@@ -1870,6 +1870,97 @@ async def test_tool_batch_runner_observation_handoff_pushes_mutation_step( |
| 1870 | 1870 | ) |
| 1871 | 1871 | |
| 1872 | 1872 | |
| 1873 | +@pytest.mark.asyncio |
| 1874 | +async def test_tool_batch_runner_discovery_completion_handoff_stays_persistent( |
| 1875 | + temp_dir: Path, |
| 1876 | +) -> None: |
| 1877 | + async def assess_confidence( |
| 1878 | + tool_name: str, |
| 1879 | + tool_args: dict, |
| 1880 | + context: str, |
| 1881 | + ) -> ConfidenceAssessment: |
| 1882 | + raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 1883 | + |
| 1884 | + async def verify_action( |
| 1885 | + tool_name: str, |
| 1886 | + tool_args: dict, |
| 1887 | + result: str, |
| 1888 | + expected: str = "", |
| 1889 | + ) -> ActionVerification: |
| 1890 | + raise AssertionError("Verification should not run for this scenario") |
| 1891 | + |
| 1892 | + reference = temp_dir / "fortran" / "chapters" / "01-introduction.html" |
| 1893 | + reference.parent.mkdir(parents=True) |
| 1894 | + reference.write_text("<h1>Introduction</h1>\n<p>Guide cadence.</p>\n") |
| 1895 | + |
| 1896 | + context = build_context( |
| 1897 | + temp_dir=temp_dir, |
| 1898 | + messages=[], |
| 1899 | + safeguards=FakeSafeguards(), |
| 1900 | + assess_confidence=assess_confidence, |
| 1901 | + verify_action=verify_action, |
| 1902 | + auto_recover=False, |
| 1903 | + ) |
| 1904 | + persistent_messages: list[str] = [] |
| 1905 | + ephemeral_messages: list[str] = [] |
| 1906 | + context.queue_steering_message_callback = persistent_messages.append |
| 1907 | + context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 1908 | + runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 1909 | + dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 1910 | + sync_todos_to_definition_of_done( |
| 1911 | + dod, |
| 1912 | + [ |
| 1913 | + { |
| 1914 | + "content": "First, examine the existing fortran guide structure and content", |
| 1915 | + "active_form": "Working on: First, examine the existing fortran guide structure and content", |
| 1916 | + "status": "pending", |
| 1917 | + }, |
| 1918 | + { |
| 1919 | + "content": "Create the nginx directory structure", |
| 1920 | + "active_form": "Working on: Create the nginx directory structure", |
| 1921 | + "status": "pending", |
| 1922 | + }, |
| 1923 | + ], |
| 1924 | + ) |
| 1925 | + tool_call = ToolCall( |
| 1926 | + id="read-reference", |
| 1927 | + name="read", |
| 1928 | + arguments={"file_path": str(reference)}, |
| 1929 | + ) |
| 1930 | + executor = FakeExecutor( |
| 1931 | + [ |
| 1932 | + tool_outcome( |
| 1933 | + tool_call=tool_call, |
| 1934 | + output="<h1>Introduction</h1>\n<p>Guide cadence.</p>\n", |
| 1935 | + is_error=False, |
| 1936 | + ) |
| 1937 | + ] |
| 1938 | + ) |
| 1939 | + |
| 1940 | + summary = TurnSummary(final_response="") |
| 1941 | + await runner.execute_batch( |
| 1942 | + tool_calls=[tool_call], |
| 1943 | + tool_source="assistant", |
| 1944 | + pending_tool_calls_seen=set(), |
| 1945 | + emit=_noop_emit, |
| 1946 | + summary=summary, |
| 1947 | + dod=dod, |
| 1948 | + executor=executor, # type: ignore[arg-type] |
| 1949 | + on_confirmation=None, |
| 1950 | + on_user_question=None, |
| 1951 | + emit_confirmation=None, |
| 1952 | + consecutive_errors=0, |
| 1953 | + ) |
| 1954 | + |
| 1955 | + assert persistent_messages |
| 1956 | + assert any( |
| 1957 | + "Continue with the next pending item: `Create the nginx directory structure`" |
| 1958 | + in message |
| 1959 | + for message in persistent_messages |
| 1960 | + ) |
| 1961 | + assert ephemeral_messages == [] |
| 1962 | + |
| 1963 | + |
| 1873 | 1964 | @pytest.mark.asyncio |
| 1874 | 1965 | async def test_tool_batch_runner_missing_artifact_nudge_prefers_pending_index_after_mkdir( |
| 1875 | 1966 | temp_dir: Path, |
