Skip to content

fix: block only normalized AgentSentry V3 BLOCK decisions #1213

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
yaozheng-fang merged 1 commit into from
Nov 6, 2026
Merged

fix: block only normalized AgentSentry V3 BLOCK decisions #1213

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
77 changes: 67 additions & 10 deletions tests/test_agentsentry_v3_transport.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,17 +43,26 @@ def client(mod):


@pytest.mark.parametrize(
"result,blocked",
"result,expected",
[
({"Degraded": False, "Decision": {"DecisionType": 1}}, False),
({"Degraded": False, "Decision": {"DecisionType": 2}}, True),
({"Degraded": True, "Decision": {"DecisionType": 1}}, True),
({"Decision": {"DecisionType": 1}}, True),
({"Degraded": False, "Decision": {"DecisionType": 4}}, True),
({"Degraded": False, "Decision": {"DecisionType": True}}, True),
({"Degraded": False, "Decision": {"DecisionType": 1}}, None),
({"Degraded": False, "Decision": {"DecisionType": 2}}, "BLOCK"),
({"Decision": {"DecisionType": 1}}, None),
({"Degraded": "false", "Decision": {"DecisionType": 1}}, None),
({"Degraded": False, "Decision": {"DecisionType": 4}}, None),
({"Degraded": False, "Decision": {"DecisionType": True}}, None),
({"Degraded": True}, None),
({"Decision": {"DecisionType": 2}}, "BLOCK"),
({"Decision": {"DecisionType": 0}}, None),
({"Decision": {"DecisionType": 3}}, None),
({"Decision": {"DecisionType": 99}}, None),
({"Decision": {"DecisionType": None}}, None),
({"Decision": {"DecisionType": "unknown"}}, None),
({"Decision": {"DecisionType": "2"}}, "BLOCK"),
({"Decision": {"DecisionType": 2.0}}, "BLOCK"),
],
)
def test_decisions(mod, result, blocked):
def test_decisions(mod, result, expected):
response = SimpleNamespace(status_code=200, json=lambda: {"Result": result})
with patch.object(mod.requests, "post", return_value=response) as post:
value = client(mod).moderate(
Expand All @@ -63,7 +72,7 @@ def test_decisions(mod, result, blocked):
session_info={"SessionID": "session", "RunID": "run"},
tool_calls=[{"ID": "call"}],
)
assert bool(value) == blocked
assert value == (getattr(mod, expected) if expected else None)
request = post.call_args
body = json.loads(request.kwargs["data"])
assert body["SessionID"] == "session"
Expand All @@ -75,6 +84,33 @@ def test_decisions(mod, result, blocked):
assert request.kwargs["allow_redirects"] is False


@pytest.mark.parametrize(
"action,decision,expected",
[
("pass_and_writeback", 1, None),
("pass_and_writeback", 2, "BLOCK"),
("block_and_writeback", 1, None),
("block_and_writeback", 2, "BLOCK"),
(None, 1, None),
("unknown", 1, None),
("pass_and_writeback", 4, None),
],
)
def test_decision_takes_precedence_over_degraded_action(
mod, action, decision, expected
):
result = {
"Decision": {"DecisionType": decision},
"Degraded": True,
}
if action is not None:
result["DegradeAction"] = action
response = SimpleNamespace(status_code=200, json=lambda: {"Result": result})
with patch.object(mod.requests, "post", return_value=response):
value = client(mod).moderate("hello", role="user", hook_name="before_dispatch")
assert value == (getattr(mod, expected) if expected else None)


def test_missing_plan_and_wrong_hook(mod):
with patch.object(mod.requests, "post") as post:
assert (
Expand All @@ -88,7 +124,28 @@ def test_missing_plan_and_wrong_hook(mod):

def test_timeout(mod):
with patch.object(mod.requests, "post", side_effect=mod.requests.Timeout):
assert client(mod).moderate("hello", role="user", hook_name="before_dispatch")
assert (
client(mod).moderate("hello", role="user", hook_name="before_dispatch")
== mod.UNAVAILABLE
)


@pytest.mark.parametrize("status,api_error", [(503, False), (200, True)])
def test_service_errors_do_not_become_allow(mod, status, api_error):
response = SimpleNamespace(
status_code=status,
json=lambda: {
"ResponseMetadata": {"Error": {"Code": "ServiceUnavailable"}}
if api_error
else {},
"Result": {"Decision": {"DecisionType": 1}},
},
)
with patch.object(mod.requests, "post", return_value=response):
assert (
client(mod).moderate("hello", role="user", hook_name="before_dispatch")
== mod.UNAVAILABLE
)


@pytest.mark.parametrize(
Expand Down
25 changes: 14 additions & 11 deletions veadk/tools/builtin_tools/agentsentry_v3.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -152,24 +152,27 @@ def moderate(self, text, *, role, hook_name, session_info=None, tool_calls=None)
return UNAVAILABLE
data = response.json()
result = data.get("Result") or {}
metadata = data.get("ResponseMetadata") or {}
request_id = metadata.get("RequestId")
if metadata.get("Error"):
logger.warning("AgentSentry V3 API error request_id=%s", request_id)
return UNAVAILABLE
decision = (result.get("Decision") or {}).get("DecisionType")
logger.info(
"AgentSentry V3 hook=%s request_id=%s decision=%s degraded=%s",
hook_name,
(data.get("ResponseMetadata") or {}).get("RequestId"),
(result.get("Decision") or {}).get("DecisionType"),
request_id,
decision,
result.get("Degraded"),
)
if (data.get("ResponseMetadata") or {}).get("Error"):
return UNAVAILABLE
if result.get("Degraded") is not False:
return UNAVAILABLE
decision = (result.get("Decision") or {}).get("DecisionType")
if type(decision) is int and decision == 1:
try:
decision = int(decision)
except (ValueError, TypeError, OverflowError):
return None
if type(decision) is int and decision == 2:
if decision == 2:
return BLOCK
# REPLACE/unknown decisions are stopped rather than executing unreviewed args.
return UNAVAILABLE
# Only an explicit BLOCK decision stops a successful moderation response.
return None
except (requests.RequestException, ValueError, TypeError, AttributeError):
logger.warning("AgentSentry V3 transport or response error")
return UNAVAILABLE
Loading