fix(strix): treat ModelBehaviorError with zero findings as backend flake

.github/workflows/strix.yml

@@ -844,7 +844,8 @@ jobs:
           # exit 1 both for genuine blocking vulnerabilities AND for
           # LLM-backend-unavailable outcomes (GitHub Models "Too many requests"
           # rate limits, OpenAI quota starvation, 413 tokens_limit_reached
-          # token-cap, connection/warm-up failures) that could not complete a scan. A backend outage is CI
+          # token-cap, context-window overflow, connection/warm-up failures, and
+          # scanner ModelBehaviorError flakes) that could not complete a scan. A backend outage is CI
           # infrastructure noise, not a security finding, so it must not fail
           # the required check and block merges.
           strix_run_log="$RUNNER_TEMP/strix_gate_console.log"
@@ -866,7 +867,7 @@ jobs:
           fi
 
           # Recognized signals that the LLM backend was unavailable / starved.
-          backend_unavailable_signal='RateLimitError|Too many requests\. For more on scraping GitHub|exceeded your current quota|insufficient_quota|billing details|"status"[[:space:]]*:[[:space:]]*"RESOURCE_EXHAUSTED"|tokens_limit_reached|Request body too large|Max size:[[:space:]]*[0-9]+[[:space:]]+tokens|Error code:[[:space:]]*413|LLM CONNECTION FAILED|Could not establish connection to the language model|LLM warm-up failed|Configured model and fallback models were unavailable|Configured Vertex model and fallback models were unavailable|emitted provider infrastructure or failure-signal output|before provider infrastructure failure|litellm(\.exceptions)?\.NotFoundError[^[:cntrl:]]*Nvidia_nimException[^[:cntrl:]]*Error code:[[:space:]]*404'
+          backend_unavailable_signal='RateLimitError|Too many requests\. For more on scraping GitHub|exceeded your current quota|insufficient_quota|billing details|"status"[[:space:]]*:[[:space:]]*"RESOURCE_EXHAUSTED"|tokens_limit_reached|Request body too large|Max size:[[:space:]]*[0-9]+[[:space:]]+tokens|Error code:[[:space:]]*413|ContextWindowExceededError|maximum context length|messages resulted in[[:space:]]+[0-9]+[[:space:]]+tokens|LLM CONNECTION FAILED|Could not establish connection to the language model|LLM warm-up failed|Configured model and fallback models were unavailable|Configured Vertex model and fallback models were unavailable|emitted provider infrastructure or failure-signal output|before provider infrastructure failure|litellm(\.exceptions)?\.NotFoundError[^[:cntrl:]]*Nvidia_nimException[^[:cntrl:]]*Error code:[[:space:]]*404|agents\.exceptions\.ModelBehaviorError:[[:space:]]*'
           # Any evidence that a vulnerability was actually reported. Its presence
           # forces a hard failure so real findings are NEVER downgraded. Keep the
           # severity branch anchored away from identifiers so environment lines

CHANGELOG.md

@@ -8,6 +8,11 @@ Semantic Versioning where the repository publishes a release.
 
 ### Added
 
+- Classify Strix `ModelBehaviorError` with zero reported vulnerabilities as a
+  backend-unavailable flake so a scanner-model failure does not block the
+  required check. `Vulnerabilities [1-9]` and a `severity:` finding still
+  fail closed.
+
 - Added an hourly organization commercial-readiness coordinator that discovers writable repositories, honors enabled dedicated writer leases and fully paginated live writer runs, refetches exact repository/workflow/run/PR state before dispatch, rotates bounded review-repair and opt-in NVIDIA OpenCode product-development targets, fails nonzero on fleet-wide inspection or dispatch outages, retains three-day JSON receipts, and keeps the existing 15-minute merge scheduler authoritative.
 - Added a dedicated Quarantine Sandbox Runtime hourly caller at minute 14 that targets protected `develop`, dispatches at most one exact-head repair, applies a two-hour same-head retry floor, preserves non-cancelling single-flight execution, and maps only the established scheduler credentials with job-scoped OIDC.
 - Added a dedicated Quarantine Sandbox Runtime hourly caller at minute 14 that targets protected `develop`, dispatches at most one exact-head repair, applies a two-hour same-head retry floor, preserves non-cancelling single-flight execution, and maps only the established scheduler credentials with job-scoped OIDC.

docs/doctoring/strix-modelbehaviorerror-classifier.md

@@ -0,0 +1,21 @@
+# Strix ModelBehaviorError classifier
+
+Observed required-check flake: Strix exits 1 with `ModelBehaviorError` and
+`Vulnerabilities 0` after the scanner model fails to complete a turn.
+
+The outer `strix.yml` backend-unavailable signal now includes
+`ModelBehaviorError`. Neutral skip still requires the absence of
+`Vulnerabilities [1-9]` and of a `severity:` finding. A scan that reports
+any numbered vulnerability stays fail-closed even when the model also
+emits `ModelBehaviorError`.
+
+## Operator action
+
+If this check repeats, inspect the exact Strix log for the qualified exception
+and confirm that no numbered vulnerability or `severity:` signal is present;
+any finding remains fail-closed before retrying.
+
+## References
+
+OpenAI. (n.d.). *Exceptions*. OpenAI Agents SDK. Retrieved August 19, 2026,
+from https://openai.github.io/openai-agents-python/ref/exceptions/

tests/test_required_workflow_queue_contract.py

@@ -1099,7 +1099,12 @@ def test_strix_provider_outage_without_findings_is_neutralized() -> None:
     assert "exceeded your current quota" in workflow
     assert "billing details" in workflow
     assert "LLM warm-up failed" in workflow
+    assert "agents\\.exceptions\\.ModelBehaviorError:[[:space:]]*" in workflow
+    assert "ContextWindowExceededError" in workflow
+    assert "maximum context length" in workflow
+    assert "messages resulted in[[:space:]]+[0-9]+[[:space:]]+tokens" in workflow
     assert "zero_vulnerabilities_signal" not in workflow
+    assert "Vulnerabilities[[:space:]]+[1-9]" in workflow
     assert "(^|[^A-Za-z0-9_])severity[[:space:]]*:" in workflow
     assert "STRIX_FAIL_ON_MIN_SEVERITY: MEDIUM" in workflow
     assert "before producing a vulnerability report" in workflow

tests/test_strix_nvidia_nim_not_found_fallback.py

@@ -251,11 +251,73 @@ def test_workflow_neutralizes_only_nvidia_404_without_findings(self) -> None:
         self.assertIn("Error code:[[:space:]]*404", workflow)
         self.assertIn("reported_vulnerability_signal", workflow)
         self.assertIn("Vulnerabilities[[:space:]]+[1-9]", workflow)
+        self.assertIn(
+            "agents\\.exceptions\\.ModelBehaviorError:[[:space:]]*",
+            workflow,
+        )
         self.assertIn(
             '! grep -Eiq "$reported_vulnerability_signal"',
             workflow,
         )
 
+    def test_outer_workflow_neutralizes_model_behavior_error_without_findings(
+        self,
+    ) -> None:
+        """Require the actual scanner ModelBehaviorError format before neutralizing."""
+
+        self.assertFalse(
+            _workflow_neutralizes("ModelBehaviorError\nVulnerabilities 0\n")
+        )
+        self.assertTrue(
+            _workflow_neutralizes(
+                "agents.exceptions.ModelBehaviorError: provider response failed\n"
+                "Vulnerabilities 0\n"
+            )
+        )
+
+    def test_outer_workflow_never_neutralizes_model_behavior_error_with_findings(
+        self,
+    ) -> None:
+        """Keep Vulnerabilities [1-9] fail-closed for the actual model exception."""
+
+        self.assertFalse(
+            _workflow_neutralizes(
+                "agents.exceptions.ModelBehaviorError: provider response failed\n"
+                "Vulnerabilities 1\n"
+            )
+        )
+        self.assertFalse(
+            _workflow_neutralizes(
+                "agents.exceptions.ModelBehaviorError: provider response failed\n"
+                "Vulnerabilities 9\n"
+            )
+        )
+
+    def test_outer_workflow_neutralizes_context_window_overflow_without_findings(
+        self,
+    ) -> None:
+        """Treat a provider context overflow as backend failure, not a finding."""
+
+        self.assertTrue(
+            _workflow_neutralizes(
+                "openai.BadRequestError: maximum context length is 1000000 "
+                "tokens; messages resulted in 1438805 tokens\n"
+                "Vulnerabilities 0\n"
+            )
+        )
+
+    def test_outer_workflow_never_neutralizes_context_overflow_with_findings(
+        self,
+    ) -> None:
+        """Keep context overflow plus a reported vulnerability fail-closed."""
+
+        self.assertFalse(
+            _workflow_neutralizes(
+                "openai.BadRequestError: ContextWindowExceededError\n"
+                "Vulnerabilities 1\n"
+            )
+        )
+
 
 if __name__ == "__main__":
     unittest.main()