test(runtime): isolate tool execution module state

tests/test_fenced_inline_args.py

@@ -8,19 +8,38 @@ language hint, not a "python" tool call with content "3\n...".
 import sys
 from unittest.mock import MagicMock
 
-for mod in ['src.agent_tools', 'src.tool_parsing', 'src.tool_schemas', 'src.tool_execution']:
-    sys.modules.pop(mod, None)
-for mod in [
-    'sqlalchemy', 'sqlalchemy.orm', 'sqlalchemy.ext', 'sqlalchemy.ext.declarative',
-    'sqlalchemy.ext.hybrid', 'sqlalchemy.sql', 'sqlalchemy.sql.expression',
-    'src.database', 'core.models', 'core.database', 'core.auth'
-]:
-    if mod not in sys.modules:
-        sys.modules[mod] = MagicMock()
+# This module needs the real agent-tool stack; importing it pulls in heavy
+# DB/auth deps, so we stub those just long enough to import, then restore them.
+# We deliberately do NOT pop src.tool_execution: popping and re-importing it
+# rebinds the `src` package's `tool_execution` attribute, so a later
+# `import src.tool_execution as te` resolves to a different module object than
+# the one its functions live in - which silently breaks tests that monkeypatch
+# it (e.g. test_edit_file's admin gate) and breaks request-scoped ContextVars.
+_ABSENT = object()
+_AGENT_MODULES = ["src.agent_tools", "src.tool_parsing", "src.tool_schemas"]
+_STUBBED = [
+    "sqlalchemy", "sqlalchemy.orm", "sqlalchemy.ext", "sqlalchemy.ext.declarative",
+    "sqlalchemy.ext.hybrid", "sqlalchemy.sql", "sqlalchemy.sql.expression",
+    "src.database", "core.models", "core.database", "core.auth",
+]
+_saved_stubs = {name: sys.modules.get(name, _ABSENT) for name in _STUBBED}
+
+for _mod in _AGENT_MODULES:
+    sys.modules.pop(_mod, None)
+for _mod in _STUBBED:
+    if _mod not in sys.modules:
+        sys.modules[_mod] = MagicMock()
 
 import src.agent_tools  # noqa: E402, F401
 from src.tool_parsing import parse_tool_blocks, strip_tool_blocks  # noqa: E402
 
+# Drop the stubs we installed so they do not leak into later tests.
+for _name, _original in _saved_stubs.items():
+    if _original is _ABSENT:
+        sys.modules.pop(_name, None)
+    else:
+        sys.modules[_name] = _original
+
 
 def test_inline_args_on_tag_line_parse():
     # The original bug: ```list_email_accounts {}  (args on the tag line)

tests/test_fenced_invoke_no_raw_xml.py

@@ -6,19 +6,38 @@ to the code executor as if it were python/bash.
 import sys
 from unittest.mock import MagicMock
 
-for mod in ['src.agent_tools', 'src.tool_parsing', 'src.tool_schemas', 'src.tool_execution']:
-    sys.modules.pop(mod, None)
-for mod in [
-    'sqlalchemy', 'sqlalchemy.orm', 'sqlalchemy.ext', 'sqlalchemy.ext.declarative',
-    'sqlalchemy.ext.hybrid', 'sqlalchemy.sql', 'sqlalchemy.sql.expression',
-    'src.database', 'core.models', 'core.database', 'core.auth'
-]:
-    if mod not in sys.modules:
-        sys.modules[mod] = MagicMock()
+# This module needs the real agent-tool stack; importing it pulls in heavy
+# DB/auth deps, so we stub those just long enough to import, then restore them.
+# We deliberately do NOT pop src.tool_execution: popping and re-importing it
+# rebinds the `src` package's `tool_execution` attribute, so a later
+# `import src.tool_execution as te` resolves to a different module object than
+# the one its functions live in - which silently breaks tests that monkeypatch
+# it (e.g. test_edit_file's admin gate) and breaks request-scoped ContextVars.
+_ABSENT = object()
+_AGENT_MODULES = ["src.agent_tools", "src.tool_parsing", "src.tool_schemas"]
+_STUBBED = [
+    "sqlalchemy", "sqlalchemy.orm", "sqlalchemy.ext", "sqlalchemy.ext.declarative",
+    "sqlalchemy.ext.hybrid", "sqlalchemy.sql", "sqlalchemy.sql.expression",
+    "src.database", "core.models", "core.database", "core.auth",
+]
+_saved_stubs = {name: sys.modules.get(name, _ABSENT) for name in _STUBBED}
+
+for _mod in _AGENT_MODULES:
+    sys.modules.pop(_mod, None)
+for _mod in _STUBBED:
+    if _mod not in sys.modules:
+        sys.modules[_mod] = MagicMock()
 
 import src.agent_tools  # noqa: E402, F401
 from src.tool_parsing import parse_tool_blocks  # noqa: E402
 
+# Drop the stubs we installed so they do not leak into later tests.
+for _name, _original in _saved_stubs.items():
+    if _original is _ABSENT:
+        sys.modules.pop(_name, None)
+    else:
+        sys.modules[_name] = _original
+
 
 def test_unconvertible_invoke_in_fence_is_not_executed_as_code():
     text = '```python\n<invoke name="foo-bar">\n<parameter name="x">1</parameter>\n</invoke>\n```'

tests/test_web_search_raw_json_tool_call.py

@@ -12,19 +12,38 @@ import json
 import sys
 from unittest.mock import MagicMock
 
-for mod in ['src.agent_tools', 'src.tool_parsing', 'src.tool_schemas', 'src.tool_execution']:
-    sys.modules.pop(mod, None)
-for mod in [
-    'sqlalchemy', 'sqlalchemy.orm', 'sqlalchemy.ext', 'sqlalchemy.ext.declarative',
-    'sqlalchemy.ext.hybrid', 'sqlalchemy.sql', 'sqlalchemy.sql.expression',
-    'src.database', 'core.models', 'core.database', 'core.auth'
-]:
-    if mod not in sys.modules:
-        sys.modules[mod] = MagicMock()
+# This module needs the real agent-tool stack; importing it pulls in heavy
+# DB/auth deps, so we stub those just long enough to import, then restore them.
+# We deliberately do NOT pop src.tool_execution: popping and re-importing it
+# rebinds the `src` package's `tool_execution` attribute, so a later
+# `import src.tool_execution as te` resolves to a different module object than
+# the one its functions live in - which silently breaks tests that monkeypatch
+# it (e.g. test_edit_file's admin gate) and breaks request-scoped ContextVars.
+_ABSENT = object()
+_AGENT_MODULES = ["src.agent_tools", "src.tool_parsing", "src.tool_schemas"]
+_STUBBED = [
+    "sqlalchemy", "sqlalchemy.orm", "sqlalchemy.ext", "sqlalchemy.ext.declarative",
+    "sqlalchemy.ext.hybrid", "sqlalchemy.sql", "sqlalchemy.sql.expression",
+    "src.database", "core.models", "core.database", "core.auth",
+]
+_saved_stubs = {name: sys.modules.get(name, _ABSENT) for name in _STUBBED}
+
+for _mod in _AGENT_MODULES:
+    sys.modules.pop(_mod, None)
+for _mod in _STUBBED:
+    if _mod not in sys.modules:
+        sys.modules[_mod] = MagicMock()
 
 import src.agent_tools  # noqa: E402, F401
 from src.tool_parsing import parse_tool_blocks, strip_tool_blocks  # noqa: E402
 
+# Drop the stubs we installed so they do not leak into later tests.
+for _name, _original in _saved_stubs.items():
+    if _original is _ABSENT:
+        sys.modules.pop(_name, None)
+    else:
+        sys.modules[_name] = _original
+
 
 def test_raw_json_after_web_search_phrase_runs_as_web_search():
     text = (

tests/test_web_search_time_filter.py

@@ -9,25 +9,40 @@ bare string (back-compat).
 import sys
 from unittest.mock import MagicMock
 
-# Clean up any mocks from previous tests to ensure we load real modules.
-for mod in ['src.agent_tools', 'src.tool_parsing', 'src.tool_schemas', 'src.tool_execution']:
-    sys.modules.pop(mod, None)
+# This module needs the real agent-tool stack; importing it pulls in heavy
+# DB/auth deps, so we stub those just long enough to import, then restore them.
+# We deliberately do NOT pop src.tool_execution: popping and re-importing it
+# rebinds the `src` package's `tool_execution` attribute, so a later
+# `import src.tool_execution as te` resolves to a different module object than
+# the one its functions live in - which silently breaks tests that monkeypatch
+# it (e.g. test_edit_file's admin gate) and breaks request-scoped ContextVars.
+_ABSENT = object()
+_AGENT_MODULES = ["src.agent_tools", "src.tool_parsing", "src.tool_schemas"]
+_STUBBED = [
+    "sqlalchemy", "sqlalchemy.orm", "sqlalchemy.ext", "sqlalchemy.ext.declarative",
+    "sqlalchemy.ext.hybrid", "sqlalchemy.sql", "sqlalchemy.sql.expression",
+    "src.database", "core.models", "core.database", "core.auth",
+]
+_saved_stubs = {name: sys.modules.get(name, _ABSENT) for name in _STUBBED}
 
-# Mock heavy database/model dependencies before importing (avoids the
-# src.tool_schemas <-> src.agent_tools circular import pulling in the DB layer).
-for mod in [
-    'sqlalchemy', 'sqlalchemy.orm', 'sqlalchemy.ext', 'sqlalchemy.ext.declarative',
-    'sqlalchemy.ext.hybrid', 'sqlalchemy.sql', 'sqlalchemy.sql.expression',
-    'src.database', 'core.models', 'core.database', 'core.auth'
-]:
-    if mod not in sys.modules:
-        sys.modules[mod] = MagicMock()
+for _mod in _AGENT_MODULES:
+    sys.modules.pop(_mod, None)
+for _mod in _STUBBED:
+    if _mod not in sys.modules:
+        sys.modules[_mod] = MagicMock()
 
 import json  # noqa: E402
 
 import src.agent_tools  # noqa: E402, F401
 from src.tool_schemas import function_call_to_tool_block  # noqa: E402
 
+# Drop the stubs we installed so they do not leak into later tests.
+for _name, _original in _saved_stubs.items():
+    if _original is _ABSENT:
+        sys.modules.pop(_name, None)
+    else:
+        sys.modules[_name] = _original
+
 
 def test_time_filter_is_preserved_as_json():
     block = function_call_to_tool_block(