diff --git a/hermes_cli/plugins_loader.py b/hermes_cli/plugins_loader.py
index e916a8633b..4ca05947f3 100644
--- a/hermes_cli/plugins_loader.py
+++ b/hermes_cli/plugins_loader.py
@@ -116,7 +116,10 @@ def run_with_load_deadline(plugin_key: str, ctx: "PluginContext", fn: Callable[[
     scope = _IN_PLUGIN_LOAD.get()
     if scope is not None:
         scope.append(ctx)
-        return fn()
+        try:
+            return fn()
+        finally:
+            scope.remove(ctx)
     timeout = _resolve_plugin_load_timeout()
     if timeout <= 0:
         return fn()
diff --git a/tests/hermes_cli/test_plugins_loader.py b/tests/hermes_cli/test_plugins_loader.py
index dede98d3b2..82ca101db9 100644
--- a/tests/hermes_cli/test_plugins_loader.py
+++ b/tests/hermes_cli/test_plugins_loader.py
@@ -1,24 +1,36 @@
-from hermes_cli import plugins_loader
 import threading
 
+import pytest
+
+from hermes_cli import plugins_loader
+
 
 def test_nested_plugin_load_runs_inline_on_deadline_worker(monkeypatch):
-    monkeypatch.setattr(plugins_loader, "_resolve_plugin_load_timeout", lambda: 1.0)
+    monkeypatch.setattr(plugins_loader, "_resolve_plugin_load_timeout", lambda: 0.3)
+    abandoned = []
 
     class Context:
+        def __init__(self, name):
+            self.name = name
+
         def _abandon_load(self):
-            raise AssertionError("unexpected timeout")
+            abandoned.append(self.name)
 
     observed_threads = []
-
-    def nested_load():
-        observed_threads.append(threading.current_thread())
+    release = threading.Event()
 
     def outer_load():
         observed_threads.append(threading.current_thread())
-        plugins_loader.run_with_load_deadline("nested", Context(), nested_load)
+        plugins_loader.run_with_load_deadline(
+            "done", Context("done"), lambda: observed_threads.append(threading.current_thread()),
+        )
+        plugins_loader.run_with_load_deadline("hung", Context("hung"), release.wait)
 
-    plugins_loader.run_with_load_deadline("outer", Context(), outer_load)
+    with pytest.raises(plugins_loader.PluginLoadTimeout):
+        plugins_loader.run_with_load_deadline("outer", Context("outer"), outer_load)
+    release.set()
 
     assert len(observed_threads) == 2
     assert observed_threads[0] is observed_threads[1]
+    # The outer timeout abandons only contexts still loading, not a nested load that already finished.
+    assert abandoned == ["outer", "hung"]