR
 from cmdstanpy.model import CmdStanModel
 from cmdstanpy.utils import EXTENSION
 
@@ -472,3 +473,36 @@ def test_diagnose() -> None:
         require_gradients_ok=False,
     )
     assert np.abs(gradients["error"]).max() > 1e-3
+
+
+def test_diagnose_error_argument(caplog: pytest.LogCaptureFixture) -> None:
+    model = CmdStanModel(stan_file=BERN_STAN)
+    with caplog.at_level(logging.DEBUG, logger="cmdstanpy"):
+        model.diagnose(data=BERN_DATA, epsilon=1e-5, error=1e-3)
+    cmds = [r.getMessage() for r in caplog.records if "Cmd:" in r.message]
+    assert len(cmds) == 1
+    assert "'epsilon=1e-05'" in cmds[0]
+    assert "'error=0.001'" in cmds[0]
+    assert cmds[0].count("epsilon=") == 1
+
+
+@pytest.mark.parametrize("fails", [False, True])
+def test_diagnose_removes_output_dir(
+    fails: bool, caplog: pytest.LogCaptureFixture
+) -> None:
+    model = CmdStanModel(stan_file=BERN_STAN)
+    cmdstanpy_tmp = set(os.listdir(_TMPDIR))
+
+    with caplog.at_level(logging.DEBUG, logger="cmdstanpy"):
+        if fails:
+            with pytest.raises(RuntimeError, match="may exceed"):
+                model.diagnose(data=BERN_DATA, epsilon=3)
+        else:
+            model.diagnose(data=BERN_DATA)
+
+    assert set(os.listdir(_TMPDIR)) == cmdstanpy_tmp
+    # the command is logged as a list repr, which escapes Windows separators
+    output_arg = "file=" + repr(str(_TMPDIR))[1:-1]
+    cmds = [r.getMessage() for r in caplog.records if "Cmd:" in r.message]
+    assert len(cmds) == 1
+    assert output_arg in cmds[0]