ouroboros-ai 0.40.2.dev10__tar.gz → 0.40.2.dev12__tar.gz

{ouroboros_ai-0.40.2.dev10 → ouroboros_ai-0.40.2.dev12}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ouroboros-ai
-Version: 0.40.2.dev10
+Version: 0.40.2.dev12
 Summary: Specification-first workflow engine for AI coding agents. Works with Claude Code and Codex CLI.
 Author-email: Q00 <jqyu.lee@gmail.com>
 License-File: LICENSE

{ouroboros_ai-0.40.2.dev10 → ouroboros_ai-0.40.2.dev12}/src/ouroboros/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '0.40.2.dev10'
-__version_tuple__ = version_tuple = (0, 40, 2, 'dev10')
+__version__ = version = '0.40.2.dev12'
+__version_tuple__ = version_tuple = (0, 40, 2, 'dev12')
 
 __commit_id__ = commit_id = None

{ouroboros_ai-0.40.2.dev10 → ouroboros_ai-0.40.2.dev12}/src/ouroboros/cli/commands/setup.py

@@ -351,7 +351,7 @@ _CODEX_ROLE_MODEL_OVERRIDE_DEFAULTS: dict[str, tuple[tuple[tuple[str, ...], obje
         (("consensus", "devil_model"), "openrouter/openai/gpt-4o"),
     ),
     "pm_interview": ((("clarification", "default_model"), "claude-opus-4-6"),),
-    "qa": ((("llm", "qa_model"), "claude-sonnet-4-20250514"),),
+    "qa": ((("llm", "qa_model"), "claude-sonnet-4-6"),),
     "reflect": ((("resilience", "reflect_model"), "claude-opus-4-6"),),
     "seed_generation": ((("clarification", "default_model"), "claude-opus-4-6"),),
     "semantic_evaluation": ((("evaluation", "semantic_model"), "claude-opus-4-6"),),
@@ -1588,7 +1588,7 @@ def _register_copilot_mcp_server() -> None:
 
 
 _COPILOT_DEFAULT_MODEL_TARGETS: tuple[tuple[str, str, str], ...] = (
-    ("llm", "qa_model", "claude-sonnet-4-20250514"),
+    ("llm", "qa_model", "claude-sonnet-4-6"),
     ("llm", "dependency_analysis_model", "claude-opus-4-6"),
     ("llm", "ontology_analysis_model", "claude-opus-4-6"),
     ("llm", "context_compression_model", "gpt-4"),

{ouroboros_ai-0.40.2.dev10 → ouroboros_ai-0.40.2.dev12}/src/ouroboros/config/loader.py

@@ -1336,11 +1336,11 @@ def get_qa_model(backend: str | None = None) -> str:
         config = load_config()
         return _normalize_configured_model_for_backend(
             config.llm.qa_model,
-            default_model="claude-sonnet-4-20250514",
+            default_model="claude-sonnet-4-6",
             backend=backend,
         )
     except ConfigError:
-        return _default_model_for_backend("claude-sonnet-4-20250514", backend=backend)
+        return _default_model_for_backend("claude-sonnet-4-6", backend=backend)
 
 
 def get_dependency_analysis_model(backend: str | None = None) -> str:

{ouroboros_ai-0.40.2.dev10 → ouroboros_ai-0.40.2.dev12}/src/ouroboros/config/models.py

@@ -131,7 +131,7 @@ class LLMConfig(BaseModel, frozen=True):
     ] = "claude_code"
     permission_mode: Literal["default", "acceptEdits", "bypassPermissions"] = "default"
     opencode_permission_mode: Literal["default", "acceptEdits", "bypassPermissions"] = "acceptEdits"
-    qa_model: str = "claude-sonnet-4-20250514"
+    qa_model: str = "claude-sonnet-4-6"
     dependency_analysis_model: str = "claude-opus-4-6"
     ontology_analysis_model: str = "claude-opus-4-6"
     context_compression_model: str = "gpt-4"

{ouroboros_ai-0.40.2.dev10 → ouroboros_ai-0.40.2.dev12}/src/ouroboros/orchestrator/pi_runtime.py

@@ -262,6 +262,12 @@ class PiRuntime:
             return None
         return event if isinstance(event, dict) else None
 
+    def _malformed_event_message(self, line: str) -> str:
+        preview = line.strip()
+        if len(preview) > 240:
+            preview = f"{preview[:237]}..."
+        return f"Malformed {self._display_name} JSON event: {preview}"
+
     def _extract_session_id(self, event: dict[str, Any]) -> str | None:
         """Extract session ID from the pi session header event."""
         if event.get("type") == "session":
@@ -504,7 +510,19 @@ class PiRuntime:
                         continue
                     event = self._parse_event(line)
                     if event is None:
-                        continue
+                        if process is not None:
+                            await self._terminate_process(process)
+                        process_terminated = True
+                        yield AgentMessage(
+                            type="result",
+                            content=self._malformed_event_message(line),
+                            data={
+                                "subtype": "error",
+                                "error_type": "MalformedPiEvent",
+                            },
+                            resume_handle=current_handle,
+                        )
+                        return
 
                     # Session header — extract ID
                     sid = self._extract_session_id(event)

{ouroboros_ai-0.40.2.dev10 → ouroboros_ai-0.40.2.dev12}/tests/unit/config/test_models.py

@@ -201,7 +201,7 @@ class TestLLMConfig:
         assert config.backend == "claude_code"
         assert config.permission_mode == "default"
         assert config.opencode_permission_mode == "acceptEdits"
-        assert config.qa_model == "claude-sonnet-4-20250514"
+        assert config.qa_model == "claude-sonnet-4-6"
         assert config.dependency_analysis_model == "claude-opus-4-6"
         assert config.ontology_analysis_model == "claude-opus-4-6"
         assert config.context_compression_model == "gpt-4"

{ouroboros_ai-0.40.2.dev10 → ouroboros_ai-0.40.2.dev12}/tests/unit/orchestrator/test_pi_runtime.py

@@ -257,6 +257,50 @@ async def test_agent_end_does_not_mask_nonzero_exit() -> None:
     assert result.data["error_type"] == "PiError"
 
 
+@pytest.mark.asyncio
+async def test_execute_task_reports_malformed_json_event() -> None:
+    process = _FakeProcess(
+        stdout_lines=[
+            _jsonl_event({"type": "session", "id": "session-1"}),
+            "not-json",
+        ],
+        stderr_lines=[],
+        returncode=0,
+    )
+    runtime = PiRuntime(cli_path="/tmp/pi", cwd="/tmp/project")
+
+    with patch("asyncio.create_subprocess_exec", return_value=process):
+        messages = [msg async for msg in runtime.execute_task("Do it")]
+
+    result = [m for m in messages if m.type == "result"][-1]
+    assert result.is_error
+    assert result.content == "Malformed Pi JSON event: not-json"
+    assert result.data == {
+        "subtype": "error",
+        "error_type": "MalformedPiEvent",
+    }
+    assert result.resume_handle is not None
+    assert result.resume_handle.native_session_id == "session-1"
+    assert process.terminated
+
+
+@pytest.mark.asyncio
+async def test_execute_task_to_result_maps_malformed_event_to_provider_error() -> None:
+    process = _FakeProcess(stdout_lines=["[bad-json]"], stderr_lines=[], returncode=0)
+    runtime = PiRuntime(cli_path="/tmp/pi", cwd="/tmp/project")
+
+    with patch("asyncio.create_subprocess_exec", return_value=process):
+        result = await runtime.execute_task_to_result("Do it")
+
+    assert result.is_err
+    error = result.error
+    assert error is not None
+    assert error.provider == "pi"
+    assert error.message == "Malformed Pi JSON event: [bad-json]"
+    assert error.details == {"messages": ["Malformed Pi JSON event: [bad-json]"]}
+    assert process.terminated
+
+
 def test_runtime_factory_constructs_pi_runtime() -> None:
     from ouroboros.orchestrator.runtime_factory import create_agent_runtime