Merge 5afc96f998 into 87f2d783d6

strix/report/coverage.py

@@ -55,7 +55,12 @@ OUTCOME_LABELS: dict[str, str] = {
 }
 
 #: Statuses that mean the agent stopped early rather than finishing its task.
-_INCOMPLETE_AGENT_STATUSES = frozenset({"crashed", "stopped", "running", "waiting"})
+#: ``failed`` and ``budget_paused`` are in ``core.agents.Status`` too, and an
+#: agent wearing either never finished the surfaces it held, so they count as
+#: incomplete here just like ``crashed`` does.
+_INCOMPLETE_AGENT_STATUSES = frozenset(
+    {"crashed", "stopped", "running", "waiting", "failed", "budget_paused"}
+)
 
 #: Run statuses that mean the scan itself did not run to completion.
 _INCOMPLETE_RUN_STATUSES = frozenset({"failed", "interrupted", "stopped", "running"})

tests/test_report_coverage.py

@@ -154,6 +154,27 @@ def test_unfinished_agent_makes_the_record_partial() -> None:
     assert "authz-tester" in doc["completeness"]["caveats"][0]
 
 
+def test_failed_agent_makes_the_record_partial() -> None:
+    """A failed agent must not let the record read as complete (#1458)."""
+    doc = _document(
+        agent_graph=_graph(statuses={"agent-1": "failed"}),
+        exit_reason="finished_by_tool",
+    )
+
+    assert doc["completeness"]["complete"] is False
+    assert "authz-tester" in doc["completeness"]["caveats"][0]
+
+
+def test_budget_paused_agent_makes_the_record_partial() -> None:
+    doc = _document(
+        agent_graph=_graph(statuses={"agent-1": "budget_paused"}),
+        exit_reason="finished_by_tool",
+    )
+
+    assert doc["completeness"]["complete"] is False
+    assert "authz-tester" in doc["completeness"]["caveats"][0]
+
+
 def test_failed_run_status_makes_the_record_partial() -> None:
     doc = _document(
         run_record={"run_id": "r1", "status": "failed"},