| 1 | """Tests for tool-batch execution on RuntimeContext.""" |
| 2 | |
| 3 | from __future__ import annotations |
| 4 | |
| 5 | from pathlib import Path |
| 6 | from types import SimpleNamespace |
| 7 | |
| 8 | import pytest |
| 9 | |
| 10 | from loader.llm.base import Message, Role, ToolCall |
| 11 | from loader.runtime.context import RuntimeContext |
| 12 | from loader.runtime.dod import ( |
| 13 | DefinitionOfDoneStore, |
| 14 | VerificationEvidence, |
| 15 | create_definition_of_done, |
| 16 | ) |
| 17 | from loader.runtime.events import AgentEvent, TurnSummary |
| 18 | from loader.runtime.executor import ToolExecutionOutcome, ToolExecutionState |
| 19 | from loader.runtime.path_display import display_runtime_path |
| 20 | from loader.runtime.permissions import ( |
| 21 | PermissionMode, |
| 22 | build_permission_policy, |
| 23 | load_permission_rules, |
| 24 | ) |
| 25 | from loader.runtime.reasoning_types import ( |
| 26 | ActionVerification, |
| 27 | ConfidenceAssessment, |
| 28 | ConfidenceLevel, |
| 29 | ) |
| 30 | from loader.runtime.recovery import RecoveryContext |
| 31 | from loader.runtime.tool_batches import ( |
| 32 | ToolBatchRunner, |
| 33 | ) |
| 34 | from loader.runtime.tool_batches import ( |
| 35 | _should_prioritize_missing_artifact as tool_batches_should_prioritize_missing_artifact, |
| 36 | ) |
| 37 | from loader.runtime.workflow import sync_todos_to_definition_of_done |
| 38 | from loader.tools.base import ToolResult as RegistryToolResult |
| 39 | from loader.tools.base import create_default_registry |
| 40 | from tests.helpers.runtime_harness import ScriptedBackend |
| 41 | |
| 42 | |
| 43 | class FakeSession: |
| 44 | def __init__(self, messages: list[Message]) -> None: |
| 45 | self.messages = list(messages) |
| 46 | self.workflow_timeline = [] |
| 47 | |
| 48 | def append(self, message: Message) -> None: |
| 49 | self.messages.append(message) |
| 50 | |
| 51 | def append_workflow_timeline_entry(self, entry) -> None: |
| 52 | self.workflow_timeline.append(entry) |
| 53 | |
| 54 | |
| 55 | class FakeCodeFilter: |
| 56 | def reset(self) -> None: |
| 57 | return None |
| 58 | |
| 59 | |
| 60 | class FakeSafeguards: |
| 61 | def __init__(self, *, detect_loop_result: tuple[bool, str] = (False, "")) -> None: |
| 62 | self.action_tracker = object() |
| 63 | self.validator = object() |
| 64 | self.code_filter = FakeCodeFilter() |
| 65 | self._detect_loop_result = detect_loop_result |
| 66 | |
| 67 | def filter_stream_chunk(self, content: str) -> str: |
| 68 | return content |
| 69 | |
| 70 | def filter_complete_content(self, content: str) -> str: |
| 71 | return content |
| 72 | |
| 73 | def should_steer(self) -> bool: |
| 74 | return False |
| 75 | |
| 76 | def get_steering_message(self) -> str | None: |
| 77 | return None |
| 78 | |
| 79 | def record_response(self, content: str) -> None: |
| 80 | return None |
| 81 | |
| 82 | def detect_text_loop(self, content: str) -> tuple[bool, str]: |
| 83 | return False, "" |
| 84 | |
| 85 | def detect_loop(self) -> tuple[bool, str]: |
| 86 | return self._detect_loop_result |
| 87 | |
| 88 | |
| 89 | class FakeExecutor: |
| 90 | def __init__(self, outcomes: list[ToolExecutionOutcome]) -> None: |
| 91 | self._outcomes = list(outcomes) |
| 92 | self.calls: list[ToolCall] = [] |
| 93 | |
| 94 | async def execute_tool_call(self, tool_call: ToolCall, **_: object) -> ToolExecutionOutcome: |
| 95 | self.calls.append(tool_call) |
| 96 | if not self._outcomes: |
| 97 | raise AssertionError("No fake tool outcome queued") |
| 98 | return self._outcomes.pop(0) |
| 99 | |
| 100 | |
| 101 | def build_context( |
| 102 | *, |
| 103 | temp_dir: Path, |
| 104 | messages: list[Message], |
| 105 | safeguards: FakeSafeguards, |
| 106 | assess_confidence, |
| 107 | verify_action, |
| 108 | recovery_context: RecoveryContext | None = None, |
| 109 | confidence_scoring: bool = False, |
| 110 | verification: bool = False, |
| 111 | auto_recover: bool = True, |
| 112 | min_confidence_for_action: int = 3, |
| 113 | ) -> RuntimeContext: |
| 114 | registry = create_default_registry(temp_dir) |
| 115 | registry.configure_workspace_root(temp_dir) |
| 116 | rule_status = load_permission_rules(temp_dir) |
| 117 | policy = build_permission_policy( |
| 118 | active_mode=PermissionMode.WORKSPACE_WRITE, |
| 119 | workspace_root=temp_dir, |
| 120 | tool_requirements=registry.get_tool_requirements(), |
| 121 | rules=rule_status.rules, |
| 122 | ) |
| 123 | context = RuntimeContext( |
| 124 | project_root=temp_dir, |
| 125 | backend=ScriptedBackend(), |
| 126 | registry=registry, |
| 127 | session=FakeSession(messages), # type: ignore[arg-type] |
| 128 | config=SimpleNamespace( |
| 129 | force_react=False, |
| 130 | max_recovery_attempts=2, |
| 131 | auto_recover=auto_recover, |
| 132 | reasoning=SimpleNamespace( |
| 133 | rollback=False, |
| 134 | show_rollback_plan=False, |
| 135 | completion_check=True, |
| 136 | max_continuation_prompts=5, |
| 137 | self_critique=False, |
| 138 | confidence_scoring=confidence_scoring, |
| 139 | min_confidence_for_action=min_confidence_for_action, |
| 140 | verification=verification, |
| 141 | ), |
| 142 | ), |
| 143 | capability_profile=SimpleNamespace(supports_native_tools=True), # type: ignore[arg-type] |
| 144 | project_context=None, |
| 145 | permission_policy=policy, |
| 146 | permission_config_status=rule_status, |
| 147 | workflow_mode="execute", |
| 148 | safeguards=safeguards, |
| 149 | reasoning=SimpleNamespace( |
| 150 | assess_confidence=assess_confidence, |
| 151 | verify_action=verify_action, |
| 152 | ), |
| 153 | recovery_context=recovery_context, |
| 154 | ) |
| 155 | return context |
| 156 | |
| 157 | |
| 158 | def tool_outcome( |
| 159 | *, |
| 160 | tool_call: ToolCall, |
| 161 | output: str, |
| 162 | is_error: bool, |
| 163 | state: ToolExecutionState = ToolExecutionState.EXECUTED, |
| 164 | metadata: dict[str, object] | None = None, |
| 165 | ) -> ToolExecutionOutcome: |
| 166 | return ToolExecutionOutcome( |
| 167 | tool_call=tool_call, |
| 168 | state=state, |
| 169 | message=Message.tool_result_message( |
| 170 | tool_call_id=tool_call.id, |
| 171 | display_content=output, |
| 172 | result_content=output, |
| 173 | is_error=is_error, |
| 174 | ), |
| 175 | event_content=output, |
| 176 | is_error=is_error, |
| 177 | result_output=output, |
| 178 | registry_result=RegistryToolResult( |
| 179 | output=output, |
| 180 | is_error=is_error, |
| 181 | metadata=metadata or {}, |
| 182 | ), |
| 183 | ) |
| 184 | |
| 185 | |
| 186 | @pytest.mark.asyncio |
| 187 | async def test_tool_batch_runner_uses_context_for_confidence_gate(temp_dir: Path) -> None: |
| 188 | captured: dict[str, str] = {} |
| 189 | |
| 190 | async def assess_confidence(tool_name: str, tool_args: dict, context: str) -> ConfidenceAssessment: |
| 191 | captured["context"] = context |
| 192 | return ConfidenceAssessment( |
| 193 | action=f"{tool_name} with {tool_args}", |
| 194 | tool_name=tool_name, |
| 195 | tool_args=tool_args, |
| 196 | level=ConfidenceLevel.LOW, |
| 197 | reasoning="Need to inspect the target first.", |
| 198 | risks=["Unknown target file"], |
| 199 | ) |
| 200 | |
| 201 | async def verify_action(tool_name: str, tool_args: dict, result: str, expected: str = "") -> ActionVerification: |
| 202 | raise AssertionError("Verification should not run for skipped actions") |
| 203 | |
| 204 | context = build_context( |
| 205 | temp_dir=temp_dir, |
| 206 | messages=[ |
| 207 | Message(role=Role.USER, content="Please inspect the project."), |
| 208 | Message(role=Role.ASSISTANT, content="I will read the file next."), |
| 209 | ], |
| 210 | safeguards=FakeSafeguards(), |
| 211 | assess_confidence=assess_confidence, |
| 212 | verify_action=verify_action, |
| 213 | confidence_scoring=True, |
| 214 | min_confidence_for_action=3, |
| 215 | ) |
| 216 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 217 | tool_call = ToolCall(id="read-1", name="read", arguments={"file_path": "README.md"}) |
| 218 | events: list[AgentEvent] = [] |
| 219 | |
| 220 | async def emit(event: AgentEvent) -> None: |
| 221 | events.append(event) |
| 222 | |
| 223 | executor = FakeExecutor([tool_outcome(tool_call=tool_call, output="unused", is_error=False)]) |
| 224 | result = await runner.execute_batch( |
| 225 | tool_calls=[tool_call], |
| 226 | tool_source="assistant", |
| 227 | pending_tool_calls_seen=set(), |
| 228 | emit=emit, |
| 229 | summary=TurnSummary(final_response=""), |
| 230 | dod=create_definition_of_done("Read the docs"), |
| 231 | executor=executor, # type: ignore[arg-type] |
| 232 | on_confirmation=None, |
| 233 | on_user_question=None, |
| 234 | emit_confirmation=None, |
| 235 | consecutive_errors=0, |
| 236 | ) |
| 237 | |
| 238 | assert result.actions_taken == [] |
| 239 | assert executor.calls == [] |
| 240 | assert "Please inspect the project." in captured["context"] |
| 241 | assert context.session.messages[-1].role == Role.USER |
| 242 | assert "[LOW CONFIDENCE WARNING]" in context.session.messages[-1].content |
| 243 | event_types = [event.type for event in events] |
| 244 | assert "confidence" in event_types |
| 245 | |
| 246 | |
| 247 | @pytest.mark.asyncio |
| 248 | async def test_tool_batch_runner_tracks_recovery_with_legacy_context(temp_dir: Path) -> None: |
| 249 | async def assess_confidence(tool_name: str, tool_args: dict, context: str) -> ConfidenceAssessment: |
| 250 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 251 | |
| 252 | async def verify_action(tool_name: str, tool_args: dict, result: str, expected: str = "") -> ActionVerification: |
| 253 | raise AssertionError("Verification should not run for failed actions") |
| 254 | |
| 255 | context = build_context( |
| 256 | temp_dir=temp_dir, |
| 257 | messages=[], |
| 258 | safeguards=FakeSafeguards(), |
| 259 | assess_confidence=assess_confidence, |
| 260 | verify_action=verify_action, |
| 261 | auto_recover=True, |
| 262 | ) |
| 263 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 264 | tool_call = ToolCall(id="bash-1", name="bash", arguments={"command": "pytest"}) |
| 265 | executor = FakeExecutor([tool_outcome(tool_call=tool_call, output="command failed", is_error=True)]) |
| 266 | summary = TurnSummary(final_response="") |
| 267 | events: list[AgentEvent] = [] |
| 268 | |
| 269 | async def emit(event: AgentEvent) -> None: |
| 270 | events.append(event) |
| 271 | |
| 272 | await runner.execute_batch( |
| 273 | tool_calls=[tool_call], |
| 274 | tool_source="assistant", |
| 275 | pending_tool_calls_seen=set(), |
| 276 | emit=emit, |
| 277 | summary=summary, |
| 278 | dod=create_definition_of_done("Run tests"), |
| 279 | executor=executor, # type: ignore[arg-type] |
| 280 | on_confirmation=None, |
| 281 | on_user_question=None, |
| 282 | emit_confirmation=None, |
| 283 | consecutive_errors=0, |
| 284 | ) |
| 285 | |
| 286 | assert context.recovery_context is not None |
| 287 | assert summary.tool_result_messages |
| 288 | assert context.session.messages[-1] == summary.tool_result_messages[-1] |
| 289 | assert any(event.type == "recovery" for event in events) |
| 290 | |
| 291 | |
| 292 | @pytest.mark.asyncio |
| 293 | async def test_tool_batch_runner_emits_tool_metadata(temp_dir: Path) -> None: |
| 294 | async def assess_confidence(tool_name: str, tool_args: dict, context: str) -> ConfidenceAssessment: |
| 295 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 296 | |
| 297 | async def verify_action(tool_name: str, tool_args: dict, result: str, expected: str = "") -> ActionVerification: |
| 298 | raise AssertionError("Verification should not run for this scenario") |
| 299 | |
| 300 | context = build_context( |
| 301 | temp_dir=temp_dir, |
| 302 | messages=[], |
| 303 | safeguards=FakeSafeguards(), |
| 304 | assess_confidence=assess_confidence, |
| 305 | verify_action=verify_action, |
| 306 | auto_recover=False, |
| 307 | ) |
| 308 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 309 | tool_call = ToolCall( |
| 310 | id="bash-1", |
| 311 | name="bash", |
| 312 | arguments={"command": "python -m http.server 8000", "background": True}, |
| 313 | ) |
| 314 | metadata = { |
| 315 | "job_id": "bash-1", |
| 316 | "status": "running", |
| 317 | "background": True, |
| 318 | } |
| 319 | executor = FakeExecutor( |
| 320 | [ |
| 321 | tool_outcome( |
| 322 | tool_call=tool_call, |
| 323 | output="Started bash job bash-1", |
| 324 | is_error=False, |
| 325 | metadata=metadata, |
| 326 | ) |
| 327 | ] |
| 328 | ) |
| 329 | events: list[AgentEvent] = [] |
| 330 | |
| 331 | async def emit(event: AgentEvent) -> None: |
| 332 | events.append(event) |
| 333 | |
| 334 | await runner.execute_batch( |
| 335 | tool_calls=[tool_call], |
| 336 | tool_source="assistant", |
| 337 | pending_tool_calls_seen=set(), |
| 338 | emit=emit, |
| 339 | summary=TurnSummary(final_response=""), |
| 340 | dod=create_definition_of_done("Launch a preview server"), |
| 341 | executor=executor, # type: ignore[arg-type] |
| 342 | on_confirmation=None, |
| 343 | on_user_question=None, |
| 344 | emit_confirmation=None, |
| 345 | consecutive_errors=0, |
| 346 | ) |
| 347 | |
| 348 | tool_result = next(event for event in events if event.type == "tool_result") |
| 349 | assert tool_result.tool_metadata == metadata |
| 350 | |
| 351 | |
| 352 | @pytest.mark.asyncio |
| 353 | async def test_tool_batch_runner_verifies_with_context_services(temp_dir: Path) -> None: |
| 354 | verification_calls: list[str] = [] |
| 355 | |
| 356 | async def assess_confidence(tool_name: str, tool_args: dict, context: str) -> ConfidenceAssessment: |
| 357 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 358 | |
| 359 | async def verify_action(tool_name: str, tool_args: dict, result: str, expected: str = "") -> ActionVerification: |
| 360 | verification_calls.append(result) |
| 361 | return ActionVerification( |
| 362 | tool_name=tool_name, |
| 363 | tool_args=tool_args, |
| 364 | expected_outcome="Success", |
| 365 | actual_result=result, |
| 366 | verified=False, |
| 367 | discrepancies=["File contents did not match"], |
| 368 | needs_correction=True, |
| 369 | correction_suggestion="Read the file before editing again.", |
| 370 | ) |
| 371 | |
| 372 | existing_recovery = RecoveryContext( |
| 373 | original_tool="edit", |
| 374 | original_args={"file_path": "README.md"}, |
| 375 | ) |
| 376 | context = build_context( |
| 377 | temp_dir=temp_dir, |
| 378 | messages=[], |
| 379 | safeguards=FakeSafeguards(), |
| 380 | assess_confidence=assess_confidence, |
| 381 | verify_action=verify_action, |
| 382 | recovery_context=existing_recovery, |
| 383 | verification=True, |
| 384 | ) |
| 385 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 386 | tool_call = ToolCall(id="read-1", name="read", arguments={"file_path": "README.md"}) |
| 387 | executor = FakeExecutor([tool_outcome(tool_call=tool_call, output="file contents", is_error=False)]) |
| 388 | events: list[AgentEvent] = [] |
| 389 | |
| 390 | async def emit(event: AgentEvent) -> None: |
| 391 | events.append(event) |
| 392 | |
| 393 | await runner.execute_batch( |
| 394 | tool_calls=[tool_call], |
| 395 | tool_source="assistant", |
| 396 | pending_tool_calls_seen=set(), |
| 397 | emit=emit, |
| 398 | summary=TurnSummary(final_response=""), |
| 399 | dod=create_definition_of_done("Read the docs"), |
| 400 | executor=executor, # type: ignore[arg-type] |
| 401 | on_confirmation=None, |
| 402 | on_user_question=None, |
| 403 | emit_confirmation=None, |
| 404 | consecutive_errors=0, |
| 405 | ) |
| 406 | |
| 407 | assert verification_calls == ["file contents"] |
| 408 | assert context.recovery_context is existing_recovery |
| 409 | assert existing_recovery.successful_steps == [ |
| 410 | ("read", {"file_path": "README.md"}) |
| 411 | ] |
| 412 | assert context.session.messages[-1].role == Role.TOOL |
| 413 | assert context.session.messages[-1].content == "file contents" |
| 414 | assert any(event.type == "verification" for event in events) |
| 415 | |
| 416 | |
| 417 | @pytest.mark.asyncio |
| 418 | async def test_tool_batch_runner_preserves_recovery_context_across_diagnostic_success( |
| 419 | temp_dir: Path, |
| 420 | ) -> None: |
| 421 | async def assess_confidence( |
| 422 | tool_name: str, |
| 423 | tool_args: dict, |
| 424 | context: str, |
| 425 | ) -> ConfidenceAssessment: |
| 426 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 427 | |
| 428 | async def verify_action( |
| 429 | tool_name: str, |
| 430 | tool_args: dict, |
| 431 | result: str, |
| 432 | expected: str = "", |
| 433 | ) -> ActionVerification: |
| 434 | raise AssertionError("Verification should not run for this scenario") |
| 435 | |
| 436 | existing_recovery = RecoveryContext( |
| 437 | original_tool="read", |
| 438 | original_args={"file_path": "chapters/04-data-types.html"}, |
| 439 | ) |
| 440 | existing_recovery.add_attempt( |
| 441 | "read", |
| 442 | {"file_path": "chapters/04-data-types.html"}, |
| 443 | "File not found", |
| 444 | ) |
| 445 | context = build_context( |
| 446 | temp_dir=temp_dir, |
| 447 | messages=[], |
| 448 | safeguards=FakeSafeguards(), |
| 449 | assess_confidence=assess_confidence, |
| 450 | verify_action=verify_action, |
| 451 | recovery_context=existing_recovery, |
| 452 | auto_recover=False, |
| 453 | ) |
| 454 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 455 | tool_call = ToolCall( |
| 456 | id="bash-1", |
| 457 | name="bash", |
| 458 | arguments={"command": "ls chapters"}, |
| 459 | ) |
| 460 | executor = FakeExecutor( |
| 461 | [tool_outcome(tool_call=tool_call, output="01-introduction.html", is_error=False)] |
| 462 | ) |
| 463 | |
| 464 | summary = TurnSummary(final_response="") |
| 465 | await runner.execute_batch( |
| 466 | tool_calls=[tool_call], |
| 467 | tool_source="assistant", |
| 468 | pending_tool_calls_seen=set(), |
| 469 | emit=_noop_emit, |
| 470 | summary=summary, |
| 471 | dod=create_definition_of_done("Fix the chapter links"), |
| 472 | executor=executor, # type: ignore[arg-type] |
| 473 | on_confirmation=None, |
| 474 | on_user_question=None, |
| 475 | emit_confirmation=None, |
| 476 | consecutive_errors=0, |
| 477 | ) |
| 478 | |
| 479 | assert context.recovery_context is existing_recovery |
| 480 | assert existing_recovery.successful_steps == [ |
| 481 | ("bash", {"command": "ls chapters"}) |
| 482 | ] |
| 483 | |
| 484 | |
| 485 | @pytest.mark.asyncio |
| 486 | async def test_tool_batch_runner_clears_recovery_context_after_successful_mutation( |
| 487 | temp_dir: Path, |
| 488 | ) -> None: |
| 489 | async def assess_confidence( |
| 490 | tool_name: str, |
| 491 | tool_args: dict, |
| 492 | context: str, |
| 493 | ) -> ConfidenceAssessment: |
| 494 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 495 | |
| 496 | async def verify_action( |
| 497 | tool_name: str, |
| 498 | tool_args: dict, |
| 499 | result: str, |
| 500 | expected: str = "", |
| 501 | ) -> ActionVerification: |
| 502 | raise AssertionError("Verification should not run for this scenario") |
| 503 | |
| 504 | existing_recovery = RecoveryContext( |
| 505 | original_tool="read", |
| 506 | original_args={"file_path": "chapters/04-data-types.html"}, |
| 507 | ) |
| 508 | existing_recovery.add_attempt( |
| 509 | "read", |
| 510 | {"file_path": "chapters/04-data-types.html"}, |
| 511 | "File not found", |
| 512 | ) |
| 513 | context = build_context( |
| 514 | temp_dir=temp_dir, |
| 515 | messages=[], |
| 516 | safeguards=FakeSafeguards(), |
| 517 | assess_confidence=assess_confidence, |
| 518 | verify_action=verify_action, |
| 519 | recovery_context=existing_recovery, |
| 520 | auto_recover=False, |
| 521 | ) |
| 522 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 523 | tool_call = ToolCall( |
| 524 | id="patch-1", |
| 525 | name="patch", |
| 526 | arguments={ |
| 527 | "file_path": "index.html", |
| 528 | "hunks": [{"old_start": 1, "old_lines": 1, "new_start": 1, "new_lines": 1, "lines": ["-a", "+b"]}], |
| 529 | }, |
| 530 | ) |
| 531 | executor = FakeExecutor( |
| 532 | [tool_outcome(tool_call=tool_call, output="Patched index.html", is_error=False)] |
| 533 | ) |
| 534 | |
| 535 | summary = TurnSummary(final_response="") |
| 536 | await runner.execute_batch( |
| 537 | tool_calls=[tool_call], |
| 538 | tool_source="assistant", |
| 539 | pending_tool_calls_seen=set(), |
| 540 | emit=_noop_emit, |
| 541 | summary=summary, |
| 542 | dod=create_definition_of_done("Fix the chapter links"), |
| 543 | executor=executor, # type: ignore[arg-type] |
| 544 | on_confirmation=None, |
| 545 | on_user_question=None, |
| 546 | emit_confirmation=None, |
| 547 | consecutive_errors=0, |
| 548 | ) |
| 549 | |
| 550 | assert context.recovery_context is None |
| 551 | |
| 552 | |
| 553 | @pytest.mark.asyncio |
| 554 | async def test_tool_batch_runner_queues_duplicate_observation_nudge( |
| 555 | temp_dir: Path, |
| 556 | ) -> None: |
| 557 | async def assess_confidence( |
| 558 | tool_name: str, |
| 559 | tool_args: dict, |
| 560 | context: str, |
| 561 | ) -> ConfidenceAssessment: |
| 562 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 563 | |
| 564 | async def verify_action( |
| 565 | tool_name: str, |
| 566 | tool_args: dict, |
| 567 | result: str, |
| 568 | expected: str = "", |
| 569 | ) -> ActionVerification: |
| 570 | raise AssertionError("Verification should not run for this scenario") |
| 571 | |
| 572 | messages = [ |
| 573 | Message( |
| 574 | role=Role.TOOL, |
| 575 | content=( |
| 576 | "Observation [glob]: Result: " |
| 577 | f"{temp_dir}/chapters/01-introduction.html\n" |
| 578 | f"{temp_dir}/chapters/02-setup.html\n" |
| 579 | f"{temp_dir}/chapters/03-basics.html" |
| 580 | ), |
| 581 | tool_results=[], |
| 582 | ), |
| 583 | Message( |
| 584 | role=Role.ASSISTANT, |
| 585 | content="I already inspected the first chapter title.", |
| 586 | tool_calls=[ |
| 587 | ToolCall( |
| 588 | id="read-ch1", |
| 589 | name="read", |
| 590 | arguments={"file_path": str(temp_dir / 'chapters' / '01-introduction.html')}, |
| 591 | ) |
| 592 | ], |
| 593 | ), |
| 594 | Message.tool_result_message( |
| 595 | tool_call_id="read-ch1", |
| 596 | display_content="<h1>Chapter 1: Introduction to Fortran</h1>\n", |
| 597 | result_content="<h1>Chapter 1: Introduction to Fortran</h1>\n", |
| 598 | ), |
| 599 | Message( |
| 600 | role=Role.ASSISTANT, |
| 601 | content="I should update the index now.", |
| 602 | tool_calls=[ |
| 603 | ToolCall( |
| 604 | id="read-index", |
| 605 | name="read", |
| 606 | arguments={"file_path": str(temp_dir / 'index.html')}, |
| 607 | ) |
| 608 | ], |
| 609 | ), |
| 610 | ] |
| 611 | context = build_context( |
| 612 | temp_dir=temp_dir, |
| 613 | messages=messages, |
| 614 | safeguards=FakeSafeguards(), |
| 615 | assess_confidence=assess_confidence, |
| 616 | verify_action=verify_action, |
| 617 | auto_recover=False, |
| 618 | ) |
| 619 | (temp_dir / "chapters").mkdir() |
| 620 | (temp_dir / "index.html").write_text("<ul></ul>\n") |
| 621 | (temp_dir / "chapters" / "01-introduction.html").write_text("<h1>Intro</h1>\n") |
| 622 | (temp_dir / "chapters" / "02-setup.html").write_text("<h1>Setup</h1>\n") |
| 623 | (temp_dir / "chapters" / "03-basics.html").write_text("<h1>Basics</h1>\n") |
| 624 | implementation_plan = temp_dir / "implementation.md" |
| 625 | implementation_plan.write_text( |
| 626 | "\n".join( |
| 627 | [ |
| 628 | "# Implementation Plan", |
| 629 | "", |
| 630 | "## File Changes", |
| 631 | f"- `{temp_dir / 'index.html'}`", |
| 632 | f"- `{temp_dir / 'chapters' / '01-introduction.html'}`", |
| 633 | f"- `{temp_dir / 'chapters' / '02-setup.html'}`", |
| 634 | f"- `{temp_dir / 'chapters' / '03-basics.html'}`", |
| 635 | f"- `{temp_dir / 'chapters' / '04-variables.html'}`", |
| 636 | ] |
| 637 | ) |
| 638 | ) |
| 639 | context.session.current_task = ( |
| 640 | f"Update {temp_dir / 'index.html'} with the right chapter links." |
| 641 | ) |
| 642 | persistent_messages: list[str] = [] |
| 643 | ephemeral_messages: list[str] = [] |
| 644 | context.queue_steering_message_callback = persistent_messages.append |
| 645 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 646 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 647 | tool_call = ToolCall( |
| 648 | id="read-dup", |
| 649 | name="read", |
| 650 | arguments={"file_path": str(temp_dir / "index.html")}, |
| 651 | ) |
| 652 | duplicate_message = ( |
| 653 | "[Skipped - duplicate action: Already read " |
| 654 | f"{temp_dir / 'index.html'} recently without any intervening changes; " |
| 655 | "reuse the earlier read result instead of rereading]" |
| 656 | ) |
| 657 | executor = FakeExecutor( |
| 658 | [ |
| 659 | ToolExecutionOutcome( |
| 660 | tool_call=tool_call, |
| 661 | state=ToolExecutionState.DUPLICATE, |
| 662 | message=Message.tool_result_message( |
| 663 | tool_call_id=tool_call.id, |
| 664 | display_content=duplicate_message, |
| 665 | result_content=duplicate_message, |
| 666 | ), |
| 667 | event_content=duplicate_message, |
| 668 | is_error=False, |
| 669 | result_output=duplicate_message, |
| 670 | ) |
| 671 | ] |
| 672 | ) |
| 673 | |
| 674 | summary = TurnSummary(final_response="") |
| 675 | dod = create_definition_of_done("Fix the chapter links") |
| 676 | dod.implementation_plan = str(implementation_plan) |
| 677 | dod.pending_items.append("Create the remaining chapter files") |
| 678 | await runner.execute_batch( |
| 679 | tool_calls=[tool_call], |
| 680 | tool_source="assistant", |
| 681 | pending_tool_calls_seen=set(), |
| 682 | emit=_noop_emit, |
| 683 | summary=summary, |
| 684 | dod=dod, |
| 685 | executor=executor, # type: ignore[arg-type] |
| 686 | on_confirmation=None, |
| 687 | on_user_question=None, |
| 688 | emit_confirmation=None, |
| 689 | consecutive_errors=0, |
| 690 | ) |
| 691 | |
| 692 | assert len(persistent_messages) == 1 |
| 693 | assert "Reuse the earlier observation instead of repeating it." in persistent_messages[0] |
| 694 | assert "A declared output artifact is still missing." in persistent_messages[0] |
| 695 | assert "Resume by creating `04-variables.html` now." in persistent_messages[0] |
| 696 | assert ( |
| 697 | "Prefer one `write` call for " |
| 698 | f"`{display_runtime_path(temp_dir / 'chapters' / '04-variables.html')}` instead of more rereads." |
| 699 | in persistent_messages[0] |
| 700 | ) |
| 701 | assert ephemeral_messages == [] |
| 702 | |
| 703 | |
| 704 | @pytest.mark.asyncio |
| 705 | async def test_tool_batch_runner_duplicate_read_keeps_root_declared_missing_html_output_active( |
| 706 | temp_dir: Path, |
| 707 | ) -> None: |
| 708 | async def assess_confidence( |
| 709 | tool_name: str, |
| 710 | tool_args: dict, |
| 711 | context: str, |
| 712 | ) -> ConfidenceAssessment: |
| 713 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 714 | |
| 715 | async def verify_action( |
| 716 | tool_name: str, |
| 717 | tool_args: dict, |
| 718 | result: str, |
| 719 | expected: str = "", |
| 720 | ) -> ActionVerification: |
| 721 | raise AssertionError("Verification should not run for this scenario") |
| 722 | |
| 723 | guide_root = temp_dir / "guide" |
| 724 | chapters = guide_root / "chapters" |
| 725 | chapters.mkdir(parents=True) |
| 726 | index = guide_root / "index.html" |
| 727 | chapter_one = chapters / "01-introduction.html" |
| 728 | index.write_text( |
| 729 | '<a href="chapters/01-introduction.html">Intro</a>\n' |
| 730 | '<a href="chapters/02-installation.html">Install</a>\n' |
| 731 | ) |
| 732 | chapter_one.write_text("<h1>Intro</h1>\n") |
| 733 | |
| 734 | implementation_plan = temp_dir / "implementation.md" |
| 735 | implementation_plan.write_text( |
| 736 | "\n".join( |
| 737 | [ |
| 738 | "# Implementation Plan", |
| 739 | "", |
| 740 | "## File Changes", |
| 741 | f"- `{index}`", |
| 742 | f"- `{chapters}/` (directory for chapter files)", |
| 743 | ] |
| 744 | ) |
| 745 | ) |
| 746 | |
| 747 | messages = [ |
| 748 | Message( |
| 749 | role=Role.ASSISTANT, |
| 750 | content="I should keep building the guide.", |
| 751 | tool_calls=[ |
| 752 | ToolCall( |
| 753 | id="read-index", |
| 754 | name="read", |
| 755 | arguments={"file_path": str(index)}, |
| 756 | ) |
| 757 | ], |
| 758 | ), |
| 759 | ] |
| 760 | context = build_context( |
| 761 | temp_dir=temp_dir, |
| 762 | messages=messages, |
| 763 | safeguards=FakeSafeguards(), |
| 764 | assess_confidence=assess_confidence, |
| 765 | verify_action=verify_action, |
| 766 | auto_recover=False, |
| 767 | ) |
| 768 | context.session.current_task = f"Build the guide rooted at {index}." |
| 769 | persistent_messages: list[str] = [] |
| 770 | ephemeral_messages: list[str] = [] |
| 771 | context.queue_steering_message_callback = persistent_messages.append |
| 772 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 773 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 774 | tool_call = ToolCall( |
| 775 | id="read-dup-rooted", |
| 776 | name="read", |
| 777 | arguments={"file_path": str(index)}, |
| 778 | ) |
| 779 | duplicate_message = ( |
| 780 | "[Skipped - duplicate action: Already read " |
| 781 | f"{index} recently without any intervening changes; " |
| 782 | "reuse the earlier read result instead of rereading]" |
| 783 | ) |
| 784 | executor = FakeExecutor( |
| 785 | [ |
| 786 | ToolExecutionOutcome( |
| 787 | tool_call=tool_call, |
| 788 | state=ToolExecutionState.DUPLICATE, |
| 789 | message=Message.tool_result_message( |
| 790 | tool_call_id=tool_call.id, |
| 791 | display_content=duplicate_message, |
| 792 | result_content=duplicate_message, |
| 793 | ), |
| 794 | event_content=duplicate_message, |
| 795 | is_error=False, |
| 796 | result_output=duplicate_message, |
| 797 | ) |
| 798 | ] |
| 799 | ) |
| 800 | |
| 801 | summary = TurnSummary(final_response="") |
| 802 | dod = create_definition_of_done("Create a multi-file HTML guide with chapters.") |
| 803 | dod.implementation_plan = str(implementation_plan) |
| 804 | dod.touched_files = [str(index), str(chapter_one)] |
| 805 | dod.completed_items = ["Create chapter files with appropriate content"] |
| 806 | dod.pending_items.append("Create the remaining chapter files") |
| 807 | |
| 808 | await runner.execute_batch( |
| 809 | tool_calls=[tool_call], |
| 810 | tool_source="assistant", |
| 811 | pending_tool_calls_seen=set(), |
| 812 | emit=_noop_emit, |
| 813 | summary=summary, |
| 814 | dod=dod, |
| 815 | executor=executor, # type: ignore[arg-type] |
| 816 | on_confirmation=None, |
| 817 | on_user_question=None, |
| 818 | emit_confirmation=None, |
| 819 | consecutive_errors=0, |
| 820 | ) |
| 821 | |
| 822 | assert len(persistent_messages) == 1 |
| 823 | assert "Create the remaining chapter files" in persistent_messages[0] |
| 824 | assert "Resume by creating `02-installation.html` now." in persistent_messages[0] |
| 825 | assert "All explicitly planned artifacts already exist on disk." not in persistent_messages[0] |
| 826 | assert ephemeral_messages == [] |
| 827 | |
| 828 | |
| 829 | @pytest.mark.asyncio |
| 830 | async def test_tool_batch_runner_duplicate_read_after_edit_mismatch_steers_to_mutation( |
| 831 | temp_dir: Path, |
| 832 | ) -> None: |
| 833 | async def assess_confidence( |
| 834 | tool_name: str, |
| 835 | tool_args: dict, |
| 836 | context: str, |
| 837 | ) -> ConfidenceAssessment: |
| 838 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 839 | |
| 840 | async def verify_action( |
| 841 | tool_name: str, |
| 842 | tool_args: dict, |
| 843 | result: str, |
| 844 | expected: str = "", |
| 845 | ) -> ActionVerification: |
| 846 | raise AssertionError("Verification should not run for this scenario") |
| 847 | |
| 848 | target = temp_dir / "guide" / "chapters" / "02-installation.html" |
| 849 | target.parent.mkdir(parents=True) |
| 850 | target.write_text( |
| 851 | "<h1>Chapter 2: Installation Guide</h1>\n" |
| 852 | "<p>This chapter is still too thin.</p>\n" |
| 853 | ) |
| 854 | recovery_context = RecoveryContext( |
| 855 | original_tool="edit", |
| 856 | original_args={ |
| 857 | "file_path": str(target), |
| 858 | "old_string": "<h1>Installation</h1>", |
| 859 | "new_string": "<h1>Installation</h1><p>Expanded.</p>", |
| 860 | }, |
| 861 | max_retries=2, |
| 862 | ) |
| 863 | recovery_context.add_attempt( |
| 864 | "edit", |
| 865 | { |
| 866 | "file_path": str(target), |
| 867 | "old_string": "<h1>Installation</h1>", |
| 868 | "new_string": "<h1>Installation</h1><p>Expanded.</p>", |
| 869 | }, |
| 870 | "old_string not found in file. Make sure it matches exactly.", |
| 871 | ) |
| 872 | context = build_context( |
| 873 | temp_dir=temp_dir, |
| 874 | messages=[], |
| 875 | safeguards=FakeSafeguards(), |
| 876 | assess_confidence=assess_confidence, |
| 877 | verify_action=verify_action, |
| 878 | recovery_context=recovery_context, |
| 879 | auto_recover=False, |
| 880 | ) |
| 881 | persistent_messages: list[str] = [] |
| 882 | context.queue_steering_message_callback = persistent_messages.append |
| 883 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 884 | tool_call = ToolCall( |
| 885 | id="read-dup-after-edit-miss", |
| 886 | name="read", |
| 887 | arguments={"file_path": str(target)}, |
| 888 | ) |
| 889 | duplicate_message = ( |
| 890 | "[Skipped - duplicate action: Already read " |
| 891 | f"{target} recently without any intervening changes; " |
| 892 | "reuse the earlier read result instead of rereading]" |
| 893 | ) |
| 894 | executor = FakeExecutor( |
| 895 | [ |
| 896 | ToolExecutionOutcome( |
| 897 | tool_call=tool_call, |
| 898 | state=ToolExecutionState.DUPLICATE, |
| 899 | message=Message.tool_result_message( |
| 900 | tool_call_id=tool_call.id, |
| 901 | display_content=duplicate_message, |
| 902 | result_content=duplicate_message, |
| 903 | ), |
| 904 | event_content=duplicate_message, |
| 905 | is_error=False, |
| 906 | result_output=duplicate_message, |
| 907 | ) |
| 908 | ] |
| 909 | ) |
| 910 | dod = create_definition_of_done("Expand thin generated guide chapters.") |
| 911 | |
| 912 | await runner.execute_batch( |
| 913 | tool_calls=[tool_call], |
| 914 | tool_source="assistant", |
| 915 | pending_tool_calls_seen=set(), |
| 916 | emit=_noop_emit, |
| 917 | summary=TurnSummary(final_response=""), |
| 918 | dod=dod, |
| 919 | executor=executor, # type: ignore[arg-type] |
| 920 | on_confirmation=None, |
| 921 | on_user_question=None, |
| 922 | emit_confirmation=None, |
| 923 | consecutive_errors=0, |
| 924 | ) |
| 925 | |
| 926 | assert len(persistent_messages) == 1 |
| 927 | assert "last edit" in persistent_messages[0] |
| 928 | assert "`old_string` did not exactly match" in persistent_messages[0] |
| 929 | assert "send one concrete mutation now" in persistent_messages[0] |
| 930 | assert "`write` with the complete replacement content" in persistent_messages[0] |
| 931 | |
| 932 | |
| 933 | @pytest.mark.asyncio |
| 934 | async def test_tool_batch_runner_todo_write_does_not_regress_completed_file_todo( |
| 935 | temp_dir: Path, |
| 936 | ) -> None: |
| 937 | async def assess_confidence( |
| 938 | tool_name: str, |
| 939 | tool_args: dict, |
| 940 | context: str, |
| 941 | ) -> ConfidenceAssessment: |
| 942 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 943 | |
| 944 | async def verify_action( |
| 945 | tool_name: str, |
| 946 | tool_args: dict, |
| 947 | result: str, |
| 948 | expected: str = "", |
| 949 | ) -> ActionVerification: |
| 950 | raise AssertionError("Verification should not run for this scenario") |
| 951 | |
| 952 | context = build_context( |
| 953 | temp_dir=temp_dir, |
| 954 | messages=[], |
| 955 | safeguards=FakeSafeguards(), |
| 956 | assess_confidence=assess_confidence, |
| 957 | verify_action=verify_action, |
| 958 | auto_recover=False, |
| 959 | ) |
| 960 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 961 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 962 | sync_todos_to_definition_of_done( |
| 963 | dod, |
| 964 | [ |
| 965 | { |
| 966 | "content": "Create 03-first-website.html", |
| 967 | "active_form": "Creating 03-first-website.html", |
| 968 | "status": "pending", |
| 969 | }, |
| 970 | { |
| 971 | "content": "Create 04-configuration-basics.html", |
| 972 | "active_form": "Creating 04-configuration-basics.html", |
| 973 | "status": "pending", |
| 974 | }, |
| 975 | ], |
| 976 | ) |
| 977 | |
| 978 | chapter_path = temp_dir / "guides" / "nginx" / "chapters" / "03-first-website.html" |
| 979 | chapter_path.parent.mkdir(parents=True) |
| 980 | write_call = ToolCall( |
| 981 | id="write-ch3", |
| 982 | name="write", |
| 983 | arguments={"file_path": str(chapter_path), "content": "<html></html>\n"}, |
| 984 | ) |
| 985 | stale_todo_call = ToolCall( |
| 986 | id="todo-stale", |
| 987 | name="TodoWrite", |
| 988 | arguments={ |
| 989 | "todos": [ |
| 990 | { |
| 991 | "content": "Create 03-first-website.html", |
| 992 | "active_form": "Creating 03-first-website.html", |
| 993 | "status": "pending", |
| 994 | }, |
| 995 | { |
| 996 | "content": "Create 04-configuration-basics.html", |
| 997 | "active_form": "Creating 04-configuration-basics.html", |
| 998 | "status": "pending", |
| 999 | }, |
| 1000 | ] |
| 1001 | }, |
| 1002 | ) |
| 1003 | executor = FakeExecutor( |
| 1004 | [ |
| 1005 | tool_outcome( |
| 1006 | tool_call=write_call, |
| 1007 | output=f"Successfully wrote {chapter_path}", |
| 1008 | is_error=False, |
| 1009 | ), |
| 1010 | tool_outcome( |
| 1011 | tool_call=stale_todo_call, |
| 1012 | output="Todos updated", |
| 1013 | is_error=False, |
| 1014 | metadata={ |
| 1015 | "new_todos": [ |
| 1016 | { |
| 1017 | "content": "Create 03-first-website.html", |
| 1018 | "active_form": "Creating 03-first-website.html", |
| 1019 | "status": "pending", |
| 1020 | }, |
| 1021 | { |
| 1022 | "content": "Create 04-configuration-basics.html", |
| 1023 | "active_form": "Creating 04-configuration-basics.html", |
| 1024 | "status": "pending", |
| 1025 | }, |
| 1026 | ] |
| 1027 | }, |
| 1028 | ), |
| 1029 | ] |
| 1030 | ) |
| 1031 | |
| 1032 | summary = TurnSummary(final_response="") |
| 1033 | await runner.execute_batch( |
| 1034 | tool_calls=[write_call, stale_todo_call], |
| 1035 | tool_source="assistant", |
| 1036 | pending_tool_calls_seen=set(), |
| 1037 | emit=_noop_emit, |
| 1038 | summary=summary, |
| 1039 | dod=dod, |
| 1040 | executor=executor, # type: ignore[arg-type] |
| 1041 | on_confirmation=None, |
| 1042 | on_user_question=None, |
| 1043 | emit_confirmation=None, |
| 1044 | consecutive_errors=0, |
| 1045 | ) |
| 1046 | |
| 1047 | assert "Create 03-first-website.html" in dod.completed_items |
| 1048 | assert "Create 03-first-website.html" not in dod.pending_items |
| 1049 | assert "Create 04-configuration-basics.html" in dod.pending_items |
| 1050 | |
| 1051 | |
| 1052 | @pytest.mark.asyncio |
| 1053 | async def test_tool_batch_runner_proactively_queues_verified_html_inventory( |
| 1054 | temp_dir: Path, |
| 1055 | ) -> None: |
| 1056 | async def assess_confidence( |
| 1057 | tool_name: str, |
| 1058 | tool_args: dict, |
| 1059 | context: str, |
| 1060 | ) -> ConfidenceAssessment: |
| 1061 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 1062 | |
| 1063 | async def verify_action( |
| 1064 | tool_name: str, |
| 1065 | tool_args: dict, |
| 1066 | result: str, |
| 1067 | expected: str = "", |
| 1068 | ) -> ActionVerification: |
| 1069 | raise AssertionError("Verification should not run for this scenario") |
| 1070 | |
| 1071 | chapters = temp_dir / "chapters" |
| 1072 | chapters.mkdir() |
| 1073 | (chapters / "01-introduction.html").write_text( |
| 1074 | "<h1>Chapter 1: Introduction to Fortran</h1>\n" |
| 1075 | ) |
| 1076 | (chapters / "02-setup.html").write_text( |
| 1077 | "<h1>Chapter 2: Setting Up Your Environment</h1>\n" |
| 1078 | ) |
| 1079 | (temp_dir / "index.html").write_text("<ul></ul>\n") |
| 1080 | |
| 1081 | context = build_context( |
| 1082 | temp_dir=temp_dir, |
| 1083 | messages=[], |
| 1084 | safeguards=FakeSafeguards(), |
| 1085 | assess_confidence=assess_confidence, |
| 1086 | verify_action=verify_action, |
| 1087 | auto_recover=False, |
| 1088 | ) |
| 1089 | context.session.current_task = ( |
| 1090 | f"Update {temp_dir / 'index.html'} so the chapter links match the sibling files." |
| 1091 | ) |
| 1092 | persistent_messages: list[str] = [] |
| 1093 | ephemeral_messages: list[str] = [] |
| 1094 | context.queue_steering_message_callback = persistent_messages.append |
| 1095 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 1096 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 1097 | tool_call = ToolCall( |
| 1098 | id="glob-1", |
| 1099 | name="glob", |
| 1100 | arguments={"path": str(chapters), "pattern": "*.html"}, |
| 1101 | ) |
| 1102 | executor = FakeExecutor( |
| 1103 | [ |
| 1104 | tool_outcome( |
| 1105 | tool_call=tool_call, |
| 1106 | output="\n".join( |
| 1107 | [ |
| 1108 | str(chapters / "01-introduction.html"), |
| 1109 | str(chapters / "02-setup.html"), |
| 1110 | ] |
| 1111 | ), |
| 1112 | is_error=False, |
| 1113 | ) |
| 1114 | ] |
| 1115 | ) |
| 1116 | |
| 1117 | summary = TurnSummary(final_response="") |
| 1118 | await runner.execute_batch( |
| 1119 | tool_calls=[tool_call], |
| 1120 | tool_source="assistant", |
| 1121 | pending_tool_calls_seen=set(), |
| 1122 | emit=_noop_emit, |
| 1123 | summary=summary, |
| 1124 | dod=create_definition_of_done("Fix the chapter links"), |
| 1125 | executor=executor, # type: ignore[arg-type] |
| 1126 | on_confirmation=None, |
| 1127 | on_user_question=None, |
| 1128 | emit_confirmation=None, |
| 1129 | consecutive_errors=0, |
| 1130 | ) |
| 1131 | |
| 1132 | assert persistent_messages == [] |
| 1133 | assert ephemeral_messages == [] |
| 1134 | assert len(summary.tool_result_messages) == 1 |
| 1135 | assert "Verified chapter inventory:" not in summary.tool_result_messages[0].content |
| 1136 | |
| 1137 | |
| 1138 | @pytest.mark.asyncio |
| 1139 | async def test_tool_batch_runner_marks_validated_html_toc_completion_after_successful_edit( |
| 1140 | temp_dir: Path, |
| 1141 | ) -> None: |
| 1142 | async def assess_confidence( |
| 1143 | tool_name: str, |
| 1144 | tool_args: dict, |
| 1145 | context: str, |
| 1146 | ) -> ConfidenceAssessment: |
| 1147 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 1148 | |
| 1149 | async def verify_action( |
| 1150 | tool_name: str, |
| 1151 | tool_args: dict, |
| 1152 | result: str, |
| 1153 | expected: str = "", |
| 1154 | ) -> ActionVerification: |
| 1155 | raise AssertionError("Verification should not run for this scenario") |
| 1156 | |
| 1157 | chapters = temp_dir / "chapters" |
| 1158 | chapters.mkdir() |
| 1159 | (chapters / "01-introduction.html").write_text( |
| 1160 | "<h1>Chapter 1: Introduction to Fortran</h1>\n" |
| 1161 | ) |
| 1162 | (chapters / "02-setup.html").write_text( |
| 1163 | "<h1>Chapter 2: Setting Up Your Environment</h1>\n" |
| 1164 | ) |
| 1165 | index_path = temp_dir / "index.html" |
| 1166 | old_block = ( |
| 1167 | '<ul class="chapter-list">\n' |
| 1168 | ' <li><a href="chapters/01-old.html">Chapter 1: Old</a></li>\n' |
| 1169 | ' <li><a href="chapters/02-old.html">Chapter 2: Old</a></li>\n' |
| 1170 | "</ul>\n" |
| 1171 | ) |
| 1172 | new_block = ( |
| 1173 | '<ul class="chapter-list">\n' |
| 1174 | ' <li><a href="chapters/01-introduction.html">Chapter 1: Introduction to Fortran</a></li>\n' |
| 1175 | ' <li><a href="chapters/02-setup.html">Chapter 2: Setting Up Your Environment</a></li>\n' |
| 1176 | "</ul>\n" |
| 1177 | ) |
| 1178 | index_path.write_text(new_block) |
| 1179 | |
| 1180 | context = build_context( |
| 1181 | temp_dir=temp_dir, |
| 1182 | messages=[], |
| 1183 | safeguards=FakeSafeguards(), |
| 1184 | assess_confidence=assess_confidence, |
| 1185 | verify_action=verify_action, |
| 1186 | auto_recover=False, |
| 1187 | ) |
| 1188 | context.session.current_task = ( |
| 1189 | "Update index.html so every chapter link and title matches the real HTML files in chapters/." |
| 1190 | ) |
| 1191 | persistent_messages: list[str] = [] |
| 1192 | ephemeral_messages: list[str] = [] |
| 1193 | context.queue_steering_message_callback = persistent_messages.append |
| 1194 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 1195 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 1196 | tool_call = ToolCall( |
| 1197 | id="edit-1", |
| 1198 | name="edit", |
| 1199 | arguments={ |
| 1200 | "file_path": str(index_path), |
| 1201 | "old_string": old_block, |
| 1202 | "new_string": new_block, |
| 1203 | }, |
| 1204 | ) |
| 1205 | executor = FakeExecutor( |
| 1206 | [ |
| 1207 | tool_outcome( |
| 1208 | tool_call=tool_call, |
| 1209 | output=f"Successfully edited {index_path}", |
| 1210 | is_error=False, |
| 1211 | ) |
| 1212 | ] |
| 1213 | ) |
| 1214 | |
| 1215 | summary = TurnSummary(final_response="") |
| 1216 | await runner.execute_batch( |
| 1217 | tool_calls=[tool_call], |
| 1218 | tool_source="assistant", |
| 1219 | pending_tool_calls_seen=set(), |
| 1220 | emit=_noop_emit, |
| 1221 | summary=summary, |
| 1222 | dod=create_definition_of_done( |
| 1223 | "Update index.html so every chapter link and title matches the real HTML files in chapters/." |
| 1224 | ), |
| 1225 | executor=executor, # type: ignore[arg-type] |
| 1226 | on_confirmation=None, |
| 1227 | on_user_question=None, |
| 1228 | emit_confirmation=None, |
| 1229 | consecutive_errors=0, |
| 1230 | ) |
| 1231 | |
| 1232 | assert all( |
| 1233 | "Semantic verification preview:" not in message.content |
| 1234 | for message in summary.tool_result_messages |
| 1235 | ) |
| 1236 | assert persistent_messages == [] |
| 1237 | assert ephemeral_messages == [] |
| 1238 | |
| 1239 | |
| 1240 | @pytest.mark.asyncio |
| 1241 | async def test_tool_batch_runner_does_not_apply_html_toc_handoff_to_reference_read( |
| 1242 | temp_dir: Path, |
| 1243 | ) -> None: |
| 1244 | async def assess_confidence( |
| 1245 | tool_name: str, |
| 1246 | tool_args: dict, |
| 1247 | context: str, |
| 1248 | ) -> ConfidenceAssessment: |
| 1249 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 1250 | |
| 1251 | async def verify_action( |
| 1252 | tool_name: str, |
| 1253 | tool_args: dict, |
| 1254 | result: str, |
| 1255 | expected: str = "", |
| 1256 | ) -> ActionVerification: |
| 1257 | raise AssertionError("Verification should not run for this scenario") |
| 1258 | |
| 1259 | chapters = temp_dir / "chapters" |
| 1260 | chapters.mkdir() |
| 1261 | (chapters / "01-introduction.html").write_text( |
| 1262 | "<h1>Chapter 1: Introduction to Fortran</h1>\n" |
| 1263 | ) |
| 1264 | (chapters / "02-setup.html").write_text( |
| 1265 | "<h1>Chapter 2: Setting Up Your Environment</h1>\n" |
| 1266 | ) |
| 1267 | index_path = temp_dir / "index.html" |
| 1268 | index_path.write_text( |
| 1269 | "<h2>Table of Contents</h2>\n" |
| 1270 | '<ul class="chapter-list">\n' |
| 1271 | ' <li><a href="chapters/01-introduction.html">Chapter 1: Introduction to Fortran</a></li>\n' |
| 1272 | ' <li><a href="chapters/02-setup.html">Chapter 2: Setting Up Your Environment</a></li>\n' |
| 1273 | "</ul>\n" |
| 1274 | ) |
| 1275 | |
| 1276 | prompt = ( |
| 1277 | "Have a look at ~/Loader/guides/fortran and chapters/ within. Get a feel " |
| 1278 | "for the structure and cadence of the guide. We are going to make an all " |
| 1279 | "new equally thorough guide on how to use the nginx tool." |
| 1280 | ) |
| 1281 | |
| 1282 | context = build_context( |
| 1283 | temp_dir=temp_dir, |
| 1284 | messages=[], |
| 1285 | safeguards=FakeSafeguards(), |
| 1286 | assess_confidence=assess_confidence, |
| 1287 | verify_action=verify_action, |
| 1288 | auto_recover=False, |
| 1289 | ) |
| 1290 | context.session.current_task = prompt # type: ignore[attr-defined] |
| 1291 | persistent_messages: list[str] = [] |
| 1292 | ephemeral_messages: list[str] = [] |
| 1293 | context.queue_steering_message_callback = persistent_messages.append |
| 1294 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 1295 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 1296 | tool_call = ToolCall( |
| 1297 | id="read-index", |
| 1298 | name="read", |
| 1299 | arguments={"file_path": str(index_path)}, |
| 1300 | ) |
| 1301 | executor = FakeExecutor( |
| 1302 | [ |
| 1303 | tool_outcome( |
| 1304 | tool_call=tool_call, |
| 1305 | output=index_path.read_text(), |
| 1306 | is_error=False, |
| 1307 | ) |
| 1308 | ] |
| 1309 | ) |
| 1310 | |
| 1311 | summary = TurnSummary(final_response="") |
| 1312 | await runner.execute_batch( |
| 1313 | tool_calls=[tool_call], |
| 1314 | tool_source="assistant", |
| 1315 | pending_tool_calls_seen=set(), |
| 1316 | emit=_noop_emit, |
| 1317 | summary=summary, |
| 1318 | dod=create_definition_of_done(prompt), |
| 1319 | executor=executor, # type: ignore[arg-type] |
| 1320 | on_confirmation=None, |
| 1321 | on_user_question=None, |
| 1322 | emit_confirmation=None, |
| 1323 | consecutive_errors=0, |
| 1324 | ) |
| 1325 | |
| 1326 | assert persistent_messages == [] |
| 1327 | assert ephemeral_messages == [] |
| 1328 | assert all( |
| 1329 | "Semantic verification preview:" not in message.content |
| 1330 | for message in summary.tool_result_messages |
| 1331 | ) |
| 1332 | |
| 1333 | |
| 1334 | @pytest.mark.asyncio |
| 1335 | async def test_tool_batch_runner_queues_next_pending_todo_after_discovery_progress( |
| 1336 | temp_dir: Path, |
| 1337 | ) -> None: |
| 1338 | async def assess_confidence( |
| 1339 | tool_name: str, |
| 1340 | tool_args: dict, |
| 1341 | context: str, |
| 1342 | ) -> ConfidenceAssessment: |
| 1343 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 1344 | |
| 1345 | async def verify_action( |
| 1346 | tool_name: str, |
| 1347 | tool_args: dict, |
| 1348 | result: str, |
| 1349 | expected: str = "", |
| 1350 | ) -> ActionVerification: |
| 1351 | raise AssertionError("Verification should not run for this scenario") |
| 1352 | |
| 1353 | reference = temp_dir / "fortran" / "chapters" / "01-introduction.html" |
| 1354 | reference.parent.mkdir(parents=True) |
| 1355 | reference.write_text("<h1>Introduction</h1>\n<p>Guide cadence.</p>\n") |
| 1356 | nginx_root = temp_dir / "Loader" / "guides" / "nginx" |
| 1357 | chapters = nginx_root / "chapters" |
| 1358 | implementation_plan = temp_dir / "implementation.md" |
| 1359 | implementation_plan.write_text( |
| 1360 | "\n".join( |
| 1361 | [ |
| 1362 | "# Implementation Plan", |
| 1363 | "", |
| 1364 | "## File Changes", |
| 1365 | f"- `{chapters}/`", |
| 1366 | f"- `{nginx_root / 'index.html'}`", |
| 1367 | "", |
| 1368 | ] |
| 1369 | ) |
| 1370 | ) |
| 1371 | |
| 1372 | context = build_context( |
| 1373 | temp_dir=temp_dir, |
| 1374 | messages=[], |
| 1375 | safeguards=FakeSafeguards(), |
| 1376 | assess_confidence=assess_confidence, |
| 1377 | verify_action=verify_action, |
| 1378 | auto_recover=False, |
| 1379 | ) |
| 1380 | persistent_messages: list[str] = [] |
| 1381 | ephemeral_messages: list[str] = [] |
| 1382 | context.queue_steering_message_callback = persistent_messages.append |
| 1383 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 1384 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 1385 | dod = create_definition_of_done("Create an equally thorough nginx guide.") |
| 1386 | dod.implementation_plan = str(implementation_plan) |
| 1387 | sync_todos_to_definition_of_done( |
| 1388 | dod, |
| 1389 | [ |
| 1390 | { |
| 1391 | "content": "Examine the existing Fortran guide structure to understand the cadence and format", |
| 1392 | "active_form": "Working on: Examine the existing Fortran guide structure to understand the cadence and format", |
| 1393 | "status": "pending", |
| 1394 | }, |
| 1395 | { |
| 1396 | "content": "Create the nginx directory structure", |
| 1397 | "active_form": "Working on: Create the nginx directory structure", |
| 1398 | "status": "pending", |
| 1399 | }, |
| 1400 | { |
| 1401 | "content": "Create the nginx index.html file", |
| 1402 | "active_form": "Working on: Create the nginx index.html file", |
| 1403 | "status": "pending", |
| 1404 | }, |
| 1405 | ], |
| 1406 | ) |
| 1407 | tool_call = ToolCall( |
| 1408 | id="read-reference", |
| 1409 | name="read", |
| 1410 | arguments={"file_path": str(reference)}, |
| 1411 | ) |
| 1412 | executor = FakeExecutor( |
| 1413 | [ |
| 1414 | tool_outcome( |
| 1415 | tool_call=tool_call, |
| 1416 | output="<h1>Introduction</h1>\n<p>Guide cadence.</p>\n", |
| 1417 | is_error=False, |
| 1418 | ) |
| 1419 | ] |
| 1420 | ) |
| 1421 | |
| 1422 | summary = TurnSummary(final_response="") |
| 1423 | await runner.execute_batch( |
| 1424 | tool_calls=[tool_call], |
| 1425 | tool_source="assistant", |
| 1426 | pending_tool_calls_seen=set(), |
| 1427 | emit=_noop_emit, |
| 1428 | summary=summary, |
| 1429 | dod=dod, |
| 1430 | executor=executor, # type: ignore[arg-type] |
| 1431 | on_confirmation=None, |
| 1432 | on_user_question=None, |
| 1433 | emit_confirmation=None, |
| 1434 | consecutive_errors=0, |
| 1435 | ) |
| 1436 | |
| 1437 | assert ( |
| 1438 | "Examine the existing Fortran guide structure to understand the cadence and format" |
| 1439 | in dod.completed_items |
| 1440 | ) |
| 1441 | assert any( |
| 1442 | "Continue with the next pending item: `Create the nginx directory structure`" |
| 1443 | in message |
| 1444 | for message in persistent_messages |
| 1445 | ) |
| 1446 | assert any( |
| 1447 | "Resume by creating `chapters/` now." in message |
| 1448 | for message in persistent_messages |
| 1449 | ) |
| 1450 | assert all("01-introduction.html" not in message for message in persistent_messages) |
| 1451 | assert ephemeral_messages == [] |
| 1452 | |
| 1453 | |
| 1454 | @pytest.mark.asyncio |
| 1455 | async def test_tool_batch_runner_queues_setup_directory_before_file_when_plan_lists_index_first( |
| 1456 | temp_dir: Path, |
| 1457 | ) -> None: |
| 1458 | async def assess_confidence( |
| 1459 | tool_name: str, |
| 1460 | tool_args: dict, |
| 1461 | context: str, |
| 1462 | ) -> ConfidenceAssessment: |
| 1463 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 1464 | |
| 1465 | async def verify_action( |
| 1466 | tool_name: str, |
| 1467 | tool_args: dict, |
| 1468 | result: str, |
| 1469 | expected: str = "", |
| 1470 | ) -> ActionVerification: |
| 1471 | raise AssertionError("Verification should not run for this scenario") |
| 1472 | |
| 1473 | reference = temp_dir / "fortran" / "chapters" / "01-introduction.html" |
| 1474 | reference.parent.mkdir(parents=True) |
| 1475 | reference.write_text("<h1>Introduction</h1>\n<p>Guide cadence.</p>\n") |
| 1476 | nginx_root = temp_dir / "Loader" / "guides" / "nginx" |
| 1477 | chapters = nginx_root / "chapters" |
| 1478 | implementation_plan = temp_dir / "implementation.md" |
| 1479 | implementation_plan.write_text( |
| 1480 | "\n".join( |
| 1481 | [ |
| 1482 | "# Implementation Plan", |
| 1483 | "", |
| 1484 | "## File Changes", |
| 1485 | f"- `{nginx_root / 'index.html'}`", |
| 1486 | f"- `{chapters}/`", |
| 1487 | "", |
| 1488 | ] |
| 1489 | ) |
| 1490 | ) |
| 1491 | |
| 1492 | context = build_context( |
| 1493 | temp_dir=temp_dir, |
| 1494 | messages=[], |
| 1495 | safeguards=FakeSafeguards(), |
| 1496 | assess_confidence=assess_confidence, |
| 1497 | verify_action=verify_action, |
| 1498 | auto_recover=False, |
| 1499 | ) |
| 1500 | persistent_messages: list[str] = [] |
| 1501 | ephemeral_messages: list[str] = [] |
| 1502 | context.queue_steering_message_callback = persistent_messages.append |
| 1503 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 1504 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 1505 | dod = create_definition_of_done("Create an equally thorough nginx guide.") |
| 1506 | dod.implementation_plan = str(implementation_plan) |
| 1507 | sync_todos_to_definition_of_done( |
| 1508 | dod, |
| 1509 | [ |
| 1510 | { |
| 1511 | "content": "Examine the existing Fortran guide structure to understand the cadence and format", |
| 1512 | "active_form": "Working on: Examine the existing Fortran guide structure to understand the cadence and format", |
| 1513 | "status": "pending", |
| 1514 | }, |
| 1515 | { |
| 1516 | "content": "Create the nginx directory structure", |
| 1517 | "active_form": "Working on: Create the nginx directory structure", |
| 1518 | "status": "pending", |
| 1519 | }, |
| 1520 | { |
| 1521 | "content": "Create the nginx index.html file", |
| 1522 | "active_form": "Working on: Create the nginx index.html file", |
| 1523 | "status": "pending", |
| 1524 | }, |
| 1525 | ], |
| 1526 | project_root=temp_dir, |
| 1527 | ) |
| 1528 | tool_call = ToolCall( |
| 1529 | id="read-reference-index-first", |
| 1530 | name="read", |
| 1531 | arguments={"file_path": str(reference)}, |
| 1532 | ) |
| 1533 | executor = FakeExecutor( |
| 1534 | [ |
| 1535 | tool_outcome( |
| 1536 | tool_call=tool_call, |
| 1537 | output="<h1>Introduction</h1>\n<p>Guide cadence.</p>\n", |
| 1538 | is_error=False, |
| 1539 | ) |
| 1540 | ] |
| 1541 | ) |
| 1542 | |
| 1543 | summary = TurnSummary(final_response="") |
| 1544 | await runner.execute_batch( |
| 1545 | tool_calls=[tool_call], |
| 1546 | tool_source="assistant", |
| 1547 | pending_tool_calls_seen=set(), |
| 1548 | emit=_noop_emit, |
| 1549 | summary=summary, |
| 1550 | dod=dod, |
| 1551 | executor=executor, # type: ignore[arg-type] |
| 1552 | on_confirmation=None, |
| 1553 | on_user_question=None, |
| 1554 | emit_confirmation=None, |
| 1555 | consecutive_errors=0, |
| 1556 | ) |
| 1557 | |
| 1558 | assert persistent_messages |
| 1559 | assert any( |
| 1560 | "Continue with the next pending item: `Create the nginx directory structure`" |
| 1561 | in message |
| 1562 | for message in persistent_messages |
| 1563 | ) |
| 1564 | assert any( |
| 1565 | "Resume by creating `chapters/` now." in message |
| 1566 | for message in persistent_messages |
| 1567 | ) |
| 1568 | assert all( |
| 1569 | "Next step: create `index.html`." not in message |
| 1570 | for message in persistent_messages |
| 1571 | ) |
| 1572 | assert ephemeral_messages == [] |
| 1573 | |
| 1574 | |
| 1575 | @pytest.mark.asyncio |
| 1576 | async def test_tool_batch_runner_duplicate_reference_read_prefers_next_pending_todo( |
| 1577 | temp_dir: Path, |
| 1578 | ) -> None: |
| 1579 | async def assess_confidence( |
| 1580 | tool_name: str, |
| 1581 | tool_args: dict, |
| 1582 | context: str, |
| 1583 | ) -> ConfidenceAssessment: |
| 1584 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 1585 | |
| 1586 | async def verify_action( |
| 1587 | tool_name: str, |
| 1588 | tool_args: dict, |
| 1589 | result: str, |
| 1590 | expected: str = "", |
| 1591 | ) -> ActionVerification: |
| 1592 | raise AssertionError("Verification should not run for this scenario") |
| 1593 | |
| 1594 | reference = temp_dir / "fortran" / "index.html" |
| 1595 | reference.parent.mkdir(parents=True) |
| 1596 | reference.write_text("<h1>Fortran Beginner's Guide</h1>\n") |
| 1597 | |
| 1598 | messages = [ |
| 1599 | Message( |
| 1600 | role=Role.TOOL, |
| 1601 | content=( |
| 1602 | "Observation [read]: Result: " |
| 1603 | "<h1>Fortran Beginner's Guide</h1>\n" |
| 1604 | ), |
| 1605 | ) |
| 1606 | ] |
| 1607 | context = build_context( |
| 1608 | temp_dir=temp_dir, |
| 1609 | messages=messages, |
| 1610 | safeguards=FakeSafeguards(), |
| 1611 | assess_confidence=assess_confidence, |
| 1612 | verify_action=verify_action, |
| 1613 | auto_recover=False, |
| 1614 | ) |
| 1615 | prompt = ( |
| 1616 | "Have a look at ~/Loader/guides/fortran and chapters/ within. Get a feel " |
| 1617 | "for the structure and cadence of the guide. We are going to make an all " |
| 1618 | "new equally thorough guide on how to use the nginx tool." |
| 1619 | ) |
| 1620 | context.session.current_task = prompt |
| 1621 | persistent_messages: list[str] = [] |
| 1622 | ephemeral_messages: list[str] = [] |
| 1623 | context.queue_steering_message_callback = persistent_messages.append |
| 1624 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 1625 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 1626 | dod = create_definition_of_done(prompt) |
| 1627 | sync_todos_to_definition_of_done( |
| 1628 | dod, |
| 1629 | [ |
| 1630 | { |
| 1631 | "content": "Examine the existing Fortran guide structure to understand the cadence and format", |
| 1632 | "active_form": "Working on: Examine the existing Fortran guide structure to understand the cadence and format", |
| 1633 | "status": "completed", |
| 1634 | }, |
| 1635 | { |
| 1636 | "content": "Create the nginx directory structure", |
| 1637 | "active_form": "Working on: Create the nginx directory structure", |
| 1638 | "status": "pending", |
| 1639 | }, |
| 1640 | { |
| 1641 | "content": "Create the nginx index.html file", |
| 1642 | "active_form": "Working on: Create the nginx index.html file", |
| 1643 | "status": "pending", |
| 1644 | }, |
| 1645 | ], |
| 1646 | ) |
| 1647 | tool_call = ToolCall( |
| 1648 | id="read-dup", |
| 1649 | name="read", |
| 1650 | arguments={"file_path": str(reference)}, |
| 1651 | ) |
| 1652 | duplicate_message = ( |
| 1653 | "[Skipped - duplicate action: Already read " |
| 1654 | f"{reference} recently without any intervening changes; " |
| 1655 | "reuse the earlier read result instead of rereading]" |
| 1656 | ) |
| 1657 | executor = FakeExecutor( |
| 1658 | [ |
| 1659 | ToolExecutionOutcome( |
| 1660 | tool_call=tool_call, |
| 1661 | state=ToolExecutionState.DUPLICATE, |
| 1662 | message=Message.tool_result_message( |
| 1663 | tool_call_id=tool_call.id, |
| 1664 | display_content=duplicate_message, |
| 1665 | result_content=duplicate_message, |
| 1666 | ), |
| 1667 | event_content=duplicate_message, |
| 1668 | is_error=False, |
| 1669 | result_output=duplicate_message, |
| 1670 | ) |
| 1671 | ] |
| 1672 | ) |
| 1673 | |
| 1674 | summary = TurnSummary(final_response="") |
| 1675 | await runner.execute_batch( |
| 1676 | tool_calls=[tool_call], |
| 1677 | tool_source="assistant", |
| 1678 | pending_tool_calls_seen=set(), |
| 1679 | emit=_noop_emit, |
| 1680 | summary=summary, |
| 1681 | dod=dod, |
| 1682 | executor=executor, # type: ignore[arg-type] |
| 1683 | on_confirmation=None, |
| 1684 | on_user_question=None, |
| 1685 | emit_confirmation=None, |
| 1686 | consecutive_errors=0, |
| 1687 | ) |
| 1688 | |
| 1689 | assert len(persistent_messages) == 1 |
| 1690 | assert "Reuse the earlier observation instead of repeating it." in persistent_messages[0] |
| 1691 | assert ( |
| 1692 | "Continue with the next pending item: `Create the nginx directory structure`" |
| 1693 | in persistent_messages[0] |
| 1694 | ) |
| 1695 | assert "Update `" not in persistent_messages[0] |
| 1696 | assert ephemeral_messages == [] |
| 1697 | |
| 1698 | |
| 1699 | @pytest.mark.asyncio |
| 1700 | async def test_tool_batch_runner_successful_reference_read_prioritizes_concrete_missing_artifact( |
| 1701 | temp_dir: Path, |
| 1702 | ) -> None: |
| 1703 | async def assess_confidence( |
| 1704 | tool_name: str, |
| 1705 | tool_args: dict, |
| 1706 | context: str, |
| 1707 | ) -> ConfidenceAssessment: |
| 1708 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 1709 | |
| 1710 | async def verify_action( |
| 1711 | tool_name: str, |
| 1712 | tool_args: dict, |
| 1713 | result: str, |
| 1714 | expected: str = "", |
| 1715 | ) -> ActionVerification: |
| 1716 | raise AssertionError("Verification should not run for this scenario") |
| 1717 | |
| 1718 | guide_root = temp_dir / "Loader" / "guides" / "nginx" |
| 1719 | chapters = guide_root / "chapters" |
| 1720 | chapters.mkdir(parents=True) |
| 1721 | chapter_one = chapters / "01-introduction.html" |
| 1722 | chapter_one.write_text("<html></html>\n") |
| 1723 | index_path = guide_root / "index.html" |
| 1724 | |
| 1725 | reference = temp_dir / "Loader" / "guides" / "fortran" / "chapters" / "01-introduction.html" |
| 1726 | reference.parent.mkdir(parents=True, exist_ok=True) |
| 1727 | reference.write_text("<h1>Introduction</h1>\n<p>Guide cadence.</p>\n") |
| 1728 | |
| 1729 | implementation_plan = temp_dir / "implementation.md" |
| 1730 | implementation_plan.write_text( |
| 1731 | "\n".join( |
| 1732 | [ |
| 1733 | "# Implementation Plan", |
| 1734 | "", |
| 1735 | "## File Changes", |
| 1736 | f"- `{guide_root}/`", |
| 1737 | f"- `{chapters}/`", |
| 1738 | f"- `{index_path}`", |
| 1739 | f"- `{chapter_one}`", |
| 1740 | f"- `{chapters / '02-installation.html'}`", |
| 1741 | "", |
| 1742 | ] |
| 1743 | ) |
| 1744 | ) |
| 1745 | |
| 1746 | context = build_context( |
| 1747 | temp_dir=temp_dir, |
| 1748 | messages=[], |
| 1749 | safeguards=FakeSafeguards(), |
| 1750 | assess_confidence=assess_confidence, |
| 1751 | verify_action=verify_action, |
| 1752 | auto_recover=False, |
| 1753 | ) |
| 1754 | persistent_messages: list[str] = [] |
| 1755 | ephemeral_messages: list[str] = [] |
| 1756 | context.queue_steering_message_callback = persistent_messages.append |
| 1757 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 1758 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 1759 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 1760 | dod.implementation_plan = str(implementation_plan) |
| 1761 | dod.touched_files.append(str(chapter_one)) |
| 1762 | sync_todos_to_definition_of_done( |
| 1763 | dod, |
| 1764 | [ |
| 1765 | { |
| 1766 | "content": "Examine the existing Fortran guide structure to understand the format and cadence", |
| 1767 | "active_form": "Working on: Examine the existing Fortran guide structure to understand the format and cadence", |
| 1768 | "status": "pending", |
| 1769 | }, |
| 1770 | { |
| 1771 | "content": "Create each chapter file with appropriate content", |
| 1772 | "active_form": "Working on: Create each chapter file with appropriate content", |
| 1773 | "status": "pending", |
| 1774 | }, |
| 1775 | { |
| 1776 | "content": "Ensure all files follow the same structure and style as the Fortran guide", |
| 1777 | "active_form": "Working on: Ensure all files follow the same structure and style as the Fortran guide", |
| 1778 | "status": "pending", |
| 1779 | }, |
| 1780 | ], |
| 1781 | ) |
| 1782 | tool_call = ToolCall( |
| 1783 | id="read-reference-chapter", |
| 1784 | name="read", |
| 1785 | arguments={"file_path": str(reference)}, |
| 1786 | ) |
| 1787 | read_output = "Observation [read]: Result: <h1>Introduction</h1>\n<p>Guide cadence.</p>\n" |
| 1788 | executor = FakeExecutor( |
| 1789 | [ |
| 1790 | ToolExecutionOutcome( |
| 1791 | tool_call=tool_call, |
| 1792 | state=ToolExecutionState.EXECUTED, |
| 1793 | message=Message.tool_result_message( |
| 1794 | tool_call_id=tool_call.id, |
| 1795 | display_content=read_output, |
| 1796 | result_content=read_output, |
| 1797 | ), |
| 1798 | event_content=read_output, |
| 1799 | is_error=False, |
| 1800 | result_output=read_output, |
| 1801 | ) |
| 1802 | ] |
| 1803 | ) |
| 1804 | |
| 1805 | summary = TurnSummary(final_response="") |
| 1806 | await runner.execute_batch( |
| 1807 | tool_calls=[tool_call], |
| 1808 | tool_source="assistant", |
| 1809 | pending_tool_calls_seen=set(), |
| 1810 | emit=_noop_emit, |
| 1811 | summary=summary, |
| 1812 | dod=dod, |
| 1813 | executor=executor, # type: ignore[arg-type] |
| 1814 | on_confirmation=None, |
| 1815 | on_user_question=None, |
| 1816 | emit_confirmation=None, |
| 1817 | consecutive_errors=0, |
| 1818 | ) |
| 1819 | |
| 1820 | assert persistent_messages |
| 1821 | assert any( |
| 1822 | "Confirmed progress: `Examine the existing Fortran guide structure to understand the format and cadence`" |
| 1823 | in message |
| 1824 | for message in persistent_messages |
| 1825 | ) |
| 1826 | assert any("Resume by creating `index.html` now." in message for message in persistent_messages) |
| 1827 | assert not any( |
| 1828 | "Continue with the next pending item: `Create each chapter file with appropriate content`" |
| 1829 | in message |
| 1830 | for message in persistent_messages |
| 1831 | ) |
| 1832 | assert ephemeral_messages == [] |
| 1833 | |
| 1834 | |
| 1835 | @pytest.mark.asyncio |
| 1836 | async def test_tool_batch_runner_duplicate_read_ignores_unplanned_expansion_after_plan_complete( |
| 1837 | temp_dir: Path, |
| 1838 | ) -> None: |
| 1839 | async def assess_confidence( |
| 1840 | tool_name: str, |
| 1841 | tool_args: dict, |
| 1842 | context: str, |
| 1843 | ) -> ConfidenceAssessment: |
| 1844 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 1845 | |
| 1846 | async def verify_action( |
| 1847 | tool_name: str, |
| 1848 | tool_args: dict, |
| 1849 | result: str, |
| 1850 | expected: str = "", |
| 1851 | ) -> ActionVerification: |
| 1852 | raise AssertionError("Verification should not run for this scenario") |
| 1853 | |
| 1854 | guide_root = temp_dir / "guides" / "nginx" |
| 1855 | chapters = guide_root / "chapters" |
| 1856 | guide_root.mkdir(parents=True) |
| 1857 | chapters.mkdir() |
| 1858 | index_path = guide_root / "index.html" |
| 1859 | chapter_one = chapters / "01-getting-started.html" |
| 1860 | chapter_two = chapters / "02-installation.html" |
| 1861 | index_path.write_text("<html></html>\n") |
| 1862 | chapter_one.write_text("<h1>One</h1>\n") |
| 1863 | chapter_two.write_text("<h1>Two</h1>\n") |
| 1864 | |
| 1865 | implementation_plan = temp_dir / "implementation.md" |
| 1866 | implementation_plan.write_text( |
| 1867 | "\n".join( |
| 1868 | [ |
| 1869 | "# Implementation Plan", |
| 1870 | "", |
| 1871 | "## File Changes", |
| 1872 | f"- `{guide_root}/`", |
| 1873 | f"- `{chapters}/`", |
| 1874 | f"- `{index_path}`", |
| 1875 | f"- `{chapter_one}`", |
| 1876 | f"- `{chapter_two}`", |
| 1877 | "", |
| 1878 | ] |
| 1879 | ) |
| 1880 | ) |
| 1881 | |
| 1882 | context = build_context( |
| 1883 | temp_dir=temp_dir, |
| 1884 | messages=[], |
| 1885 | safeguards=FakeSafeguards(), |
| 1886 | assess_confidence=assess_confidence, |
| 1887 | verify_action=verify_action, |
| 1888 | auto_recover=False, |
| 1889 | ) |
| 1890 | persistent_messages: list[str] = [] |
| 1891 | ephemeral_messages: list[str] = [] |
| 1892 | context.queue_steering_message_callback = persistent_messages.append |
| 1893 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 1894 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 1895 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 1896 | dod.implementation_plan = str(implementation_plan) |
| 1897 | dod.pending_items = [ |
| 1898 | "Create 07-performance-tuning.html", |
| 1899 | "Verify all guide files are linked and complete", |
| 1900 | "Complete the requested work", |
| 1901 | ] |
| 1902 | |
| 1903 | tool_call = ToolCall( |
| 1904 | id="read-dup", |
| 1905 | name="read", |
| 1906 | arguments={"file_path": str(chapter_one)}, |
| 1907 | ) |
| 1908 | duplicate_message = ( |
| 1909 | "[Skipped - duplicate action: Already read " |
| 1910 | f"{chapter_one} recently without any intervening changes; " |
| 1911 | "reuse the earlier read result instead of rereading]" |
| 1912 | ) |
| 1913 | executor = FakeExecutor( |
| 1914 | [ |
| 1915 | ToolExecutionOutcome( |
| 1916 | tool_call=tool_call, |
| 1917 | state=ToolExecutionState.DUPLICATE, |
| 1918 | message=Message.tool_result_message( |
| 1919 | tool_call_id=tool_call.id, |
| 1920 | display_content=duplicate_message, |
| 1921 | result_content=duplicate_message, |
| 1922 | ), |
| 1923 | event_content=duplicate_message, |
| 1924 | is_error=False, |
| 1925 | result_output=duplicate_message, |
| 1926 | ) |
| 1927 | ] |
| 1928 | ) |
| 1929 | |
| 1930 | summary = TurnSummary(final_response="") |
| 1931 | await runner.execute_batch( |
| 1932 | tool_calls=[tool_call], |
| 1933 | tool_source="assistant", |
| 1934 | pending_tool_calls_seen=set(), |
| 1935 | emit=_noop_emit, |
| 1936 | summary=summary, |
| 1937 | dod=dod, |
| 1938 | executor=executor, # type: ignore[arg-type] |
| 1939 | on_confirmation=None, |
| 1940 | on_user_question=None, |
| 1941 | emit_confirmation=None, |
| 1942 | consecutive_errors=0, |
| 1943 | ) |
| 1944 | |
| 1945 | assert len(persistent_messages) == 1 |
| 1946 | assert "Verify all guide files are linked and complete" in persistent_messages[0] |
| 1947 | assert "Create 07-performance-tuning.html" not in persistent_messages[0] |
| 1948 | assert ephemeral_messages == [] |
| 1949 | |
| 1950 | |
| 1951 | @pytest.mark.asyncio |
| 1952 | async def test_tool_batch_runner_duplicate_read_after_plan_complete_pushes_verification_handoff( |
| 1953 | temp_dir: Path, |
| 1954 | ) -> None: |
| 1955 | async def assess_confidence( |
| 1956 | tool_name: str, |
| 1957 | tool_args: dict, |
| 1958 | context: str, |
| 1959 | ) -> ConfidenceAssessment: |
| 1960 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 1961 | |
| 1962 | async def verify_action( |
| 1963 | tool_name: str, |
| 1964 | tool_args: dict, |
| 1965 | result: str, |
| 1966 | expected: str = "", |
| 1967 | ) -> ActionVerification: |
| 1968 | raise AssertionError("Verification should not run for this scenario") |
| 1969 | |
| 1970 | guide_root = temp_dir / "guides" / "nginx" |
| 1971 | chapters = guide_root / "chapters" |
| 1972 | guide_root.mkdir(parents=True) |
| 1973 | chapters.mkdir() |
| 1974 | index_path = guide_root / "index.html" |
| 1975 | chapter_one = chapters / "01-getting-started.html" |
| 1976 | chapter_two = chapters / "02-installation.html" |
| 1977 | index_path.write_text("<html></html>\n") |
| 1978 | chapter_one.write_text("<h1>One</h1>\n") |
| 1979 | chapter_two.write_text("<h1>Two</h1>\n") |
| 1980 | |
| 1981 | implementation_plan = temp_dir / "implementation.md" |
| 1982 | implementation_plan.write_text( |
| 1983 | "\n".join( |
| 1984 | [ |
| 1985 | "# Implementation Plan", |
| 1986 | "", |
| 1987 | "## File Changes", |
| 1988 | f"- `{guide_root}/`", |
| 1989 | f"- `{chapters}/`", |
| 1990 | f"- `{index_path}`", |
| 1991 | f"- `{chapter_one}`", |
| 1992 | f"- `{chapter_two}`", |
| 1993 | "", |
| 1994 | ] |
| 1995 | ) |
| 1996 | ) |
| 1997 | |
| 1998 | context = build_context( |
| 1999 | temp_dir=temp_dir, |
| 2000 | messages=[], |
| 2001 | safeguards=FakeSafeguards(), |
| 2002 | assess_confidence=assess_confidence, |
| 2003 | verify_action=verify_action, |
| 2004 | auto_recover=False, |
| 2005 | ) |
| 2006 | persistent_messages: list[str] = [] |
| 2007 | ephemeral_messages: list[str] = [] |
| 2008 | context.queue_steering_message_callback = persistent_messages.append |
| 2009 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 2010 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 2011 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 2012 | dod.implementation_plan = str(implementation_plan) |
| 2013 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 2014 | dod.pending_items = [ |
| 2015 | "Create 07-performance-tuning.html", |
| 2016 | "Complete the requested work", |
| 2017 | ] |
| 2018 | |
| 2019 | tool_call = ToolCall( |
| 2020 | id="read-dup", |
| 2021 | name="read", |
| 2022 | arguments={"file_path": str(chapter_one)}, |
| 2023 | ) |
| 2024 | duplicate_message = ( |
| 2025 | "[Skipped - duplicate action: Already read " |
| 2026 | f"{chapter_one} recently without any intervening changes; " |
| 2027 | "reuse the earlier read result instead of rereading]" |
| 2028 | ) |
| 2029 | executor = FakeExecutor( |
| 2030 | [ |
| 2031 | ToolExecutionOutcome( |
| 2032 | tool_call=tool_call, |
| 2033 | state=ToolExecutionState.DUPLICATE, |
| 2034 | message=Message.tool_result_message( |
| 2035 | tool_call_id=tool_call.id, |
| 2036 | display_content=duplicate_message, |
| 2037 | result_content=duplicate_message, |
| 2038 | ), |
| 2039 | event_content=duplicate_message, |
| 2040 | is_error=False, |
| 2041 | result_output=duplicate_message, |
| 2042 | ) |
| 2043 | ] |
| 2044 | ) |
| 2045 | |
| 2046 | summary = TurnSummary(final_response="") |
| 2047 | await runner.execute_batch( |
| 2048 | tool_calls=[tool_call], |
| 2049 | tool_source="assistant", |
| 2050 | pending_tool_calls_seen=set(), |
| 2051 | emit=_noop_emit, |
| 2052 | summary=summary, |
| 2053 | dod=dod, |
| 2054 | executor=executor, # type: ignore[arg-type] |
| 2055 | on_confirmation=None, |
| 2056 | on_user_question=None, |
| 2057 | emit_confirmation=None, |
| 2058 | consecutive_errors=0, |
| 2059 | ) |
| 2060 | |
| 2061 | assert len(persistent_messages) == 1 |
| 2062 | assert "All explicitly planned artifacts already exist on disk." in persistent_messages[0] |
| 2063 | assert ( |
| 2064 | "Finish with a final response now so Loader can run verification automatically." |
| 2065 | in persistent_messages[0] |
| 2066 | ) |
| 2067 | assert "Create 07-performance-tuning.html" not in persistent_messages[0] |
| 2068 | assert ephemeral_messages == [] |
| 2069 | |
| 2070 | |
| 2071 | @pytest.mark.asyncio |
| 2072 | async def test_tool_batch_runner_duplicate_read_after_plan_complete_ignores_stale_creation_todos( |
| 2073 | temp_dir: Path, |
| 2074 | ) -> None: |
| 2075 | async def assess_confidence( |
| 2076 | tool_name: str, |
| 2077 | tool_args: dict, |
| 2078 | context: str, |
| 2079 | ) -> ConfidenceAssessment: |
| 2080 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 2081 | |
| 2082 | async def verify_action( |
| 2083 | tool_name: str, |
| 2084 | tool_args: dict, |
| 2085 | result: str, |
| 2086 | expected: str = "", |
| 2087 | ) -> ActionVerification: |
| 2088 | raise AssertionError("Verification should not run for this scenario") |
| 2089 | |
| 2090 | guide_root = temp_dir / "guides" / "nginx" |
| 2091 | chapters = guide_root / "chapters" |
| 2092 | guide_root.mkdir(parents=True) |
| 2093 | chapters.mkdir() |
| 2094 | index_path = guide_root / "index.html" |
| 2095 | chapter_one = chapters / "01-getting-started.html" |
| 2096 | chapter_two = chapters / "02-installation.html" |
| 2097 | index_path.write_text("<html></html>\n") |
| 2098 | chapter_one.write_text("<h1>One</h1>\n") |
| 2099 | chapter_two.write_text("<h1>Two</h1>\n") |
| 2100 | |
| 2101 | implementation_plan = temp_dir / "implementation.md" |
| 2102 | implementation_plan.write_text( |
| 2103 | "\n".join( |
| 2104 | [ |
| 2105 | "# Implementation Plan", |
| 2106 | "", |
| 2107 | "## File Changes", |
| 2108 | f"- `{guide_root}/`", |
| 2109 | f"- `{chapters}/`", |
| 2110 | f"- `{index_path}`", |
| 2111 | f"- `{chapter_one}`", |
| 2112 | f"- `{chapter_two}`", |
| 2113 | "", |
| 2114 | ] |
| 2115 | ) |
| 2116 | ) |
| 2117 | |
| 2118 | context = build_context( |
| 2119 | temp_dir=temp_dir, |
| 2120 | messages=[], |
| 2121 | safeguards=FakeSafeguards(), |
| 2122 | assess_confidence=assess_confidence, |
| 2123 | verify_action=verify_action, |
| 2124 | auto_recover=False, |
| 2125 | ) |
| 2126 | persistent_messages: list[str] = [] |
| 2127 | ephemeral_messages: list[str] = [] |
| 2128 | context.queue_steering_message_callback = persistent_messages.append |
| 2129 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 2130 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 2131 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 2132 | dod.implementation_plan = str(implementation_plan) |
| 2133 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 2134 | dod.pending_items = [ |
| 2135 | "Create 01-getting-started.html", |
| 2136 | "Creating 02-installation.html", |
| 2137 | "Complete the requested work", |
| 2138 | ] |
| 2139 | |
| 2140 | tool_call = ToolCall( |
| 2141 | id="read-dup-built-stale", |
| 2142 | name="read", |
| 2143 | arguments={"file_path": str(chapter_one)}, |
| 2144 | ) |
| 2145 | duplicate_message = ( |
| 2146 | "[Skipped - duplicate action: Already read " |
| 2147 | f"{chapter_one} recently without any intervening changes; " |
| 2148 | "reuse the earlier read result instead of rereading]" |
| 2149 | ) |
| 2150 | executor = FakeExecutor( |
| 2151 | [ |
| 2152 | ToolExecutionOutcome( |
| 2153 | tool_call=tool_call, |
| 2154 | state=ToolExecutionState.DUPLICATE, |
| 2155 | message=Message.tool_result_message( |
| 2156 | tool_call_id=tool_call.id, |
| 2157 | display_content=duplicate_message, |
| 2158 | result_content=duplicate_message, |
| 2159 | ), |
| 2160 | event_content=duplicate_message, |
| 2161 | is_error=False, |
| 2162 | result_output=duplicate_message, |
| 2163 | ) |
| 2164 | ] |
| 2165 | ) |
| 2166 | |
| 2167 | summary = TurnSummary(final_response="") |
| 2168 | await runner.execute_batch( |
| 2169 | tool_calls=[tool_call], |
| 2170 | tool_source="assistant", |
| 2171 | pending_tool_calls_seen=set(), |
| 2172 | emit=_noop_emit, |
| 2173 | summary=summary, |
| 2174 | dod=dod, |
| 2175 | executor=executor, # type: ignore[arg-type] |
| 2176 | on_confirmation=None, |
| 2177 | on_user_question=None, |
| 2178 | emit_confirmation=None, |
| 2179 | consecutive_errors=0, |
| 2180 | ) |
| 2181 | |
| 2182 | assert len(persistent_messages) == 1 |
| 2183 | assert "All explicitly planned artifacts already exist on disk." in persistent_messages[0] |
| 2184 | assert ( |
| 2185 | "Finish with a final response now so Loader can run verification automatically." |
| 2186 | in persistent_messages[0] |
| 2187 | ) |
| 2188 | assert "Create 01-getting-started.html" not in persistent_messages[0] |
| 2189 | assert "Creating 02-installation.html" not in persistent_messages[0] |
| 2190 | assert ephemeral_messages == [] |
| 2191 | |
| 2192 | |
| 2193 | @pytest.mark.asyncio |
| 2194 | async def test_tool_batch_runner_successful_read_after_plan_complete_pushes_review_handoff( |
| 2195 | temp_dir: Path, |
| 2196 | ) -> None: |
| 2197 | async def assess_confidence( |
| 2198 | tool_name: str, |
| 2199 | tool_args: dict, |
| 2200 | context: str, |
| 2201 | ) -> ConfidenceAssessment: |
| 2202 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 2203 | |
| 2204 | async def verify_action( |
| 2205 | tool_name: str, |
| 2206 | tool_args: dict, |
| 2207 | result: str, |
| 2208 | expected: str = "", |
| 2209 | ) -> ActionVerification: |
| 2210 | raise AssertionError("Verification should not run for this scenario") |
| 2211 | |
| 2212 | guide_root = temp_dir / "guides" / "nginx" |
| 2213 | chapters = guide_root / "chapters" |
| 2214 | guide_root.mkdir(parents=True) |
| 2215 | chapters.mkdir() |
| 2216 | index_path = guide_root / "index.html" |
| 2217 | chapter_one = chapters / "01-getting-started.html" |
| 2218 | chapter_two = chapters / "02-installation.html" |
| 2219 | index_path.write_text("<html></html>\n") |
| 2220 | chapter_one.write_text("<h1>One</h1>\n") |
| 2221 | chapter_two.write_text("<h1>Two</h1>\n") |
| 2222 | |
| 2223 | implementation_plan = temp_dir / "implementation.md" |
| 2224 | implementation_plan.write_text( |
| 2225 | "\n".join( |
| 2226 | [ |
| 2227 | "# Implementation Plan", |
| 2228 | "", |
| 2229 | "## File Changes", |
| 2230 | f"- `{guide_root}/`", |
| 2231 | f"- `{chapters}/`", |
| 2232 | f"- `{index_path}`", |
| 2233 | f"- `{chapter_one}`", |
| 2234 | f"- `{chapter_two}`", |
| 2235 | "", |
| 2236 | ] |
| 2237 | ) |
| 2238 | ) |
| 2239 | |
| 2240 | context = build_context( |
| 2241 | temp_dir=temp_dir, |
| 2242 | messages=[], |
| 2243 | safeguards=FakeSafeguards(), |
| 2244 | assess_confidence=assess_confidence, |
| 2245 | verify_action=verify_action, |
| 2246 | auto_recover=False, |
| 2247 | ) |
| 2248 | persistent_messages: list[str] = [] |
| 2249 | ephemeral_messages: list[str] = [] |
| 2250 | context.queue_steering_message_callback = persistent_messages.append |
| 2251 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 2252 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 2253 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 2254 | dod.implementation_plan = str(implementation_plan) |
| 2255 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 2256 | sync_todos_to_definition_of_done( |
| 2257 | dod, |
| 2258 | [ |
| 2259 | { |
| 2260 | "content": "Create 01-getting-started.html", |
| 2261 | "active_form": "Creating 01-getting-started.html", |
| 2262 | "status": "pending", |
| 2263 | }, |
| 2264 | { |
| 2265 | "content": "Ensure all files are properly linked and formatted consistently", |
| 2266 | "active_form": "Reviewing guide consistency and linkage", |
| 2267 | "status": "pending", |
| 2268 | }, |
| 2269 | ], |
| 2270 | ) |
| 2271 | |
| 2272 | tool_call = ToolCall( |
| 2273 | id="read-built-review", |
| 2274 | name="read", |
| 2275 | arguments={"file_path": str(chapter_one)}, |
| 2276 | ) |
| 2277 | executor = FakeExecutor( |
| 2278 | [tool_outcome(tool_call=tool_call, output=chapter_one.read_text(), is_error=False)] |
| 2279 | ) |
| 2280 | |
| 2281 | summary = TurnSummary(final_response="") |
| 2282 | await runner.execute_batch( |
| 2283 | tool_calls=[tool_call], |
| 2284 | tool_source="assistant", |
| 2285 | pending_tool_calls_seen=set(), |
| 2286 | emit=_noop_emit, |
| 2287 | summary=summary, |
| 2288 | dod=dod, |
| 2289 | executor=executor, # type: ignore[arg-type] |
| 2290 | on_confirmation=None, |
| 2291 | on_user_question=None, |
| 2292 | emit_confirmation=None, |
| 2293 | consecutive_errors=0, |
| 2294 | ) |
| 2295 | |
| 2296 | assert persistent_messages == [] |
| 2297 | assert len(ephemeral_messages) == 1 |
| 2298 | message = ephemeral_messages[0] |
| 2299 | assert "All explicitly planned artifacts already exist." in message |
| 2300 | assert "Ensure all files are properly linked and formatted consistently" in message |
| 2301 | assert "Create 01-getting-started.html" not in message |
| 2302 | assert "do not keep broad-rereading the output set" in message |
| 2303 | assert "If no specific mismatch remains, finish with a final response so Loader can verify." in message |
| 2304 | |
| 2305 | |
| 2306 | @pytest.mark.asyncio |
| 2307 | async def test_tool_batch_runner_successful_read_after_plan_complete_switches_to_verify( |
| 2308 | temp_dir: Path, |
| 2309 | ) -> None: |
| 2310 | async def assess_confidence( |
| 2311 | tool_name: str, |
| 2312 | tool_args: dict, |
| 2313 | context: str, |
| 2314 | ) -> ConfidenceAssessment: |
| 2315 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 2316 | |
| 2317 | async def verify_action( |
| 2318 | tool_name: str, |
| 2319 | tool_args: dict, |
| 2320 | result: str, |
| 2321 | expected: str = "", |
| 2322 | ) -> ActionVerification: |
| 2323 | raise AssertionError("Verification should not run for this scenario") |
| 2324 | |
| 2325 | guide_root = temp_dir / "guides" / "nginx" |
| 2326 | chapters = guide_root / "chapters" |
| 2327 | guide_root.mkdir(parents=True) |
| 2328 | chapters.mkdir() |
| 2329 | index_path = guide_root / "index.html" |
| 2330 | chapter_one = chapters / "01-getting-started.html" |
| 2331 | chapter_two = chapters / "02-installation.html" |
| 2332 | index_path.write_text("<html></html>\n") |
| 2333 | chapter_one.write_text("<h1>One</h1>\n") |
| 2334 | chapter_two.write_text("<h1>Two</h1>\n") |
| 2335 | |
| 2336 | implementation_plan = temp_dir / "implementation.md" |
| 2337 | implementation_plan.write_text( |
| 2338 | "\n".join( |
| 2339 | [ |
| 2340 | "# Implementation Plan", |
| 2341 | "", |
| 2342 | "## File Changes", |
| 2343 | f"- `{guide_root}/`", |
| 2344 | f"- `{chapters}/`", |
| 2345 | f"- `{index_path}`", |
| 2346 | f"- `{chapter_one}`", |
| 2347 | f"- `{chapter_two}`", |
| 2348 | "", |
| 2349 | ] |
| 2350 | ) |
| 2351 | ) |
| 2352 | |
| 2353 | context = build_context( |
| 2354 | temp_dir=temp_dir, |
| 2355 | messages=[], |
| 2356 | safeguards=FakeSafeguards(), |
| 2357 | assess_confidence=assess_confidence, |
| 2358 | verify_action=verify_action, |
| 2359 | auto_recover=False, |
| 2360 | ) |
| 2361 | persistent_messages: list[str] = [] |
| 2362 | ephemeral_messages: list[str] = [] |
| 2363 | context.queue_steering_message_callback = persistent_messages.append |
| 2364 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 2365 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 2366 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 2367 | dod.implementation_plan = str(implementation_plan) |
| 2368 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 2369 | |
| 2370 | tool_call = ToolCall( |
| 2371 | id="read-built-verify", |
| 2372 | name="read", |
| 2373 | arguments={"file_path": str(chapter_one)}, |
| 2374 | ) |
| 2375 | executor = FakeExecutor( |
| 2376 | [tool_outcome(tool_call=tool_call, output=chapter_one.read_text(), is_error=False)] |
| 2377 | ) |
| 2378 | |
| 2379 | summary = TurnSummary(final_response="") |
| 2380 | await runner.execute_batch( |
| 2381 | tool_calls=[tool_call], |
| 2382 | tool_source="assistant", |
| 2383 | pending_tool_calls_seen=set(), |
| 2384 | emit=_noop_emit, |
| 2385 | summary=summary, |
| 2386 | dod=dod, |
| 2387 | executor=executor, # type: ignore[arg-type] |
| 2388 | on_confirmation=None, |
| 2389 | on_user_question=None, |
| 2390 | emit_confirmation=None, |
| 2391 | consecutive_errors=0, |
| 2392 | ) |
| 2393 | |
| 2394 | assert len(persistent_messages) == 1 |
| 2395 | assert "All explicitly planned artifacts already exist." in persistent_messages[0] |
| 2396 | assert "Finish with a final response now so Loader can run verification automatically." in persistent_messages[0] |
| 2397 | assert "stop broad rereads" in persistent_messages[0] |
| 2398 | assert ephemeral_messages == [] |
| 2399 | assert context.workflow_mode == "verify" |
| 2400 | |
| 2401 | |
| 2402 | @pytest.mark.asyncio |
| 2403 | async def test_tool_batch_runner_observation_handoff_pushes_mutation_step( |
| 2404 | temp_dir: Path, |
| 2405 | ) -> None: |
| 2406 | async def assess_confidence( |
| 2407 | tool_name: str, |
| 2408 | tool_args: dict, |
| 2409 | context: str, |
| 2410 | ) -> ConfidenceAssessment: |
| 2411 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 2412 | |
| 2413 | async def verify_action( |
| 2414 | tool_name: str, |
| 2415 | tool_args: dict, |
| 2416 | result: str, |
| 2417 | expected: str = "", |
| 2418 | ) -> ActionVerification: |
| 2419 | raise AssertionError("Verification should not run for this scenario") |
| 2420 | |
| 2421 | reference = temp_dir / "fortran" / "chapters" / "01-introduction.html" |
| 2422 | reference.parent.mkdir(parents=True) |
| 2423 | reference.write_text("<h1>Introduction</h1>\n<p>Guide cadence.</p>\n") |
| 2424 | |
| 2425 | context = build_context( |
| 2426 | temp_dir=temp_dir, |
| 2427 | messages=[], |
| 2428 | safeguards=FakeSafeguards(), |
| 2429 | assess_confidence=assess_confidence, |
| 2430 | verify_action=verify_action, |
| 2431 | auto_recover=False, |
| 2432 | ) |
| 2433 | persistent_messages: list[str] = [] |
| 2434 | ephemeral_messages: list[str] = [] |
| 2435 | context.queue_steering_message_callback = persistent_messages.append |
| 2436 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 2437 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 2438 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 2439 | sync_todos_to_definition_of_done( |
| 2440 | dod, |
| 2441 | [ |
| 2442 | { |
| 2443 | "content": "Examine the existing Fortran guide structure to understand the cadence and format", |
| 2444 | "active_form": "Working on: Examine the existing Fortran guide structure to understand the cadence and format", |
| 2445 | "status": "pending", |
| 2446 | }, |
| 2447 | { |
| 2448 | "content": "Create the nginx index.html file", |
| 2449 | "active_form": "Working on: Create the nginx index.html file", |
| 2450 | "status": "pending", |
| 2451 | }, |
| 2452 | ], |
| 2453 | ) |
| 2454 | tool_call = ToolCall( |
| 2455 | id="read-reference", |
| 2456 | name="read", |
| 2457 | arguments={"file_path": str(reference)}, |
| 2458 | ) |
| 2459 | executor = FakeExecutor( |
| 2460 | [ |
| 2461 | tool_outcome( |
| 2462 | tool_call=tool_call, |
| 2463 | output="<h1>Introduction</h1>\n<p>Guide cadence.</p>\n", |
| 2464 | is_error=False, |
| 2465 | ) |
| 2466 | ] |
| 2467 | ) |
| 2468 | |
| 2469 | summary = TurnSummary(final_response="") |
| 2470 | await runner.execute_batch( |
| 2471 | tool_calls=[tool_call], |
| 2472 | tool_source="assistant", |
| 2473 | pending_tool_calls_seen=set(), |
| 2474 | emit=_noop_emit, |
| 2475 | summary=summary, |
| 2476 | dod=dod, |
| 2477 | executor=executor, # type: ignore[arg-type] |
| 2478 | on_confirmation=None, |
| 2479 | on_user_question=None, |
| 2480 | emit_confirmation=None, |
| 2481 | consecutive_errors=0, |
| 2482 | ) |
| 2483 | |
| 2484 | assert any( |
| 2485 | "Continue with the next pending item: `Create the nginx index.html file`" |
| 2486 | in message |
| 2487 | for message in persistent_messages |
| 2488 | ) |
| 2489 | assert any( |
| 2490 | "stop gathering more reference material and perform the change now" in message |
| 2491 | for message in persistent_messages |
| 2492 | ) |
| 2493 | assert ephemeral_messages == [] |
| 2494 | |
| 2495 | |
| 2496 | @pytest.mark.asyncio |
| 2497 | async def test_tool_batch_runner_discovery_completion_handoff_stays_persistent( |
| 2498 | temp_dir: Path, |
| 2499 | ) -> None: |
| 2500 | async def assess_confidence( |
| 2501 | tool_name: str, |
| 2502 | tool_args: dict, |
| 2503 | context: str, |
| 2504 | ) -> ConfidenceAssessment: |
| 2505 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 2506 | |
| 2507 | async def verify_action( |
| 2508 | tool_name: str, |
| 2509 | tool_args: dict, |
| 2510 | result: str, |
| 2511 | expected: str = "", |
| 2512 | ) -> ActionVerification: |
| 2513 | raise AssertionError("Verification should not run for this scenario") |
| 2514 | |
| 2515 | reference = temp_dir / "fortran" / "chapters" / "01-introduction.html" |
| 2516 | reference.parent.mkdir(parents=True) |
| 2517 | reference.write_text("<h1>Introduction</h1>\n<p>Guide cadence.</p>\n") |
| 2518 | |
| 2519 | context = build_context( |
| 2520 | temp_dir=temp_dir, |
| 2521 | messages=[], |
| 2522 | safeguards=FakeSafeguards(), |
| 2523 | assess_confidence=assess_confidence, |
| 2524 | verify_action=verify_action, |
| 2525 | auto_recover=False, |
| 2526 | ) |
| 2527 | persistent_messages: list[str] = [] |
| 2528 | ephemeral_messages: list[str] = [] |
| 2529 | context.queue_steering_message_callback = persistent_messages.append |
| 2530 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 2531 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 2532 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 2533 | sync_todos_to_definition_of_done( |
| 2534 | dod, |
| 2535 | [ |
| 2536 | { |
| 2537 | "content": "First, examine the existing fortran guide structure and content", |
| 2538 | "active_form": "Working on: First, examine the existing fortran guide structure and content", |
| 2539 | "status": "pending", |
| 2540 | }, |
| 2541 | { |
| 2542 | "content": "Create the nginx directory structure", |
| 2543 | "active_form": "Working on: Create the nginx directory structure", |
| 2544 | "status": "pending", |
| 2545 | }, |
| 2546 | ], |
| 2547 | ) |
| 2548 | tool_call = ToolCall( |
| 2549 | id="read-reference", |
| 2550 | name="read", |
| 2551 | arguments={"file_path": str(reference)}, |
| 2552 | ) |
| 2553 | executor = FakeExecutor( |
| 2554 | [ |
| 2555 | tool_outcome( |
| 2556 | tool_call=tool_call, |
| 2557 | output="<h1>Introduction</h1>\n<p>Guide cadence.</p>\n", |
| 2558 | is_error=False, |
| 2559 | ) |
| 2560 | ] |
| 2561 | ) |
| 2562 | |
| 2563 | summary = TurnSummary(final_response="") |
| 2564 | await runner.execute_batch( |
| 2565 | tool_calls=[tool_call], |
| 2566 | tool_source="assistant", |
| 2567 | pending_tool_calls_seen=set(), |
| 2568 | emit=_noop_emit, |
| 2569 | summary=summary, |
| 2570 | dod=dod, |
| 2571 | executor=executor, # type: ignore[arg-type] |
| 2572 | on_confirmation=None, |
| 2573 | on_user_question=None, |
| 2574 | emit_confirmation=None, |
| 2575 | consecutive_errors=0, |
| 2576 | ) |
| 2577 | |
| 2578 | assert persistent_messages |
| 2579 | assert any( |
| 2580 | "Continue with the next pending item: `Create the nginx directory structure`" |
| 2581 | in message |
| 2582 | for message in persistent_messages |
| 2583 | ) |
| 2584 | assert ephemeral_messages == [] |
| 2585 | |
| 2586 | |
| 2587 | @pytest.mark.asyncio |
| 2588 | async def test_tool_batch_runner_missing_artifact_nudge_names_next_file_after_setup_mkdir( |
| 2589 | temp_dir: Path, |
| 2590 | ) -> None: |
| 2591 | async def assess_confidence( |
| 2592 | tool_name: str, |
| 2593 | tool_args: dict, |
| 2594 | context: str, |
| 2595 | ) -> ConfidenceAssessment: |
| 2596 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 2597 | |
| 2598 | async def verify_action( |
| 2599 | tool_name: str, |
| 2600 | tool_args: dict, |
| 2601 | result: str, |
| 2602 | expected: str = "", |
| 2603 | ) -> ActionVerification: |
| 2604 | raise AssertionError("Verification should not run for this scenario") |
| 2605 | |
| 2606 | nginx_root = temp_dir / "Loader" / "guides" / "nginx" |
| 2607 | chapters = nginx_root / "chapters" |
| 2608 | implementation_plan = temp_dir / "implementation.md" |
| 2609 | implementation_plan.write_text( |
| 2610 | "\n".join( |
| 2611 | [ |
| 2612 | "# Implementation Plan", |
| 2613 | "", |
| 2614 | "## File Changes", |
| 2615 | f"- `{chapters}/`", |
| 2616 | f"- `{nginx_root / 'index.html'}`", |
| 2617 | "", |
| 2618 | ] |
| 2619 | ) |
| 2620 | ) |
| 2621 | |
| 2622 | context = build_context( |
| 2623 | temp_dir=temp_dir, |
| 2624 | messages=[], |
| 2625 | safeguards=FakeSafeguards(), |
| 2626 | assess_confidence=assess_confidence, |
| 2627 | verify_action=verify_action, |
| 2628 | auto_recover=False, |
| 2629 | ) |
| 2630 | persistent_messages: list[str] = [] |
| 2631 | ephemeral_messages: list[str] = [] |
| 2632 | context.queue_steering_message_callback = persistent_messages.append |
| 2633 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 2634 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 2635 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 2636 | dod.implementation_plan = str(implementation_plan) |
| 2637 | sync_todos_to_definition_of_done( |
| 2638 | dod, |
| 2639 | [ |
| 2640 | { |
| 2641 | "content": "Create the nginx directory structure", |
| 2642 | "active_form": "Creating the nginx directory structure", |
| 2643 | "status": "pending", |
| 2644 | }, |
| 2645 | { |
| 2646 | "content": "Develop the main index.html file with proper structure", |
| 2647 | "active_form": "Developing the main index.html file with proper structure", |
| 2648 | "status": "pending", |
| 2649 | }, |
| 2650 | ], |
| 2651 | ) |
| 2652 | |
| 2653 | tool_call = ToolCall( |
| 2654 | id="mkdir-nginx", |
| 2655 | name="bash", |
| 2656 | arguments={"command": f"mkdir -p {chapters}"}, |
| 2657 | ) |
| 2658 | executor = FakeExecutor( |
| 2659 | [ |
| 2660 | tool_outcome( |
| 2661 | tool_call=tool_call, |
| 2662 | output="", |
| 2663 | is_error=False, |
| 2664 | ) |
| 2665 | ] |
| 2666 | ) |
| 2667 | |
| 2668 | summary = TurnSummary(final_response="") |
| 2669 | await runner.execute_batch( |
| 2670 | tool_calls=[tool_call], |
| 2671 | tool_source="assistant", |
| 2672 | pending_tool_calls_seen=set(), |
| 2673 | emit=_noop_emit, |
| 2674 | summary=summary, |
| 2675 | dod=dod, |
| 2676 | executor=executor, # type: ignore[arg-type] |
| 2677 | on_confirmation=None, |
| 2678 | on_user_question=None, |
| 2679 | emit_confirmation=None, |
| 2680 | consecutive_errors=0, |
| 2681 | ) |
| 2682 | |
| 2683 | assert persistent_messages |
| 2684 | message = persistent_messages[-1] |
| 2685 | assert "Directory setup is complete." in message |
| 2686 | assert "Next step: create `index.html`." in message |
| 2687 | assert "Write a compact but real initial version of that file now" in message |
| 2688 | assert ephemeral_messages == [] |
| 2689 | |
| 2690 | |
| 2691 | @pytest.mark.asyncio |
| 2692 | async def test_tool_batch_runner_first_chapter_handoff_stays_persistent_until_substantive_output_exists( |
| 2693 | temp_dir: Path, |
| 2694 | ) -> None: |
| 2695 | async def assess_confidence( |
| 2696 | tool_name: str, |
| 2697 | tool_args: dict, |
| 2698 | context: str, |
| 2699 | ) -> ConfidenceAssessment: |
| 2700 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 2701 | |
| 2702 | async def verify_action( |
| 2703 | tool_name: str, |
| 2704 | tool_args: dict, |
| 2705 | result: str, |
| 2706 | expected: str = "", |
| 2707 | ) -> ActionVerification: |
| 2708 | raise AssertionError("Verification should not run for this scenario") |
| 2709 | |
| 2710 | nginx_root = temp_dir / "guides" / "nginx" |
| 2711 | chapters = nginx_root / "chapters" |
| 2712 | chapters.mkdir(parents=True) |
| 2713 | index_path = nginx_root / "index.html" |
| 2714 | |
| 2715 | implementation_plan = temp_dir / "implementation.md" |
| 2716 | implementation_plan.write_text( |
| 2717 | "\n".join( |
| 2718 | [ |
| 2719 | "# Implementation Plan", |
| 2720 | "", |
| 2721 | "## File Changes", |
| 2722 | f"- `{chapters}/`", |
| 2723 | f"- `{index_path}`", |
| 2724 | f"- `{chapters / '01-introduction.html'}`", |
| 2725 | "", |
| 2726 | ] |
| 2727 | ) |
| 2728 | ) |
| 2729 | |
| 2730 | context = build_context( |
| 2731 | temp_dir=temp_dir, |
| 2732 | messages=[], |
| 2733 | safeguards=FakeSafeguards(), |
| 2734 | assess_confidence=assess_confidence, |
| 2735 | verify_action=verify_action, |
| 2736 | auto_recover=False, |
| 2737 | ) |
| 2738 | persistent_messages: list[str] = [] |
| 2739 | ephemeral_messages: list[str] = [] |
| 2740 | context.queue_steering_message_callback = persistent_messages.append |
| 2741 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 2742 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 2743 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 2744 | dod.implementation_plan = str(implementation_plan) |
| 2745 | sync_todos_to_definition_of_done( |
| 2746 | dod, |
| 2747 | [ |
| 2748 | { |
| 2749 | "content": "Create the main index.html file with proper structure", |
| 2750 | "active_form": "Creating the main index.html file with proper structure", |
| 2751 | "status": "pending", |
| 2752 | }, |
| 2753 | { |
| 2754 | "content": "Create each chapter file with appropriate content", |
| 2755 | "active_form": "Creating each chapter file with appropriate content", |
| 2756 | "status": "pending", |
| 2757 | }, |
| 2758 | ], |
| 2759 | ) |
| 2760 | |
| 2761 | tool_call = ToolCall( |
| 2762 | id="write-index", |
| 2763 | name="write", |
| 2764 | arguments={ |
| 2765 | "file_path": str(index_path), |
| 2766 | "content": "<html></html>\n", |
| 2767 | }, |
| 2768 | ) |
| 2769 | executor = FakeExecutor( |
| 2770 | [ |
| 2771 | tool_outcome( |
| 2772 | tool_call=tool_call, |
| 2773 | output=f"Successfully wrote 14 bytes to {index_path}", |
| 2774 | is_error=False, |
| 2775 | ) |
| 2776 | ] |
| 2777 | ) |
| 2778 | |
| 2779 | summary = TurnSummary(final_response="") |
| 2780 | await runner.execute_batch( |
| 2781 | tool_calls=[tool_call], |
| 2782 | tool_source="assistant", |
| 2783 | pending_tool_calls_seen=set(), |
| 2784 | emit=_noop_emit, |
| 2785 | summary=summary, |
| 2786 | dod=dod, |
| 2787 | executor=executor, # type: ignore[arg-type] |
| 2788 | on_confirmation=None, |
| 2789 | on_user_question=None, |
| 2790 | emit_confirmation=None, |
| 2791 | consecutive_errors=0, |
| 2792 | ) |
| 2793 | |
| 2794 | assert persistent_messages |
| 2795 | assert ephemeral_messages == [] |
| 2796 | message = persistent_messages[-1] |
| 2797 | assert "Confirmed progress:" in message |
| 2798 | assert "Next step: create `01-introduction.html`." in message |
| 2799 | assert ( |
| 2800 | f"Prefer one `write(file_path=..., content=...)` call for `{(chapters / '01-introduction.html').resolve(strict=False)}` now." |
| 2801 | in message |
| 2802 | ) |
| 2803 | assert "Write a compact but real initial version of that file now" not in message |
| 2804 | assert "Do not reread reference material or spend the next turn on bookkeeping." in message |
| 2805 | |
| 2806 | |
| 2807 | @pytest.mark.asyncio |
| 2808 | async def test_tool_batch_runner_directory_handoff_uses_home_relative_path( |
| 2809 | temp_dir: Path, |
| 2810 | monkeypatch: pytest.MonkeyPatch, |
| 2811 | ) -> None: |
| 2812 | monkeypatch.setenv("HOME", str(temp_dir.resolve(strict=False))) |
| 2813 | |
| 2814 | async def assess_confidence( |
| 2815 | tool_name: str, |
| 2816 | tool_args: dict, |
| 2817 | context: str, |
| 2818 | ) -> ConfidenceAssessment: |
| 2819 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 2820 | |
| 2821 | async def verify_action( |
| 2822 | tool_name: str, |
| 2823 | tool_args: dict, |
| 2824 | result: str, |
| 2825 | expected: str = "", |
| 2826 | ) -> ActionVerification: |
| 2827 | raise AssertionError("Verification should not run for this scenario") |
| 2828 | |
| 2829 | nginx_root = temp_dir / "Loader" / "guides" / "nginx" |
| 2830 | chapters = nginx_root / "chapters" |
| 2831 | index_path = nginx_root / "index.html" |
| 2832 | |
| 2833 | implementation_plan = temp_dir / "implementation.md" |
| 2834 | implementation_plan.write_text( |
| 2835 | "\n".join( |
| 2836 | [ |
| 2837 | "# Implementation Plan", |
| 2838 | "", |
| 2839 | "## File Changes", |
| 2840 | f"- `{chapters}/`", |
| 2841 | f"- `{index_path}`", |
| 2842 | "", |
| 2843 | ] |
| 2844 | ) |
| 2845 | ) |
| 2846 | |
| 2847 | context = build_context( |
| 2848 | temp_dir=temp_dir, |
| 2849 | messages=[], |
| 2850 | safeguards=FakeSafeguards(), |
| 2851 | assess_confidence=assess_confidence, |
| 2852 | verify_action=verify_action, |
| 2853 | auto_recover=False, |
| 2854 | ) |
| 2855 | persistent_messages: list[str] = [] |
| 2856 | context.queue_steering_message_callback = persistent_messages.append |
| 2857 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 2858 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 2859 | dod.implementation_plan = str(implementation_plan) |
| 2860 | sync_todos_to_definition_of_done( |
| 2861 | dod, |
| 2862 | [ |
| 2863 | { |
| 2864 | "content": "Create the nginx directory structure", |
| 2865 | "active_form": "Creating the nginx directory structure", |
| 2866 | "status": "pending", |
| 2867 | }, |
| 2868 | { |
| 2869 | "content": "Develop the main index.html file with proper structure", |
| 2870 | "active_form": "Developing the main index.html file with proper structure", |
| 2871 | "status": "pending", |
| 2872 | }, |
| 2873 | ], |
| 2874 | ) |
| 2875 | |
| 2876 | tool_call = ToolCall( |
| 2877 | id="mkdir-nginx-home", |
| 2878 | name="bash", |
| 2879 | arguments={"command": f"mkdir -p {chapters}"}, |
| 2880 | ) |
| 2881 | executor = FakeExecutor( |
| 2882 | [ |
| 2883 | tool_outcome( |
| 2884 | tool_call=tool_call, |
| 2885 | output="", |
| 2886 | is_error=False, |
| 2887 | ) |
| 2888 | ] |
| 2889 | ) |
| 2890 | |
| 2891 | summary = TurnSummary(final_response="") |
| 2892 | await runner.execute_batch( |
| 2893 | tool_calls=[tool_call], |
| 2894 | tool_source="assistant", |
| 2895 | pending_tool_calls_seen=set(), |
| 2896 | emit=_noop_emit, |
| 2897 | summary=summary, |
| 2898 | dod=dod, |
| 2899 | executor=executor, # type: ignore[arg-type] |
| 2900 | on_confirmation=None, |
| 2901 | on_user_question=None, |
| 2902 | emit_confirmation=None, |
| 2903 | consecutive_errors=0, |
| 2904 | ) |
| 2905 | |
| 2906 | assert persistent_messages |
| 2907 | message = persistent_messages[-1] |
| 2908 | assert "Next step: create `index.html`." in message |
| 2909 | assert "`~/Loader/guides/nginx/index.html`" in message |
| 2910 | assert "Write a compact but real initial version of that file now" in message |
| 2911 | |
| 2912 | |
| 2913 | @pytest.mark.asyncio |
| 2914 | async def test_tool_batch_runner_redirects_post_write_self_audit_to_next_missing_artifact( |
| 2915 | temp_dir: Path, |
| 2916 | ) -> None: |
| 2917 | async def assess_confidence( |
| 2918 | tool_name: str, |
| 2919 | tool_args: dict, |
| 2920 | context: str, |
| 2921 | ) -> ConfidenceAssessment: |
| 2922 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 2923 | |
| 2924 | async def verify_action( |
| 2925 | tool_name: str, |
| 2926 | tool_args: dict, |
| 2927 | result: str, |
| 2928 | expected: str = "", |
| 2929 | ) -> ActionVerification: |
| 2930 | raise AssertionError("Verification should not run in this scenario") |
| 2931 | |
| 2932 | nginx_root = temp_dir / "guides" / "nginx" |
| 2933 | chapters = nginx_root / "chapters" |
| 2934 | chapters.mkdir(parents=True) |
| 2935 | index_path = nginx_root / "index.html" |
| 2936 | index_path.write_text( |
| 2937 | "\n".join( |
| 2938 | [ |
| 2939 | "<html>", |
| 2940 | '<a href="chapters/01-introduction.html">Chapter 1: Introduction to Nginx</a>', |
| 2941 | '<a href="chapters/02-installation.html">Chapter 2: Installation and Setup</a>', |
| 2942 | "</html>", |
| 2943 | ] |
| 2944 | ) |
| 2945 | + "\n" |
| 2946 | ) |
| 2947 | |
| 2948 | implementation_plan = temp_dir / "implementation.md" |
| 2949 | implementation_plan.write_text( |
| 2950 | "\n".join( |
| 2951 | [ |
| 2952 | "# Implementation Plan", |
| 2953 | "", |
| 2954 | "## File Changes", |
| 2955 | f"- `{nginx_root}/`", |
| 2956 | f"- `{chapters}/`", |
| 2957 | f"- `{index_path}`", |
| 2958 | f"- `{chapters / '01-introduction.html'}`", |
| 2959 | "", |
| 2960 | ] |
| 2961 | ) |
| 2962 | ) |
| 2963 | |
| 2964 | context = build_context( |
| 2965 | temp_dir=temp_dir, |
| 2966 | messages=[], |
| 2967 | safeguards=FakeSafeguards(), |
| 2968 | assess_confidence=assess_confidence, |
| 2969 | verify_action=verify_action, |
| 2970 | auto_recover=False, |
| 2971 | ) |
| 2972 | persistent_messages: list[str] = [] |
| 2973 | ephemeral_messages: list[str] = [] |
| 2974 | context.queue_steering_message_callback = persistent_messages.append |
| 2975 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 2976 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 2977 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 2978 | dod.implementation_plan = str(implementation_plan) |
| 2979 | dod.touched_files.append(str(index_path)) |
| 2980 | dod.completed_items.append("Develop the main index.html file for the nginx guide") |
| 2981 | dod.pending_items.append("Create chapter files for the nginx guide") |
| 2982 | |
| 2983 | tool_call = ToolCall( |
| 2984 | id="read-index-self-audit", |
| 2985 | name="read", |
| 2986 | arguments={"file_path": str(index_path)}, |
| 2987 | ) |
| 2988 | executor = FakeExecutor( |
| 2989 | [ |
| 2990 | tool_outcome( |
| 2991 | tool_call=tool_call, |
| 2992 | output="1\t<html>\n", |
| 2993 | is_error=False, |
| 2994 | ) |
| 2995 | ] |
| 2996 | ) |
| 2997 | |
| 2998 | summary = TurnSummary(final_response="") |
| 2999 | await runner.execute_batch( |
| 3000 | tool_calls=[tool_call], |
| 3001 | tool_source="assistant", |
| 3002 | pending_tool_calls_seen=set(), |
| 3003 | emit=_noop_emit, |
| 3004 | summary=summary, |
| 3005 | dod=dod, |
| 3006 | executor=executor, # type: ignore[arg-type] |
| 3007 | on_confirmation=None, |
| 3008 | on_user_question=None, |
| 3009 | emit_confirmation=None, |
| 3010 | consecutive_errors=0, |
| 3011 | ) |
| 3012 | |
| 3013 | assert persistent_messages |
| 3014 | message = persistent_messages[-1] |
| 3015 | assert "You already have the current contents of `index.html` from the successful write." in message |
| 3016 | assert "Resume by creating `01-introduction.html` now." in message |
| 3017 | assert "Do not spend another turn rereading the file you just wrote or on TodoWrite alone." in message |
| 3018 | assert ephemeral_messages == [] |
| 3019 | |
| 3020 | |
| 3021 | @pytest.mark.asyncio |
| 3022 | async def test_tool_batch_runner_preserves_first_file_handoff_after_recovery_prompt( |
| 3023 | temp_dir: Path, |
| 3024 | ) -> None: |
| 3025 | async def assess_confidence( |
| 3026 | tool_name: str, |
| 3027 | tool_args: dict, |
| 3028 | context: str, |
| 3029 | ) -> ConfidenceAssessment: |
| 3030 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 3031 | |
| 3032 | async def verify_action( |
| 3033 | tool_name: str, |
| 3034 | tool_args: dict, |
| 3035 | result: str, |
| 3036 | expected: str = "", |
| 3037 | ) -> ActionVerification: |
| 3038 | raise AssertionError("Verification should not run for this scenario") |
| 3039 | |
| 3040 | nginx_root = temp_dir / "guides" / "nginx" |
| 3041 | chapters = nginx_root / "chapters" |
| 3042 | chapters.mkdir(parents=True) |
| 3043 | index_path = nginx_root / "index.html" |
| 3044 | |
| 3045 | implementation_plan = temp_dir / "implementation.md" |
| 3046 | implementation_plan.write_text( |
| 3047 | "\n".join( |
| 3048 | [ |
| 3049 | "# Implementation Plan", |
| 3050 | "", |
| 3051 | "## File Changes", |
| 3052 | f"- `{chapters}/`", |
| 3053 | f"- `{index_path}`", |
| 3054 | f"- `{chapters / '01-introduction.html'}`", |
| 3055 | "", |
| 3056 | ] |
| 3057 | ) |
| 3058 | ) |
| 3059 | |
| 3060 | context = build_context( |
| 3061 | temp_dir=temp_dir, |
| 3062 | messages=[ |
| 3063 | Message( |
| 3064 | role=Role.USER, |
| 3065 | content=( |
| 3066 | "[EMPTY ASSISTANT RESPONSE]\n" |
| 3067 | "Respond with that concrete mutation tool call now. Do not return an empty response." |
| 3068 | ), |
| 3069 | ) |
| 3070 | ], |
| 3071 | safeguards=FakeSafeguards(), |
| 3072 | assess_confidence=assess_confidence, |
| 3073 | verify_action=verify_action, |
| 3074 | auto_recover=False, |
| 3075 | ) |
| 3076 | persistent_messages: list[str] = [] |
| 3077 | ephemeral_messages: list[str] = [] |
| 3078 | context.queue_steering_message_callback = persistent_messages.append |
| 3079 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 3080 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 3081 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 3082 | dod.implementation_plan = str(implementation_plan) |
| 3083 | sync_todos_to_definition_of_done( |
| 3084 | dod, |
| 3085 | [ |
| 3086 | { |
| 3087 | "content": "Create the main index.html file with proper structure", |
| 3088 | "active_form": "Creating the main index.html file with proper structure", |
| 3089 | "status": "pending", |
| 3090 | }, |
| 3091 | { |
| 3092 | "content": "Create each chapter file with appropriate content", |
| 3093 | "active_form": "Creating each chapter file with appropriate content", |
| 3094 | "status": "pending", |
| 3095 | }, |
| 3096 | ], |
| 3097 | ) |
| 3098 | |
| 3099 | tool_call = ToolCall( |
| 3100 | id="write-index-recovered", |
| 3101 | name="write", |
| 3102 | arguments={ |
| 3103 | "file_path": str(index_path), |
| 3104 | "content": "<html></html>\n", |
| 3105 | }, |
| 3106 | ) |
| 3107 | executor = FakeExecutor( |
| 3108 | [ |
| 3109 | tool_outcome( |
| 3110 | tool_call=tool_call, |
| 3111 | output=f"Successfully wrote 14 bytes to {index_path}", |
| 3112 | is_error=False, |
| 3113 | ) |
| 3114 | ] |
| 3115 | ) |
| 3116 | |
| 3117 | summary = TurnSummary(final_response="") |
| 3118 | await runner.execute_batch( |
| 3119 | tool_calls=[tool_call], |
| 3120 | tool_source="assistant", |
| 3121 | pending_tool_calls_seen=set(), |
| 3122 | emit=_noop_emit, |
| 3123 | summary=summary, |
| 3124 | dod=dod, |
| 3125 | executor=executor, # type: ignore[arg-type] |
| 3126 | on_confirmation=None, |
| 3127 | on_user_question=None, |
| 3128 | emit_confirmation=None, |
| 3129 | consecutive_errors=0, |
| 3130 | ) |
| 3131 | |
| 3132 | assert persistent_messages |
| 3133 | assert ephemeral_messages == [] |
| 3134 | message = persistent_messages[-1] |
| 3135 | assert "Next step: create `01-introduction.html`." in message |
| 3136 | assert "Write a compact but real initial version of that file now" not in message |
| 3137 | |
| 3138 | |
| 3139 | @pytest.mark.asyncio |
| 3140 | async def test_tool_batch_runner_todowrite_uses_concrete_output_language_for_aggregate_chapter_step( |
| 3141 | temp_dir: Path, |
| 3142 | ) -> None: |
| 3143 | async def assess_confidence( |
| 3144 | tool_name: str, |
| 3145 | tool_args: dict, |
| 3146 | context: str, |
| 3147 | ) -> ConfidenceAssessment: |
| 3148 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 3149 | |
| 3150 | async def verify_action( |
| 3151 | tool_name: str, |
| 3152 | tool_args: dict, |
| 3153 | result: str, |
| 3154 | expected: str = "", |
| 3155 | ) -> ActionVerification: |
| 3156 | raise AssertionError("Verification should not run in this scenario") |
| 3157 | |
| 3158 | guide_root = temp_dir / "guides" / "nginx" |
| 3159 | chapters = guide_root / "chapters" |
| 3160 | chapters.mkdir(parents=True) |
| 3161 | index_path = guide_root / "index.html" |
| 3162 | index_path.write_text( |
| 3163 | "\n".join( |
| 3164 | [ |
| 3165 | "<html>", |
| 3166 | '<a href="chapters/01-introduction.html">Chapter 1: Introduction to Nginx</a>', |
| 3167 | '<a href="chapters/02-installation.html">Chapter 2: Installation and Setup</a>', |
| 3168 | "</html>", |
| 3169 | ] |
| 3170 | ) |
| 3171 | + "\n" |
| 3172 | ) |
| 3173 | |
| 3174 | implementation_plan = temp_dir / "implementation.md" |
| 3175 | implementation_plan.write_text( |
| 3176 | "\n".join( |
| 3177 | [ |
| 3178 | "# Implementation Plan", |
| 3179 | "", |
| 3180 | "## File Changes", |
| 3181 | f"- `{guide_root}/`", |
| 3182 | f"- `{chapters}/`", |
| 3183 | f"- `{index_path}`", |
| 3184 | "", |
| 3185 | ] |
| 3186 | ) |
| 3187 | ) |
| 3188 | |
| 3189 | context = build_context( |
| 3190 | temp_dir=temp_dir, |
| 3191 | messages=[], |
| 3192 | safeguards=FakeSafeguards(), |
| 3193 | assess_confidence=assess_confidence, |
| 3194 | verify_action=verify_action, |
| 3195 | ) |
| 3196 | queued_messages: list[str] = [] |
| 3197 | context.queue_steering_message_callback = queued_messages.append |
| 3198 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 3199 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 3200 | dod.implementation_plan = str(implementation_plan) |
| 3201 | dod.touched_files.append(str(index_path)) |
| 3202 | sync_todos_to_definition_of_done( |
| 3203 | dod, |
| 3204 | [ |
| 3205 | { |
| 3206 | "content": "Develop the main index.html file with proper structure", |
| 3207 | "active_form": "Developing the main index.html file with proper structure", |
| 3208 | "status": "completed", |
| 3209 | }, |
| 3210 | { |
| 3211 | "content": "Create chapter files with content and structure", |
| 3212 | "active_form": "Creating chapter files with content and structure", |
| 3213 | "status": "pending", |
| 3214 | }, |
| 3215 | ], |
| 3216 | ) |
| 3217 | |
| 3218 | todos = [ |
| 3219 | { |
| 3220 | "content": "Develop the main index.html file with proper structure", |
| 3221 | "active_form": "Developing the main index.html file with proper structure", |
| 3222 | "status": "completed", |
| 3223 | }, |
| 3224 | { |
| 3225 | "content": "Create chapter files with content and structure", |
| 3226 | "active_form": "Creating chapter files with content and structure", |
| 3227 | "status": "pending", |
| 3228 | }, |
| 3229 | ] |
| 3230 | tool_call = ToolCall( |
| 3231 | id="todo-aggregate", |
| 3232 | name="TodoWrite", |
| 3233 | arguments={"todos": todos}, |
| 3234 | ) |
| 3235 | executor = FakeExecutor( |
| 3236 | [ |
| 3237 | tool_outcome( |
| 3238 | tool_call=tool_call, |
| 3239 | output="Todos updated", |
| 3240 | is_error=False, |
| 3241 | metadata={"new_todos": todos}, |
| 3242 | ) |
| 3243 | ] |
| 3244 | ) |
| 3245 | |
| 3246 | summary = TurnSummary(final_response="") |
| 3247 | await runner.execute_batch( |
| 3248 | tool_calls=[tool_call], |
| 3249 | tool_source="assistant", |
| 3250 | pending_tool_calls_seen=set(), |
| 3251 | emit=_noop_emit, |
| 3252 | summary=summary, |
| 3253 | dod=dod, |
| 3254 | executor=executor, # type: ignore[arg-type] |
| 3255 | on_confirmation=None, |
| 3256 | on_user_question=None, |
| 3257 | emit_confirmation=None, |
| 3258 | consecutive_errors=0, |
| 3259 | ) |
| 3260 | |
| 3261 | assert queued_messages |
| 3262 | message = queued_messages[-1] |
| 3263 | assert "Todo tracking is updated." in message |
| 3264 | assert "Next step: create `01-introduction.html`." in message |
| 3265 | assert ( |
| 3266 | "Continue with the next pending item: `Create chapter files with content and structure`." |
| 3267 | not in message |
| 3268 | ) |
| 3269 | |
| 3270 | |
| 3271 | @pytest.mark.asyncio |
| 3272 | async def test_duplicate_observation_nudge_prioritizes_missing_artifact_over_review( |
| 3273 | temp_dir: Path, |
| 3274 | ) -> None: |
| 3275 | async def assess_confidence( |
| 3276 | tool_name: str, |
| 3277 | tool_args: dict, |
| 3278 | context: str, |
| 3279 | ) -> ConfidenceAssessment: |
| 3280 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 3281 | |
| 3282 | async def verify_action( |
| 3283 | tool_name: str, |
| 3284 | tool_args: dict, |
| 3285 | result: str, |
| 3286 | expected: str = "", |
| 3287 | ) -> ActionVerification: |
| 3288 | raise AssertionError("Verification should not run for this scenario") |
| 3289 | |
| 3290 | guide_root = temp_dir / "guides" / "nginx" |
| 3291 | chapters = guide_root / "chapters" |
| 3292 | chapters.mkdir(parents=True) |
| 3293 | index_path = guide_root / "index.html" |
| 3294 | chapter_one = chapters / "01-getting-started.html" |
| 3295 | chapter_one.write_text("<h1>One</h1>\n") |
| 3296 | index_path.write_text("<a href=\"chapters/01-getting-started.html\">One</a>\n") |
| 3297 | |
| 3298 | implementation_plan = temp_dir / "implementation.md" |
| 3299 | implementation_plan.write_text( |
| 3300 | "\n".join( |
| 3301 | [ |
| 3302 | "# Implementation Plan", |
| 3303 | "", |
| 3304 | "## File Changes", |
| 3305 | f"- `{index_path}`", |
| 3306 | f"- `{chapter_one}`", |
| 3307 | f"- `{chapters / '06-ssl-configuration.html'}`", |
| 3308 | "", |
| 3309 | ] |
| 3310 | ) |
| 3311 | ) |
| 3312 | |
| 3313 | context = build_context( |
| 3314 | temp_dir=temp_dir, |
| 3315 | messages=[], |
| 3316 | safeguards=FakeSafeguards(), |
| 3317 | assess_confidence=assess_confidence, |
| 3318 | verify_action=verify_action, |
| 3319 | auto_recover=False, |
| 3320 | ) |
| 3321 | persistent_messages: list[str] = [] |
| 3322 | ephemeral_messages: list[str] = [] |
| 3323 | context.queue_steering_message_callback = persistent_messages.append |
| 3324 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 3325 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 3326 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 3327 | dod.implementation_plan = str(implementation_plan) |
| 3328 | sync_todos_to_definition_of_done( |
| 3329 | dod, |
| 3330 | [ |
| 3331 | { |
| 3332 | "content": "Ensure all files are properly linked and formatted consistently", |
| 3333 | "active_form": "Working on: Ensure all files are properly linked and formatted consistently", |
| 3334 | "status": "pending", |
| 3335 | }, |
| 3336 | { |
| 3337 | "content": "Create the final chapter (06-ssl-configuration.html)", |
| 3338 | "active_form": "Working on: Create the final chapter (06-ssl-configuration.html)", |
| 3339 | "status": "pending", |
| 3340 | }, |
| 3341 | ], |
| 3342 | ) |
| 3343 | assert tool_batches_should_prioritize_missing_artifact( |
| 3344 | dod=dod, |
| 3345 | next_pending=dod.pending_items[0], |
| 3346 | missing_artifact=(chapters / "06-ssl-configuration.html", False), |
| 3347 | project_root=temp_dir, |
| 3348 | ) |
| 3349 | |
| 3350 | tool_call = ToolCall( |
| 3351 | id="dup-read", |
| 3352 | name="read", |
| 3353 | arguments={"file_path": str(index_path)}, |
| 3354 | ) |
| 3355 | runner._queue_duplicate_observation_nudge(tool_call, dod=dod) # type: ignore[attr-defined] |
| 3356 | |
| 3357 | assert persistent_messages |
| 3358 | message = persistent_messages[-1] |
| 3359 | assert "06-ssl-configuration.html" in message |
| 3360 | assert "Do not switch into review or consistency-check mode" in message |
| 3361 | assert ( |
| 3362 | "Continue with the next pending item: `Ensure all files are properly linked and formatted consistently`" |
| 3363 | not in message |
| 3364 | ) |
| 3365 | |
| 3366 | |
| 3367 | @pytest.mark.asyncio |
| 3368 | async def test_tool_batch_runner_hands_off_to_verification_once_planned_artifacts_exist( |
| 3369 | temp_dir: Path, |
| 3370 | ) -> None: |
| 3371 | async def assess_confidence( |
| 3372 | tool_name: str, |
| 3373 | tool_args: dict, |
| 3374 | context: str, |
| 3375 | ) -> ConfidenceAssessment: |
| 3376 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 3377 | |
| 3378 | async def verify_action( |
| 3379 | tool_name: str, |
| 3380 | tool_args: dict, |
| 3381 | result: str, |
| 3382 | expected: str = "", |
| 3383 | ) -> ActionVerification: |
| 3384 | raise AssertionError("Verification should not run for this scenario") |
| 3385 | |
| 3386 | guide_root = temp_dir / "guides" / "nginx" |
| 3387 | chapters = guide_root / "chapters" |
| 3388 | chapters.mkdir(parents=True) |
| 3389 | index_path = guide_root / "index.html" |
| 3390 | chapter_one = chapters / "01-getting-started.html" |
| 3391 | chapter_two = chapters / "02-installation.html" |
| 3392 | index_path.write_text("<a href=\"chapters/01-getting-started.html\">One</a>\n") |
| 3393 | chapter_one.write_text("<h1>One</h1>\n") |
| 3394 | chapter_two.write_text("<h1>Two</h1>\n") |
| 3395 | |
| 3396 | implementation_plan = temp_dir / "implementation.md" |
| 3397 | implementation_plan.write_text( |
| 3398 | "\n".join( |
| 3399 | [ |
| 3400 | "# Implementation Plan", |
| 3401 | "", |
| 3402 | "## File Changes", |
| 3403 | f"- `{chapters}/`", |
| 3404 | f"- `{index_path}`", |
| 3405 | f"- `{chapter_one}`", |
| 3406 | f"- `{chapter_two}`", |
| 3407 | "", |
| 3408 | ] |
| 3409 | ) |
| 3410 | ) |
| 3411 | |
| 3412 | context = build_context( |
| 3413 | temp_dir=temp_dir, |
| 3414 | messages=[], |
| 3415 | safeguards=FakeSafeguards(), |
| 3416 | assess_confidence=assess_confidence, |
| 3417 | verify_action=verify_action, |
| 3418 | auto_recover=False, |
| 3419 | ) |
| 3420 | persistent_messages: list[str] = [] |
| 3421 | ephemeral_messages: list[str] = [] |
| 3422 | context.queue_steering_message_callback = persistent_messages.append |
| 3423 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 3424 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 3425 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 3426 | dod.implementation_plan = str(implementation_plan) |
| 3427 | sync_todos_to_definition_of_done( |
| 3428 | dod, |
| 3429 | [ |
| 3430 | { |
| 3431 | "content": "Create the guide files", |
| 3432 | "active_form": "Working on: Create the guide files", |
| 3433 | "status": "completed", |
| 3434 | }, |
| 3435 | { |
| 3436 | "content": "Ensure all files are properly linked and formatted consistently", |
| 3437 | "active_form": "Working on: Ensure all files are properly linked and formatted consistently", |
| 3438 | "status": "pending", |
| 3439 | }, |
| 3440 | ], |
| 3441 | ) |
| 3442 | tool_call = ToolCall( |
| 3443 | id="write-final", |
| 3444 | name="write", |
| 3445 | arguments={ |
| 3446 | "file_path": str(chapter_two), |
| 3447 | "content": "<h1>Two</h1>\n", |
| 3448 | }, |
| 3449 | ) |
| 3450 | executor = FakeExecutor( |
| 3451 | [ |
| 3452 | tool_outcome( |
| 3453 | tool_call=tool_call, |
| 3454 | output=f"Successfully wrote {chapter_two}", |
| 3455 | is_error=False, |
| 3456 | ) |
| 3457 | ] |
| 3458 | ) |
| 3459 | |
| 3460 | summary = TurnSummary(final_response="") |
| 3461 | await runner.execute_batch( |
| 3462 | tool_calls=[tool_call], |
| 3463 | tool_source="assistant", |
| 3464 | pending_tool_calls_seen=set(), |
| 3465 | emit=_noop_emit, |
| 3466 | summary=summary, |
| 3467 | dod=dod, |
| 3468 | executor=executor, # type: ignore[arg-type] |
| 3469 | on_confirmation=None, |
| 3470 | on_user_question=None, |
| 3471 | emit_confirmation=None, |
| 3472 | consecutive_errors=0, |
| 3473 | ) |
| 3474 | |
| 3475 | assert any( |
| 3476 | "All explicitly planned artifacts now exist on disk." in message |
| 3477 | for message in persistent_messages |
| 3478 | ) |
| 3479 | assert any( |
| 3480 | "Ensure all files are properly linked and formatted consistently" in message |
| 3481 | for message in persistent_messages |
| 3482 | ) |
| 3483 | assert any( |
| 3484 | "Finish with a final response once no specific mismatch remains so Loader can verify." |
| 3485 | in message |
| 3486 | for message in persistent_messages |
| 3487 | ) |
| 3488 | |
| 3489 | |
| 3490 | @pytest.mark.asyncio |
| 3491 | async def test_tool_batch_runner_mutation_handoff_points_at_next_missing_artifact( |
| 3492 | temp_dir: Path, |
| 3493 | ) -> None: |
| 3494 | async def assess_confidence( |
| 3495 | tool_name: str, |
| 3496 | tool_args: dict, |
| 3497 | context: str, |
| 3498 | ) -> ConfidenceAssessment: |
| 3499 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 3500 | |
| 3501 | async def verify_action( |
| 3502 | tool_name: str, |
| 3503 | tool_args: dict, |
| 3504 | result: str, |
| 3505 | expected: str = "", |
| 3506 | ) -> ActionVerification: |
| 3507 | raise AssertionError("Verification should not run in this scenario") |
| 3508 | |
| 3509 | guide_root = temp_dir / "guides" / "nginx" |
| 3510 | chapters = guide_root / "chapters" |
| 3511 | guide_root.mkdir(parents=True) |
| 3512 | chapters.mkdir() |
| 3513 | index_path = guide_root / "index.html" |
| 3514 | index_path.write_text("<html></html>\n") |
| 3515 | chapter_one = chapters / "01-getting-started.html" |
| 3516 | chapter_two = chapters / "02-installation.html" |
| 3517 | implementation_plan = temp_dir / "implementation.md" |
| 3518 | implementation_plan.write_text( |
| 3519 | "\n".join( |
| 3520 | [ |
| 3521 | "# Implementation Plan", |
| 3522 | "", |
| 3523 | "## File Changes", |
| 3524 | f"- `{guide_root}/`", |
| 3525 | f"- `{index_path}`", |
| 3526 | f"- `{chapter_one}`", |
| 3527 | f"- `{chapter_two}`", |
| 3528 | "", |
| 3529 | ] |
| 3530 | ) |
| 3531 | ) |
| 3532 | |
| 3533 | context = build_context( |
| 3534 | temp_dir=temp_dir, |
| 3535 | messages=[], |
| 3536 | safeguards=FakeSafeguards(), |
| 3537 | assess_confidence=assess_confidence, |
| 3538 | verify_action=verify_action, |
| 3539 | auto_recover=False, |
| 3540 | ) |
| 3541 | persistent_messages: list[str] = [] |
| 3542 | ephemeral_messages: list[str] = [] |
| 3543 | context.queue_steering_message_callback = persistent_messages.append |
| 3544 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 3545 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 3546 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 3547 | dod.implementation_plan = str(implementation_plan) |
| 3548 | sync_todos_to_definition_of_done( |
| 3549 | dod, |
| 3550 | [ |
| 3551 | { |
| 3552 | "content": "Create the main index.html file with proper structure", |
| 3553 | "active_form": "Working on: Create the main index.html file with proper structure", |
| 3554 | "status": "pending", |
| 3555 | }, |
| 3556 | { |
| 3557 | "content": "Create each chapter file in sequence, following the established pattern", |
| 3558 | "active_form": "Working on: Create each chapter file in sequence, following the established pattern", |
| 3559 | "status": "pending", |
| 3560 | }, |
| 3561 | { |
| 3562 | "content": "Ensure all files are properly linked and formatted consistently", |
| 3563 | "active_form": "Working on: Ensure all files are properly linked and formatted consistently", |
| 3564 | "status": "pending", |
| 3565 | }, |
| 3566 | ], |
| 3567 | ) |
| 3568 | tool_call = ToolCall( |
| 3569 | id="write-index", |
| 3570 | name="write", |
| 3571 | arguments={"file_path": str(index_path), "content": "<html></html>\n"}, |
| 3572 | ) |
| 3573 | executor = FakeExecutor( |
| 3574 | [tool_outcome(tool_call=tool_call, output=f"Successfully wrote {index_path}", is_error=False)] |
| 3575 | ) |
| 3576 | |
| 3577 | summary = TurnSummary(final_response="") |
| 3578 | await runner.execute_batch( |
| 3579 | tool_calls=[tool_call], |
| 3580 | tool_source="assistant", |
| 3581 | pending_tool_calls_seen=set(), |
| 3582 | emit=_noop_emit, |
| 3583 | summary=summary, |
| 3584 | dod=dod, |
| 3585 | executor=executor, # type: ignore[arg-type] |
| 3586 | on_confirmation=None, |
| 3587 | on_user_question=None, |
| 3588 | emit_confirmation=None, |
| 3589 | consecutive_errors=0, |
| 3590 | ) |
| 3591 | |
| 3592 | assert persistent_messages |
| 3593 | assert ephemeral_messages == [] |
| 3594 | message = persistent_messages[-1] |
| 3595 | assert "Next step: create `01-getting-started.html`." in message |
| 3596 | assert "Write a compact but real initial version of that file now" not in message |
| 3597 | assert "refresh `TodoWrite`" not in message |
| 3598 | assert "Do not reread reference material or spend the next turn on bookkeeping." in message |
| 3599 | |
| 3600 | |
| 3601 | @pytest.mark.asyncio |
| 3602 | async def test_tool_batch_runner_large_plan_does_not_claim_completion_early( |
| 3603 | temp_dir: Path, |
| 3604 | ) -> None: |
| 3605 | async def assess_confidence( |
| 3606 | tool_name: str, |
| 3607 | tool_args: dict, |
| 3608 | context: str, |
| 3609 | ) -> ConfidenceAssessment: |
| 3610 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 3611 | |
| 3612 | async def verify_action( |
| 3613 | tool_name: str, |
| 3614 | tool_args: dict, |
| 3615 | result: str, |
| 3616 | expected: str = "", |
| 3617 | ) -> ActionVerification: |
| 3618 | raise AssertionError("Verification should not run in this scenario") |
| 3619 | |
| 3620 | guide_root = temp_dir / "guides" / "nginx" |
| 3621 | chapters = guide_root / "chapters" |
| 3622 | guide_root.mkdir(parents=True) |
| 3623 | chapters.mkdir() |
| 3624 | index_path = guide_root / "index.html" |
| 3625 | index_path.write_text("<html></html>\n") |
| 3626 | |
| 3627 | chapter_paths = [ |
| 3628 | chapters / "01-getting-started.html", |
| 3629 | chapters / "02-installation.html", |
| 3630 | chapters / "03-first-website.html", |
| 3631 | chapters / "04-configuration-basics.html", |
| 3632 | chapters / "05-advanced-configurations.html", |
| 3633 | chapters / "06-performance-tuning.html", |
| 3634 | chapters / "07-security-best-practices.html", |
| 3635 | ] |
| 3636 | for chapter in chapter_paths[:4]: |
| 3637 | chapter.write_text(f"<h1>{chapter.stem}</h1>\n") |
| 3638 | chapter_paths[4].write_text("<h1>Advanced configurations</h1>\n") |
| 3639 | |
| 3640 | implementation_plan = temp_dir / "implementation.md" |
| 3641 | implementation_plan.write_text( |
| 3642 | "\n".join( |
| 3643 | [ |
| 3644 | "# Implementation Plan", |
| 3645 | "", |
| 3646 | "## File Changes", |
| 3647 | f"- `{guide_root}/`", |
| 3648 | f"- `{chapters}/`", |
| 3649 | f"- `{index_path}`", |
| 3650 | *[f"- `{path}`" for path in chapter_paths], |
| 3651 | "", |
| 3652 | ] |
| 3653 | ) |
| 3654 | ) |
| 3655 | |
| 3656 | context = build_context( |
| 3657 | temp_dir=temp_dir, |
| 3658 | messages=[], |
| 3659 | safeguards=FakeSafeguards(), |
| 3660 | assess_confidence=assess_confidence, |
| 3661 | verify_action=verify_action, |
| 3662 | auto_recover=False, |
| 3663 | ) |
| 3664 | persistent_messages: list[str] = [] |
| 3665 | ephemeral_messages: list[str] = [] |
| 3666 | context.queue_steering_message_callback = persistent_messages.append |
| 3667 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 3668 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 3669 | dod = create_definition_of_done("Create a thorough nginx guide.") |
| 3670 | dod.implementation_plan = str(implementation_plan) |
| 3671 | sync_todos_to_definition_of_done( |
| 3672 | dod, |
| 3673 | [ |
| 3674 | { |
| 3675 | "content": "Create the nginx guide artifacts", |
| 3676 | "active_form": "Creating nginx guide artifacts", |
| 3677 | "status": "pending", |
| 3678 | }, |
| 3679 | { |
| 3680 | "content": "Verify all guide files are linked and complete", |
| 3681 | "active_form": "Verifying guide linkage and completeness", |
| 3682 | "status": "pending", |
| 3683 | }, |
| 3684 | ], |
| 3685 | ) |
| 3686 | tool_call = ToolCall( |
| 3687 | id="write-chapter-05", |
| 3688 | name="write", |
| 3689 | arguments={ |
| 3690 | "file_path": str(chapter_paths[4]), |
| 3691 | "content": "<h1>Advanced configurations</h1>\n", |
| 3692 | }, |
| 3693 | ) |
| 3694 | executor = FakeExecutor( |
| 3695 | [ |
| 3696 | tool_outcome( |
| 3697 | tool_call=tool_call, |
| 3698 | output=f"Successfully wrote {chapter_paths[4]}", |
| 3699 | is_error=False, |
| 3700 | ) |
| 3701 | ] |
| 3702 | ) |
| 3703 | |
| 3704 | summary = TurnSummary(final_response="") |
| 3705 | await runner.execute_batch( |
| 3706 | tool_calls=[tool_call], |
| 3707 | tool_source="assistant", |
| 3708 | pending_tool_calls_seen=set(), |
| 3709 | emit=_noop_emit, |
| 3710 | summary=summary, |
| 3711 | dod=dod, |
| 3712 | executor=executor, # type: ignore[arg-type] |
| 3713 | on_confirmation=None, |
| 3714 | on_user_question=None, |
| 3715 | emit_confirmation=None, |
| 3716 | consecutive_errors=0, |
| 3717 | ) |
| 3718 | |
| 3719 | assert any( |
| 3720 | "Next step: create `06-performance-tuning.html`." in message |
| 3721 | for message in ephemeral_messages |
| 3722 | ) |
| 3723 | assert not any( |
| 3724 | "All explicitly planned artifacts now exist on disk." in message |
| 3725 | for message in ephemeral_messages |
| 3726 | ) |
| 3727 | |
| 3728 | |
| 3729 | @pytest.mark.asyncio |
| 3730 | async def test_tool_batch_runner_uses_compact_missing_artifact_nudge_after_substantial_progress( |
| 3731 | temp_dir: Path, |
| 3732 | ) -> None: |
| 3733 | async def assess_confidence( |
| 3734 | tool_name: str, |
| 3735 | tool_args: dict, |
| 3736 | context: str, |
| 3737 | ) -> ConfidenceAssessment: |
| 3738 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 3739 | |
| 3740 | async def verify_action( |
| 3741 | tool_name: str, |
| 3742 | tool_args: dict, |
| 3743 | result: str, |
| 3744 | expected: str = "", |
| 3745 | ) -> ActionVerification: |
| 3746 | raise AssertionError("Verification should not run in this scenario") |
| 3747 | |
| 3748 | guide_root = temp_dir / "guides" / "nginx" |
| 3749 | chapters = guide_root / "chapters" |
| 3750 | guide_root.mkdir(parents=True) |
| 3751 | chapters.mkdir() |
| 3752 | index_path = guide_root / "index.html" |
| 3753 | chapter_paths = [ |
| 3754 | chapters / "01-introduction.html", |
| 3755 | chapters / "02-installation.html", |
| 3756 | chapters / "03-configuration.html", |
| 3757 | chapters / "04-basic-usage.html", |
| 3758 | chapters / "05-advanced-features.html", |
| 3759 | ] |
| 3760 | for path in (index_path, *chapter_paths[:4]): |
| 3761 | path.write_text("<html></html>\n") |
| 3762 | |
| 3763 | implementation_plan = temp_dir / "implementation.md" |
| 3764 | implementation_plan.write_text( |
| 3765 | "\n".join( |
| 3766 | [ |
| 3767 | "# Implementation Plan", |
| 3768 | "", |
| 3769 | "## File Changes", |
| 3770 | f"- `{guide_root}/`", |
| 3771 | f"- `{chapters}/`", |
| 3772 | f"- `{index_path}`", |
| 3773 | *[f"- `{path}`" for path in chapter_paths], |
| 3774 | "", |
| 3775 | ] |
| 3776 | ) |
| 3777 | ) |
| 3778 | |
| 3779 | context = build_context( |
| 3780 | temp_dir=temp_dir, |
| 3781 | messages=[], |
| 3782 | safeguards=FakeSafeguards(), |
| 3783 | assess_confidence=assess_confidence, |
| 3784 | verify_action=verify_action, |
| 3785 | auto_recover=False, |
| 3786 | ) |
| 3787 | persistent_messages: list[str] = [] |
| 3788 | ephemeral_messages: list[str] = [] |
| 3789 | context.queue_steering_message_callback = persistent_messages.append |
| 3790 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 3791 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 3792 | dod = create_definition_of_done("Create a thorough nginx guide.") |
| 3793 | dod.implementation_plan = str(implementation_plan) |
| 3794 | dod.touched_files.extend(str(path) for path in (index_path, *chapter_paths[:4])) |
| 3795 | dod.completed_items.extend( |
| 3796 | [ |
| 3797 | "Create the nginx directory structure", |
| 3798 | "Create the main index.html file with proper structure", |
| 3799 | ] |
| 3800 | ) |
| 3801 | sync_todos_to_definition_of_done( |
| 3802 | dod, |
| 3803 | [ |
| 3804 | { |
| 3805 | "content": "Create each chapter file with appropriate content", |
| 3806 | "active_form": "Creating each chapter file with appropriate content", |
| 3807 | "status": "pending", |
| 3808 | } |
| 3809 | ], |
| 3810 | ) |
| 3811 | tool_call = ToolCall( |
| 3812 | id="write-chapter-04", |
| 3813 | name="write", |
| 3814 | arguments={ |
| 3815 | "file_path": str(chapter_paths[3]), |
| 3816 | "content": "<html>updated</html>\n", |
| 3817 | }, |
| 3818 | ) |
| 3819 | executor = FakeExecutor( |
| 3820 | [ |
| 3821 | tool_outcome( |
| 3822 | tool_call=tool_call, |
| 3823 | output=f"Successfully wrote {chapter_paths[3]}", |
| 3824 | is_error=False, |
| 3825 | ) |
| 3826 | ] |
| 3827 | ) |
| 3828 | |
| 3829 | summary = TurnSummary(final_response="") |
| 3830 | await runner.execute_batch( |
| 3831 | tool_calls=[tool_call], |
| 3832 | tool_source="assistant", |
| 3833 | pending_tool_calls_seen=set(), |
| 3834 | emit=_noop_emit, |
| 3835 | summary=summary, |
| 3836 | dod=dod, |
| 3837 | executor=executor, # type: ignore[arg-type] |
| 3838 | on_confirmation=None, |
| 3839 | on_user_question=None, |
| 3840 | emit_confirmation=None, |
| 3841 | consecutive_errors=0, |
| 3842 | ) |
| 3843 | |
| 3844 | assert ephemeral_messages |
| 3845 | message = ephemeral_messages[-1] |
| 3846 | assert "Next step: create `05-advanced-features.html`." in message |
| 3847 | assert "Do not reread reference material or spend the next turn on bookkeeping." in message |
| 3848 | assert "refresh `TodoWrite`" not in message |
| 3849 | |
| 3850 | |
| 3851 | @pytest.mark.asyncio |
| 3852 | async def test_tool_batch_runner_todowrite_with_missing_artifact_requeues_exact_resume_step( |
| 3853 | temp_dir: Path, |
| 3854 | ) -> None: |
| 3855 | async def assess_confidence( |
| 3856 | tool_name: str, |
| 3857 | tool_args: dict, |
| 3858 | context: str, |
| 3859 | ) -> ConfidenceAssessment: |
| 3860 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 3861 | |
| 3862 | async def verify_action( |
| 3863 | tool_name: str, |
| 3864 | tool_args: dict, |
| 3865 | result: str, |
| 3866 | expected: str = "", |
| 3867 | ) -> ActionVerification: |
| 3868 | raise AssertionError("Verification should not run in this scenario") |
| 3869 | |
| 3870 | guide_root = temp_dir / "guides" / "nginx" |
| 3871 | chapters = guide_root / "chapters" |
| 3872 | guide_root.mkdir(parents=True) |
| 3873 | chapters.mkdir() |
| 3874 | index_path = guide_root / "index.html" |
| 3875 | index_path.write_text("<html></html>\n") |
| 3876 | chapter_one = chapters / "01-getting-started.html" |
| 3877 | chapter_two = chapters / "02-installation.html" |
| 3878 | chapter_one.write_text("<h1>One</h1>\n") |
| 3879 | |
| 3880 | implementation_plan = temp_dir / "implementation.md" |
| 3881 | implementation_plan.write_text( |
| 3882 | "\n".join( |
| 3883 | [ |
| 3884 | "# Implementation Plan", |
| 3885 | "", |
| 3886 | "## File Changes", |
| 3887 | f"- `{guide_root}/`", |
| 3888 | f"- `{chapters}/`", |
| 3889 | f"- `{index_path}`", |
| 3890 | f"- `{chapter_one}`", |
| 3891 | f"- `{chapter_two}`", |
| 3892 | "", |
| 3893 | ] |
| 3894 | ) |
| 3895 | ) |
| 3896 | |
| 3897 | context = build_context( |
| 3898 | temp_dir=temp_dir, |
| 3899 | messages=[], |
| 3900 | safeguards=FakeSafeguards(), |
| 3901 | assess_confidence=assess_confidence, |
| 3902 | verify_action=verify_action, |
| 3903 | auto_recover=False, |
| 3904 | ) |
| 3905 | persistent_messages: list[str] = [] |
| 3906 | ephemeral_messages: list[str] = [] |
| 3907 | context.queue_steering_message_callback = persistent_messages.append |
| 3908 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 3909 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 3910 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 3911 | dod.implementation_plan = str(implementation_plan) |
| 3912 | sync_todos_to_definition_of_done( |
| 3913 | dod, |
| 3914 | [ |
| 3915 | { |
| 3916 | "content": "Create 01-getting-started.html", |
| 3917 | "active_form": "Creating 01-getting-started.html", |
| 3918 | "status": "completed", |
| 3919 | }, |
| 3920 | { |
| 3921 | "content": "Create 02-installation.html", |
| 3922 | "active_form": "Creating 02-installation.html", |
| 3923 | "status": "pending", |
| 3924 | }, |
| 3925 | ], |
| 3926 | ) |
| 3927 | dod.touched_files.extend([str(index_path), str(chapter_one)]) |
| 3928 | |
| 3929 | tool_call = ToolCall( |
| 3930 | id="todo-only", |
| 3931 | name="TodoWrite", |
| 3932 | arguments={ |
| 3933 | "todos": [ |
| 3934 | { |
| 3935 | "content": "Create 01-getting-started.html", |
| 3936 | "active_form": "Creating 01-getting-started.html", |
| 3937 | "status": "completed", |
| 3938 | }, |
| 3939 | { |
| 3940 | "content": "Create 02-installation.html", |
| 3941 | "active_form": "Creating 02-installation.html", |
| 3942 | "status": "pending", |
| 3943 | }, |
| 3944 | ] |
| 3945 | }, |
| 3946 | ) |
| 3947 | executor = FakeExecutor( |
| 3948 | [ |
| 3949 | tool_outcome( |
| 3950 | tool_call=tool_call, |
| 3951 | output="Todos updated", |
| 3952 | is_error=False, |
| 3953 | metadata={ |
| 3954 | "new_todos": [ |
| 3955 | { |
| 3956 | "content": "Create 01-getting-started.html", |
| 3957 | "active_form": "Creating 01-getting-started.html", |
| 3958 | "status": "completed", |
| 3959 | }, |
| 3960 | { |
| 3961 | "content": "Create 02-installation.html", |
| 3962 | "active_form": "Creating 02-installation.html", |
| 3963 | "status": "pending", |
| 3964 | }, |
| 3965 | ] |
| 3966 | }, |
| 3967 | ) |
| 3968 | ] |
| 3969 | ) |
| 3970 | |
| 3971 | summary = TurnSummary(final_response="") |
| 3972 | await runner.execute_batch( |
| 3973 | tool_calls=[tool_call], |
| 3974 | tool_source="assistant", |
| 3975 | pending_tool_calls_seen=set(), |
| 3976 | emit=_noop_emit, |
| 3977 | summary=summary, |
| 3978 | dod=dod, |
| 3979 | executor=executor, # type: ignore[arg-type] |
| 3980 | on_confirmation=None, |
| 3981 | on_user_question=None, |
| 3982 | emit_confirmation=None, |
| 3983 | consecutive_errors=0, |
| 3984 | ) |
| 3985 | |
| 3986 | assert persistent_messages |
| 3987 | message = persistent_messages[-1] |
| 3988 | assert "Todo tracking is updated. Next step: create `02-installation.html`." in message |
| 3989 | assert "Prefer one `write(file_path=..., content=...)` call" in message |
| 3990 | assert "Make your next response the concrete mutation tool call itself." in message |
| 3991 | assert ephemeral_messages == [] |
| 3992 | |
| 3993 | |
| 3994 | @pytest.mark.asyncio |
| 3995 | async def test_tool_batch_runner_todowrite_after_artifacts_exist_pushes_verification_handoff( |
| 3996 | temp_dir: Path, |
| 3997 | ) -> None: |
| 3998 | async def assess_confidence( |
| 3999 | tool_name: str, |
| 4000 | tool_args: dict, |
| 4001 | context: str, |
| 4002 | ) -> ConfidenceAssessment: |
| 4003 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 4004 | |
| 4005 | async def verify_action( |
| 4006 | tool_name: str, |
| 4007 | tool_args: dict, |
| 4008 | result: str, |
| 4009 | expected: str = "", |
| 4010 | ) -> ActionVerification: |
| 4011 | raise AssertionError("Verification should not run in this scenario") |
| 4012 | |
| 4013 | guide_root = temp_dir / "guides" / "nginx" |
| 4014 | chapters = guide_root / "chapters" |
| 4015 | guide_root.mkdir(parents=True) |
| 4016 | chapters.mkdir() |
| 4017 | index_path = guide_root / "index.html" |
| 4018 | chapter_one = chapters / "01-getting-started.html" |
| 4019 | chapter_two = chapters / "02-installation.html" |
| 4020 | index_path.write_text("<html></html>\n") |
| 4021 | chapter_one.write_text("<h1>One</h1>\n") |
| 4022 | chapter_two.write_text("<h1>Two</h1>\n") |
| 4023 | |
| 4024 | implementation_plan = temp_dir / "implementation.md" |
| 4025 | implementation_plan.write_text( |
| 4026 | "\n".join( |
| 4027 | [ |
| 4028 | "# Implementation Plan", |
| 4029 | "", |
| 4030 | "## File Changes", |
| 4031 | f"- `{guide_root}/`", |
| 4032 | f"- `{chapters}/`", |
| 4033 | f"- `{index_path}`", |
| 4034 | f"- `{chapter_one}`", |
| 4035 | f"- `{chapter_two}`", |
| 4036 | "", |
| 4037 | ] |
| 4038 | ) |
| 4039 | ) |
| 4040 | |
| 4041 | context = build_context( |
| 4042 | temp_dir=temp_dir, |
| 4043 | messages=[], |
| 4044 | safeguards=FakeSafeguards(), |
| 4045 | assess_confidence=assess_confidence, |
| 4046 | verify_action=verify_action, |
| 4047 | auto_recover=False, |
| 4048 | ) |
| 4049 | queued_messages: list[str] = [] |
| 4050 | context.queue_steering_message_callback = queued_messages.append |
| 4051 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 4052 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 4053 | dod.implementation_plan = str(implementation_plan) |
| 4054 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 4055 | sync_todos_to_definition_of_done( |
| 4056 | dod, |
| 4057 | [ |
| 4058 | { |
| 4059 | "content": "First, examine the existing Fortran guide structure to understand the format and content organization", |
| 4060 | "active_form": "Working on: First, examine the existing Fortran guide structure to understand the format and content organization", |
| 4061 | "status": "pending", |
| 4062 | }, |
| 4063 | { |
| 4064 | "content": "Verify all guide files are linked and complete", |
| 4065 | "active_form": "Working on: Verify all guide files are linked and complete", |
| 4066 | "status": "pending", |
| 4067 | }, |
| 4068 | ], |
| 4069 | project_root=temp_dir, |
| 4070 | ) |
| 4071 | |
| 4072 | tool_call = ToolCall( |
| 4073 | id="todo-only", |
| 4074 | name="TodoWrite", |
| 4075 | arguments={ |
| 4076 | "todos": [ |
| 4077 | { |
| 4078 | "content": "First, examine the existing Fortran guide structure to understand the format and content organization", |
| 4079 | "active_form": "Working on: First, examine the existing Fortran guide structure to understand the format and content organization", |
| 4080 | "status": "pending", |
| 4081 | }, |
| 4082 | { |
| 4083 | "content": "Verify all guide files are linked and complete", |
| 4084 | "active_form": "Working on: Verify all guide files are linked and complete", |
| 4085 | "status": "pending", |
| 4086 | }, |
| 4087 | ] |
| 4088 | }, |
| 4089 | ) |
| 4090 | executor = FakeExecutor( |
| 4091 | [ |
| 4092 | tool_outcome( |
| 4093 | tool_call=tool_call, |
| 4094 | output="Todos updated", |
| 4095 | is_error=False, |
| 4096 | metadata={ |
| 4097 | "new_todos": [ |
| 4098 | { |
| 4099 | "content": "First, examine the existing Fortran guide structure to understand the format and content organization", |
| 4100 | "active_form": "Working on: First, examine the existing Fortran guide structure to understand the format and content organization", |
| 4101 | "status": "pending", |
| 4102 | }, |
| 4103 | { |
| 4104 | "content": "Verify all guide files are linked and complete", |
| 4105 | "active_form": "Working on: Verify all guide files are linked and complete", |
| 4106 | "status": "pending", |
| 4107 | }, |
| 4108 | ] |
| 4109 | }, |
| 4110 | ) |
| 4111 | ] |
| 4112 | ) |
| 4113 | |
| 4114 | summary = TurnSummary(final_response="") |
| 4115 | await runner.execute_batch( |
| 4116 | tool_calls=[tool_call], |
| 4117 | tool_source="assistant", |
| 4118 | pending_tool_calls_seen=set(), |
| 4119 | emit=_noop_emit, |
| 4120 | summary=summary, |
| 4121 | dod=dod, |
| 4122 | executor=executor, # type: ignore[arg-type] |
| 4123 | on_confirmation=None, |
| 4124 | on_user_question=None, |
| 4125 | emit_confirmation=None, |
| 4126 | consecutive_errors=0, |
| 4127 | ) |
| 4128 | |
| 4129 | assert queued_messages |
| 4130 | message = queued_messages[-1] |
| 4131 | assert "Todo tracking is updated. All explicitly planned artifacts now exist on disk." in message |
| 4132 | assert "Verify all guide files are linked and complete" in message |
| 4133 | assert ( |
| 4134 | "Finish with a final response once no specific mismatch remains so Loader can verify." |
| 4135 | in message |
| 4136 | ) |
| 4137 | assert "reopen reference materials" in message |
| 4138 | assert "Fortran guide structure" not in message |
| 4139 | assert context.workflow_mode == "execute" |
| 4140 | |
| 4141 | |
| 4142 | @pytest.mark.asyncio |
| 4143 | async def test_tool_batch_runner_todowrite_after_outputs_exist_but_links_missing_still_handoffs_to_verify( |
| 4144 | temp_dir: Path, |
| 4145 | ) -> None: |
| 4146 | async def assess_confidence( |
| 4147 | tool_name: str, |
| 4148 | tool_args: dict, |
| 4149 | context: str, |
| 4150 | ) -> ConfidenceAssessment: |
| 4151 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 4152 | |
| 4153 | async def verify_action( |
| 4154 | tool_name: str, |
| 4155 | tool_args: dict, |
| 4156 | result: str, |
| 4157 | expected: str = "", |
| 4158 | ) -> ActionVerification: |
| 4159 | raise AssertionError("Verification should not run for this scenario") |
| 4160 | |
| 4161 | guide_root = temp_dir / "guides" / "nginx" |
| 4162 | chapters = guide_root / "chapters" |
| 4163 | guide_root.mkdir(parents=True) |
| 4164 | chapters.mkdir() |
| 4165 | index_path = guide_root / "index.html" |
| 4166 | chapter_one = chapters / "01-introduction.html" |
| 4167 | chapter_two = chapters / "02-installation.html" |
| 4168 | index_path.write_text( |
| 4169 | "\n".join( |
| 4170 | [ |
| 4171 | '<a href="chapters/01-introduction.html">Intro</a>', |
| 4172 | '<a href="chapters/02-installation.html">Install</a>', |
| 4173 | '<a href="../index.html">Back</a>', |
| 4174 | "", |
| 4175 | ] |
| 4176 | ) |
| 4177 | ) |
| 4178 | chapter_one.write_text("<html></html>\n") |
| 4179 | chapter_two.write_text("<html></html>\n") |
| 4180 | |
| 4181 | implementation_plan = temp_dir / "implementation.md" |
| 4182 | implementation_plan.write_text( |
| 4183 | "\n".join( |
| 4184 | [ |
| 4185 | "# Implementation Plan", |
| 4186 | "", |
| 4187 | "## File Changes", |
| 4188 | f"- `{guide_root}/`", |
| 4189 | f"- `{chapters}/`", |
| 4190 | f"- `{index_path}`", |
| 4191 | f"- `{chapter_one}`", |
| 4192 | f"- `{chapter_two}`", |
| 4193 | "", |
| 4194 | ] |
| 4195 | ) |
| 4196 | ) |
| 4197 | |
| 4198 | context = build_context( |
| 4199 | temp_dir=temp_dir, |
| 4200 | messages=[], |
| 4201 | safeguards=FakeSafeguards(), |
| 4202 | assess_confidence=assess_confidence, |
| 4203 | verify_action=verify_action, |
| 4204 | auto_recover=False, |
| 4205 | ) |
| 4206 | queued_messages: list[str] = [] |
| 4207 | context.queue_steering_message_callback = queued_messages.append |
| 4208 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 4209 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 4210 | dod.implementation_plan = str(implementation_plan) |
| 4211 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 4212 | sync_todos_to_definition_of_done( |
| 4213 | dod, |
| 4214 | [ |
| 4215 | { |
| 4216 | "content": "Create chapter files following the established pattern", |
| 4217 | "active_form": "Creating chapter files", |
| 4218 | "status": "in_progress", |
| 4219 | } |
| 4220 | ], |
| 4221 | project_root=temp_dir, |
| 4222 | ) |
| 4223 | |
| 4224 | tool_call = ToolCall( |
| 4225 | id="todo-post-build", |
| 4226 | name="TodoWrite", |
| 4227 | arguments={ |
| 4228 | "todos": [ |
| 4229 | { |
| 4230 | "content": "Create chapter files following the established pattern", |
| 4231 | "active_form": "Creating chapter files", |
| 4232 | "status": "in_progress", |
| 4233 | } |
| 4234 | ] |
| 4235 | }, |
| 4236 | ) |
| 4237 | executor = FakeExecutor( |
| 4238 | [ |
| 4239 | tool_outcome( |
| 4240 | tool_call=tool_call, |
| 4241 | output="Todos updated", |
| 4242 | is_error=False, |
| 4243 | metadata={ |
| 4244 | "new_todos": [ |
| 4245 | { |
| 4246 | "content": "Create chapter files following the established pattern", |
| 4247 | "active_form": "Creating chapter files", |
| 4248 | "status": "in_progress", |
| 4249 | } |
| 4250 | ] |
| 4251 | }, |
| 4252 | ) |
| 4253 | ] |
| 4254 | ) |
| 4255 | |
| 4256 | summary = TurnSummary(final_response="") |
| 4257 | await runner.execute_batch( |
| 4258 | tool_calls=[tool_call], |
| 4259 | tool_source="assistant", |
| 4260 | pending_tool_calls_seen=set(), |
| 4261 | emit=_noop_emit, |
| 4262 | summary=summary, |
| 4263 | dod=dod, |
| 4264 | executor=executor, # type: ignore[arg-type] |
| 4265 | on_confirmation=None, |
| 4266 | on_user_question=None, |
| 4267 | emit_confirmation=None, |
| 4268 | consecutive_errors=0, |
| 4269 | ) |
| 4270 | |
| 4271 | assert queued_messages |
| 4272 | message = queued_messages[-1] |
| 4273 | assert "Todo tracking is updated. All explicitly planned artifacts now exist on disk." in message |
| 4274 | assert "Finish with a final response now so Loader can run verification automatically." in message |
| 4275 | assert "Repair or verify the current files instead of expanding the artifact set." not in message |
| 4276 | assert context.workflow_mode == "verify" |
| 4277 | |
| 4278 | |
| 4279 | @pytest.mark.asyncio |
| 4280 | async def test_tool_batch_runner_todowrite_during_quality_repair_requires_mutation( |
| 4281 | temp_dir: Path, |
| 4282 | ) -> None: |
| 4283 | async def assess_confidence( |
| 4284 | tool_name: str, |
| 4285 | tool_args: dict, |
| 4286 | context: str, |
| 4287 | ) -> ConfidenceAssessment: |
| 4288 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 4289 | |
| 4290 | async def verify_action( |
| 4291 | tool_name: str, |
| 4292 | tool_args: dict, |
| 4293 | result: str, |
| 4294 | expected: str = "", |
| 4295 | ) -> ActionVerification: |
| 4296 | raise AssertionError("Verification should not run for this scenario") |
| 4297 | |
| 4298 | guide_root = temp_dir / "guides" / "nginx" |
| 4299 | chapters = guide_root / "chapters" |
| 4300 | chapters.mkdir(parents=True) |
| 4301 | index_path = guide_root / "index.html" |
| 4302 | chapter_one = chapters / "01-introduction.html" |
| 4303 | index_path.write_text("<html></html>\n") |
| 4304 | chapter_one.write_text("<html></html>\n") |
| 4305 | |
| 4306 | implementation_plan = temp_dir / "implementation.md" |
| 4307 | implementation_plan.write_text( |
| 4308 | "\n".join( |
| 4309 | [ |
| 4310 | "# Implementation Plan", |
| 4311 | "", |
| 4312 | "## File Changes", |
| 4313 | f"- `{guide_root}/`", |
| 4314 | f"- `{chapters}/`", |
| 4315 | f"- `{index_path}`", |
| 4316 | f"- `{chapter_one}`", |
| 4317 | "", |
| 4318 | ] |
| 4319 | ) |
| 4320 | ) |
| 4321 | |
| 4322 | context = build_context( |
| 4323 | temp_dir=temp_dir, |
| 4324 | messages=[ |
| 4325 | Message( |
| 4326 | role=Role.USER, |
| 4327 | content=( |
| 4328 | "Repair focus:\n" |
| 4329 | f"- Improve `{chapter_one}`: thin content (409 text chars, expected at least 1758).\n" |
| 4330 | f"- Improve `{chapter_one}`: insufficient structured content (6 blocks, expected at least 18).\n" |
| 4331 | f"- Immediate next step: edit `{chapter_one}`.\n" |
| 4332 | ), |
| 4333 | ) |
| 4334 | ], |
| 4335 | safeguards=FakeSafeguards(), |
| 4336 | assess_confidence=assess_confidence, |
| 4337 | verify_action=verify_action, |
| 4338 | auto_recover=False, |
| 4339 | ) |
| 4340 | context.set_workflow_mode("verify") |
| 4341 | queued_messages: list[str] = [] |
| 4342 | emitted_responses: list[str] = [] |
| 4343 | context.queue_steering_message_callback = queued_messages.append |
| 4344 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 4345 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 4346 | dod.implementation_plan = str(implementation_plan) |
| 4347 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 4348 | sync_todos_to_definition_of_done( |
| 4349 | dod, |
| 4350 | [ |
| 4351 | { |
| 4352 | "content": "Expand generated chapters to satisfy quality verification", |
| 4353 | "active_form": "Expanding generated chapters", |
| 4354 | "status": "in_progress", |
| 4355 | } |
| 4356 | ], |
| 4357 | project_root=temp_dir, |
| 4358 | ) |
| 4359 | pending_before_todowrite = list(dod.pending_items) |
| 4360 | completed_before_todowrite = list(dod.completed_items) |
| 4361 | |
| 4362 | tool_call = ToolCall( |
| 4363 | id="todo-quality", |
| 4364 | name="TodoWrite", |
| 4365 | arguments={ |
| 4366 | "todos": [ |
| 4367 | { |
| 4368 | "content": "Expand generated chapters to satisfy quality verification", |
| 4369 | "active_form": "Expanding generated chapters", |
| 4370 | "status": "completed", |
| 4371 | } |
| 4372 | ] |
| 4373 | }, |
| 4374 | ) |
| 4375 | executor = FakeExecutor( |
| 4376 | [ |
| 4377 | tool_outcome( |
| 4378 | tool_call=tool_call, |
| 4379 | output="Todos updated", |
| 4380 | is_error=False, |
| 4381 | metadata={ |
| 4382 | "new_todos": [ |
| 4383 | { |
| 4384 | "content": "Expand generated chapters to satisfy quality verification", |
| 4385 | "active_form": "Expanding generated chapters", |
| 4386 | "status": "completed", |
| 4387 | } |
| 4388 | ] |
| 4389 | }, |
| 4390 | ) |
| 4391 | ] |
| 4392 | ) |
| 4393 | |
| 4394 | async def emit(event: AgentEvent) -> None: |
| 4395 | if event.type == "response": |
| 4396 | emitted_responses.append(str(event.content)) |
| 4397 | |
| 4398 | summary = TurnSummary(final_response="") |
| 4399 | result = await runner.execute_batch( |
| 4400 | tool_calls=[tool_call], |
| 4401 | tool_source="assistant", |
| 4402 | pending_tool_calls_seen=set(), |
| 4403 | emit=emit, |
| 4404 | summary=summary, |
| 4405 | dod=dod, |
| 4406 | executor=executor, # type: ignore[arg-type] |
| 4407 | on_confirmation=None, |
| 4408 | on_user_question=None, |
| 4409 | emit_confirmation=None, |
| 4410 | consecutive_errors=0, |
| 4411 | ) |
| 4412 | |
| 4413 | assert queued_messages |
| 4414 | message = queued_messages[-1] |
| 4415 | assert "verification still has an active HTML content-quality repair" in message |
| 4416 | assert "TodoWrite cannot satisfy that verifier" in message |
| 4417 | assert f"Immediate next step: edit `{chapter_one.resolve(strict=False)}`" in message |
| 4418 | assert "thin content" in message |
| 4419 | assert "Finish with a final response now" not in message |
| 4420 | assert context.workflow_mode == "execute" |
| 4421 | assert result.halted is False |
| 4422 | assert summary.final_response == "" |
| 4423 | assert not emitted_responses |
| 4424 | assert dod.pending_items == pending_before_todowrite |
| 4425 | assert dod.completed_items == completed_before_todowrite |
| 4426 | |
| 4427 | |
| 4428 | def test_todowrite_quality_repair_nudge_uses_exact_anchor_after_stale_context( |
| 4429 | temp_dir: Path, |
| 4430 | ) -> None: |
| 4431 | async def assess_confidence( |
| 4432 | tool_name: str, |
| 4433 | tool_args: dict, |
| 4434 | context: str, |
| 4435 | ) -> ConfidenceAssessment: |
| 4436 | raise AssertionError("Confidence should not run for direct nudge test") |
| 4437 | |
| 4438 | async def verify_action( |
| 4439 | tool_name: str, |
| 4440 | tool_args: dict, |
| 4441 | result: str, |
| 4442 | expected: str = "", |
| 4443 | ) -> ActionVerification: |
| 4444 | raise AssertionError("Verification should not run for direct nudge test") |
| 4445 | |
| 4446 | guide_root = temp_dir / "guides" / "nginx" |
| 4447 | chapters = guide_root / "chapters" |
| 4448 | chapters.mkdir(parents=True) |
| 4449 | chapter_one = chapters / "05-load-balancing.html" |
| 4450 | chapter_one.write_text("<html><body><h1>Load Balancing</h1></body></html>\n") |
| 4451 | context = build_context( |
| 4452 | temp_dir=temp_dir, |
| 4453 | messages=[ |
| 4454 | Message( |
| 4455 | role=Role.USER, |
| 4456 | content=( |
| 4457 | "Repair focus:\n" |
| 4458 | f"- Improve `{chapter_one}`: thin content " |
| 4459 | "(846 text chars, expected at least 1758).\n" |
| 4460 | f"- Immediate next step: edit `{chapter_one}`.\n" |
| 4461 | ), |
| 4462 | ), |
| 4463 | Message( |
| 4464 | role=Role.TOOL, |
| 4465 | content=( |
| 4466 | "Observation [edit]: Error: Failed to complete the operation " |
| 4467 | f"after 2 attempts for {chapter_one}. old_string not found in file." |
| 4468 | ), |
| 4469 | ), |
| 4470 | ], |
| 4471 | safeguards=FakeSafeguards(), |
| 4472 | assess_confidence=assess_confidence, |
| 4473 | verify_action=verify_action, |
| 4474 | auto_recover=False, |
| 4475 | ) |
| 4476 | queued_messages: list[str] = [] |
| 4477 | context.queue_steering_message_callback = queued_messages.append |
| 4478 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 4479 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 4480 | |
| 4481 | runner._queue_todowrite_resume_nudge(dod=dod) |
| 4482 | |
| 4483 | assert queued_messages |
| 4484 | message = queued_messages[-1] |
| 4485 | assert f"Immediate next step: edit `{chapter_one.resolve(strict=False)}`" in message |
| 4486 | assert "`edit(file_path=..., old_string=..., new_string=...)`" in message |
| 4487 | assert "Use this exact current closing-tail anchor as `old_string`" in message |
| 4488 | assert "```html\n</body></html>\n```" in message |
| 4489 | assert "do not call `read`, `patch`, `write`, or TodoWrite again first" in message |
| 4490 | |
| 4491 | |
| 4492 | @pytest.mark.asyncio |
| 4493 | async def test_tool_batch_runner_preempts_post_build_audit_after_todowrite_verify_handoff( |
| 4494 | temp_dir: Path, |
| 4495 | ) -> None: |
| 4496 | async def assess_confidence( |
| 4497 | tool_name: str, |
| 4498 | tool_args: dict, |
| 4499 | context: str, |
| 4500 | ) -> ConfidenceAssessment: |
| 4501 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 4502 | |
| 4503 | async def verify_action( |
| 4504 | tool_name: str, |
| 4505 | tool_args: dict, |
| 4506 | result: str, |
| 4507 | expected: str = "", |
| 4508 | ) -> ActionVerification: |
| 4509 | raise AssertionError("Verification should not run for this scenario") |
| 4510 | |
| 4511 | guide_root = temp_dir / "guides" / "nginx" |
| 4512 | chapters = guide_root / "chapters" |
| 4513 | guide_root.mkdir(parents=True) |
| 4514 | chapters.mkdir() |
| 4515 | index_path = guide_root / "index.html" |
| 4516 | chapter_one = chapters / "01-introduction.html" |
| 4517 | chapter_two = chapters / "02-installation.html" |
| 4518 | index_path.write_text( |
| 4519 | "\n".join( |
| 4520 | [ |
| 4521 | '<li><a href="chapters/01-introduction.html">Chapter 1: Introduction</a></li>', |
| 4522 | '<li><a href="chapters/02-installation.html">Chapter 2: Installation</a></li>', |
| 4523 | "", |
| 4524 | ] |
| 4525 | ) |
| 4526 | ) |
| 4527 | chapter_one.write_text("<html></html>\n") |
| 4528 | chapter_two.write_text("<html></html>\n") |
| 4529 | |
| 4530 | implementation_plan = temp_dir / "implementation.md" |
| 4531 | implementation_plan.write_text( |
| 4532 | "\n".join( |
| 4533 | [ |
| 4534 | "# Implementation Plan", |
| 4535 | "", |
| 4536 | "## File Changes", |
| 4537 | f"- `{guide_root}/`", |
| 4538 | f"- `{chapters}/`", |
| 4539 | f"- `{index_path}`", |
| 4540 | f"- `{chapter_one}`", |
| 4541 | f"- `{chapter_two}`", |
| 4542 | "", |
| 4543 | ] |
| 4544 | ) |
| 4545 | ) |
| 4546 | |
| 4547 | context = build_context( |
| 4548 | temp_dir=temp_dir, |
| 4549 | messages=[], |
| 4550 | safeguards=FakeSafeguards(), |
| 4551 | assess_confidence=assess_confidence, |
| 4552 | verify_action=verify_action, |
| 4553 | auto_recover=False, |
| 4554 | ) |
| 4555 | queued_messages: list[str] = [] |
| 4556 | context.queue_steering_message_callback = queued_messages.append |
| 4557 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 4558 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 4559 | dod.implementation_plan = str(implementation_plan) |
| 4560 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 4561 | |
| 4562 | todo_call = ToolCall( |
| 4563 | id="todo-post-build-preempt", |
| 4564 | name="TodoWrite", |
| 4565 | arguments={"todos": []}, |
| 4566 | ) |
| 4567 | audit_read = ToolCall( |
| 4568 | id="read-after-todo", |
| 4569 | name="read", |
| 4570 | arguments={"file_path": str(index_path)}, |
| 4571 | ) |
| 4572 | executor = FakeExecutor( |
| 4573 | [ |
| 4574 | tool_outcome( |
| 4575 | tool_call=todo_call, |
| 4576 | output="Todos updated", |
| 4577 | is_error=False, |
| 4578 | metadata={"new_todos": []}, |
| 4579 | ), |
| 4580 | tool_outcome( |
| 4581 | tool_call=audit_read, |
| 4582 | output=index_path.read_text(), |
| 4583 | is_error=False, |
| 4584 | ), |
| 4585 | ] |
| 4586 | ) |
| 4587 | |
| 4588 | summary = TurnSummary(final_response="") |
| 4589 | result = await runner.execute_batch( |
| 4590 | tool_calls=[todo_call, audit_read], |
| 4591 | tool_source="assistant", |
| 4592 | pending_tool_calls_seen=set(), |
| 4593 | emit=_noop_emit, |
| 4594 | summary=summary, |
| 4595 | dod=dod, |
| 4596 | executor=executor, # type: ignore[arg-type] |
| 4597 | on_confirmation=None, |
| 4598 | on_user_question=None, |
| 4599 | emit_confirmation=None, |
| 4600 | consecutive_errors=0, |
| 4601 | ) |
| 4602 | |
| 4603 | assert result.continue_after_batch is True |
| 4604 | assert result.halted is False |
| 4605 | assert [call.id for call in executor.calls] == ["todo-post-build-preempt"] |
| 4606 | assert len(summary.tool_result_messages) == 1 |
| 4607 | assert context.workflow_mode == "verify" |
| 4608 | assert queued_messages |
| 4609 | assert "Finish with a final response now so Loader can run verification automatically." in queued_messages[-1] |
| 4610 | |
| 4611 | |
| 4612 | @pytest.mark.asyncio |
| 4613 | async def test_tool_batch_runner_todowrite_complete_directory_plan_does_not_reinfer_first_child( |
| 4614 | temp_dir: Path, |
| 4615 | ) -> None: |
| 4616 | async def assess_confidence( |
| 4617 | tool_name: str, |
| 4618 | tool_args: dict, |
| 4619 | context: str, |
| 4620 | ) -> ConfidenceAssessment: |
| 4621 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 4622 | |
| 4623 | async def verify_action( |
| 4624 | tool_name: str, |
| 4625 | tool_args: dict, |
| 4626 | result: str, |
| 4627 | expected: str = "", |
| 4628 | ) -> ActionVerification: |
| 4629 | raise AssertionError("Verification should not run for this scenario") |
| 4630 | |
| 4631 | reference = temp_dir / "fortran" / "chapters" / "01-introduction.html" |
| 4632 | reference.parent.mkdir(parents=True) |
| 4633 | reference.write_text("<h1>Introduction</h1>\n") |
| 4634 | |
| 4635 | guide_root = temp_dir / "Loader" / "guides" / "nginx" |
| 4636 | chapters = guide_root / "chapters" |
| 4637 | guide_root.mkdir(parents=True) |
| 4638 | chapters.mkdir() |
| 4639 | index_path = guide_root / "index.html" |
| 4640 | chapter_one = chapters / "01-introduction.html" |
| 4641 | chapter_two = chapters / "02-installation.html" |
| 4642 | chapter_three = chapters / "03-basic-configuration.html" |
| 4643 | index_path.write_text( |
| 4644 | "\n".join( |
| 4645 | [ |
| 4646 | '<a href="chapters/01-introduction.html">Introduction</a>', |
| 4647 | '<a href="chapters/02-installation.html">Installation</a>', |
| 4648 | '<a href="chapters/03-basic-configuration.html">Configuration</a>', |
| 4649 | "", |
| 4650 | ] |
| 4651 | ) |
| 4652 | ) |
| 4653 | chapter_one.write_text("<html></html>\n") |
| 4654 | chapter_two.write_text("<html></html>\n") |
| 4655 | chapter_three.write_text("<html></html>\n") |
| 4656 | |
| 4657 | implementation_plan = temp_dir / "implementation.md" |
| 4658 | implementation_plan.write_text( |
| 4659 | "\n".join( |
| 4660 | [ |
| 4661 | "# Implementation Plan", |
| 4662 | "", |
| 4663 | "## File Changes", |
| 4664 | f"- `{guide_root / 'index.html'}`", |
| 4665 | f"- `{chapters}/`", |
| 4666 | "", |
| 4667 | ] |
| 4668 | ) |
| 4669 | ) |
| 4670 | |
| 4671 | messages = [ |
| 4672 | Message( |
| 4673 | role=Role.ASSISTANT, |
| 4674 | content="I examined the reference guide structure.", |
| 4675 | tool_calls=[ |
| 4676 | ToolCall( |
| 4677 | id="read-reference-child", |
| 4678 | name="read", |
| 4679 | arguments={"file_path": str(reference)}, |
| 4680 | ) |
| 4681 | ], |
| 4682 | ) |
| 4683 | ] |
| 4684 | context = build_context( |
| 4685 | temp_dir=temp_dir, |
| 4686 | messages=messages, |
| 4687 | safeguards=FakeSafeguards(), |
| 4688 | assess_confidence=assess_confidence, |
| 4689 | verify_action=verify_action, |
| 4690 | auto_recover=False, |
| 4691 | ) |
| 4692 | queued_messages: list[str] = [] |
| 4693 | context.queue_steering_message_callback = queued_messages.append |
| 4694 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 4695 | dod = create_definition_of_done("Create an equally thorough nginx guide.") |
| 4696 | dod.implementation_plan = str(implementation_plan) |
| 4697 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 4698 | |
| 4699 | todo_call = ToolCall( |
| 4700 | id="todo-complete-directory-plan", |
| 4701 | name="TodoWrite", |
| 4702 | arguments={"todos": []}, |
| 4703 | ) |
| 4704 | executor = FakeExecutor( |
| 4705 | [ |
| 4706 | tool_outcome( |
| 4707 | tool_call=todo_call, |
| 4708 | output="Todos updated", |
| 4709 | is_error=False, |
| 4710 | metadata={"new_todos": []}, |
| 4711 | ) |
| 4712 | ] |
| 4713 | ) |
| 4714 | |
| 4715 | summary = TurnSummary(final_response="") |
| 4716 | result = await runner.execute_batch( |
| 4717 | tool_calls=[todo_call], |
| 4718 | tool_source="assistant", |
| 4719 | pending_tool_calls_seen=set(), |
| 4720 | emit=_noop_emit, |
| 4721 | summary=summary, |
| 4722 | dod=dod, |
| 4723 | executor=executor, # type: ignore[arg-type] |
| 4724 | on_confirmation=None, |
| 4725 | on_user_question=None, |
| 4726 | emit_confirmation=None, |
| 4727 | consecutive_errors=0, |
| 4728 | ) |
| 4729 | |
| 4730 | assert result.halted is True |
| 4731 | assert result.final_response == ( |
| 4732 | "Todo tracking is complete; running Loader verification on the generated " |
| 4733 | "files now." |
| 4734 | ) |
| 4735 | assert summary.final_response == result.final_response |
| 4736 | assert context.workflow_mode == "verify" |
| 4737 | assert summary.tool_result_messages |
| 4738 | assert ( |
| 4739 | "final response should be provided next for Loader verification" |
| 4740 | in summary.tool_result_messages[-1].content |
| 4741 | ) |
| 4742 | assert "01-introduction.html" not in summary.tool_result_messages[-1].content |
| 4743 | assert "chapter files" not in summary.tool_result_messages[-1].content.lower() |
| 4744 | assert "fortran guide structure" not in summary.tool_result_messages[-1].content.lower() |
| 4745 | |
| 4746 | |
| 4747 | @pytest.mark.asyncio |
| 4748 | async def test_tool_batch_runner_preempts_post_build_observation_batch_for_verify_handoff( |
| 4749 | temp_dir: Path, |
| 4750 | ) -> None: |
| 4751 | async def assess_confidence( |
| 4752 | tool_name: str, |
| 4753 | tool_args: dict, |
| 4754 | context: str, |
| 4755 | ) -> ConfidenceAssessment: |
| 4756 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 4757 | |
| 4758 | async def verify_action( |
| 4759 | tool_name: str, |
| 4760 | tool_args: dict, |
| 4761 | result: str, |
| 4762 | expected: str = "", |
| 4763 | ) -> ActionVerification: |
| 4764 | raise AssertionError("Verification should not run for this scenario") |
| 4765 | |
| 4766 | guide_root = temp_dir / "guides" / "nginx" |
| 4767 | chapters = guide_root / "chapters" |
| 4768 | guide_root.mkdir(parents=True) |
| 4769 | chapters.mkdir() |
| 4770 | index_path = guide_root / "index.html" |
| 4771 | chapter_one = chapters / "01-introduction.html" |
| 4772 | chapter_two = chapters / "02-installation.html" |
| 4773 | chapter_three = chapters / "03-configuration.html" |
| 4774 | index_path.write_text( |
| 4775 | "\n".join( |
| 4776 | [ |
| 4777 | '<li><a href="chapters/01-introduction.html">Chapter 1: Introduction</a></li>', |
| 4778 | '<li><a href="chapters/02-installation.html">Chapter 2: Installation</a></li>', |
| 4779 | "", |
| 4780 | ] |
| 4781 | ) |
| 4782 | ) |
| 4783 | chapter_one.write_text("<html></html>\n") |
| 4784 | chapter_two.write_text("<html></html>\n") |
| 4785 | chapter_three.write_text("<html></html>\n") |
| 4786 | |
| 4787 | implementation_plan = temp_dir / "implementation.md" |
| 4788 | implementation_plan.write_text( |
| 4789 | "\n".join( |
| 4790 | [ |
| 4791 | "# Implementation Plan", |
| 4792 | "", |
| 4793 | "## File Changes", |
| 4794 | f"- `{guide_root}/`", |
| 4795 | f"- `{chapters}/`", |
| 4796 | f"- `{index_path}`", |
| 4797 | "", |
| 4798 | ] |
| 4799 | ) |
| 4800 | ) |
| 4801 | |
| 4802 | context = build_context( |
| 4803 | temp_dir=temp_dir, |
| 4804 | messages=[], |
| 4805 | safeguards=FakeSafeguards(), |
| 4806 | assess_confidence=assess_confidence, |
| 4807 | verify_action=verify_action, |
| 4808 | auto_recover=False, |
| 4809 | ) |
| 4810 | queued_messages: list[str] = [] |
| 4811 | context.queue_steering_message_callback = queued_messages.append |
| 4812 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 4813 | dod = create_definition_of_done("Create an equally thorough nginx guide.") |
| 4814 | dod.implementation_plan = str(implementation_plan) |
| 4815 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 4816 | |
| 4817 | audit_bash = ToolCall( |
| 4818 | id="bash-post-build-audit", |
| 4819 | name="bash", |
| 4820 | arguments={"command": f"ls -la {guide_root}"}, |
| 4821 | ) |
| 4822 | audit_read = ToolCall( |
| 4823 | id="read-index-after-audit", |
| 4824 | name="read", |
| 4825 | arguments={"file_path": str(index_path)}, |
| 4826 | ) |
| 4827 | executor = FakeExecutor( |
| 4828 | [ |
| 4829 | tool_outcome( |
| 4830 | tool_call=audit_bash, |
| 4831 | output="total 8\n", |
| 4832 | is_error=False, |
| 4833 | ), |
| 4834 | tool_outcome( |
| 4835 | tool_call=audit_read, |
| 4836 | output=index_path.read_text(), |
| 4837 | is_error=False, |
| 4838 | ), |
| 4839 | ] |
| 4840 | ) |
| 4841 | |
| 4842 | summary = TurnSummary(final_response="") |
| 4843 | result = await runner.execute_batch( |
| 4844 | tool_calls=[audit_bash, audit_read], |
| 4845 | tool_source="assistant", |
| 4846 | pending_tool_calls_seen=set(), |
| 4847 | emit=_noop_emit, |
| 4848 | summary=summary, |
| 4849 | dod=dod, |
| 4850 | executor=executor, # type: ignore[arg-type] |
| 4851 | on_confirmation=None, |
| 4852 | on_user_question=None, |
| 4853 | emit_confirmation=None, |
| 4854 | consecutive_errors=0, |
| 4855 | ) |
| 4856 | |
| 4857 | assert result.continue_after_batch is True |
| 4858 | assert [call.id for call in executor.calls] == ["bash-post-build-audit"] |
| 4859 | assert context.workflow_mode == "verify" |
| 4860 | assert queued_messages |
| 4861 | assert "Finish with a final response now so Loader can run verification automatically." in queued_messages[-1] |
| 4862 | |
| 4863 | |
| 4864 | @pytest.mark.asyncio |
| 4865 | async def test_tool_batch_runner_preempts_post_build_observation_batch_during_consistency_review( |
| 4866 | temp_dir: Path, |
| 4867 | ) -> None: |
| 4868 | async def assess_confidence( |
| 4869 | tool_name: str, |
| 4870 | tool_args: dict, |
| 4871 | context: str, |
| 4872 | ) -> ConfidenceAssessment: |
| 4873 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 4874 | |
| 4875 | async def verify_action( |
| 4876 | tool_name: str, |
| 4877 | tool_args: dict, |
| 4878 | result: str, |
| 4879 | expected: str = "", |
| 4880 | ) -> ActionVerification: |
| 4881 | raise AssertionError("Verification should not run for this scenario") |
| 4882 | |
| 4883 | guide_root = temp_dir / "guides" / "nginx" |
| 4884 | chapters = guide_root / "chapters" |
| 4885 | guide_root.mkdir(parents=True) |
| 4886 | chapters.mkdir() |
| 4887 | index_path = guide_root / "index.html" |
| 4888 | chapter_one = chapters / "01-introduction.html" |
| 4889 | chapter_two = chapters / "02-installation.html" |
| 4890 | chapter_three = chapters / "03-basic-configuration.html" |
| 4891 | index_path.write_text("<html></html>\n") |
| 4892 | chapter_one.write_text("<html></html>\n") |
| 4893 | chapter_two.write_text("<html></html>\n") |
| 4894 | chapter_three.write_text("<html></html>\n") |
| 4895 | |
| 4896 | implementation_plan = temp_dir / "implementation.md" |
| 4897 | implementation_plan.write_text( |
| 4898 | "\n".join( |
| 4899 | [ |
| 4900 | "# Implementation Plan", |
| 4901 | "", |
| 4902 | "## File Changes", |
| 4903 | f"- `{guide_root}/`", |
| 4904 | f"- `{chapters}/`", |
| 4905 | f"- `{index_path}`", |
| 4906 | "", |
| 4907 | ] |
| 4908 | ) |
| 4909 | ) |
| 4910 | |
| 4911 | context = build_context( |
| 4912 | temp_dir=temp_dir, |
| 4913 | messages=[], |
| 4914 | safeguards=FakeSafeguards(), |
| 4915 | assess_confidence=assess_confidence, |
| 4916 | verify_action=verify_action, |
| 4917 | auto_recover=False, |
| 4918 | ) |
| 4919 | queued_messages: list[str] = [] |
| 4920 | queued_ephemeral: list[str] = [] |
| 4921 | context.queue_steering_message_callback = queued_messages.append |
| 4922 | context.queue_ephemeral_steering_message_callback = queued_ephemeral.append |
| 4923 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 4924 | dod = create_definition_of_done("Create an equally thorough nginx guide.") |
| 4925 | dod.implementation_plan = str(implementation_plan) |
| 4926 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 4927 | sync_todos_to_definition_of_done( |
| 4928 | dod, |
| 4929 | [ |
| 4930 | { |
| 4931 | "content": "Review the generated guide for consistency and completeness", |
| 4932 | "active_form": "Reviewing the generated guide for consistency and completeness", |
| 4933 | "status": "pending", |
| 4934 | } |
| 4935 | ], |
| 4936 | project_root=temp_dir, |
| 4937 | ) |
| 4938 | |
| 4939 | audit_read = ToolCall( |
| 4940 | id="read-index-during-review", |
| 4941 | name="read", |
| 4942 | arguments={"file_path": str(index_path)}, |
| 4943 | ) |
| 4944 | second_read = ToolCall( |
| 4945 | id="read-chapter-after-review", |
| 4946 | name="read", |
| 4947 | arguments={"file_path": str(chapter_one)}, |
| 4948 | ) |
| 4949 | executor = FakeExecutor( |
| 4950 | [ |
| 4951 | tool_outcome( |
| 4952 | tool_call=audit_read, |
| 4953 | output=index_path.read_text(), |
| 4954 | is_error=False, |
| 4955 | ), |
| 4956 | tool_outcome( |
| 4957 | tool_call=second_read, |
| 4958 | output=chapter_one.read_text(), |
| 4959 | is_error=False, |
| 4960 | ), |
| 4961 | ] |
| 4962 | ) |
| 4963 | |
| 4964 | summary = TurnSummary(final_response="") |
| 4965 | result = await runner.execute_batch( |
| 4966 | tool_calls=[audit_read, second_read], |
| 4967 | tool_source="assistant", |
| 4968 | pending_tool_calls_seen=set(), |
| 4969 | emit=_noop_emit, |
| 4970 | summary=summary, |
| 4971 | dod=dod, |
| 4972 | executor=executor, # type: ignore[arg-type] |
| 4973 | on_confirmation=None, |
| 4974 | on_user_question=None, |
| 4975 | emit_confirmation=None, |
| 4976 | consecutive_errors=0, |
| 4977 | ) |
| 4978 | |
| 4979 | assert result.continue_after_batch is True |
| 4980 | assert [call.id for call in executor.calls] == ["read-index-during-review"] |
| 4981 | queued = queued_ephemeral or queued_messages |
| 4982 | assert queued |
| 4983 | assert "All explicitly planned artifacts already exist." in queued[-1] |
| 4984 | assert "generated files" in queued[-1] |
| 4985 | |
| 4986 | |
| 4987 | @pytest.mark.asyncio |
| 4988 | async def test_tool_batch_runner_skips_post_build_user_question_during_consistency_review( |
| 4989 | temp_dir: Path, |
| 4990 | ) -> None: |
| 4991 | async def assess_confidence( |
| 4992 | tool_name: str, |
| 4993 | tool_args: dict, |
| 4994 | context: str, |
| 4995 | ) -> ConfidenceAssessment: |
| 4996 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 4997 | |
| 4998 | async def verify_action( |
| 4999 | tool_name: str, |
| 5000 | tool_args: dict, |
| 5001 | result: str, |
| 5002 | expected: str = "", |
| 5003 | ) -> ActionVerification: |
| 5004 | raise AssertionError("Verification should not run for this scenario") |
| 5005 | |
| 5006 | guide_root = temp_dir / "guides" / "nginx" |
| 5007 | chapters = guide_root / "chapters" |
| 5008 | guide_root.mkdir(parents=True) |
| 5009 | chapters.mkdir() |
| 5010 | index_path = guide_root / "index.html" |
| 5011 | chapter_one = chapters / "01-introduction.html" |
| 5012 | chapter_two = chapters / "02-installation.html" |
| 5013 | index_path.write_text( |
| 5014 | "\n".join( |
| 5015 | [ |
| 5016 | '<li><a href="chapters/01-introduction.html">Chapter 1: Introduction</a></li>', |
| 5017 | '<li><a href="chapters/02-installation.html">Chapter 2: Installation</a></li>', |
| 5018 | "", |
| 5019 | ] |
| 5020 | ) |
| 5021 | ) |
| 5022 | chapter_one.write_text("<html></html>\n") |
| 5023 | chapter_two.write_text("<html></html>\n") |
| 5024 | |
| 5025 | implementation_plan = temp_dir / "implementation.md" |
| 5026 | implementation_plan.write_text( |
| 5027 | "\n".join( |
| 5028 | [ |
| 5029 | "# Implementation Plan", |
| 5030 | "", |
| 5031 | "## File Changes", |
| 5032 | f"- `{guide_root}/`", |
| 5033 | f"- `{chapters}/`", |
| 5034 | f"- `{index_path}`", |
| 5035 | f"- `{chapter_one}`", |
| 5036 | f"- `{chapter_two}`", |
| 5037 | "", |
| 5038 | ] |
| 5039 | ) |
| 5040 | ) |
| 5041 | |
| 5042 | context = build_context( |
| 5043 | temp_dir=temp_dir, |
| 5044 | messages=[], |
| 5045 | safeguards=FakeSafeguards(), |
| 5046 | assess_confidence=assess_confidence, |
| 5047 | verify_action=verify_action, |
| 5048 | auto_recover=False, |
| 5049 | ) |
| 5050 | queued_messages: list[str] = [] |
| 5051 | context.queue_steering_message_callback = queued_messages.append |
| 5052 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 5053 | dod = create_definition_of_done("Create an equally thorough nginx guide.") |
| 5054 | dod.implementation_plan = str(implementation_plan) |
| 5055 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 5056 | dod.pending_items = ["Ensure all files are properly linked and formatted"] |
| 5057 | |
| 5058 | question_call = ToolCall( |
| 5059 | id="ask-post-build-review", |
| 5060 | name="AskUserQuestion", |
| 5061 | arguments={ |
| 5062 | "question": "Which specific aspects of the reference guide should I copy?", |
| 5063 | "context": "I already created the output files and want to ensure they match.", |
| 5064 | }, |
| 5065 | ) |
| 5066 | executor = FakeExecutor([]) |
| 5067 | |
| 5068 | summary = TurnSummary(final_response="") |
| 5069 | result = await runner.execute_batch( |
| 5070 | tool_calls=[question_call], |
| 5071 | tool_source="assistant", |
| 5072 | pending_tool_calls_seen=set(), |
| 5073 | emit=_noop_emit, |
| 5074 | summary=summary, |
| 5075 | dod=dod, |
| 5076 | executor=executor, # type: ignore[arg-type] |
| 5077 | on_confirmation=None, |
| 5078 | on_user_question=None, |
| 5079 | emit_confirmation=None, |
| 5080 | consecutive_errors=0, |
| 5081 | ) |
| 5082 | |
| 5083 | assert result.continue_after_batch is True |
| 5084 | assert executor.calls == [] |
| 5085 | assert queued_messages |
| 5086 | assert "The remaining work is review/verification of the generated files." in queued_messages[-1] |
| 5087 | assert "Do not ask the user for more clarification about the reference pattern now." in queued_messages[-1] |
| 5088 | assert "Finish with a final response now so Loader can run verification automatically." in queued_messages[-1] |
| 5089 | assert context.workflow_mode == "verify" |
| 5090 | assert summary.tool_result_messages |
| 5091 | assert "Skipped - stale post-build user question" in summary.tool_result_messages[-1].content |
| 5092 | |
| 5093 | |
| 5094 | @pytest.mark.asyncio |
| 5095 | async def test_tool_batch_runner_rewrites_stale_todowrite_summary_from_reconciled_dod( |
| 5096 | temp_dir: Path, |
| 5097 | ) -> None: |
| 5098 | async def assess_confidence( |
| 5099 | tool_name: str, |
| 5100 | tool_args: dict, |
| 5101 | context: str, |
| 5102 | ) -> ConfidenceAssessment: |
| 5103 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 5104 | |
| 5105 | async def verify_action( |
| 5106 | tool_name: str, |
| 5107 | tool_args: dict, |
| 5108 | result: str, |
| 5109 | expected: str = "", |
| 5110 | ) -> ActionVerification: |
| 5111 | raise AssertionError("Verification should not run for this scenario") |
| 5112 | |
| 5113 | guide_root = temp_dir / "guides" / "nginx" |
| 5114 | chapters = guide_root / "chapters" |
| 5115 | guide_root.mkdir(parents=True) |
| 5116 | chapters.mkdir() |
| 5117 | index_path = guide_root / "index.html" |
| 5118 | for name in ( |
| 5119 | "01-introduction.html", |
| 5120 | "02-installation.html", |
| 5121 | "03-basic-configuration.html", |
| 5122 | "04-advanced-usage.html", |
| 5123 | "05-troubleshooting.html", |
| 5124 | ): |
| 5125 | (chapters / name).write_text("<html></html>\n") |
| 5126 | index_path.write_text("<html></html>\n") |
| 5127 | |
| 5128 | implementation_plan = temp_dir / "implementation.md" |
| 5129 | implementation_plan.write_text( |
| 5130 | "\n".join( |
| 5131 | [ |
| 5132 | "# Implementation Plan", |
| 5133 | "", |
| 5134 | "## File Changes", |
| 5135 | f"- `{guide_root}/`", |
| 5136 | f"- `{chapters}/`", |
| 5137 | f"- `{index_path}`", |
| 5138 | "", |
| 5139 | ] |
| 5140 | ) |
| 5141 | ) |
| 5142 | |
| 5143 | context = build_context( |
| 5144 | temp_dir=temp_dir, |
| 5145 | messages=[], |
| 5146 | safeguards=FakeSafeguards(), |
| 5147 | assess_confidence=assess_confidence, |
| 5148 | verify_action=verify_action, |
| 5149 | auto_recover=False, |
| 5150 | ) |
| 5151 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 5152 | dod = create_definition_of_done("Create an equally thorough nginx guide.") |
| 5153 | dod.implementation_plan = str(implementation_plan) |
| 5154 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 5155 | |
| 5156 | tool_call = ToolCall( |
| 5157 | id="todo-stale-summary", |
| 5158 | name="TodoWrite", |
| 5159 | arguments={ |
| 5160 | "todos": [ |
| 5161 | { |
| 5162 | "content": "First, examine the existing fortran guide structure and content to understand the format", |
| 5163 | "active_form": "Working on: First, examine the existing fortran guide structure and content to understand the format", |
| 5164 | "status": "pending", |
| 5165 | } |
| 5166 | ] |
| 5167 | }, |
| 5168 | ) |
| 5169 | executor = FakeExecutor( |
| 5170 | [ |
| 5171 | tool_outcome( |
| 5172 | tool_call=tool_call, |
| 5173 | output="Todos updated", |
| 5174 | is_error=False, |
| 5175 | metadata={ |
| 5176 | "new_todos": [ |
| 5177 | { |
| 5178 | "content": "First, examine the existing fortran guide structure and content to understand the format", |
| 5179 | "active_form": "Working on: First, examine the existing fortran guide structure and content to understand the format", |
| 5180 | "status": "pending", |
| 5181 | } |
| 5182 | ] |
| 5183 | }, |
| 5184 | ) |
| 5185 | ] |
| 5186 | ) |
| 5187 | |
| 5188 | summary = TurnSummary(final_response="") |
| 5189 | result = await runner.execute_batch( |
| 5190 | tool_calls=[tool_call], |
| 5191 | tool_source="assistant", |
| 5192 | pending_tool_calls_seen=set(), |
| 5193 | emit=_noop_emit, |
| 5194 | summary=summary, |
| 5195 | dod=dod, |
| 5196 | executor=executor, # type: ignore[arg-type] |
| 5197 | on_confirmation=None, |
| 5198 | on_user_question=None, |
| 5199 | emit_confirmation=None, |
| 5200 | consecutive_errors=0, |
| 5201 | ) |
| 5202 | |
| 5203 | assert result.halted is True |
| 5204 | assert result.final_response == ( |
| 5205 | "Todo tracking is complete; running Loader verification on the generated " |
| 5206 | "files now." |
| 5207 | ) |
| 5208 | assert summary.final_response == result.final_response |
| 5209 | assert summary.tool_result_messages |
| 5210 | message = summary.tool_result_messages[-1].content |
| 5211 | assert "updated todo list" in message |
| 5212 | assert "final response should be provided next for Loader verification" in message |
| 5213 | assert "next pending:" not in message |
| 5214 | assert "fortran guide structure" not in message.lower() |
| 5215 | |
| 5216 | |
| 5217 | @pytest.mark.asyncio |
| 5218 | async def test_tool_batch_runner_todowrite_drops_unplanned_expansion_after_outputs_exist( |
| 5219 | temp_dir: Path, |
| 5220 | ) -> None: |
| 5221 | async def assess_confidence( |
| 5222 | tool_name: str, |
| 5223 | tool_args: dict, |
| 5224 | context: str, |
| 5225 | ) -> ConfidenceAssessment: |
| 5226 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 5227 | |
| 5228 | async def verify_action( |
| 5229 | tool_name: str, |
| 5230 | tool_args: dict, |
| 5231 | result: str, |
| 5232 | expected: str = "", |
| 5233 | ) -> ActionVerification: |
| 5234 | raise AssertionError("Verification should not run for this scenario") |
| 5235 | |
| 5236 | guide_root = temp_dir / "guides" / "nginx" |
| 5237 | chapters = guide_root / "chapters" |
| 5238 | guide_root.mkdir(parents=True) |
| 5239 | chapters.mkdir() |
| 5240 | index_path = guide_root / "index.html" |
| 5241 | chapter_one = chapters / "01-introduction.html" |
| 5242 | chapter_two = chapters / "02-installation.html" |
| 5243 | index_path.write_text( |
| 5244 | "\n".join( |
| 5245 | [ |
| 5246 | '<a href="chapters/01-introduction.html">Intro</a>', |
| 5247 | '<a href="chapters/02-installation.html">Install</a>', |
| 5248 | '<a href="../index.html">Back</a>', |
| 5249 | "", |
| 5250 | ] |
| 5251 | ) |
| 5252 | ) |
| 5253 | chapter_one.write_text("<html></html>\n") |
| 5254 | chapter_two.write_text("<html></html>\n") |
| 5255 | |
| 5256 | implementation_plan = temp_dir / "implementation.md" |
| 5257 | implementation_plan.write_text( |
| 5258 | "\n".join( |
| 5259 | [ |
| 5260 | "# Implementation Plan", |
| 5261 | "", |
| 5262 | "## File Changes", |
| 5263 | f"- `{guide_root}/`", |
| 5264 | f"- `{chapters}/`", |
| 5265 | f"- `{index_path}`", |
| 5266 | f"- `{chapter_one}`", |
| 5267 | f"- `{chapter_two}`", |
| 5268 | "", |
| 5269 | ] |
| 5270 | ) |
| 5271 | ) |
| 5272 | |
| 5273 | context = build_context( |
| 5274 | temp_dir=temp_dir, |
| 5275 | messages=[], |
| 5276 | safeguards=FakeSafeguards(), |
| 5277 | assess_confidence=assess_confidence, |
| 5278 | verify_action=verify_action, |
| 5279 | auto_recover=False, |
| 5280 | ) |
| 5281 | queued_messages: list[str] = [] |
| 5282 | context.queue_steering_message_callback = queued_messages.append |
| 5283 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 5284 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 5285 | dod.implementation_plan = str(implementation_plan) |
| 5286 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 5287 | |
| 5288 | tool_call = ToolCall( |
| 5289 | id="todo-post-build-expansion", |
| 5290 | name="TodoWrite", |
| 5291 | arguments={ |
| 5292 | "todos": [ |
| 5293 | { |
| 5294 | "content": "Create index.html for nginx guide", |
| 5295 | "activeForm": "Creating index.html", |
| 5296 | "status": "in_progress", |
| 5297 | }, |
| 5298 | { |
| 5299 | "content": "Create chapter 01-introduction.html", |
| 5300 | "activeForm": "Creating chapter 01-introduction.html", |
| 5301 | "status": "completed", |
| 5302 | }, |
| 5303 | { |
| 5304 | "content": "Create chapter 02-installation.html", |
| 5305 | "activeForm": "Creating chapter 02-installation.html", |
| 5306 | "status": "completed", |
| 5307 | }, |
| 5308 | { |
| 5309 | "content": "Create chapter 08-troubleshooting.html", |
| 5310 | "activeForm": "Creating chapter 08-troubleshooting.html", |
| 5311 | "status": "pending", |
| 5312 | }, |
| 5313 | ] |
| 5314 | }, |
| 5315 | ) |
| 5316 | executor = FakeExecutor( |
| 5317 | [ |
| 5318 | tool_outcome( |
| 5319 | tool_call=tool_call, |
| 5320 | output="Todos updated", |
| 5321 | is_error=False, |
| 5322 | metadata={ |
| 5323 | "new_todos": [ |
| 5324 | { |
| 5325 | "content": "Create index.html for nginx guide", |
| 5326 | "active_form": "Creating index.html", |
| 5327 | "status": "in_progress", |
| 5328 | }, |
| 5329 | { |
| 5330 | "content": "Create chapter 01-introduction.html", |
| 5331 | "active_form": "Creating chapter 01-introduction.html", |
| 5332 | "status": "completed", |
| 5333 | }, |
| 5334 | { |
| 5335 | "content": "Create chapter 02-installation.html", |
| 5336 | "active_form": "Creating chapter 02-installation.html", |
| 5337 | "status": "completed", |
| 5338 | }, |
| 5339 | { |
| 5340 | "content": "Create chapter 08-troubleshooting.html", |
| 5341 | "active_form": "Creating chapter 08-troubleshooting.html", |
| 5342 | "status": "pending", |
| 5343 | }, |
| 5344 | ] |
| 5345 | }, |
| 5346 | ) |
| 5347 | ] |
| 5348 | ) |
| 5349 | |
| 5350 | summary = TurnSummary(final_response="") |
| 5351 | await runner.execute_batch( |
| 5352 | tool_calls=[tool_call], |
| 5353 | tool_source="assistant", |
| 5354 | pending_tool_calls_seen=set(), |
| 5355 | emit=_noop_emit, |
| 5356 | summary=summary, |
| 5357 | dod=dod, |
| 5358 | executor=executor, # type: ignore[arg-type] |
| 5359 | on_confirmation=None, |
| 5360 | on_user_question=None, |
| 5361 | emit_confirmation=None, |
| 5362 | consecutive_errors=0, |
| 5363 | ) |
| 5364 | |
| 5365 | assert queued_messages |
| 5366 | message = queued_messages[-1] |
| 5367 | assert "Todo tracking is updated. All explicitly planned artifacts now exist on disk." in message |
| 5368 | assert "Finish with a final response now so Loader can run verification automatically." in message |
| 5369 | assert "Repair or verify the current files instead of expanding the artifact set." not in message |
| 5370 | assert "08-troubleshooting.html" not in message |
| 5371 | assert context.workflow_mode == "verify" |
| 5372 | |
| 5373 | |
| 5374 | @pytest.mark.asyncio |
| 5375 | async def test_tool_batch_runner_todowrite_with_existing_output_roots_requeues_next_mutation( |
| 5376 | temp_dir: Path, |
| 5377 | ) -> None: |
| 5378 | async def assess_confidence( |
| 5379 | tool_name: str, |
| 5380 | tool_args: dict, |
| 5381 | context: str, |
| 5382 | ) -> ConfidenceAssessment: |
| 5383 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 5384 | |
| 5385 | async def verify_action( |
| 5386 | tool_name: str, |
| 5387 | tool_args: dict, |
| 5388 | result: str, |
| 5389 | expected: str = "", |
| 5390 | ) -> ActionVerification: |
| 5391 | raise AssertionError("Verification should not run in this scenario") |
| 5392 | |
| 5393 | guide_root = temp_dir / "guides" / "nginx" |
| 5394 | chapters = guide_root / "chapters" |
| 5395 | guide_root.mkdir(parents=True) |
| 5396 | chapters.mkdir() |
| 5397 | index_path = guide_root / "index.html" |
| 5398 | index_path.write_text( |
| 5399 | "\n".join( |
| 5400 | [ |
| 5401 | "<!DOCTYPE html>", |
| 5402 | "<html>", |
| 5403 | "<body>", |
| 5404 | '<a href="chapters/01-introduction.html">Introduction</a>', |
| 5405 | "</body>", |
| 5406 | "</html>", |
| 5407 | "", |
| 5408 | ] |
| 5409 | ) |
| 5410 | ) |
| 5411 | |
| 5412 | implementation_plan = temp_dir / "implementation.md" |
| 5413 | implementation_plan.write_text( |
| 5414 | "\n".join( |
| 5415 | [ |
| 5416 | "# Implementation Plan", |
| 5417 | "", |
| 5418 | "## File Changes", |
| 5419 | f"- `{guide_root}/`", |
| 5420 | f"- `{chapters}/`", |
| 5421 | f"- `{index_path}`", |
| 5422 | "", |
| 5423 | ] |
| 5424 | ) |
| 5425 | ) |
| 5426 | |
| 5427 | context = build_context( |
| 5428 | temp_dir=temp_dir, |
| 5429 | messages=[], |
| 5430 | safeguards=FakeSafeguards(), |
| 5431 | assess_confidence=assess_confidence, |
| 5432 | verify_action=verify_action, |
| 5433 | auto_recover=False, |
| 5434 | ) |
| 5435 | queued_messages: list[str] = [] |
| 5436 | context.queue_steering_message_callback = queued_messages.append |
| 5437 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 5438 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 5439 | dod.implementation_plan = str(implementation_plan) |
| 5440 | dod.touched_files.append(str(index_path)) |
| 5441 | sync_todos_to_definition_of_done( |
| 5442 | dod, |
| 5443 | [ |
| 5444 | { |
| 5445 | "content": "Examine the existing Fortran guide structure", |
| 5446 | "active_form": "Examining the existing Fortran guide structure", |
| 5447 | "status": "completed", |
| 5448 | }, |
| 5449 | { |
| 5450 | "content": "Create the nginx directory structure", |
| 5451 | "active_form": "Creating the nginx directory structure", |
| 5452 | "status": "completed", |
| 5453 | }, |
| 5454 | { |
| 5455 | "content": "Write the introduction chapter", |
| 5456 | "active_form": "Writing the introduction chapter", |
| 5457 | "status": "pending", |
| 5458 | }, |
| 5459 | ], |
| 5460 | project_root=temp_dir, |
| 5461 | ) |
| 5462 | |
| 5463 | tool_call = ToolCall( |
| 5464 | id="todo-next-mutation", |
| 5465 | name="TodoWrite", |
| 5466 | arguments={ |
| 5467 | "todos": [ |
| 5468 | { |
| 5469 | "content": "Examine the existing Fortran guide structure", |
| 5470 | "active_form": "Examining the existing Fortran guide structure", |
| 5471 | "status": "completed", |
| 5472 | }, |
| 5473 | { |
| 5474 | "content": "Create the nginx directory structure", |
| 5475 | "active_form": "Creating the nginx directory structure", |
| 5476 | "status": "completed", |
| 5477 | }, |
| 5478 | { |
| 5479 | "content": "Write the introduction chapter", |
| 5480 | "active_form": "Writing the introduction chapter", |
| 5481 | "status": "pending", |
| 5482 | }, |
| 5483 | ] |
| 5484 | }, |
| 5485 | ) |
| 5486 | executor = FakeExecutor( |
| 5487 | [ |
| 5488 | tool_outcome( |
| 5489 | tool_call=tool_call, |
| 5490 | output="Todos updated", |
| 5491 | is_error=False, |
| 5492 | metadata={ |
| 5493 | "new_todos": [ |
| 5494 | { |
| 5495 | "content": "Examine the existing Fortran guide structure", |
| 5496 | "active_form": "Examining the existing Fortran guide structure", |
| 5497 | "status": "completed", |
| 5498 | }, |
| 5499 | { |
| 5500 | "content": "Create the nginx directory structure", |
| 5501 | "active_form": "Creating the nginx directory structure", |
| 5502 | "status": "completed", |
| 5503 | }, |
| 5504 | { |
| 5505 | "content": "Write the introduction chapter", |
| 5506 | "active_form": "Writing the introduction chapter", |
| 5507 | "status": "pending", |
| 5508 | }, |
| 5509 | ] |
| 5510 | }, |
| 5511 | ) |
| 5512 | ] |
| 5513 | ) |
| 5514 | |
| 5515 | summary = TurnSummary(final_response="") |
| 5516 | await runner.execute_batch( |
| 5517 | tool_calls=[tool_call], |
| 5518 | tool_source="assistant", |
| 5519 | pending_tool_calls_seen=set(), |
| 5520 | emit=_noop_emit, |
| 5521 | summary=summary, |
| 5522 | dod=dod, |
| 5523 | executor=executor, # type: ignore[arg-type] |
| 5524 | on_confirmation=None, |
| 5525 | on_user_question=None, |
| 5526 | emit_confirmation=None, |
| 5527 | consecutive_errors=0, |
| 5528 | ) |
| 5529 | |
| 5530 | assert queued_messages |
| 5531 | message = queued_messages[-1] |
| 5532 | assert "Todo tracking is updated. Next step: create `01-introduction.html`." in message |
| 5533 | assert "Prefer one `write(file_path=..., content=...)` call" in message |
| 5534 | assert "Make your next response the concrete mutation tool call itself." in message |
| 5535 | |
| 5536 | |
| 5537 | @pytest.mark.asyncio |
| 5538 | async def test_tool_batch_runner_todowrite_prefers_pending_index_over_empty_output_directory( |
| 5539 | temp_dir: Path, |
| 5540 | ) -> None: |
| 5541 | async def assess_confidence( |
| 5542 | tool_name: str, |
| 5543 | tool_args: dict, |
| 5544 | context: str, |
| 5545 | ) -> ConfidenceAssessment: |
| 5546 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 5547 | |
| 5548 | async def verify_action( |
| 5549 | tool_name: str, |
| 5550 | tool_args: dict, |
| 5551 | result: str, |
| 5552 | expected: str = "", |
| 5553 | ) -> ActionVerification: |
| 5554 | raise AssertionError("Verification should not run in this scenario") |
| 5555 | |
| 5556 | guide_root = temp_dir / "Loader" / "guides" / "nginx" |
| 5557 | chapters = guide_root / "chapters" |
| 5558 | chapters.mkdir(parents=True) |
| 5559 | index_path = guide_root / "index.html" |
| 5560 | implementation_plan = temp_dir / "implementation.md" |
| 5561 | implementation_plan.write_text( |
| 5562 | "\n".join( |
| 5563 | [ |
| 5564 | "# Implementation Plan", |
| 5565 | "", |
| 5566 | "## File Changes", |
| 5567 | f"- `{chapters}/`", |
| 5568 | f"- `{index_path}`", |
| 5569 | "", |
| 5570 | ] |
| 5571 | ) |
| 5572 | ) |
| 5573 | |
| 5574 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 5575 | dod.implementation_plan = str(implementation_plan) |
| 5576 | sync_todos_to_definition_of_done( |
| 5577 | dod, |
| 5578 | [ |
| 5579 | { |
| 5580 | "content": "Examine the existing Fortran guide structure to understand the format and depth", |
| 5581 | "active_form": "Examining the existing Fortran guide structure", |
| 5582 | "status": "completed", |
| 5583 | }, |
| 5584 | { |
| 5585 | "content": "Create the new nginx guide directory structure", |
| 5586 | "active_form": "Creating the new nginx guide directory structure", |
| 5587 | "status": "completed", |
| 5588 | }, |
| 5589 | { |
| 5590 | "content": "Create a new index.html for the nginx guide", |
| 5591 | "active_form": "Creating a new index.html for the nginx guide", |
| 5592 | "status": "pending", |
| 5593 | }, |
| 5594 | { |
| 5595 | "content": "Create the first chapter for the nginx guide", |
| 5596 | "active_form": "Creating the first chapter for the nginx guide", |
| 5597 | "status": "pending", |
| 5598 | }, |
| 5599 | ], |
| 5600 | project_root=temp_dir, |
| 5601 | ) |
| 5602 | |
| 5603 | queued_messages: list[str] = [] |
| 5604 | context = build_context( |
| 5605 | temp_dir=temp_dir, |
| 5606 | messages=[], |
| 5607 | safeguards=FakeSafeguards(), |
| 5608 | assess_confidence=assess_confidence, |
| 5609 | verify_action=verify_action, |
| 5610 | auto_recover=False, |
| 5611 | ) |
| 5612 | context.queue_steering_message_callback = queued_messages.append |
| 5613 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 5614 | |
| 5615 | todos = [ |
| 5616 | { |
| 5617 | "content": "Examine the existing Fortran guide structure to understand the format and depth", |
| 5618 | "active_form": "Examining the existing Fortran guide structure", |
| 5619 | "status": "completed", |
| 5620 | }, |
| 5621 | { |
| 5622 | "content": "Create the new nginx guide directory structure", |
| 5623 | "active_form": "Creating the new nginx guide directory structure", |
| 5624 | "status": "completed", |
| 5625 | }, |
| 5626 | { |
| 5627 | "content": "Create a new index.html for the nginx guide", |
| 5628 | "active_form": "Creating a new index.html for the nginx guide", |
| 5629 | "status": "pending", |
| 5630 | }, |
| 5631 | { |
| 5632 | "content": "Create the first chapter for the nginx guide", |
| 5633 | "active_form": "Creating the first chapter for the nginx guide", |
| 5634 | "status": "pending", |
| 5635 | }, |
| 5636 | ] |
| 5637 | tool_call = ToolCall( |
| 5638 | id="todo-index-before-chapter", |
| 5639 | name="TodoWrite", |
| 5640 | arguments={"todos": todos}, |
| 5641 | ) |
| 5642 | executor = FakeExecutor( |
| 5643 | [ |
| 5644 | tool_outcome( |
| 5645 | tool_call=tool_call, |
| 5646 | output="Todos updated", |
| 5647 | is_error=False, |
| 5648 | metadata={"new_todos": todos}, |
| 5649 | ) |
| 5650 | ] |
| 5651 | ) |
| 5652 | |
| 5653 | summary = TurnSummary(final_response="") |
| 5654 | await runner.execute_batch( |
| 5655 | tool_calls=[tool_call], |
| 5656 | tool_source="assistant", |
| 5657 | pending_tool_calls_seen=set(), |
| 5658 | emit=_noop_emit, |
| 5659 | summary=summary, |
| 5660 | dod=dod, |
| 5661 | executor=executor, # type: ignore[arg-type] |
| 5662 | on_confirmation=None, |
| 5663 | on_user_question=None, |
| 5664 | emit_confirmation=None, |
| 5665 | consecutive_errors=0, |
| 5666 | ) |
| 5667 | |
| 5668 | assert queued_messages |
| 5669 | message = queued_messages[-1] |
| 5670 | assert "Todo tracking is updated. Next step: create `index.html`." in message |
| 5671 | assert f"Prefer one `write(file_path=..., content=...)` call for `{index_path.resolve(strict=False)}`" in message |
| 5672 | assert "01-introduction.html" not in message |
| 5673 | |
| 5674 | |
| 5675 | @pytest.mark.asyncio |
| 5676 | async def test_tool_batch_runner_todowrite_with_declared_child_targets_names_next_missing_file( |
| 5677 | temp_dir: Path, |
| 5678 | ) -> None: |
| 5679 | async def assess_confidence( |
| 5680 | tool_name: str, |
| 5681 | tool_args: dict, |
| 5682 | context: str, |
| 5683 | ) -> ConfidenceAssessment: |
| 5684 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 5685 | |
| 5686 | async def verify_action( |
| 5687 | tool_name: str, |
| 5688 | tool_args: dict, |
| 5689 | result: str, |
| 5690 | expected: str = "", |
| 5691 | ) -> ActionVerification: |
| 5692 | raise AssertionError("Verification should not run in this scenario") |
| 5693 | |
| 5694 | guide_root = temp_dir / "guides" / "nginx" |
| 5695 | chapters = guide_root / "chapters" |
| 5696 | guide_root.mkdir(parents=True) |
| 5697 | chapters.mkdir() |
| 5698 | index_path = guide_root / "index.html" |
| 5699 | index_path.write_text( |
| 5700 | "\n".join( |
| 5701 | [ |
| 5702 | "<html>", |
| 5703 | '<a href="chapters/introduction.html">Introduction</a>', |
| 5704 | '<a href="chapters/installation.html">Installation</a>', |
| 5705 | "</html>", |
| 5706 | ] |
| 5707 | ) |
| 5708 | + "\n" |
| 5709 | ) |
| 5710 | |
| 5711 | implementation_plan = temp_dir / "implementation.md" |
| 5712 | implementation_plan.write_text( |
| 5713 | "\n".join( |
| 5714 | [ |
| 5715 | "# Implementation Plan", |
| 5716 | "", |
| 5717 | "## File Changes", |
| 5718 | f"- `{guide_root}/`", |
| 5719 | f"- `{chapters}/`", |
| 5720 | f"- `{index_path}`", |
| 5721 | "", |
| 5722 | ] |
| 5723 | ) |
| 5724 | ) |
| 5725 | |
| 5726 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 5727 | dod.implementation_plan = str(implementation_plan) |
| 5728 | dod.pending_items = [ |
| 5729 | "Write the introduction chapter", |
| 5730 | "Complete the requested work", |
| 5731 | ] |
| 5732 | dod.touched_files.append(str(index_path)) |
| 5733 | |
| 5734 | queued_messages: list[str] = [] |
| 5735 | context = build_context( |
| 5736 | temp_dir=temp_dir, |
| 5737 | messages=[], |
| 5738 | safeguards=FakeSafeguards(), |
| 5739 | assess_confidence=assess_confidence, |
| 5740 | verify_action=verify_action, |
| 5741 | auto_recover=False, |
| 5742 | ) |
| 5743 | context.queue_steering_message_callback = queued_messages.append |
| 5744 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 5745 | |
| 5746 | tool_call = ToolCall( |
| 5747 | id="todo-1", |
| 5748 | name="TodoWrite", |
| 5749 | arguments={ |
| 5750 | "todos": [ |
| 5751 | { |
| 5752 | "content": "Write the introduction chapter", |
| 5753 | "activeForm": "Writing the introduction chapter", |
| 5754 | "status": "pending", |
| 5755 | } |
| 5756 | ] |
| 5757 | }, |
| 5758 | ) |
| 5759 | executor = FakeExecutor( |
| 5760 | [ |
| 5761 | tool_outcome( |
| 5762 | tool_call=tool_call, |
| 5763 | output="Todos updated", |
| 5764 | is_error=False, |
| 5765 | metadata={ |
| 5766 | "new_todos": [ |
| 5767 | { |
| 5768 | "content": "Write the introduction chapter", |
| 5769 | "active_form": "Writing the introduction chapter", |
| 5770 | "status": "pending", |
| 5771 | } |
| 5772 | ] |
| 5773 | }, |
| 5774 | ) |
| 5775 | ] |
| 5776 | ) |
| 5777 | |
| 5778 | summary = TurnSummary(final_response="") |
| 5779 | await runner.execute_batch( |
| 5780 | tool_calls=[tool_call], |
| 5781 | tool_source="assistant", |
| 5782 | pending_tool_calls_seen=set(), |
| 5783 | emit=_noop_emit, |
| 5784 | summary=summary, |
| 5785 | dod=dod, |
| 5786 | executor=executor, # type: ignore[arg-type] |
| 5787 | on_confirmation=None, |
| 5788 | on_user_question=None, |
| 5789 | emit_confirmation=None, |
| 5790 | consecutive_errors=0, |
| 5791 | ) |
| 5792 | |
| 5793 | assert queued_messages |
| 5794 | message = queued_messages[-1] |
| 5795 | assert "Todo tracking is updated. Next step: create `introduction.html`." in message |
| 5796 | assert "Prefer one `write(file_path=..., content=...)` call" in message |
| 5797 | assert "Make your next response the concrete mutation tool call itself." in message |
| 5798 | |
| 5799 | |
| 5800 | @pytest.mark.asyncio |
| 5801 | async def test_tool_batch_runner_todowrite_names_concrete_pending_file_after_artifacts_exist( |
| 5802 | temp_dir: Path, |
| 5803 | ) -> None: |
| 5804 | async def assess_confidence( |
| 5805 | tool_name: str, |
| 5806 | tool_args: dict, |
| 5807 | context: str, |
| 5808 | ) -> ConfidenceAssessment: |
| 5809 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 5810 | |
| 5811 | async def verify_action( |
| 5812 | tool_name: str, |
| 5813 | tool_args: dict, |
| 5814 | result: str, |
| 5815 | expected: str = "", |
| 5816 | ) -> ActionVerification: |
| 5817 | raise AssertionError("Verification should not run in this scenario") |
| 5818 | |
| 5819 | guide_root = temp_dir / "guides" / "nginx" |
| 5820 | chapters = guide_root / "chapters" |
| 5821 | guide_root.mkdir(parents=True) |
| 5822 | chapters.mkdir() |
| 5823 | index_path = guide_root / "index.html" |
| 5824 | chapter_one = chapters / "01-introduction.html" |
| 5825 | index_path.write_text( |
| 5826 | "\n".join( |
| 5827 | [ |
| 5828 | "<html>", |
| 5829 | '<a href="chapters/01-introduction.html">Chapter 1: Introduction to NGINX Tool</a>', |
| 5830 | '<a href="chapters/02-installation.html">Chapter 2: Installation and Setup</a>', |
| 5831 | "</html>", |
| 5832 | ] |
| 5833 | ) |
| 5834 | + "\n" |
| 5835 | ) |
| 5836 | chapter_one.write_text("<html></html>\n") |
| 5837 | |
| 5838 | implementation_plan = temp_dir / "implementation.md" |
| 5839 | implementation_plan.write_text( |
| 5840 | "\n".join( |
| 5841 | [ |
| 5842 | "# Implementation Plan", |
| 5843 | "", |
| 5844 | "## File Changes", |
| 5845 | f"- `{guide_root}/`", |
| 5846 | f"- `{chapters}/`", |
| 5847 | f"- `{index_path}`", |
| 5848 | "", |
| 5849 | ] |
| 5850 | ) |
| 5851 | ) |
| 5852 | |
| 5853 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 5854 | dod.implementation_plan = str(implementation_plan) |
| 5855 | dod.pending_items = [ |
| 5856 | "Creating Chapter 2: Installation and Setup", |
| 5857 | "Complete the requested work", |
| 5858 | ] |
| 5859 | dod.touched_files.extend([str(index_path), str(chapter_one)]) |
| 5860 | |
| 5861 | queued_messages: list[str] = [] |
| 5862 | context = build_context( |
| 5863 | temp_dir=temp_dir, |
| 5864 | messages=[], |
| 5865 | safeguards=FakeSafeguards(), |
| 5866 | assess_confidence=assess_confidence, |
| 5867 | verify_action=verify_action, |
| 5868 | auto_recover=False, |
| 5869 | ) |
| 5870 | context.queue_steering_message_callback = queued_messages.append |
| 5871 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 5872 | |
| 5873 | tool_call = ToolCall( |
| 5874 | id="todo-1", |
| 5875 | name="TodoWrite", |
| 5876 | arguments={ |
| 5877 | "todos": [ |
| 5878 | { |
| 5879 | "content": "Creating Chapter 2: Installation and Setup", |
| 5880 | "activeForm": "Creating Chapter 2: Installation and Setup", |
| 5881 | "status": "pending", |
| 5882 | } |
| 5883 | ] |
| 5884 | }, |
| 5885 | ) |
| 5886 | executor = FakeExecutor( |
| 5887 | [ |
| 5888 | tool_outcome( |
| 5889 | tool_call=tool_call, |
| 5890 | output="Todos updated", |
| 5891 | is_error=False, |
| 5892 | metadata={ |
| 5893 | "new_todos": [ |
| 5894 | { |
| 5895 | "content": "Creating Chapter 2: Installation and Setup", |
| 5896 | "active_form": "Creating Chapter 2: Installation and Setup", |
| 5897 | "status": "pending", |
| 5898 | } |
| 5899 | ] |
| 5900 | }, |
| 5901 | ) |
| 5902 | ] |
| 5903 | ) |
| 5904 | |
| 5905 | summary = TurnSummary(final_response="") |
| 5906 | await runner.execute_batch( |
| 5907 | tool_calls=[tool_call], |
| 5908 | tool_source="assistant", |
| 5909 | pending_tool_calls_seen=set(), |
| 5910 | emit=_noop_emit, |
| 5911 | summary=summary, |
| 5912 | dod=dod, |
| 5913 | executor=executor, # type: ignore[arg-type] |
| 5914 | on_confirmation=None, |
| 5915 | on_user_question=None, |
| 5916 | emit_confirmation=None, |
| 5917 | consecutive_errors=0, |
| 5918 | ) |
| 5919 | |
| 5920 | assert queued_messages |
| 5921 | message = queued_messages[-1] |
| 5922 | assert "Todo tracking is updated. Next step: create `02-installation.html`." in message |
| 5923 | assert "Prefer one `write(file_path=..., content=...)` call" in message |
| 5924 | assert "Make your next response the concrete mutation tool call itself" in message |
| 5925 | |
| 5926 | |
| 5927 | @pytest.mark.asyncio |
| 5928 | async def test_tool_batch_runner_todowrite_uses_observed_sibling_pattern_for_next_file( |
| 5929 | temp_dir: Path, |
| 5930 | ) -> None: |
| 5931 | async def assess_confidence( |
| 5932 | tool_name: str, |
| 5933 | tool_args: dict, |
| 5934 | context: str, |
| 5935 | ) -> ConfidenceAssessment: |
| 5936 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 5937 | |
| 5938 | async def verify_action( |
| 5939 | tool_name: str, |
| 5940 | tool_args: dict, |
| 5941 | result: str, |
| 5942 | expected: str = "", |
| 5943 | ) -> ActionVerification: |
| 5944 | raise AssertionError("Verification should not run in this scenario") |
| 5945 | |
| 5946 | reference_chapters = temp_dir / "fortran" / "chapters" |
| 5947 | reference_chapters.mkdir(parents=True) |
| 5948 | (reference_chapters / "01-introduction.html").write_text("<h1>Introduction</h1>\n") |
| 5949 | |
| 5950 | guide_root = temp_dir / "guides" / "nginx" |
| 5951 | chapters = guide_root / "chapters" |
| 5952 | guide_root.mkdir(parents=True) |
| 5953 | chapters.mkdir() |
| 5954 | index_path = guide_root / "index.html" |
| 5955 | index_path.write_text("<html></html>\n") |
| 5956 | |
| 5957 | implementation_plan = temp_dir / "implementation.md" |
| 5958 | implementation_plan.write_text( |
| 5959 | "\n".join( |
| 5960 | [ |
| 5961 | "# Implementation Plan", |
| 5962 | "", |
| 5963 | "## File Changes", |
| 5964 | f"- `{guide_root}/`", |
| 5965 | f"- `{chapters}/`", |
| 5966 | f"- `{index_path}`", |
| 5967 | "", |
| 5968 | ] |
| 5969 | ) |
| 5970 | ) |
| 5971 | |
| 5972 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 5973 | dod.implementation_plan = str(implementation_plan) |
| 5974 | dod.pending_items = [ |
| 5975 | "Write the introduction chapter", |
| 5976 | "Complete the requested work", |
| 5977 | ] |
| 5978 | dod.touched_files.append(str(index_path)) |
| 5979 | |
| 5980 | queued_messages: list[str] = [] |
| 5981 | context = build_context( |
| 5982 | temp_dir=temp_dir, |
| 5983 | messages=[ |
| 5984 | Message( |
| 5985 | role=Role.ASSISTANT, |
| 5986 | content="", |
| 5987 | tool_calls=[ |
| 5988 | ToolCall( |
| 5989 | id="read-ref-1", |
| 5990 | name="read", |
| 5991 | arguments={"file_path": str(reference_chapters / "01-introduction.html")}, |
| 5992 | ) |
| 5993 | ], |
| 5994 | ) |
| 5995 | ], |
| 5996 | safeguards=FakeSafeguards(), |
| 5997 | assess_confidence=assess_confidence, |
| 5998 | verify_action=verify_action, |
| 5999 | auto_recover=False, |
| 6000 | ) |
| 6001 | context.queue_steering_message_callback = queued_messages.append |
| 6002 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 6003 | |
| 6004 | tool_call = ToolCall( |
| 6005 | id="todo-observed-1", |
| 6006 | name="TodoWrite", |
| 6007 | arguments={ |
| 6008 | "todos": [ |
| 6009 | { |
| 6010 | "content": "Write the introduction chapter", |
| 6011 | "activeForm": "Writing the introduction chapter", |
| 6012 | "status": "pending", |
| 6013 | } |
| 6014 | ] |
| 6015 | }, |
| 6016 | ) |
| 6017 | executor = FakeExecutor( |
| 6018 | [ |
| 6019 | tool_outcome( |
| 6020 | tool_call=tool_call, |
| 6021 | output="Todos updated", |
| 6022 | is_error=False, |
| 6023 | metadata={ |
| 6024 | "new_todos": [ |
| 6025 | { |
| 6026 | "content": "Write the introduction chapter", |
| 6027 | "active_form": "Writing the introduction chapter", |
| 6028 | "status": "pending", |
| 6029 | } |
| 6030 | ] |
| 6031 | }, |
| 6032 | ) |
| 6033 | ] |
| 6034 | ) |
| 6035 | |
| 6036 | summary = TurnSummary(final_response="") |
| 6037 | await runner.execute_batch( |
| 6038 | tool_calls=[tool_call], |
| 6039 | tool_source="assistant", |
| 6040 | pending_tool_calls_seen=set(), |
| 6041 | emit=_noop_emit, |
| 6042 | summary=summary, |
| 6043 | dod=dod, |
| 6044 | executor=executor, # type: ignore[arg-type] |
| 6045 | on_confirmation=None, |
| 6046 | on_user_question=None, |
| 6047 | emit_confirmation=None, |
| 6048 | consecutive_errors=0, |
| 6049 | ) |
| 6050 | |
| 6051 | assert queued_messages |
| 6052 | message = queued_messages[-1] |
| 6053 | assert "Todo tracking is updated. Next step: create `01-introduction.html`." in message |
| 6054 | assert "Prefer one `write(file_path=..., content=...)` call" in message |
| 6055 | |
| 6056 | |
| 6057 | @pytest.mark.asyncio |
| 6058 | async def test_tool_batch_runner_bookkeeping_note_with_missing_artifact_requeues_resume_step( |
| 6059 | temp_dir: Path, |
| 6060 | ) -> None: |
| 6061 | async def assess_confidence( |
| 6062 | tool_name: str, |
| 6063 | tool_args: dict, |
| 6064 | context: str, |
| 6065 | ) -> ConfidenceAssessment: |
| 6066 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 6067 | |
| 6068 | async def verify_action( |
| 6069 | tool_name: str, |
| 6070 | tool_args: dict, |
| 6071 | result: str, |
| 6072 | expected: str = "", |
| 6073 | ) -> ActionVerification: |
| 6074 | raise AssertionError("Verification should not run in this scenario") |
| 6075 | |
| 6076 | guide_root = temp_dir / "guides" / "nginx" |
| 6077 | chapters = guide_root / "chapters" |
| 6078 | guide_root.mkdir(parents=True) |
| 6079 | chapters.mkdir() |
| 6080 | index_path = guide_root / "index.html" |
| 6081 | chapter_one = chapters / "01-getting-started.html" |
| 6082 | chapter_two = chapters / "02-installation.html" |
| 6083 | index_path.write_text("<html></html>\n") |
| 6084 | chapter_one.write_text("<h1>One</h1>\n") |
| 6085 | |
| 6086 | implementation_plan = temp_dir / "implementation.md" |
| 6087 | implementation_plan.write_text( |
| 6088 | "\n".join( |
| 6089 | [ |
| 6090 | "# Implementation Plan", |
| 6091 | "", |
| 6092 | "## File Changes", |
| 6093 | f"- `{guide_root}/`", |
| 6094 | f"- `{chapters}/`", |
| 6095 | f"- `{index_path}`", |
| 6096 | f"- `{chapter_one}`", |
| 6097 | f"- `{chapter_two}`", |
| 6098 | "", |
| 6099 | ] |
| 6100 | ) |
| 6101 | ) |
| 6102 | |
| 6103 | context = build_context( |
| 6104 | temp_dir=temp_dir, |
| 6105 | messages=[], |
| 6106 | safeguards=FakeSafeguards(), |
| 6107 | assess_confidence=assess_confidence, |
| 6108 | verify_action=verify_action, |
| 6109 | auto_recover=False, |
| 6110 | ) |
| 6111 | queued_messages: list[str] = [] |
| 6112 | context.queue_steering_message_callback = queued_messages.append |
| 6113 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 6114 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 6115 | dod.implementation_plan = str(implementation_plan) |
| 6116 | sync_todos_to_definition_of_done( |
| 6117 | dod, |
| 6118 | [ |
| 6119 | { |
| 6120 | "content": "Create 01-getting-started.html", |
| 6121 | "active_form": "Creating 01-getting-started.html", |
| 6122 | "status": "completed", |
| 6123 | }, |
| 6124 | { |
| 6125 | "content": "Create 02-installation.html", |
| 6126 | "active_form": "Creating 02-installation.html", |
| 6127 | "status": "pending", |
| 6128 | }, |
| 6129 | ], |
| 6130 | project_root=temp_dir, |
| 6131 | ) |
| 6132 | dod.touched_files.extend([str(index_path), str(chapter_one)]) |
| 6133 | |
| 6134 | tool_call = ToolCall( |
| 6135 | id="working-note", |
| 6136 | name="notepad_write_working", |
| 6137 | arguments={"content": "Creating the second chapter file: Installation"}, |
| 6138 | ) |
| 6139 | executor = FakeExecutor( |
| 6140 | [ |
| 6141 | tool_outcome( |
| 6142 | tool_call=tool_call, |
| 6143 | output="Working note recorded", |
| 6144 | is_error=False, |
| 6145 | ) |
| 6146 | ] |
| 6147 | ) |
| 6148 | |
| 6149 | summary = TurnSummary(final_response="") |
| 6150 | await runner.execute_batch( |
| 6151 | tool_calls=[tool_call], |
| 6152 | tool_source="assistant", |
| 6153 | pending_tool_calls_seen=set(), |
| 6154 | emit=_noop_emit, |
| 6155 | summary=summary, |
| 6156 | dod=dod, |
| 6157 | executor=executor, # type: ignore[arg-type] |
| 6158 | on_confirmation=None, |
| 6159 | on_user_question=None, |
| 6160 | emit_confirmation=None, |
| 6161 | consecutive_errors=0, |
| 6162 | ) |
| 6163 | |
| 6164 | assert queued_messages |
| 6165 | message = queued_messages[-1] |
| 6166 | assert "Bookkeeping note is recorded. A declared output artifact is still missing." in message |
| 6167 | assert "Resume by creating `02-installation.html` now." in message |
| 6168 | assert "Make your next response the concrete mutation tool call itself" in message |
| 6169 | assert "refresh `TodoWrite`" in message |
| 6170 | assert "Do not spend the next turn on additional notes, rediscovery, verification, or final confirmation" in message |
| 6171 | |
| 6172 | |
| 6173 | @pytest.mark.asyncio |
| 6174 | async def test_tool_batch_runner_working_note_respects_discovery_first_pending_step( |
| 6175 | temp_dir: Path, |
| 6176 | ) -> None: |
| 6177 | async def assess_confidence( |
| 6178 | tool_name: str, |
| 6179 | tool_args: dict, |
| 6180 | context: str, |
| 6181 | ) -> ConfidenceAssessment: |
| 6182 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 6183 | |
| 6184 | async def verify_action( |
| 6185 | tool_name: str, |
| 6186 | tool_args: dict, |
| 6187 | result: str, |
| 6188 | expected: str = "", |
| 6189 | ) -> ActionVerification: |
| 6190 | raise AssertionError("Verification should not run in this scenario") |
| 6191 | |
| 6192 | implementation_plan = temp_dir / "implementation.md" |
| 6193 | implementation_plan.write_text( |
| 6194 | "\n".join( |
| 6195 | [ |
| 6196 | "# Implementation Plan", |
| 6197 | "", |
| 6198 | "## File Changes", |
| 6199 | f"- `{temp_dir / 'guides' / 'nginx' / 'index.html'}`", |
| 6200 | f"- `{temp_dir / 'guides' / 'nginx' / 'chapters'}`", |
| 6201 | "", |
| 6202 | ] |
| 6203 | ) |
| 6204 | ) |
| 6205 | |
| 6206 | context = build_context( |
| 6207 | temp_dir=temp_dir, |
| 6208 | messages=[], |
| 6209 | safeguards=FakeSafeguards(), |
| 6210 | assess_confidence=assess_confidence, |
| 6211 | verify_action=verify_action, |
| 6212 | auto_recover=False, |
| 6213 | ) |
| 6214 | queued_messages: list[str] = [] |
| 6215 | context.queue_steering_message_callback = queued_messages.append |
| 6216 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 6217 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 6218 | dod.implementation_plan = str(implementation_plan) |
| 6219 | dod.pending_items.extend( |
| 6220 | [ |
| 6221 | "First, examine the existing fortran guide structure and content to understand the format", |
| 6222 | "Create the nginx directory structure", |
| 6223 | "Develop the main index.html file for the nginx guide", |
| 6224 | ] |
| 6225 | ) |
| 6226 | |
| 6227 | tool_call = ToolCall( |
| 6228 | id="working-note", |
| 6229 | name="notepad_write_working", |
| 6230 | arguments={"content": "Analyzing the fortran guide structure before creating nginx guide"}, |
| 6231 | ) |
| 6232 | executor = FakeExecutor( |
| 6233 | [ |
| 6234 | tool_outcome( |
| 6235 | tool_call=tool_call, |
| 6236 | output="Working note recorded", |
| 6237 | is_error=False, |
| 6238 | ) |
| 6239 | ] |
| 6240 | ) |
| 6241 | |
| 6242 | summary = TurnSummary(final_response="") |
| 6243 | await runner.execute_batch( |
| 6244 | tool_calls=[tool_call], |
| 6245 | tool_source="assistant", |
| 6246 | pending_tool_calls_seen=set(), |
| 6247 | emit=_noop_emit, |
| 6248 | summary=summary, |
| 6249 | dod=dod, |
| 6250 | executor=executor, # type: ignore[arg-type] |
| 6251 | on_confirmation=None, |
| 6252 | on_user_question=None, |
| 6253 | emit_confirmation=None, |
| 6254 | consecutive_errors=0, |
| 6255 | ) |
| 6256 | |
| 6257 | assert queued_messages |
| 6258 | message = queued_messages[-1] |
| 6259 | assert ( |
| 6260 | "Continue with the next pending item: `First, examine the existing fortran guide structure and content to understand the format`." |
| 6261 | in message |
| 6262 | ) |
| 6263 | assert "one concrete evidence-gathering tool call" in message |
| 6264 | assert "Resume by creating `index.html` now." not in message |
| 6265 | |
| 6266 | |
| 6267 | @pytest.mark.asyncio |
| 6268 | async def test_tool_batch_runner_working_note_prefers_declared_output_gap_over_stale_discovery( |
| 6269 | temp_dir: Path, |
| 6270 | ) -> None: |
| 6271 | async def assess_confidence( |
| 6272 | tool_name: str, |
| 6273 | tool_args: dict, |
| 6274 | context: str, |
| 6275 | ) -> ConfidenceAssessment: |
| 6276 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 6277 | |
| 6278 | async def verify_action( |
| 6279 | tool_name: str, |
| 6280 | tool_args: dict, |
| 6281 | result: str, |
| 6282 | expected: str = "", |
| 6283 | ) -> ActionVerification: |
| 6284 | raise AssertionError("Verification should not run in this scenario") |
| 6285 | |
| 6286 | guide_root = temp_dir / "guides" / "nginx" |
| 6287 | chapters_dir = guide_root / "chapters" |
| 6288 | chapters_dir.mkdir(parents=True) |
| 6289 | index_path = guide_root / "index.html" |
| 6290 | first_chapter = chapters_dir / "01-introduction.html" |
| 6291 | index_path.write_text( |
| 6292 | "\n".join( |
| 6293 | [ |
| 6294 | '<a href="chapters/01-introduction.html">Introduction</a>', |
| 6295 | '<a href="chapters/02-installation.html">Installation</a>', |
| 6296 | '<a href="chapters/03-configuration.html">Configuration</a>', |
| 6297 | ] |
| 6298 | ) |
| 6299 | ) |
| 6300 | first_chapter.write_text("<h1>Introduction</h1>\n") |
| 6301 | |
| 6302 | implementation_plan = temp_dir / "implementation.md" |
| 6303 | implementation_plan.write_text( |
| 6304 | "\n".join( |
| 6305 | [ |
| 6306 | "# Implementation Plan", |
| 6307 | "", |
| 6308 | "## File Changes", |
| 6309 | f"- `{guide_root / 'index.html'}`", |
| 6310 | f"- `{chapters_dir}/`", |
| 6311 | "", |
| 6312 | ] |
| 6313 | ) |
| 6314 | ) |
| 6315 | |
| 6316 | context = build_context( |
| 6317 | temp_dir=temp_dir, |
| 6318 | messages=[], |
| 6319 | safeguards=FakeSafeguards(), |
| 6320 | assess_confidence=assess_confidence, |
| 6321 | verify_action=verify_action, |
| 6322 | auto_recover=False, |
| 6323 | ) |
| 6324 | queued_messages: list[str] = [] |
| 6325 | context.queue_steering_message_callback = queued_messages.append |
| 6326 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 6327 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 6328 | dod.implementation_plan = str(implementation_plan) |
| 6329 | dod.pending_items.extend( |
| 6330 | [ |
| 6331 | "First, examine the existing fortran guide structure and content to understand the format", |
| 6332 | "Create chapter files following the established pattern", |
| 6333 | ] |
| 6334 | ) |
| 6335 | dod.touched_files.extend([str(index_path), str(first_chapter)]) |
| 6336 | |
| 6337 | tool_call = ToolCall( |
| 6338 | id="working-note", |
| 6339 | name="notepad_write_working", |
| 6340 | arguments={"content": "Created index and first chapter; next is chapter 2"}, |
| 6341 | ) |
| 6342 | executor = FakeExecutor( |
| 6343 | [ |
| 6344 | tool_outcome( |
| 6345 | tool_call=tool_call, |
| 6346 | output="Working note recorded", |
| 6347 | is_error=False, |
| 6348 | ) |
| 6349 | ] |
| 6350 | ) |
| 6351 | |
| 6352 | summary = TurnSummary(final_response="") |
| 6353 | await runner.execute_batch( |
| 6354 | tool_calls=[tool_call], |
| 6355 | tool_source="assistant", |
| 6356 | pending_tool_calls_seen=set(), |
| 6357 | emit=_noop_emit, |
| 6358 | summary=summary, |
| 6359 | dod=dod, |
| 6360 | executor=executor, # type: ignore[arg-type] |
| 6361 | on_confirmation=None, |
| 6362 | on_user_question=None, |
| 6363 | emit_confirmation=None, |
| 6364 | consecutive_errors=0, |
| 6365 | ) |
| 6366 | |
| 6367 | assert queued_messages |
| 6368 | message = queued_messages[-1] |
| 6369 | assert "Bookkeeping note is recorded. A declared output artifact is still missing." in message |
| 6370 | assert "Resume by creating `02-installation.html` now." in message |
| 6371 | assert "Continue with the next pending item: `First, examine the existing fortran guide structure" not in message |
| 6372 | |
| 6373 | |
| 6374 | @pytest.mark.asyncio |
| 6375 | async def test_tool_batch_runner_shallow_glob_does_not_handoff_before_content_read( |
| 6376 | temp_dir: Path, |
| 6377 | ) -> None: |
| 6378 | async def assess_confidence( |
| 6379 | tool_name: str, |
| 6380 | tool_args: dict, |
| 6381 | context: str, |
| 6382 | ) -> ConfidenceAssessment: |
| 6383 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 6384 | |
| 6385 | async def verify_action( |
| 6386 | tool_name: str, |
| 6387 | tool_args: dict, |
| 6388 | result: str, |
| 6389 | expected: str = "", |
| 6390 | ) -> ActionVerification: |
| 6391 | raise AssertionError("Verification should not run in this scenario") |
| 6392 | |
| 6393 | fortran_root = temp_dir / "Loader" / "guides" / "fortran" |
| 6394 | chapters_dir = fortran_root / "chapters" |
| 6395 | chapters_dir.mkdir(parents=True) |
| 6396 | |
| 6397 | implementation_plan = temp_dir / "implementation.md" |
| 6398 | implementation_plan.write_text( |
| 6399 | "\n".join( |
| 6400 | [ |
| 6401 | "# Implementation Plan", |
| 6402 | "", |
| 6403 | "## File Changes", |
| 6404 | f"- `{temp_dir / 'Loader' / 'guides' / 'nginx' / 'index.html'}`", |
| 6405 | f"- `{temp_dir / 'Loader' / 'guides' / 'nginx' / 'chapters'}`", |
| 6406 | "", |
| 6407 | ] |
| 6408 | ) |
| 6409 | ) |
| 6410 | |
| 6411 | context = build_context( |
| 6412 | temp_dir=temp_dir, |
| 6413 | messages=[], |
| 6414 | safeguards=FakeSafeguards(), |
| 6415 | assess_confidence=assess_confidence, |
| 6416 | verify_action=verify_action, |
| 6417 | auto_recover=False, |
| 6418 | ) |
| 6419 | queued_messages: list[str] = [] |
| 6420 | context.queue_steering_message_callback = queued_messages.append |
| 6421 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 6422 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 6423 | dod.implementation_plan = str(implementation_plan) |
| 6424 | dod.pending_items.extend( |
| 6425 | [ |
| 6426 | "First, examine the existing fortran guide structure and content", |
| 6427 | "Create the nginx directory structure", |
| 6428 | "Develop the main index.html file for nginx guide", |
| 6429 | ] |
| 6430 | ) |
| 6431 | |
| 6432 | tool_call = ToolCall( |
| 6433 | id="glob-1", |
| 6434 | name="glob", |
| 6435 | arguments={"pattern": "**", "path": str(fortran_root)}, |
| 6436 | ) |
| 6437 | executor = FakeExecutor( |
| 6438 | [ |
| 6439 | tool_outcome( |
| 6440 | tool_call=tool_call, |
| 6441 | output=f"{fortran_root}\n{chapters_dir}", |
| 6442 | is_error=False, |
| 6443 | ) |
| 6444 | ] |
| 6445 | ) |
| 6446 | |
| 6447 | summary = TurnSummary(final_response="") |
| 6448 | await runner.execute_batch( |
| 6449 | tool_calls=[tool_call], |
| 6450 | tool_source="assistant", |
| 6451 | pending_tool_calls_seen=set(), |
| 6452 | emit=_noop_emit, |
| 6453 | summary=summary, |
| 6454 | dod=dod, |
| 6455 | executor=executor, # type: ignore[arg-type] |
| 6456 | on_confirmation=None, |
| 6457 | on_user_question=None, |
| 6458 | emit_confirmation=None, |
| 6459 | consecutive_errors=0, |
| 6460 | ) |
| 6461 | |
| 6462 | assert queued_messages == [] |
| 6463 | |
| 6464 | |
| 6465 | @pytest.mark.asyncio |
| 6466 | async def test_tool_batch_runner_hands_off_noop_toc_edit_when_file_is_already_valid( |
| 6467 | temp_dir: Path, |
| 6468 | ) -> None: |
| 6469 | async def assess_confidence( |
| 6470 | tool_name: str, |
| 6471 | tool_args: dict, |
| 6472 | context: str, |
| 6473 | ) -> ConfidenceAssessment: |
| 6474 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 6475 | |
| 6476 | async def verify_action( |
| 6477 | tool_name: str, |
| 6478 | tool_args: dict, |
| 6479 | result: str, |
| 6480 | expected: str = "", |
| 6481 | ) -> ActionVerification: |
| 6482 | raise AssertionError("Verification should not run in this scenario") |
| 6483 | |
| 6484 | prompt = ( |
| 6485 | "Have a look at ~/Loader/guides/fortran/index.html, then " |
| 6486 | "~/Loader/guides/fortran/chapters. The table of contents links in " |
| 6487 | "index.html are inaccurate and the href’s are wrong. Let’s update the " |
| 6488 | "links and their link texts to be correct." |
| 6489 | ) |
| 6490 | chapters = temp_dir / "chapters" |
| 6491 | chapters.mkdir() |
| 6492 | (chapters / "01-introduction.html").write_text( |
| 6493 | "<h1>Chapter 1: Introduction to Fortran</h1>\n" |
| 6494 | ) |
| 6495 | (chapters / "02-setup.html").write_text( |
| 6496 | "<h1>Chapter 2: Setting Up Your Environment</h1>\n" |
| 6497 | ) |
| 6498 | current_block = ( |
| 6499 | "<h2>Table of Contents</h2>\n" |
| 6500 | ' <ul class="chapter-list">\n' |
| 6501 | ' <li><a href="chapters/01-introduction.html">Chapter 1: Introduction to Fortran</a></li>\n' |
| 6502 | ' <li><a href="chapters/02-setup.html">Chapter 2: Setting Up Your Environment</a></li>\n' |
| 6503 | " </ul>\n" |
| 6504 | ) |
| 6505 | index_path = temp_dir / "index.html" |
| 6506 | index_path.write_text(current_block) |
| 6507 | |
| 6508 | context = build_context( |
| 6509 | temp_dir=temp_dir, |
| 6510 | messages=[], |
| 6511 | safeguards=FakeSafeguards(), |
| 6512 | assess_confidence=assess_confidence, |
| 6513 | verify_action=verify_action, |
| 6514 | auto_recover=False, |
| 6515 | ) |
| 6516 | context.session.current_task = prompt # type: ignore[attr-defined] |
| 6517 | queued_messages: list[str] = [] |
| 6518 | context.queue_steering_message_callback = queued_messages.append |
| 6519 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 6520 | tool_call = ToolCall( |
| 6521 | id="edit-1", |
| 6522 | name="edit", |
| 6523 | arguments={ |
| 6524 | "file_path": str(index_path), |
| 6525 | "old_string": current_block, |
| 6526 | "new_string": current_block, |
| 6527 | }, |
| 6528 | ) |
| 6529 | executor = FakeExecutor( |
| 6530 | [ |
| 6531 | tool_outcome( |
| 6532 | tool_call=tool_call, |
| 6533 | output=( |
| 6534 | "[Blocked - old_string and new_string are identical - no change " |
| 6535 | "would occur] Suggestion: Provide different old and new strings" |
| 6536 | ), |
| 6537 | is_error=True, |
| 6538 | state=ToolExecutionState.BLOCKED, |
| 6539 | ) |
| 6540 | ] |
| 6541 | ) |
| 6542 | |
| 6543 | await runner.execute_batch( |
| 6544 | tool_calls=[tool_call], |
| 6545 | tool_source="assistant", |
| 6546 | pending_tool_calls_seen=set(), |
| 6547 | emit=_noop_emit, |
| 6548 | summary=TurnSummary(final_response=""), |
| 6549 | dod=create_definition_of_done(prompt), |
| 6550 | executor=executor, # type: ignore[arg-type] |
| 6551 | on_confirmation=None, |
| 6552 | on_user_question=None, |
| 6553 | emit_confirmation=None, |
| 6554 | consecutive_errors=0, |
| 6555 | ) |
| 6556 | |
| 6557 | assert queued_messages == [] |
| 6558 | |
| 6559 | |
| 6560 | def test_tool_batch_runner_blocked_noop_edit_nudge_stays_on_active_repair_target( |
| 6561 | temp_dir: Path, |
| 6562 | ) -> None: |
| 6563 | async def assess_confidence( |
| 6564 | tool_name: str, |
| 6565 | tool_args: dict, |
| 6566 | context: str, |
| 6567 | ) -> ConfidenceAssessment: |
| 6568 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 6569 | |
| 6570 | async def verify_action( |
| 6571 | tool_name: str, |
| 6572 | tool_args: dict, |
| 6573 | result: str, |
| 6574 | expected: str = "", |
| 6575 | ) -> ActionVerification: |
| 6576 | raise AssertionError("Verification should not run in this scenario") |
| 6577 | |
| 6578 | repair_target = temp_dir / "guide" / "chapters" / "04-basic-usage.html" |
| 6579 | context = build_context( |
| 6580 | temp_dir=temp_dir, |
| 6581 | messages=[ |
| 6582 | Message( |
| 6583 | role=Role.ASSISTANT, |
| 6584 | content=( |
| 6585 | "Repair focus:\n" |
| 6586 | f"- Fix the broken local reference `05-advanced-topics.html` in `{repair_target}`.\n" |
| 6587 | f"- Immediate next step: edit `{repair_target}`.\n" |
| 6588 | f"- If the broken reference should remain, create `{temp_dir / 'guide' / 'chapters' / '05-advanced-topics.html'}`; otherwise remove or replace `05-advanced-topics.html`.\n" |
| 6589 | ), |
| 6590 | ) |
| 6591 | ], |
| 6592 | safeguards=FakeSafeguards(), |
| 6593 | assess_confidence=assess_confidence, |
| 6594 | verify_action=verify_action, |
| 6595 | ) |
| 6596 | queued: list[str] = [] |
| 6597 | context.queue_steering_message_callback = queued.append |
| 6598 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 6599 | dod = create_definition_of_done("Repair a guide page.") |
| 6600 | |
| 6601 | runner._queue_blocked_html_edit_nudge( |
| 6602 | ToolCall( |
| 6603 | id="edit-1", |
| 6604 | name="edit", |
| 6605 | arguments={ |
| 6606 | "file_path": str(repair_target), |
| 6607 | "old_string": "same", |
| 6608 | "new_string": "same", |
| 6609 | }, |
| 6610 | ), |
| 6611 | "[Blocked - old_string and new_string are identical - no change would occur] Suggestion: Provide different old and new strings", |
| 6612 | dod=dod, |
| 6613 | ) |
| 6614 | |
| 6615 | assert queued |
| 6616 | assert str(repair_target) in queued[0] |
| 6617 | assert "no on-disk change" in queued[0] |
| 6618 | assert "replace the surrounding block" in queued[0] |
| 6619 | assert "Do not reopen unrelated reference materials" in queued[0] |
| 6620 | |
| 6621 | |
| 6622 | def test_tool_batch_runner_blocked_noop_edit_after_full_build_prefers_verification( |
| 6623 | temp_dir: Path, |
| 6624 | ) -> None: |
| 6625 | async def assess_confidence( |
| 6626 | tool_name: str, |
| 6627 | tool_args: dict, |
| 6628 | context: str, |
| 6629 | ) -> ConfidenceAssessment: |
| 6630 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 6631 | |
| 6632 | async def verify_action( |
| 6633 | tool_name: str, |
| 6634 | tool_args: dict, |
| 6635 | result: str, |
| 6636 | expected: str = "", |
| 6637 | ) -> ActionVerification: |
| 6638 | raise AssertionError("Verification should not run in this scenario") |
| 6639 | |
| 6640 | guide_root = temp_dir / "guide" |
| 6641 | chapters = guide_root / "chapters" |
| 6642 | chapters.mkdir(parents=True) |
| 6643 | index_path = guide_root / "index.html" |
| 6644 | chapter_one = chapters / "01-introduction.html" |
| 6645 | index_path.write_text("<html></html>\n") |
| 6646 | chapter_one.write_text("<html></html>\n") |
| 6647 | |
| 6648 | implementation_plan = temp_dir / "implementation.md" |
| 6649 | implementation_plan.write_text( |
| 6650 | "\n".join( |
| 6651 | [ |
| 6652 | "# Implementation Plan", |
| 6653 | "", |
| 6654 | "## File Changes", |
| 6655 | f"- `{index_path}`", |
| 6656 | f"- `{chapter_one}`", |
| 6657 | "", |
| 6658 | ] |
| 6659 | ) |
| 6660 | ) |
| 6661 | |
| 6662 | context = build_context( |
| 6663 | temp_dir=temp_dir, |
| 6664 | messages=[ |
| 6665 | Message( |
| 6666 | role=Role.ASSISTANT, |
| 6667 | content=( |
| 6668 | "Repair focus:\n" |
| 6669 | f"- Confirm the final guide state in `{index_path}`.\n" |
| 6670 | f"- Immediate next step: verify `{index_path}` if no concrete mismatch remains.\n" |
| 6671 | ), |
| 6672 | ) |
| 6673 | ], |
| 6674 | safeguards=FakeSafeguards(), |
| 6675 | assess_confidence=assess_confidence, |
| 6676 | verify_action=verify_action, |
| 6677 | ) |
| 6678 | queued: list[str] = [] |
| 6679 | context.queue_steering_message_callback = queued.append |
| 6680 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 6681 | |
| 6682 | dod = create_definition_of_done("Create a multi-file guide.") |
| 6683 | dod.implementation_plan = str(implementation_plan) |
| 6684 | dod.touched_files.extend([str(index_path), str(chapter_one)]) |
| 6685 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 6686 | |
| 6687 | runner._queue_blocked_html_edit_nudge( |
| 6688 | ToolCall( |
| 6689 | id="edit-1", |
| 6690 | name="edit", |
| 6691 | arguments={ |
| 6692 | "file_path": str(index_path), |
| 6693 | "old_string": "same", |
| 6694 | "new_string": "same", |
| 6695 | }, |
| 6696 | ), |
| 6697 | "[Blocked - old_string and new_string are identical - no change would occur] Suggestion: Provide different old and new strings", |
| 6698 | dod=dod, |
| 6699 | ) |
| 6700 | |
| 6701 | assert queued |
| 6702 | assert "All explicitly planned artifacts already exist." in queued[0] |
| 6703 | assert "Finish with a final response now so Loader can run verification automatically." in queued[0] |
| 6704 | assert "replace the surrounding block" not in queued[0] |
| 6705 | |
| 6706 | |
| 6707 | def test_tool_batch_runner_blocked_noop_edit_keeps_quality_repair_active_after_full_build( |
| 6708 | temp_dir: Path, |
| 6709 | ) -> None: |
| 6710 | async def assess_confidence( |
| 6711 | tool_name: str, |
| 6712 | tool_args: dict, |
| 6713 | context: str, |
| 6714 | ) -> ConfidenceAssessment: |
| 6715 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 6716 | |
| 6717 | async def verify_action( |
| 6718 | tool_name: str, |
| 6719 | tool_args: dict, |
| 6720 | result: str, |
| 6721 | expected: str = "", |
| 6722 | ) -> ActionVerification: |
| 6723 | raise AssertionError("Verification should not run in this scenario") |
| 6724 | |
| 6725 | guide_root = temp_dir / "guide" |
| 6726 | chapters = guide_root / "chapters" |
| 6727 | chapters.mkdir(parents=True) |
| 6728 | index_path = guide_root / "index.html" |
| 6729 | chapter_one = chapters / "01-introduction.html" |
| 6730 | chapter_two = chapters / "02-installation.html" |
| 6731 | index_path.write_text("<html></html>\n") |
| 6732 | chapter_one.write_text("<html></html>\n") |
| 6733 | chapter_two.write_text("<html></html>\n") |
| 6734 | |
| 6735 | implementation_plan = temp_dir / "implementation.md" |
| 6736 | implementation_plan.write_text( |
| 6737 | "\n".join( |
| 6738 | [ |
| 6739 | "# Implementation Plan", |
| 6740 | "", |
| 6741 | "## File Changes", |
| 6742 | f"- `{index_path}`", |
| 6743 | f"- `{chapter_one}`", |
| 6744 | f"- `{chapter_two}`", |
| 6745 | "", |
| 6746 | ] |
| 6747 | ) |
| 6748 | ) |
| 6749 | |
| 6750 | context = build_context( |
| 6751 | temp_dir=temp_dir, |
| 6752 | messages=[ |
| 6753 | Message( |
| 6754 | role=Role.USER, |
| 6755 | content=( |
| 6756 | "Repair focus:\n" |
| 6757 | f"- Improve `{chapter_two}`: thin content (504 text chars, expected at least 1758).\n" |
| 6758 | f"- Improve `{chapter_two}`: insufficient structured content (6 blocks, expected at least 18).\n" |
| 6759 | f"- Immediate next step: edit `{chapter_two}`.\n" |
| 6760 | ), |
| 6761 | ) |
| 6762 | ], |
| 6763 | safeguards=FakeSafeguards(), |
| 6764 | assess_confidence=assess_confidence, |
| 6765 | verify_action=verify_action, |
| 6766 | ) |
| 6767 | queued: list[str] = [] |
| 6768 | context.queue_steering_message_callback = queued.append |
| 6769 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 6770 | |
| 6771 | dod = create_definition_of_done("Create a multi-file guide.") |
| 6772 | dod.implementation_plan = str(implementation_plan) |
| 6773 | dod.touched_files.extend([str(index_path), str(chapter_one), str(chapter_two)]) |
| 6774 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 6775 | |
| 6776 | runner._queue_blocked_html_edit_nudge( |
| 6777 | ToolCall( |
| 6778 | id="edit-1", |
| 6779 | name="edit", |
| 6780 | arguments={ |
| 6781 | "file_path": str(chapter_two), |
| 6782 | "old_string": "same", |
| 6783 | "new_string": "same", |
| 6784 | }, |
| 6785 | ), |
| 6786 | "[Blocked - old_string and new_string are identical - no change would occur] Suggestion: Provide different old and new strings", |
| 6787 | dod=dod, |
| 6788 | ) |
| 6789 | |
| 6790 | assert queued |
| 6791 | assert "active content-quality repair is not complete" in queued[0] |
| 6792 | assert "Repair focus:" in queued[0] |
| 6793 | assert f"Immediate next step: edit `{chapter_two}`" in queued[0] |
| 6794 | assert "thin content" in queued[0] |
| 6795 | assert "TodoWrite cannot satisfy" not in queued[0] |
| 6796 | assert "Finish with a final response now" not in queued[0] |
| 6797 | |
| 6798 | |
| 6799 | async def _noop_emit(event: AgentEvent) -> None: |
| 6800 | return None |
| 6801 | |
| 6802 | |
| 6803 | @pytest.mark.asyncio |
| 6804 | async def test_tool_batch_runner_marks_verification_planned_after_new_mutation( |
| 6805 | temp_dir: Path, |
| 6806 | ) -> None: |
| 6807 | async def assess_confidence( |
| 6808 | tool_name: str, |
| 6809 | tool_args: dict, |
| 6810 | context: str, |
| 6811 | ) -> ConfidenceAssessment: |
| 6812 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 6813 | |
| 6814 | async def verify_action( |
| 6815 | tool_name: str, |
| 6816 | tool_args: dict, |
| 6817 | result: str, |
| 6818 | expected: str = "", |
| 6819 | ) -> ActionVerification: |
| 6820 | raise AssertionError("Verification should not run for this scenario") |
| 6821 | |
| 6822 | context = build_context( |
| 6823 | temp_dir=temp_dir, |
| 6824 | messages=[], |
| 6825 | safeguards=FakeSafeguards(), |
| 6826 | assess_confidence=assess_confidence, |
| 6827 | verify_action=verify_action, |
| 6828 | ) |
| 6829 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 6830 | tool_call = ToolCall( |
| 6831 | id="write-1", |
| 6832 | name="write", |
| 6833 | arguments={"file_path": str(temp_dir / "README.md"), "content": "updated\n"}, |
| 6834 | ) |
| 6835 | executor = FakeExecutor( |
| 6836 | [tool_outcome(tool_call=tool_call, output="wrote file", is_error=False)] |
| 6837 | ) |
| 6838 | summary = TurnSummary(final_response="") |
| 6839 | dod = create_definition_of_done("Update README and verify it still works.") |
| 6840 | events: list[AgentEvent] = [] |
| 6841 | |
| 6842 | async def emit(event: AgentEvent) -> None: |
| 6843 | events.append(event) |
| 6844 | |
| 6845 | await runner.execute_batch( |
| 6846 | tool_calls=[tool_call], |
| 6847 | tool_source="assistant", |
| 6848 | pending_tool_calls_seen=set(), |
| 6849 | emit=emit, |
| 6850 | summary=summary, |
| 6851 | dod=dod, |
| 6852 | executor=executor, # type: ignore[arg-type] |
| 6853 | on_confirmation=None, |
| 6854 | on_user_question=None, |
| 6855 | emit_confirmation=None, |
| 6856 | consecutive_errors=0, |
| 6857 | ) |
| 6858 | |
| 6859 | assert dod.last_verification_result == "planned" |
| 6860 | assert dod.verification_commands |
| 6861 | assert "Collect verification evidence" in dod.pending_items |
| 6862 | assert dod.active_verification_attempt_id == "verification-attempt-1" |
| 6863 | assert dod.active_verification_attempt_number == 1 |
| 6864 | assert summary.workflow_timeline[-1].reason_code == "verification_planned" |
| 6865 | assert summary.workflow_timeline[-1].policy_outcome == "planned" |
| 6866 | assert summary.workflow_timeline[-1].verification_observations[0].status == "planned" |
| 6867 | assert ( |
| 6868 | summary.workflow_timeline[-1].verification_observations[0].attempt_id |
| 6869 | == "verification-attempt-1" |
| 6870 | ) |
| 6871 | assert ( |
| 6872 | summary.workflow_timeline[-1].verification_observations[0].attempt_number == 1 |
| 6873 | ) |
| 6874 | |
| 6875 | |
| 6876 | @pytest.mark.asyncio |
| 6877 | async def test_tool_batch_runner_does_not_mark_verification_planned_after_setup_only_mkdir( |
| 6878 | temp_dir: Path, |
| 6879 | ) -> None: |
| 6880 | async def assess_confidence( |
| 6881 | tool_name: str, |
| 6882 | tool_args: dict, |
| 6883 | context: str, |
| 6884 | ) -> ConfidenceAssessment: |
| 6885 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 6886 | |
| 6887 | async def verify_action( |
| 6888 | tool_name: str, |
| 6889 | tool_args: dict, |
| 6890 | result: str, |
| 6891 | expected: str = "", |
| 6892 | ) -> ActionVerification: |
| 6893 | raise AssertionError("Verification should not run in this scenario") |
| 6894 | |
| 6895 | context = build_context( |
| 6896 | temp_dir=temp_dir, |
| 6897 | messages=[], |
| 6898 | safeguards=FakeSafeguards(), |
| 6899 | assess_confidence=assess_confidence, |
| 6900 | verify_action=verify_action, |
| 6901 | ) |
| 6902 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 6903 | nginx_root = temp_dir / "Loader" / "guides" / "nginx" |
| 6904 | chapters = nginx_root / "chapters" |
| 6905 | implementation_plan = temp_dir / "implementation.md" |
| 6906 | implementation_plan.write_text( |
| 6907 | "\n".join( |
| 6908 | [ |
| 6909 | "# Implementation Plan", |
| 6910 | "", |
| 6911 | "## File Changes", |
| 6912 | f"- `{chapters}/`", |
| 6913 | f"- `{nginx_root / 'index.html'}`", |
| 6914 | "", |
| 6915 | ] |
| 6916 | ) |
| 6917 | ) |
| 6918 | |
| 6919 | tool_call = ToolCall( |
| 6920 | id="mkdir-1", |
| 6921 | name="bash", |
| 6922 | arguments={"command": f"mkdir -p {chapters}"}, |
| 6923 | ) |
| 6924 | executor = FakeExecutor( |
| 6925 | [tool_outcome(tool_call=tool_call, output="", is_error=False)] |
| 6926 | ) |
| 6927 | summary = TurnSummary(final_response="") |
| 6928 | dod = create_definition_of_done("Create an equally thorough nginx guide with chapters.") |
| 6929 | dod.implementation_plan = str(implementation_plan) |
| 6930 | events: list[AgentEvent] = [] |
| 6931 | |
| 6932 | async def emit(event: AgentEvent) -> None: |
| 6933 | events.append(event) |
| 6934 | |
| 6935 | await runner.execute_batch( |
| 6936 | tool_calls=[tool_call], |
| 6937 | tool_source="assistant", |
| 6938 | pending_tool_calls_seen=set(), |
| 6939 | emit=emit, |
| 6940 | summary=summary, |
| 6941 | dod=dod, |
| 6942 | executor=executor, # type: ignore[arg-type] |
| 6943 | on_confirmation=None, |
| 6944 | on_user_question=None, |
| 6945 | emit_confirmation=None, |
| 6946 | consecutive_errors=0, |
| 6947 | ) |
| 6948 | |
| 6949 | assert dod.last_verification_result is None |
| 6950 | assert "Collect verification evidence" not in dod.pending_items |
| 6951 | assert not any( |
| 6952 | entry.reason_code == "verification_planned" for entry in summary.workflow_timeline |
| 6953 | ) |
| 6954 | |
| 6955 | |
| 6956 | @pytest.mark.asyncio |
| 6957 | async def test_tool_batch_runner_does_not_mark_verification_planned_while_chapter_build_pending( |
| 6958 | temp_dir: Path, |
| 6959 | ) -> None: |
| 6960 | async def assess_confidence( |
| 6961 | tool_name: str, |
| 6962 | tool_args: dict, |
| 6963 | context: str, |
| 6964 | ) -> ConfidenceAssessment: |
| 6965 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 6966 | |
| 6967 | async def verify_action( |
| 6968 | tool_name: str, |
| 6969 | tool_args: dict, |
| 6970 | result: str, |
| 6971 | expected: str = "", |
| 6972 | ) -> ActionVerification: |
| 6973 | raise AssertionError("Verification should not run in this scenario") |
| 6974 | |
| 6975 | context = build_context( |
| 6976 | temp_dir=temp_dir, |
| 6977 | messages=[], |
| 6978 | safeguards=FakeSafeguards(), |
| 6979 | assess_confidence=assess_confidence, |
| 6980 | verify_action=verify_action, |
| 6981 | ) |
| 6982 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 6983 | nginx_root = temp_dir / "Loader" / "guides" / "nginx" |
| 6984 | chapters = nginx_root / "chapters" |
| 6985 | chapters.mkdir(parents=True) |
| 6986 | index_path = nginx_root / "index.html" |
| 6987 | implementation_plan = temp_dir / "implementation.md" |
| 6988 | implementation_plan.write_text( |
| 6989 | "\n".join( |
| 6990 | [ |
| 6991 | "# Implementation Plan", |
| 6992 | "", |
| 6993 | "## File Changes", |
| 6994 | f"- `{nginx_root}/`", |
| 6995 | f"- `{chapters}/`", |
| 6996 | f"- `{index_path}`", |
| 6997 | "", |
| 6998 | ] |
| 6999 | ) |
| 7000 | ) |
| 7001 | |
| 7002 | tool_call = ToolCall( |
| 7003 | id="write-index", |
| 7004 | name="write", |
| 7005 | arguments={"file_path": str(index_path), "content": "<html></html>\n"}, |
| 7006 | ) |
| 7007 | executor = FakeExecutor( |
| 7008 | [tool_outcome(tool_call=tool_call, output="wrote file", is_error=False)] |
| 7009 | ) |
| 7010 | summary = TurnSummary(final_response="") |
| 7011 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 7012 | dod.implementation_plan = str(implementation_plan) |
| 7013 | dod.pending_items.extend( |
| 7014 | [ |
| 7015 | "Develop the main index.html file with proper structure", |
| 7016 | "Create first nginx chapter", |
| 7017 | ] |
| 7018 | ) |
| 7019 | events: list[AgentEvent] = [] |
| 7020 | |
| 7021 | async def emit(event: AgentEvent) -> None: |
| 7022 | events.append(event) |
| 7023 | |
| 7024 | await runner.execute_batch( |
| 7025 | tool_calls=[tool_call], |
| 7026 | tool_source="assistant", |
| 7027 | pending_tool_calls_seen=set(), |
| 7028 | emit=emit, |
| 7029 | summary=summary, |
| 7030 | dod=dod, |
| 7031 | executor=executor, # type: ignore[arg-type] |
| 7032 | on_confirmation=None, |
| 7033 | on_user_question=None, |
| 7034 | emit_confirmation=None, |
| 7035 | consecutive_errors=0, |
| 7036 | ) |
| 7037 | |
| 7038 | assert dod.last_verification_result is None |
| 7039 | assert "Collect verification evidence" not in dod.pending_items |
| 7040 | assert "Create first nginx chapter" in dod.pending_items |
| 7041 | assert not any( |
| 7042 | entry.reason_code == "verification_planned" for entry in summary.workflow_timeline |
| 7043 | ) |
| 7044 | |
| 7045 | |
| 7046 | @pytest.mark.asyncio |
| 7047 | async def test_tool_batch_runner_marks_passed_verification_stale_after_new_mutation( |
| 7048 | temp_dir: Path, |
| 7049 | ) -> None: |
| 7050 | async def assess_confidence( |
| 7051 | tool_name: str, |
| 7052 | tool_args: dict, |
| 7053 | context: str, |
| 7054 | ) -> ConfidenceAssessment: |
| 7055 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 7056 | |
| 7057 | async def verify_action( |
| 7058 | tool_name: str, |
| 7059 | tool_args: dict, |
| 7060 | result: str, |
| 7061 | expected: str = "", |
| 7062 | ) -> ActionVerification: |
| 7063 | raise AssertionError("Verification should not run for this scenario") |
| 7064 | |
| 7065 | context = build_context( |
| 7066 | temp_dir=temp_dir, |
| 7067 | messages=[], |
| 7068 | safeguards=FakeSafeguards(), |
| 7069 | assess_confidence=assess_confidence, |
| 7070 | verify_action=verify_action, |
| 7071 | ) |
| 7072 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 7073 | tool_call = ToolCall( |
| 7074 | id="write-1", |
| 7075 | name="write", |
| 7076 | arguments={"file_path": str(temp_dir / "README.md"), "content": "updated\n"}, |
| 7077 | ) |
| 7078 | executor = FakeExecutor( |
| 7079 | [tool_outcome(tool_call=tool_call, output="wrote file", is_error=False)] |
| 7080 | ) |
| 7081 | summary = TurnSummary(final_response="") |
| 7082 | dod = create_definition_of_done("Update README and verify it still works.") |
| 7083 | dod.verification_commands = ["uv run pytest -q"] |
| 7084 | dod.last_verification_result = "passed" |
| 7085 | dod.verification_attempt_counter = 1 |
| 7086 | dod.active_verification_attempt_id = "verification-attempt-1" |
| 7087 | dod.active_verification_attempt_number = 1 |
| 7088 | dod.evidence = [ |
| 7089 | VerificationEvidence( |
| 7090 | command="uv run pytest -q", |
| 7091 | passed=True, |
| 7092 | stdout="401 passed", |
| 7093 | kind="test", |
| 7094 | ) |
| 7095 | ] |
| 7096 | dod.completed_items.append("Collect verification evidence") |
| 7097 | events: list[AgentEvent] = [] |
| 7098 | |
| 7099 | async def emit(event: AgentEvent) -> None: |
| 7100 | events.append(event) |
| 7101 | |
| 7102 | await runner.execute_batch( |
| 7103 | tool_calls=[tool_call], |
| 7104 | tool_source="assistant", |
| 7105 | pending_tool_calls_seen=set(), |
| 7106 | emit=emit, |
| 7107 | summary=summary, |
| 7108 | dod=dod, |
| 7109 | executor=executor, # type: ignore[arg-type] |
| 7110 | on_confirmation=None, |
| 7111 | on_user_question=None, |
| 7112 | emit_confirmation=None, |
| 7113 | consecutive_errors=0, |
| 7114 | ) |
| 7115 | |
| 7116 | assert dod.last_verification_result == "stale" |
| 7117 | assert dod.evidence == [] |
| 7118 | assert "Collect verification evidence" in dod.pending_items |
| 7119 | assert "Collect verification evidence" not in dod.completed_items |
| 7120 | assert dod.active_verification_attempt_id == "verification-attempt-2" |
| 7121 | assert dod.active_verification_attempt_number == 2 |
| 7122 | assert summary.workflow_timeline[-1].reason_code == "verification_stale" |
| 7123 | assert summary.workflow_timeline[-1].policy_outcome == "stale" |
| 7124 | assert summary.workflow_timeline[-1].verification_observations[0].status == "stale" |
| 7125 | assert ( |
| 7126 | summary.workflow_timeline[-1].verification_observations[0].attempt_id |
| 7127 | == "verification-attempt-1" |
| 7128 | ) |
| 7129 | assert ( |
| 7130 | summary.workflow_timeline[-1].verification_observations[0].attempt_number == 1 |
| 7131 | ) |
| 7132 | assert ( |
| 7133 | summary.workflow_timeline[-1].verification_observations[0].supersedes_attempt_id |
| 7134 | == "verification-attempt-2" |
| 7135 | ) |
| 7136 | assert ( |
| 7137 | summary.workflow_timeline[-1].verification_observations[0].command |
| 7138 | == "uv run pytest -q" |
| 7139 | ) |
| 7140 | |
| 7141 | |
| 7142 | def test_tool_batch_runner_blocked_active_repair_nudge_uses_repair_scope(temp_dir: Path) -> None: |
| 7143 | async def assess_confidence( |
| 7144 | tool_name: str, |
| 7145 | tool_args: dict, |
| 7146 | context: str, |
| 7147 | ) -> ConfidenceAssessment: |
| 7148 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 7149 | |
| 7150 | async def verify_action( |
| 7151 | tool_name: str, |
| 7152 | tool_args: dict, |
| 7153 | result: str, |
| 7154 | expected: str = "", |
| 7155 | ) -> ActionVerification: |
| 7156 | raise AssertionError("Verification should not run in this scenario") |
| 7157 | |
| 7158 | repair_target = temp_dir / "guide" / "index.html" |
| 7159 | context = build_context( |
| 7160 | temp_dir=temp_dir, |
| 7161 | messages=[ |
| 7162 | Message( |
| 7163 | role=Role.ASSISTANT, |
| 7164 | content=( |
| 7165 | "Repair focus:\n" |
| 7166 | f"- Fix the broken local reference `chapters/01-getting-started.html` in `{repair_target}`.\n" |
| 7167 | f"- Immediate next step: edit `{repair_target}`.\n" |
| 7168 | f"- If the broken reference should remain, create `{temp_dir / 'guide' / 'chapters' / '01-getting-started.html'}`; otherwise remove or replace `chapters/01-getting-started.html`.\n" |
| 7169 | ), |
| 7170 | ) |
| 7171 | ], |
| 7172 | safeguards=FakeSafeguards(), |
| 7173 | assess_confidence=assess_confidence, |
| 7174 | verify_action=verify_action, |
| 7175 | ) |
| 7176 | queued: list[str] = [] |
| 7177 | context.queue_steering_message_callback = queued.append |
| 7178 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 7179 | |
| 7180 | runner._queue_blocked_active_repair_nudge( |
| 7181 | "[Blocked - active repair scope: verification already identified the repair target.]" |
| 7182 | ) |
| 7183 | |
| 7184 | assert queued |
| 7185 | assert str(repair_target) in queued[0] |
| 7186 | assert str(temp_dir / "guide" / "chapters" / "01-getting-started.html") in queued[0] |
| 7187 | assert "Do not reopen unrelated reference materials" in queued[0] |
| 7188 | |
| 7189 | |
| 7190 | def test_tool_batch_runner_blocked_active_repair_mutation_nudge_uses_allowed_paths( |
| 7191 | temp_dir: Path, |
| 7192 | ) -> None: |
| 7193 | async def assess_confidence( |
| 7194 | tool_name: str, |
| 7195 | tool_args: dict, |
| 7196 | context: str, |
| 7197 | ) -> ConfidenceAssessment: |
| 7198 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 7199 | |
| 7200 | async def verify_action( |
| 7201 | tool_name: str, |
| 7202 | tool_args: dict, |
| 7203 | result: str, |
| 7204 | expected: str = "", |
| 7205 | ) -> ActionVerification: |
| 7206 | raise AssertionError("Verification should not run in this scenario") |
| 7207 | |
| 7208 | repair_target = temp_dir / "guide" / "chapters" / "05-advanced-configurations.html" |
| 7209 | stylesheet = temp_dir / "guide" / "styles.css" |
| 7210 | context = build_context( |
| 7211 | temp_dir=temp_dir, |
| 7212 | messages=[ |
| 7213 | Message( |
| 7214 | role=Role.ASSISTANT, |
| 7215 | content=( |
| 7216 | "Repair focus:\n" |
| 7217 | f"- Fix the broken local reference `../styles.css` in `{repair_target}`.\n" |
| 7218 | f"- Immediate next step: edit `{repair_target}`.\n" |
| 7219 | f"- If the broken reference should remain, create `{stylesheet}`; otherwise remove or replace `../styles.css`.\n" |
| 7220 | ), |
| 7221 | ) |
| 7222 | ], |
| 7223 | safeguards=FakeSafeguards(), |
| 7224 | assess_confidence=assess_confidence, |
| 7225 | verify_action=verify_action, |
| 7226 | ) |
| 7227 | queued: list[str] = [] |
| 7228 | context.queue_steering_message_callback = queued.append |
| 7229 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 7230 | |
| 7231 | runner._queue_blocked_active_repair_mutation_nudge( |
| 7232 | "[Blocked - active repair mutation scope: verification already identified the repair target.]" |
| 7233 | ) |
| 7234 | |
| 7235 | assert queued |
| 7236 | assert str(repair_target) in queued[0] |
| 7237 | assert str(stylesheet) in queued[0] |
| 7238 | assert "before widening the change set" in queued[0] |
| 7239 | |
| 7240 | |
| 7241 | def test_tool_batch_runner_duplicate_repair_mutation_restates_verifier_deltas( |
| 7242 | temp_dir: Path, |
| 7243 | ) -> None: |
| 7244 | async def assess_confidence( |
| 7245 | tool_name: str, |
| 7246 | tool_args: dict, |
| 7247 | context: str, |
| 7248 | ) -> ConfidenceAssessment: |
| 7249 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 7250 | |
| 7251 | async def verify_action( |
| 7252 | tool_name: str, |
| 7253 | tool_args: dict, |
| 7254 | result: str, |
| 7255 | expected: str = "", |
| 7256 | ) -> ActionVerification: |
| 7257 | raise AssertionError("Verification should not run in this scenario") |
| 7258 | |
| 7259 | index_path = temp_dir / "guide" / "index.html" |
| 7260 | chapter_path = temp_dir / "guide" / "chapters" / "02-installation.html" |
| 7261 | context = build_context( |
| 7262 | temp_dir=temp_dir, |
| 7263 | messages=[ |
| 7264 | Message( |
| 7265 | role=Role.USER, |
| 7266 | content=( |
| 7267 | "Repair focus:\n" |
| 7268 | f"- Improve `{index_path}`: insufficient structured content (9 blocks, expected at least 12).\n" |
| 7269 | f"- Improve `{chapter_path}`: thin content (526 text chars, expected at least 1758).\n" |
| 7270 | f"- Immediate next step: edit `{index_path}`.\n" |
| 7271 | "- Update the listed generated artifacts directly; do not recreate the artifact set.\n" |
| 7272 | ), |
| 7273 | ) |
| 7274 | ], |
| 7275 | safeguards=FakeSafeguards(), |
| 7276 | assess_confidence=assess_confidence, |
| 7277 | verify_action=verify_action, |
| 7278 | ) |
| 7279 | queued: list[str] = [] |
| 7280 | context.queue_steering_message_callback = queued.append |
| 7281 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 7282 | dod = create_definition_of_done("Create a multi-file guide.") |
| 7283 | |
| 7284 | runner._queue_duplicate_mutation_nudge( # type: ignore[attr-defined] |
| 7285 | ToolCall( |
| 7286 | id="dup-write", |
| 7287 | name="write", |
| 7288 | arguments={"file_path": str(index_path), "content": "<h1>same</h1>"}, |
| 7289 | ), |
| 7290 | dod=dod, |
| 7291 | ) |
| 7292 | |
| 7293 | assert queued |
| 7294 | assert "skipped because it would not change" in queued[0] |
| 7295 | assert "Do not submit the same content again" in queued[0] |
| 7296 | assert "insufficient structured content" in queued[0] |
| 7297 | assert "thin content" in queued[0] |
| 7298 | assert "make one real edit" in queued[0] |
| 7299 | |
| 7300 | |
| 7301 | @pytest.mark.asyncio |
| 7302 | async def test_tool_batch_runner_quality_repair_success_hands_to_next_target( |
| 7303 | temp_dir: Path, |
| 7304 | ) -> None: |
| 7305 | async def assess_confidence( |
| 7306 | tool_name: str, |
| 7307 | tool_args: dict, |
| 7308 | context: str, |
| 7309 | ) -> ConfidenceAssessment: |
| 7310 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 7311 | |
| 7312 | async def verify_action( |
| 7313 | tool_name: str, |
| 7314 | tool_args: dict, |
| 7315 | result: str, |
| 7316 | expected: str = "", |
| 7317 | ) -> ActionVerification: |
| 7318 | raise AssertionError("Verification should not run in this scenario") |
| 7319 | |
| 7320 | chapters = temp_dir / "guide" / "chapters" |
| 7321 | first = chapters / "01-introduction.html" |
| 7322 | second = chapters / "02-installation.html" |
| 7323 | chapters.mkdir(parents=True) |
| 7324 | first.write_text("<h1>Intro</h1>\n") |
| 7325 | second.write_text("<h1>Install</h1>\n") |
| 7326 | context = build_context( |
| 7327 | temp_dir=temp_dir, |
| 7328 | messages=[ |
| 7329 | Message( |
| 7330 | role=Role.ASSISTANT, |
| 7331 | content=( |
| 7332 | "Repair focus:\n" |
| 7333 | f"- Improve `{first}`: thin content (400 text chars, expected at least 1758).\n" |
| 7334 | f"- Improve `{second}`: insufficient structured content (6 blocks, expected at least 18).\n" |
| 7335 | f"- Immediate next step: edit `{first}` with a substantial expansion or replacement.\n" |
| 7336 | "- Repair every listed quality target in order before any final answer.\n" |
| 7337 | ), |
| 7338 | ) |
| 7339 | ], |
| 7340 | safeguards=FakeSafeguards(), |
| 7341 | assess_confidence=assess_confidence, |
| 7342 | verify_action=verify_action, |
| 7343 | ) |
| 7344 | queued: list[str] = [] |
| 7345 | context.queue_steering_message_callback = queued.append |
| 7346 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 7347 | dod = create_definition_of_done("Repair generated HTML guide quality.") |
| 7348 | tool_call = ToolCall( |
| 7349 | id="write-intro", |
| 7350 | name="write", |
| 7351 | arguments={ |
| 7352 | "file_path": str(first), |
| 7353 | "content": "<h1>Intro</h1><p>Substantial expansion.</p>\n", |
| 7354 | }, |
| 7355 | ) |
| 7356 | |
| 7357 | await runner.execute_batch( |
| 7358 | tool_calls=[tool_call], |
| 7359 | tool_source="assistant", |
| 7360 | pending_tool_calls_seen=set(), |
| 7361 | emit=_noop_emit, |
| 7362 | summary=TurnSummary(final_response=""), |
| 7363 | dod=dod, |
| 7364 | executor=FakeExecutor( |
| 7365 | [ |
| 7366 | tool_outcome( |
| 7367 | tool_call=tool_call, |
| 7368 | output=f"Successfully wrote {first}", |
| 7369 | is_error=False, |
| 7370 | ) |
| 7371 | ] |
| 7372 | ), # type: ignore[arg-type] |
| 7373 | on_confirmation=None, |
| 7374 | on_user_question=None, |
| 7375 | emit_confirmation=None, |
| 7376 | consecutive_errors=0, |
| 7377 | ) |
| 7378 | |
| 7379 | assert queued |
| 7380 | handoff = next(message for message in queued if "next listed quality target" in message) |
| 7381 | assert str(second.resolve(strict=False)) in handoff |
| 7382 | assert "Do not rerun verification" in handoff |
| 7383 | assert "Repair focus:" in handoff |
| 7384 | assert "insufficient structured content" in handoff |
| 7385 | assert f"Immediate next step: edit `{second.resolve(strict=False)}`" in handoff |
| 7386 | assert all("All explicitly planned artifacts now exist" not in message for message in queued) |
| 7387 | |
| 7388 | |
| 7389 | @pytest.mark.asyncio |
| 7390 | async def test_tool_batch_runner_hands_off_after_active_repair_support_file_write( |
| 7391 | temp_dir: Path, |
| 7392 | ) -> None: |
| 7393 | async def assess_confidence( |
| 7394 | tool_name: str, |
| 7395 | tool_args: dict, |
| 7396 | context: str, |
| 7397 | ) -> ConfidenceAssessment: |
| 7398 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 7399 | |
| 7400 | async def verify_action( |
| 7401 | tool_name: str, |
| 7402 | tool_args: dict, |
| 7403 | result: str, |
| 7404 | expected: str = "", |
| 7405 | ) -> ActionVerification: |
| 7406 | raise AssertionError("Verification should not run in this scenario") |
| 7407 | |
| 7408 | repair_target = temp_dir / "guide" / "index.html" |
| 7409 | stylesheet = temp_dir / "guide" / "style.css" |
| 7410 | repair_target.parent.mkdir(parents=True) |
| 7411 | repair_target.write_text('<link rel="stylesheet" href="style.css">\n') |
| 7412 | context = build_context( |
| 7413 | temp_dir=temp_dir, |
| 7414 | messages=[ |
| 7415 | Message( |
| 7416 | role=Role.ASSISTANT, |
| 7417 | content=( |
| 7418 | "Repair focus:\n" |
| 7419 | f"- Fix the broken local reference `style.css` in `{repair_target}`.\n" |
| 7420 | f"- Immediate next step: edit `{repair_target}`.\n" |
| 7421 | f"- If the broken reference should remain, create `{stylesheet}`; otherwise remove or replace `style.css`.\n" |
| 7422 | ), |
| 7423 | ) |
| 7424 | ], |
| 7425 | safeguards=FakeSafeguards(), |
| 7426 | assess_confidence=assess_confidence, |
| 7427 | verify_action=verify_action, |
| 7428 | ) |
| 7429 | queued: list[str] = [] |
| 7430 | context.queue_steering_message_callback = queued.append |
| 7431 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 7432 | dod = create_definition_of_done("Repair a guide stylesheet link.") |
| 7433 | tool_call = ToolCall( |
| 7434 | id="write-style", |
| 7435 | name="write", |
| 7436 | arguments={ |
| 7437 | "file_path": str(stylesheet), |
| 7438 | "content": "body { font-family: sans-serif; }\n", |
| 7439 | }, |
| 7440 | ) |
| 7441 | |
| 7442 | await runner.execute_batch( |
| 7443 | tool_calls=[tool_call], |
| 7444 | tool_source="assistant", |
| 7445 | pending_tool_calls_seen=set(), |
| 7446 | emit=_noop_emit, |
| 7447 | summary=TurnSummary(final_response=""), |
| 7448 | dod=dod, |
| 7449 | executor=FakeExecutor( |
| 7450 | [ |
| 7451 | tool_outcome( |
| 7452 | tool_call=tool_call, |
| 7453 | output=f"Successfully wrote {stylesheet}", |
| 7454 | is_error=False, |
| 7455 | ) |
| 7456 | ] |
| 7457 | ), # type: ignore[arg-type] |
| 7458 | on_confirmation=None, |
| 7459 | on_user_question=None, |
| 7460 | emit_confirmation=None, |
| 7461 | consecutive_errors=0, |
| 7462 | ) |
| 7463 | |
| 7464 | assert queued |
| 7465 | assert any("support file for the active verification repair now exists" in message for message in queued) |
| 7466 | assert any("Do not retarget" in message for message in queued) |
| 7467 | assert any("Loader can re-run verification" in message for message in queued) |
| 7468 | |
| 7469 | |
| 7470 | def test_tool_batch_runner_blocked_late_reference_drift_nudge_points_to_missing_artifact( |
| 7471 | temp_dir: Path, |
| 7472 | ) -> None: |
| 7473 | async def assess_confidence( |
| 7474 | tool_name: str, |
| 7475 | tool_args: dict, |
| 7476 | context: str, |
| 7477 | ) -> ConfidenceAssessment: |
| 7478 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 7479 | |
| 7480 | async def verify_action( |
| 7481 | tool_name: str, |
| 7482 | tool_args: dict, |
| 7483 | result: str, |
| 7484 | expected: str = "", |
| 7485 | ) -> ActionVerification: |
| 7486 | raise AssertionError("Verification should not run in this scenario") |
| 7487 | |
| 7488 | context = build_context( |
| 7489 | temp_dir=temp_dir, |
| 7490 | messages=[], |
| 7491 | safeguards=FakeSafeguards(), |
| 7492 | assess_confidence=assess_confidence, |
| 7493 | verify_action=verify_action, |
| 7494 | ) |
| 7495 | queued: list[str] = [] |
| 7496 | context.queue_steering_message_callback = queued.append |
| 7497 | store = DefinitionOfDoneStore(temp_dir) |
| 7498 | dod = create_definition_of_done("Create a multi-file guide from a reference") |
| 7499 | plan_path = temp_dir / "implementation.md" |
| 7500 | plan_path.write_text( |
| 7501 | "# File Changes\n" |
| 7502 | "- `guide/index.html`\n" |
| 7503 | "- `guide/chapters/01-getting-started.html`\n" |
| 7504 | "- `guide/chapters/02-installation.html`\n" |
| 7505 | "- `guide/chapters/03-first-website.html`\n" |
| 7506 | ) |
| 7507 | dod.implementation_plan = str(plan_path) |
| 7508 | (temp_dir / "guide" / "chapters").mkdir(parents=True, exist_ok=True) |
| 7509 | (temp_dir / "guide" / "index.html").write_text("index") |
| 7510 | (temp_dir / "guide" / "chapters" / "01-getting-started.html").write_text("one") |
| 7511 | (temp_dir / "guide" / "chapters" / "02-installation.html").write_text("two") |
| 7512 | runner = ToolBatchRunner(context, store) |
| 7513 | |
| 7514 | runner._queue_blocked_late_reference_drift_nudge( |
| 7515 | "[Blocked - late reference drift: several planned artifacts already exist.]", |
| 7516 | dod=dod, |
| 7517 | ) |
| 7518 | |
| 7519 | assert queued |
| 7520 | assert "03-first-website.html" in queued[0] |
| 7521 | assert "older reference materials" in queued[0] |
| 7522 | |
| 7523 | |
| 7524 | def test_tool_batch_runner_blocked_completed_artifact_scope_nudge_prefers_verification( |
| 7525 | temp_dir: Path, |
| 7526 | ) -> None: |
| 7527 | async def assess_confidence( |
| 7528 | tool_name: str, |
| 7529 | tool_args: dict, |
| 7530 | context: str, |
| 7531 | ) -> ConfidenceAssessment: |
| 7532 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 7533 | |
| 7534 | async def verify_action( |
| 7535 | tool_name: str, |
| 7536 | tool_args: dict, |
| 7537 | result: str, |
| 7538 | expected: str = "", |
| 7539 | ) -> ActionVerification: |
| 7540 | raise AssertionError("Verification should not run in this scenario") |
| 7541 | |
| 7542 | guide_root = temp_dir / "guide" |
| 7543 | chapters = guide_root / "chapters" |
| 7544 | guide_root.mkdir(parents=True) |
| 7545 | chapters.mkdir() |
| 7546 | index_path = guide_root / "index.html" |
| 7547 | chapter_one = chapters / "01-getting-started.html" |
| 7548 | chapter_two = chapters / "02-installation.html" |
| 7549 | index_path.write_text("index") |
| 7550 | chapter_one.write_text("one") |
| 7551 | chapter_two.write_text("two") |
| 7552 | |
| 7553 | implementation_plan = temp_dir / "implementation.md" |
| 7554 | implementation_plan.write_text( |
| 7555 | "\n".join( |
| 7556 | [ |
| 7557 | "# Implementation Plan", |
| 7558 | "", |
| 7559 | "## File Changes", |
| 7560 | f"- `{guide_root}`", |
| 7561 | f"- `{chapters}`", |
| 7562 | f"- `{index_path}`", |
| 7563 | f"- `{chapter_one}`", |
| 7564 | f"- `{chapter_two}`", |
| 7565 | "", |
| 7566 | ] |
| 7567 | ) |
| 7568 | ) |
| 7569 | |
| 7570 | context = build_context( |
| 7571 | temp_dir=temp_dir, |
| 7572 | messages=[], |
| 7573 | safeguards=FakeSafeguards(), |
| 7574 | assess_confidence=assess_confidence, |
| 7575 | verify_action=verify_action, |
| 7576 | ) |
| 7577 | queued: list[str] = [] |
| 7578 | context.queue_steering_message_callback = queued.append |
| 7579 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 7580 | dod = create_definition_of_done("Create a multi-file guide from a reference") |
| 7581 | dod.implementation_plan = str(implementation_plan) |
| 7582 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 7583 | sync_todos_to_definition_of_done( |
| 7584 | dod, |
| 7585 | [ |
| 7586 | { |
| 7587 | "content": "Verify all guide files are linked and complete", |
| 7588 | "active_form": "Working on: Verify all guide files are linked and complete", |
| 7589 | "status": "pending", |
| 7590 | } |
| 7591 | ], |
| 7592 | project_root=temp_dir, |
| 7593 | ) |
| 7594 | |
| 7595 | runner._queue_blocked_completed_artifact_scope_nudge( |
| 7596 | "[Blocked - completed artifact set scope: all explicitly planned artifacts already exist.]", |
| 7597 | dod=dod, |
| 7598 | ) |
| 7599 | |
| 7600 | assert queued |
| 7601 | assert context.workflow_mode == "verify" |
| 7602 | assert "All explicitly planned artifacts already exist." in queued[0] |
| 7603 | assert "Verify all guide files are linked and complete" in queued[0] |
| 7604 | assert "Do not reopen earlier reference materials." in queued[0] |
| 7605 | assert "Finish with a final response so Loader can verify" in queued[0] |
| 7606 | |
| 7607 | |
| 7608 | def test_tool_batch_runner_blocked_post_build_audit_nudge_switches_to_verify( |
| 7609 | temp_dir: Path, |
| 7610 | ) -> None: |
| 7611 | async def assess_confidence( |
| 7612 | tool_name: str, |
| 7613 | tool_args: dict, |
| 7614 | context: str, |
| 7615 | ) -> ConfidenceAssessment: |
| 7616 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 7617 | |
| 7618 | async def verify_action( |
| 7619 | tool_name: str, |
| 7620 | tool_args: dict, |
| 7621 | result: str, |
| 7622 | expected: str = "", |
| 7623 | ) -> ActionVerification: |
| 7624 | raise AssertionError("Verification should not run in this scenario") |
| 7625 | |
| 7626 | guide_root = temp_dir / "guide" |
| 7627 | chapters = guide_root / "chapters" |
| 7628 | guide_root.mkdir(parents=True) |
| 7629 | chapters.mkdir() |
| 7630 | index_path = guide_root / "index.html" |
| 7631 | chapter_one = chapters / "01-getting-started.html" |
| 7632 | chapter_two = chapters / "02-installation.html" |
| 7633 | index_path.write_text("index") |
| 7634 | chapter_one.write_text("one") |
| 7635 | chapter_two.write_text("two") |
| 7636 | |
| 7637 | implementation_plan = temp_dir / "implementation.md" |
| 7638 | implementation_plan.write_text( |
| 7639 | "\n".join( |
| 7640 | [ |
| 7641 | "# Implementation Plan", |
| 7642 | "", |
| 7643 | "## File Changes", |
| 7644 | f"- `{guide_root}`", |
| 7645 | f"- `{chapters}`", |
| 7646 | f"- `{index_path}`", |
| 7647 | f"- `{chapter_one}`", |
| 7648 | f"- `{chapter_two}`", |
| 7649 | "", |
| 7650 | ] |
| 7651 | ) |
| 7652 | ) |
| 7653 | |
| 7654 | context = build_context( |
| 7655 | temp_dir=temp_dir, |
| 7656 | messages=[], |
| 7657 | safeguards=FakeSafeguards(), |
| 7658 | assess_confidence=assess_confidence, |
| 7659 | verify_action=verify_action, |
| 7660 | ) |
| 7661 | queued: list[str] = [] |
| 7662 | context.queue_steering_message_callback = queued.append |
| 7663 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 7664 | dod = create_definition_of_done("Create a multi-file guide from a reference") |
| 7665 | dod.implementation_plan = str(implementation_plan) |
| 7666 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 7667 | |
| 7668 | runner._queue_blocked_completed_artifact_scope_nudge( |
| 7669 | "[Blocked - post-build audit loop: all explicitly planned artifacts already exist.]", |
| 7670 | dod=dod, |
| 7671 | ) |
| 7672 | |
| 7673 | assert queued |
| 7674 | assert context.workflow_mode == "verify" |
| 7675 | assert "All explicitly planned artifacts already exist." in queued[0] |
| 7676 | assert "finish with a final response so Loader can verify" in queued[0] |
| 7677 | |
| 7678 | |
| 7679 | @pytest.mark.asyncio |
| 7680 | async def test_tool_batch_runner_does_not_halt_on_repeated_post_build_audit_blocks( |
| 7681 | temp_dir: Path, |
| 7682 | ) -> None: |
| 7683 | async def assess_confidence( |
| 7684 | tool_name: str, |
| 7685 | tool_args: dict, |
| 7686 | context: str, |
| 7687 | ) -> ConfidenceAssessment: |
| 7688 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 7689 | |
| 7690 | async def verify_action( |
| 7691 | tool_name: str, |
| 7692 | tool_args: dict, |
| 7693 | result: str, |
| 7694 | expected: str = "", |
| 7695 | ) -> ActionVerification: |
| 7696 | raise AssertionError("Verification should not run in this scenario") |
| 7697 | |
| 7698 | guide_root = temp_dir / "guide" |
| 7699 | chapters = guide_root / "chapters" |
| 7700 | guide_root.mkdir(parents=True) |
| 7701 | chapters.mkdir() |
| 7702 | index_path = guide_root / "index.html" |
| 7703 | chapter_one = chapters / "01-getting-started.html" |
| 7704 | chapter_two = chapters / "02-installation.html" |
| 7705 | index_path.write_text("index") |
| 7706 | chapter_one.write_text("one") |
| 7707 | chapter_two.write_text("two") |
| 7708 | |
| 7709 | implementation_plan = temp_dir / "implementation.md" |
| 7710 | implementation_plan.write_text( |
| 7711 | "\n".join( |
| 7712 | [ |
| 7713 | "# Implementation Plan", |
| 7714 | "", |
| 7715 | "## File Changes", |
| 7716 | f"- `{guide_root}`", |
| 7717 | f"- `{chapters}`", |
| 7718 | f"- `{index_path}`", |
| 7719 | f"- `{chapter_one}`", |
| 7720 | f"- `{chapter_two}`", |
| 7721 | "", |
| 7722 | ] |
| 7723 | ) |
| 7724 | ) |
| 7725 | |
| 7726 | context = build_context( |
| 7727 | temp_dir=temp_dir, |
| 7728 | messages=[], |
| 7729 | safeguards=FakeSafeguards(), |
| 7730 | assess_confidence=assess_confidence, |
| 7731 | verify_action=verify_action, |
| 7732 | ) |
| 7733 | queued: list[str] = [] |
| 7734 | context.queue_steering_message_callback = queued.append |
| 7735 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 7736 | dod = create_definition_of_done("Create a multi-file guide from a reference") |
| 7737 | dod.implementation_plan = str(implementation_plan) |
| 7738 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 7739 | |
| 7740 | blocked_message = ( |
| 7741 | "[Blocked - post-build audit loop: all explicitly planned artifacts already exist.]" |
| 7742 | ) |
| 7743 | tool_calls = [ |
| 7744 | ToolCall( |
| 7745 | id=f"audit-{index}", |
| 7746 | name="bash", |
| 7747 | arguments={"command": f"cd {temp_dir} && ls -la guide/chapters/"}, |
| 7748 | ) |
| 7749 | for index in range(1, 4) |
| 7750 | ] |
| 7751 | executor = FakeExecutor( |
| 7752 | [ |
| 7753 | tool_outcome( |
| 7754 | tool_call=tool_call, |
| 7755 | output=blocked_message, |
| 7756 | is_error=True, |
| 7757 | state=ToolExecutionState.BLOCKED, |
| 7758 | ) |
| 7759 | for tool_call in tool_calls |
| 7760 | ] |
| 7761 | ) |
| 7762 | events: list[AgentEvent] = [] |
| 7763 | |
| 7764 | async def emit(event: AgentEvent) -> None: |
| 7765 | events.append(event) |
| 7766 | |
| 7767 | result = await runner.execute_batch( |
| 7768 | tool_calls=tool_calls, |
| 7769 | tool_source="native", |
| 7770 | pending_tool_calls_seen=set(), |
| 7771 | emit=emit, |
| 7772 | summary=TurnSummary(final_response=""), |
| 7773 | dod=dod, |
| 7774 | executor=executor, |
| 7775 | on_confirmation=None, |
| 7776 | on_user_question=None, |
| 7777 | emit_confirmation=None, |
| 7778 | consecutive_errors=0, |
| 7779 | ) |
| 7780 | |
| 7781 | assert result.halted is False |
| 7782 | assert result.consecutive_errors == 0 |
| 7783 | assert context.workflow_mode == "verify" |
| 7784 | assert queued |
| 7785 | assert any("finish with a final response so Loader can verify" in message for message in queued) |
| 7786 | |
| 7787 | |
| 7788 | def test_tool_batch_runner_blocked_html_declared_target_nudge_uses_closest_declared_target( |
| 7789 | temp_dir: Path, |
| 7790 | ) -> None: |
| 7791 | async def assess_confidence( |
| 7792 | tool_name: str, |
| 7793 | tool_args: dict, |
| 7794 | context: str, |
| 7795 | ) -> ConfidenceAssessment: |
| 7796 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 7797 | |
| 7798 | async def verify_action( |
| 7799 | tool_name: str, |
| 7800 | tool_args: dict, |
| 7801 | result: str, |
| 7802 | expected: str = "", |
| 7803 | ) -> ActionVerification: |
| 7804 | raise AssertionError("Verification should not run in this scenario") |
| 7805 | |
| 7806 | context = build_context( |
| 7807 | temp_dir=temp_dir, |
| 7808 | messages=[], |
| 7809 | safeguards=FakeSafeguards(), |
| 7810 | assess_confidence=assess_confidence, |
| 7811 | verify_action=verify_action, |
| 7812 | ) |
| 7813 | queued: list[str] = [] |
| 7814 | context.queue_steering_message_callback = queued.append |
| 7815 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 7816 | |
| 7817 | runner._queue_blocked_html_declared_target_nudge( |
| 7818 | ToolCall( |
| 7819 | id="write-ch1", |
| 7820 | name="write", |
| 7821 | arguments={"file_path": str(temp_dir / "guide" / "chapters" / "01-introduction.html")}, |
| 7822 | ), |
| 7823 | ( |
| 7824 | "[Blocked - HTML page introduces new local targets outside the current declared artifact set] " |
| 7825 | "Suggestion: Keep non-root HTML pages within the root-declared local-link set and avoid " |
| 7826 | "introducing new sibling targets that the guide root does not declare, for example fix: 02-setup.html. " |
| 7827 | "Already-declared local targets include: chapters/01-introduction.html, chapters/02-installation.html, " |
| 7828 | "chapters/03-configuration.html. Closest declared local targets include: chapters/02-installation.html" |
| 7829 | ), |
| 7830 | ) |
| 7831 | |
| 7832 | assert queued |
| 7833 | assert str(temp_dir / "guide" / "chapters" / "01-introduction.html") in queued[0] |
| 7834 | assert "`chapters/02-installation.html`" in queued[0] |
| 7835 | assert "same file now" in queued[0] |
| 7836 | |
| 7837 | |
| 7838 | def test_tool_batch_runner_blocked_html_declared_target_nudge_without_close_match( |
| 7839 | temp_dir: Path, |
| 7840 | ) -> None: |
| 7841 | async def assess_confidence( |
| 7842 | tool_name: str, |
| 7843 | tool_args: dict, |
| 7844 | context: str, |
| 7845 | ) -> ConfidenceAssessment: |
| 7846 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 7847 | |
| 7848 | async def verify_action( |
| 7849 | tool_name: str, |
| 7850 | tool_args: dict, |
| 7851 | result: str, |
| 7852 | expected: str = "", |
| 7853 | ) -> ActionVerification: |
| 7854 | raise AssertionError("Verification should not run in this scenario") |
| 7855 | |
| 7856 | context = build_context( |
| 7857 | temp_dir=temp_dir, |
| 7858 | messages=[], |
| 7859 | safeguards=FakeSafeguards(), |
| 7860 | assess_confidence=assess_confidence, |
| 7861 | verify_action=verify_action, |
| 7862 | ) |
| 7863 | queued: list[str] = [] |
| 7864 | context.queue_steering_message_callback = queued.append |
| 7865 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 7866 | |
| 7867 | runner._queue_blocked_html_declared_target_nudge( |
| 7868 | ToolCall( |
| 7869 | id="write-ch1", |
| 7870 | name="write", |
| 7871 | arguments={"file_path": str(temp_dir / "guide" / "chapters" / "introduction.html")}, |
| 7872 | ), |
| 7873 | ( |
| 7874 | "[Blocked - HTML page introduces new local targets outside the current declared artifact set] " |
| 7875 | "Suggestion: Keep non-root HTML pages within the root-declared local-link set and avoid " |
| 7876 | "introducing new sibling targets that the guide root does not declare; remove or replace " |
| 7877 | "undeclared hrefs like: troubleshooting.html. " |
| 7878 | "Already-declared local targets include: chapters/introduction.html, chapters/installation.html, " |
| 7879 | "chapters/configuration.html. Allowed hrefs from this file include: ../index.html, " |
| 7880 | "installation.html, configuration.html." |
| 7881 | ), |
| 7882 | ) |
| 7883 | |
| 7884 | assert queued |
| 7885 | assert "use only these exact href values" in queued[0] |
| 7886 | assert "`installation.html`" in queued[0] |
| 7887 | assert "`../index.html`" in queued[0] |
| 7888 | assert "closest declared target(s)" not in queued[0] |
| 7889 | |
| 7890 | |
| 7891 | def test_tool_batch_runner_blocked_html_declared_file_creation_nudge_points_to_root( |
| 7892 | temp_dir: Path, |
| 7893 | ) -> None: |
| 7894 | async def assess_confidence( |
| 7895 | tool_name: str, |
| 7896 | tool_args: dict, |
| 7897 | context: str, |
| 7898 | ) -> ConfidenceAssessment: |
| 7899 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 7900 | |
| 7901 | async def verify_action( |
| 7902 | tool_name: str, |
| 7903 | tool_args: dict, |
| 7904 | result: str, |
| 7905 | expected: str = "", |
| 7906 | ) -> ActionVerification: |
| 7907 | raise AssertionError("Verification should not run in this scenario") |
| 7908 | |
| 7909 | context = build_context( |
| 7910 | temp_dir=temp_dir, |
| 7911 | messages=[], |
| 7912 | safeguards=FakeSafeguards(), |
| 7913 | assess_confidence=assess_confidence, |
| 7914 | verify_action=verify_action, |
| 7915 | ) |
| 7916 | queued: list[str] = [] |
| 7917 | context.queue_steering_message_callback = queued.append |
| 7918 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 7919 | dod = create_definition_of_done("Create a guide.") |
| 7920 | |
| 7921 | target = temp_dir / "guide" / "chapters" / "troubleshooting.html" |
| 7922 | runner._queue_blocked_html_declared_file_creation_nudge( |
| 7923 | ToolCall( |
| 7924 | id="write-troubleshooting", |
| 7925 | name="write", |
| 7926 | arguments={"file_path": str(target)}, |
| 7927 | ), |
| 7928 | ( |
| 7929 | "[Blocked - HTML file creation falls outside the current declared artifact set] " |
| 7930 | "Suggestion: Keep new non-root HTML files within the root-declared artifact set and " |
| 7931 | f"update the guide root `{(temp_dir / 'guide' / 'index.html').resolve(strict=False)}` " |
| 7932 | "before creating undeclared sibling pages, for example: chapters/troubleshooting.html. " |
| 7933 | "Already-declared local targets include: chapters/advanced-topics.html, " |
| 7934 | "chapters/basic-usage.html, chapters/configuration.html" |
| 7935 | ), |
| 7936 | dod=dod, |
| 7937 | ) |
| 7938 | |
| 7939 | assert queued |
| 7940 | assert "update" in queued[0].lower() |
| 7941 | assert str((temp_dir / "guide" / "index.html").resolve(strict=False)) in queued[0] |
| 7942 | assert "`chapters/troubleshooting.html`" in queued[0] |
| 7943 | assert "retry the file creation" in queued[0] |
| 7944 | |
| 7945 | |
| 7946 | def test_tool_batch_runner_blocked_html_declared_file_creation_after_outputs_exist_prefers_verify( |
| 7947 | temp_dir: Path, |
| 7948 | ) -> None: |
| 7949 | async def assess_confidence( |
| 7950 | tool_name: str, |
| 7951 | tool_args: dict, |
| 7952 | context: str, |
| 7953 | ) -> ConfidenceAssessment: |
| 7954 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 7955 | |
| 7956 | async def verify_action( |
| 7957 | tool_name: str, |
| 7958 | tool_args: dict, |
| 7959 | result: str, |
| 7960 | expected: str = "", |
| 7961 | ) -> ActionVerification: |
| 7962 | raise AssertionError("Verification should not run in this scenario") |
| 7963 | |
| 7964 | guide = temp_dir / "guide" |
| 7965 | chapters = guide / "chapters" |
| 7966 | guide.mkdir() |
| 7967 | chapters.mkdir() |
| 7968 | index = guide / "index.html" |
| 7969 | index.write_text( |
| 7970 | "\n".join( |
| 7971 | [ |
| 7972 | '<a href="chapters/01-introduction.html">Intro</a>', |
| 7973 | '<a href="chapters/02-installation.html">Install</a>', |
| 7974 | '<a href="../index.html">Back</a>', |
| 7975 | "", |
| 7976 | ] |
| 7977 | ) |
| 7978 | ) |
| 7979 | (chapters / "01-introduction.html").write_text("<html></html>\n") |
| 7980 | (chapters / "02-installation.html").write_text("<html></html>\n") |
| 7981 | |
| 7982 | implementation_plan = temp_dir / "implementation.md" |
| 7983 | implementation_plan.write_text( |
| 7984 | "\n".join( |
| 7985 | [ |
| 7986 | "# Implementation Plan", |
| 7987 | "", |
| 7988 | "## File Changes", |
| 7989 | f"- `{index}`", |
| 7990 | f"- `{chapters / '01-introduction.html'}`", |
| 7991 | f"- `{chapters / '02-installation.html'}`", |
| 7992 | "", |
| 7993 | ] |
| 7994 | ) |
| 7995 | ) |
| 7996 | |
| 7997 | context = build_context( |
| 7998 | temp_dir=temp_dir, |
| 7999 | messages=[], |
| 8000 | safeguards=FakeSafeguards(), |
| 8001 | assess_confidence=assess_confidence, |
| 8002 | verify_action=verify_action, |
| 8003 | ) |
| 8004 | queued: list[str] = [] |
| 8005 | context.queue_steering_message_callback = queued.append |
| 8006 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 8007 | dod = create_definition_of_done("Create a guide.") |
| 8008 | dod.implementation_plan = str(implementation_plan) |
| 8009 | dod.verification_commands = [f"ls -la {guide}"] |
| 8010 | dod.touched_files = [str(index), str(chapters / "01-introduction.html"), str(chapters / "02-installation.html")] |
| 8011 | |
| 8012 | target = guide / "chapters" / "08-advanced-configuration.html" |
| 8013 | runner._queue_blocked_html_declared_file_creation_nudge( |
| 8014 | ToolCall( |
| 8015 | id="write-extra", |
| 8016 | name="write", |
| 8017 | arguments={"file_path": str(target)}, |
| 8018 | ), |
| 8019 | ( |
| 8020 | "[Blocked - HTML file creation falls outside the current declared artifact set] " |
| 8021 | "Suggestion: Keep new non-root HTML files within the root-declared artifact set and " |
| 8022 | f"update the guide root `{index.resolve(strict=False)}` before creating undeclared sibling pages, " |
| 8023 | "for example: chapters/08-advanced-configuration.html." |
| 8024 | ), |
| 8025 | dod=dod, |
| 8026 | ) |
| 8027 | |
| 8028 | assert queued |
| 8029 | assert "All explicitly planned artifacts already exist on disk." in queued[0] |
| 8030 | assert "Do not expand the output set with `chapters/08-advanced-configuration.html`." in queued[0] |
| 8031 | assert "Finish with a final response now so Loader can run verification automatically." in queued[0] |
| 8032 | assert "update the guide root" not in queued[0] |
| 8033 | |
| 8034 | |
| 8035 | def test_tool_batch_runner_blocked_html_declared_file_creation_prefers_closest_target( |
| 8036 | temp_dir: Path, |
| 8037 | ) -> None: |
| 8038 | async def assess_confidence( |
| 8039 | tool_name: str, |
| 8040 | tool_args: dict, |
| 8041 | context: str, |
| 8042 | ) -> ConfidenceAssessment: |
| 8043 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 8044 | |
| 8045 | async def verify_action( |
| 8046 | tool_name: str, |
| 8047 | tool_args: dict, |
| 8048 | result: str, |
| 8049 | expected: str = "", |
| 8050 | ) -> ActionVerification: |
| 8051 | raise AssertionError("Verification should not run in this scenario") |
| 8052 | |
| 8053 | context = build_context( |
| 8054 | temp_dir=temp_dir, |
| 8055 | messages=[], |
| 8056 | safeguards=FakeSafeguards(), |
| 8057 | assess_confidence=assess_confidence, |
| 8058 | verify_action=verify_action, |
| 8059 | ) |
| 8060 | queued: list[str] = [] |
| 8061 | context.queue_steering_message_callback = queued.append |
| 8062 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 8063 | dod = create_definition_of_done("Create a guide.") |
| 8064 | |
| 8065 | target = temp_dir / "guide" / "chapters" / "02-basics.html" |
| 8066 | runner._queue_blocked_html_declared_file_creation_nudge( |
| 8067 | ToolCall( |
| 8068 | id="write-basics", |
| 8069 | name="write", |
| 8070 | arguments={"file_path": str(target)}, |
| 8071 | ), |
| 8072 | ( |
| 8073 | "[Blocked - HTML file creation falls outside the current declared artifact set] " |
| 8074 | "Suggestion: Keep new non-root HTML files within the root-declared artifact set. " |
| 8075 | "Do not create undeclared sibling page `chapters/02-basics.html`; use the closest declared local target instead. " |
| 8076 | "Already-declared local targets include: chapters/01-introduction.html, " |
| 8077 | "chapters/02-installation.html, chapters/03-basic-configuration.html. " |
| 8078 | "Closest declared local targets include: chapters/02-installation.html" |
| 8079 | ), |
| 8080 | dod=dod, |
| 8081 | ) |
| 8082 | |
| 8083 | assert queued |
| 8084 | assert "Do not create `chapters/02-basics.html`." in queued[0] |
| 8085 | assert "closest declared target instead: `chapters/02-installation.html`" in queued[0] |
| 8086 | assert "Already-declared local targets include:" in queued[0] |
| 8087 | assert "update the guide root" not in queued[0] |
| 8088 | |
| 8089 | |
| 8090 | def test_tool_batch_runner_blocked_html_missing_target_after_outputs_exist_prefers_verify( |
| 8091 | temp_dir: Path, |
| 8092 | ) -> None: |
| 8093 | async def assess_confidence( |
| 8094 | tool_name: str, |
| 8095 | tool_args: dict, |
| 8096 | context: str, |
| 8097 | ) -> ConfidenceAssessment: |
| 8098 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 8099 | |
| 8100 | async def verify_action( |
| 8101 | tool_name: str, |
| 8102 | tool_args: dict, |
| 8103 | result: str, |
| 8104 | expected: str = "", |
| 8105 | ) -> ActionVerification: |
| 8106 | raise AssertionError("Verification should not run in this scenario") |
| 8107 | |
| 8108 | guide = temp_dir / "guide" |
| 8109 | chapters = guide / "chapters" |
| 8110 | guide.mkdir() |
| 8111 | chapters.mkdir() |
| 8112 | index = guide / "index.html" |
| 8113 | index.write_text( |
| 8114 | "\n".join( |
| 8115 | [ |
| 8116 | '<a href="chapters/01-introduction.html">Intro</a>', |
| 8117 | '<a href="chapters/02-installation.html">Install</a>', |
| 8118 | '<a href="../index.html">Back</a>', |
| 8119 | "", |
| 8120 | ] |
| 8121 | ) |
| 8122 | ) |
| 8123 | (chapters / "01-introduction.html").write_text("<html></html>\n") |
| 8124 | (chapters / "02-installation.html").write_text("<html></html>\n") |
| 8125 | |
| 8126 | implementation_plan = temp_dir / "implementation.md" |
| 8127 | implementation_plan.write_text( |
| 8128 | "\n".join( |
| 8129 | [ |
| 8130 | "# Implementation Plan", |
| 8131 | "", |
| 8132 | "## File Changes", |
| 8133 | f"- `{index}`", |
| 8134 | f"- `{chapters / '01-introduction.html'}`", |
| 8135 | f"- `{chapters / '02-installation.html'}`", |
| 8136 | "", |
| 8137 | ] |
| 8138 | ) |
| 8139 | ) |
| 8140 | |
| 8141 | context = build_context( |
| 8142 | temp_dir=temp_dir, |
| 8143 | messages=[], |
| 8144 | safeguards=FakeSafeguards(), |
| 8145 | assess_confidence=assess_confidence, |
| 8146 | verify_action=verify_action, |
| 8147 | ) |
| 8148 | queued: list[str] = [] |
| 8149 | context.queue_steering_message_callback = queued.append |
| 8150 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 8151 | dod = create_definition_of_done("Create a guide.") |
| 8152 | dod.implementation_plan = str(implementation_plan) |
| 8153 | dod.verification_commands = [f"ls -la {guide}"] |
| 8154 | dod.touched_files = [str(index), str(chapters / "01-introduction.html"), str(chapters / "02-installation.html")] |
| 8155 | |
| 8156 | runner._queue_blocked_html_missing_target_nudge( |
| 8157 | ToolCall( |
| 8158 | id="edit-root", |
| 8159 | name="edit", |
| 8160 | arguments={"file_path": str(index)}, |
| 8161 | ), |
| 8162 | ( |
| 8163 | "[Blocked - Edited HTML links point to files that do not exist] " |
| 8164 | "Suggestion: Use only existing local targets for href values and avoid introducing missing links. " |
| 8165 | "Broken href(s): chapters/08-advanced-configuration.html. " |
| 8166 | "Replace them with an existing local target or remove the broken link." |
| 8167 | ), |
| 8168 | dod=dod, |
| 8169 | ) |
| 8170 | |
| 8171 | assert queued |
| 8172 | assert "All explicitly planned artifacts already exist on disk." in queued[0] |
| 8173 | assert f"Stay on `{index}`." in queued[0] |
| 8174 | assert "Do not introduce new local-link targets beyond the current output set." in queued[0] |
| 8175 | assert "Repair the existing generated files instead of expanding the guide." in queued[0] |
| 8176 | assert "Replace broken hrefs with existing local targets or remove the broken link." in queued[0] |
| 8177 | |
| 8178 | |
| 8179 | def test_tool_batch_runner_blocked_html_asset_nudge_retries_same_file( |
| 8180 | temp_dir: Path, |
| 8181 | ) -> None: |
| 8182 | async def assess_confidence( |
| 8183 | tool_name: str, |
| 8184 | tool_args: dict, |
| 8185 | context: str, |
| 8186 | ) -> ConfidenceAssessment: |
| 8187 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 8188 | |
| 8189 | async def verify_action( |
| 8190 | tool_name: str, |
| 8191 | tool_args: dict, |
| 8192 | result: str, |
| 8193 | expected: str = "", |
| 8194 | ) -> ActionVerification: |
| 8195 | raise AssertionError("Verification should not run in this scenario") |
| 8196 | |
| 8197 | context = build_context( |
| 8198 | temp_dir=temp_dir, |
| 8199 | messages=[], |
| 8200 | safeguards=FakeSafeguards(), |
| 8201 | assess_confidence=assess_confidence, |
| 8202 | verify_action=verify_action, |
| 8203 | ) |
| 8204 | queued: list[str] = [] |
| 8205 | context.queue_steering_message_callback = queued.append |
| 8206 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 8207 | target = temp_dir / "guide" / "chapters" / "03-configuration.html" |
| 8208 | |
| 8209 | runner._queue_blocked_html_asset_nudge( |
| 8210 | ToolCall( |
| 8211 | id="write-config", |
| 8212 | name="write", |
| 8213 | arguments={"file_path": str(target)}, |
| 8214 | ), |
| 8215 | ( |
| 8216 | "[Blocked - HTML local asset references do not exist] Suggestion: " |
| 8217 | "Use only existing local assets for non-HTML href values. " |
| 8218 | "Missing local asset href(s): ../styles.css. Remove the asset link, " |
| 8219 | "create the referenced asset first, inline the styling/content, or point " |
| 8220 | "the href at an existing local file." |
| 8221 | ), |
| 8222 | ) |
| 8223 | |
| 8224 | assert queued |
| 8225 | assert str(target) in queued[0] |
| 8226 | assert "was not created or updated" in queued[0] |
| 8227 | assert "Remove or replace `../styles.css`." in queued[0] |
| 8228 | assert "Do not resend the same `<link>` tag" in queued[0] |
| 8229 | assert "do not claim completion" in queued[0] |
| 8230 | |
| 8231 | |
| 8232 | def test_tool_batch_runner_repeated_blocked_html_asset_nudge_forces_href_removal( |
| 8233 | temp_dir: Path, |
| 8234 | ) -> None: |
| 8235 | async def assess_confidence( |
| 8236 | tool_name: str, |
| 8237 | tool_args: dict, |
| 8238 | context: str, |
| 8239 | ) -> ConfidenceAssessment: |
| 8240 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 8241 | |
| 8242 | async def verify_action( |
| 8243 | tool_name: str, |
| 8244 | tool_args: dict, |
| 8245 | result: str, |
| 8246 | expected: str = "", |
| 8247 | ) -> ActionVerification: |
| 8248 | raise AssertionError("Verification should not run in this scenario") |
| 8249 | |
| 8250 | blocked_event = ( |
| 8251 | "[Blocked - HTML local asset references do not exist] Suggestion: " |
| 8252 | "Use only existing local assets for non-HTML href values. " |
| 8253 | "Missing local asset href(s): ../style.css. Remove the asset link, " |
| 8254 | "create the referenced asset first, inline the styling/content, or point " |
| 8255 | "the href at an existing local file." |
| 8256 | ) |
| 8257 | context = build_context( |
| 8258 | temp_dir=temp_dir, |
| 8259 | messages=[Message(role=Role.TOOL, content=blocked_event)], |
| 8260 | safeguards=FakeSafeguards(), |
| 8261 | assess_confidence=assess_confidence, |
| 8262 | verify_action=verify_action, |
| 8263 | ) |
| 8264 | context.session.append(Message(role=Role.TOOL, content=blocked_event)) |
| 8265 | queued: list[str] = [] |
| 8266 | context.queue_steering_message_callback = queued.append |
| 8267 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 8268 | target = temp_dir / "guide" / "chapters" / "05-troubleshooting.html" |
| 8269 | |
| 8270 | runner._queue_blocked_html_asset_nudge( |
| 8271 | ToolCall( |
| 8272 | id="write-troubleshooting", |
| 8273 | name="write", |
| 8274 | arguments={"file_path": str(target)}, |
| 8275 | ), |
| 8276 | blocked_event, |
| 8277 | ) |
| 8278 | |
| 8279 | assert queued |
| 8280 | assert "blocked 2 times" in queued[0] |
| 8281 | assert "`../style.css`" in queued[0] |
| 8282 | assert "line removed" in queued[0] |
| 8283 | assert "Do not resend another" in queued[0] |
| 8284 | |
| 8285 | |
| 8286 | @pytest.mark.asyncio |
| 8287 | async def test_tool_batch_runner_blocked_empty_file_path_nudges_concrete_next_artifact( |
| 8288 | temp_dir: Path, |
| 8289 | ) -> None: |
| 8290 | async def assess_confidence( |
| 8291 | tool_name: str, |
| 8292 | tool_args: dict, |
| 8293 | context: str, |
| 8294 | ) -> ConfidenceAssessment: |
| 8295 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 8296 | |
| 8297 | async def verify_action( |
| 8298 | tool_name: str, |
| 8299 | tool_args: dict, |
| 8300 | result: str, |
| 8301 | expected: str = "", |
| 8302 | ) -> ActionVerification: |
| 8303 | raise AssertionError("Verification should not run in this scenario") |
| 8304 | |
| 8305 | guide_root = temp_dir / "guides" / "nginx" |
| 8306 | chapters = guide_root / "chapters" |
| 8307 | chapters.mkdir(parents=True) |
| 8308 | index_path = guide_root / "index.html" |
| 8309 | chapter_one = chapters / "01-introduction.html" |
| 8310 | chapter_two = chapters / "02-installation.html" |
| 8311 | index_path.write_text("<html></html>\n") |
| 8312 | chapter_one.write_text("<h1>Intro</h1>\n") |
| 8313 | |
| 8314 | implementation_plan = temp_dir / "implementation.md" |
| 8315 | implementation_plan.write_text( |
| 8316 | "\n".join( |
| 8317 | [ |
| 8318 | "# Implementation Plan", |
| 8319 | "", |
| 8320 | "## File Changes", |
| 8321 | f"- `{index_path}`", |
| 8322 | f"- `{chapter_one}`", |
| 8323 | f"- `{chapter_two}`", |
| 8324 | "", |
| 8325 | ] |
| 8326 | ) |
| 8327 | ) |
| 8328 | |
| 8329 | context = build_context( |
| 8330 | temp_dir=temp_dir, |
| 8331 | messages=[], |
| 8332 | safeguards=FakeSafeguards(), |
| 8333 | assess_confidence=assess_confidence, |
| 8334 | verify_action=verify_action, |
| 8335 | auto_recover=False, |
| 8336 | ) |
| 8337 | queued: list[str] = [] |
| 8338 | context.queue_steering_message_callback = queued.append |
| 8339 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 8340 | tool_call = ToolCall( |
| 8341 | id="write-2", |
| 8342 | name="write", |
| 8343 | arguments={"file_path": "", "content": "<html></html>\n"}, |
| 8344 | ) |
| 8345 | blocked_message = "[Blocked - Empty file path] Suggestion: Provide a valid file path" |
| 8346 | executor = FakeExecutor( |
| 8347 | [ |
| 8348 | ToolExecutionOutcome( |
| 8349 | tool_call=tool_call, |
| 8350 | state=ToolExecutionState.BLOCKED, |
| 8351 | message=Message.tool_result_message( |
| 8352 | tool_call_id=tool_call.id, |
| 8353 | display_content=blocked_message, |
| 8354 | result_content=blocked_message, |
| 8355 | is_error=True, |
| 8356 | ), |
| 8357 | event_content=blocked_message, |
| 8358 | is_error=True, |
| 8359 | result_output=blocked_message, |
| 8360 | ) |
| 8361 | ] |
| 8362 | ) |
| 8363 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 8364 | dod.implementation_plan = str(implementation_plan) |
| 8365 | dod.touched_files.extend([str(index_path), str(chapter_one)]) |
| 8366 | dod.pending_items.append("Creating Chapter 2: Installation and Setup") |
| 8367 | |
| 8368 | await runner.execute_batch( |
| 8369 | tool_calls=[tool_call], |
| 8370 | tool_source="assistant", |
| 8371 | pending_tool_calls_seen=set(), |
| 8372 | emit=_noop_emit, |
| 8373 | summary=TurnSummary(final_response=""), |
| 8374 | dod=dod, |
| 8375 | executor=executor, # type: ignore[arg-type] |
| 8376 | on_confirmation=None, |
| 8377 | on_user_question=None, |
| 8378 | emit_confirmation=None, |
| 8379 | consecutive_errors=0, |
| 8380 | ) |
| 8381 | |
| 8382 | assert queued |
| 8383 | assert "did not provide a valid `file_path`" in queued[0] |
| 8384 | assert "Resume by creating `02-installation.html` now." in queued[0] |
| 8385 | assert ( |
| 8386 | f"Prefer one `write` call for `{display_runtime_path(chapter_two)}` instead of more rereads." |
| 8387 | in queued[0] |
| 8388 | ) |
| 8389 | assert context.recovery_context is not None |
| 8390 | assert context.recovery_context.attempts[-1].error == blocked_message |