| 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.permissions import ( |
| 20 | PermissionMode, |
| 21 | build_permission_policy, |
| 22 | load_permission_rules, |
| 23 | ) |
| 24 | from loader.runtime.reasoning_types import ( |
| 25 | ActionVerification, |
| 26 | ConfidenceAssessment, |
| 27 | ConfidenceLevel, |
| 28 | ) |
| 29 | from loader.runtime.recovery import RecoveryContext |
| 30 | from loader.runtime.tool_batches import ( |
| 31 | ToolBatchRunner, |
| 32 | ) |
| 33 | from loader.runtime.tool_batches import ( |
| 34 | _should_prioritize_missing_artifact as tool_batches_should_prioritize_missing_artifact, |
| 35 | ) |
| 36 | from loader.runtime.workflow import sync_todos_to_definition_of_done |
| 37 | from loader.tools.base import ToolResult as RegistryToolResult |
| 38 | from loader.tools.base import create_default_registry |
| 39 | from tests.helpers.runtime_harness import ScriptedBackend |
| 40 | |
| 41 | |
| 42 | class FakeSession: |
| 43 | def __init__(self, messages: list[Message]) -> None: |
| 44 | self.messages = list(messages) |
| 45 | self.workflow_timeline = [] |
| 46 | |
| 47 | def append(self, message: Message) -> None: |
| 48 | self.messages.append(message) |
| 49 | |
| 50 | def append_workflow_timeline_entry(self, entry) -> None: |
| 51 | self.workflow_timeline.append(entry) |
| 52 | |
| 53 | |
| 54 | class FakeCodeFilter: |
| 55 | def reset(self) -> None: |
| 56 | return None |
| 57 | |
| 58 | |
| 59 | class FakeSafeguards: |
| 60 | def __init__(self, *, detect_loop_result: tuple[bool, str] = (False, "")) -> None: |
| 61 | self.action_tracker = object() |
| 62 | self.validator = object() |
| 63 | self.code_filter = FakeCodeFilter() |
| 64 | self._detect_loop_result = detect_loop_result |
| 65 | |
| 66 | def filter_stream_chunk(self, content: str) -> str: |
| 67 | return content |
| 68 | |
| 69 | def filter_complete_content(self, content: str) -> str: |
| 70 | return content |
| 71 | |
| 72 | def should_steer(self) -> bool: |
| 73 | return False |
| 74 | |
| 75 | def get_steering_message(self) -> str | None: |
| 76 | return None |
| 77 | |
| 78 | def record_response(self, content: str) -> None: |
| 79 | return None |
| 80 | |
| 81 | def detect_text_loop(self, content: str) -> tuple[bool, str]: |
| 82 | return False, "" |
| 83 | |
| 84 | def detect_loop(self) -> tuple[bool, str]: |
| 85 | return self._detect_loop_result |
| 86 | |
| 87 | |
| 88 | class FakeExecutor: |
| 89 | def __init__(self, outcomes: list[ToolExecutionOutcome]) -> None: |
| 90 | self._outcomes = list(outcomes) |
| 91 | self.calls: list[ToolCall] = [] |
| 92 | |
| 93 | async def execute_tool_call(self, tool_call: ToolCall, **_: object) -> ToolExecutionOutcome: |
| 94 | self.calls.append(tool_call) |
| 95 | if not self._outcomes: |
| 96 | raise AssertionError("No fake tool outcome queued") |
| 97 | return self._outcomes.pop(0) |
| 98 | |
| 99 | |
| 100 | def build_context( |
| 101 | *, |
| 102 | temp_dir: Path, |
| 103 | messages: list[Message], |
| 104 | safeguards: FakeSafeguards, |
| 105 | assess_confidence, |
| 106 | verify_action, |
| 107 | recovery_context: RecoveryContext | None = None, |
| 108 | confidence_scoring: bool = False, |
| 109 | verification: bool = False, |
| 110 | auto_recover: bool = True, |
| 111 | min_confidence_for_action: int = 3, |
| 112 | ) -> RuntimeContext: |
| 113 | registry = create_default_registry(temp_dir) |
| 114 | registry.configure_workspace_root(temp_dir) |
| 115 | rule_status = load_permission_rules(temp_dir) |
| 116 | policy = build_permission_policy( |
| 117 | active_mode=PermissionMode.WORKSPACE_WRITE, |
| 118 | workspace_root=temp_dir, |
| 119 | tool_requirements=registry.get_tool_requirements(), |
| 120 | rules=rule_status.rules, |
| 121 | ) |
| 122 | context = RuntimeContext( |
| 123 | project_root=temp_dir, |
| 124 | backend=ScriptedBackend(), |
| 125 | registry=registry, |
| 126 | session=FakeSession(messages), # type: ignore[arg-type] |
| 127 | config=SimpleNamespace( |
| 128 | force_react=False, |
| 129 | max_recovery_attempts=2, |
| 130 | auto_recover=auto_recover, |
| 131 | reasoning=SimpleNamespace( |
| 132 | rollback=False, |
| 133 | show_rollback_plan=False, |
| 134 | completion_check=True, |
| 135 | max_continuation_prompts=5, |
| 136 | self_critique=False, |
| 137 | confidence_scoring=confidence_scoring, |
| 138 | min_confidence_for_action=min_confidence_for_action, |
| 139 | verification=verification, |
| 140 | ), |
| 141 | ), |
| 142 | capability_profile=SimpleNamespace(supports_native_tools=True), # type: ignore[arg-type] |
| 143 | project_context=None, |
| 144 | permission_policy=policy, |
| 145 | permission_config_status=rule_status, |
| 146 | workflow_mode="execute", |
| 147 | safeguards=safeguards, |
| 148 | reasoning=SimpleNamespace( |
| 149 | assess_confidence=assess_confidence, |
| 150 | verify_action=verify_action, |
| 151 | ), |
| 152 | recovery_context=recovery_context, |
| 153 | ) |
| 154 | return context |
| 155 | |
| 156 | |
| 157 | def tool_outcome( |
| 158 | *, |
| 159 | tool_call: ToolCall, |
| 160 | output: str, |
| 161 | is_error: bool, |
| 162 | state: ToolExecutionState = ToolExecutionState.EXECUTED, |
| 163 | metadata: dict[str, object] | None = None, |
| 164 | ) -> ToolExecutionOutcome: |
| 165 | return ToolExecutionOutcome( |
| 166 | tool_call=tool_call, |
| 167 | state=state, |
| 168 | message=Message.tool_result_message( |
| 169 | tool_call_id=tool_call.id, |
| 170 | display_content=output, |
| 171 | result_content=output, |
| 172 | is_error=is_error, |
| 173 | ), |
| 174 | event_content=output, |
| 175 | is_error=is_error, |
| 176 | result_output=output, |
| 177 | registry_result=RegistryToolResult( |
| 178 | output=output, |
| 179 | is_error=is_error, |
| 180 | metadata=metadata or {}, |
| 181 | ), |
| 182 | ) |
| 183 | |
| 184 | |
| 185 | @pytest.mark.asyncio |
| 186 | async def test_tool_batch_runner_uses_context_for_confidence_gate(temp_dir: Path) -> None: |
| 187 | captured: dict[str, str] = {} |
| 188 | |
| 189 | async def assess_confidence(tool_name: str, tool_args: dict, context: str) -> ConfidenceAssessment: |
| 190 | captured["context"] = context |
| 191 | return ConfidenceAssessment( |
| 192 | action=f"{tool_name} with {tool_args}", |
| 193 | tool_name=tool_name, |
| 194 | tool_args=tool_args, |
| 195 | level=ConfidenceLevel.LOW, |
| 196 | reasoning="Need to inspect the target first.", |
| 197 | risks=["Unknown target file"], |
| 198 | ) |
| 199 | |
| 200 | async def verify_action(tool_name: str, tool_args: dict, result: str, expected: str = "") -> ActionVerification: |
| 201 | raise AssertionError("Verification should not run for skipped actions") |
| 202 | |
| 203 | context = build_context( |
| 204 | temp_dir=temp_dir, |
| 205 | messages=[ |
| 206 | Message(role=Role.USER, content="Please inspect the project."), |
| 207 | Message(role=Role.ASSISTANT, content="I will read the file next."), |
| 208 | ], |
| 209 | safeguards=FakeSafeguards(), |
| 210 | assess_confidence=assess_confidence, |
| 211 | verify_action=verify_action, |
| 212 | confidence_scoring=True, |
| 213 | min_confidence_for_action=3, |
| 214 | ) |
| 215 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 216 | tool_call = ToolCall(id="read-1", name="read", arguments={"file_path": "README.md"}) |
| 217 | events: list[AgentEvent] = [] |
| 218 | |
| 219 | async def emit(event: AgentEvent) -> None: |
| 220 | events.append(event) |
| 221 | |
| 222 | executor = FakeExecutor([tool_outcome(tool_call=tool_call, output="unused", is_error=False)]) |
| 223 | result = await runner.execute_batch( |
| 224 | tool_calls=[tool_call], |
| 225 | tool_source="assistant", |
| 226 | pending_tool_calls_seen=set(), |
| 227 | emit=emit, |
| 228 | summary=TurnSummary(final_response=""), |
| 229 | dod=create_definition_of_done("Read the docs"), |
| 230 | executor=executor, # type: ignore[arg-type] |
| 231 | on_confirmation=None, |
| 232 | on_user_question=None, |
| 233 | emit_confirmation=None, |
| 234 | consecutive_errors=0, |
| 235 | ) |
| 236 | |
| 237 | assert result.actions_taken == [] |
| 238 | assert executor.calls == [] |
| 239 | assert "Please inspect the project." in captured["context"] |
| 240 | assert context.session.messages[-1].role == Role.USER |
| 241 | assert "[LOW CONFIDENCE WARNING]" in context.session.messages[-1].content |
| 242 | event_types = [event.type for event in events] |
| 243 | assert "confidence" in event_types |
| 244 | |
| 245 | |
| 246 | @pytest.mark.asyncio |
| 247 | async def test_tool_batch_runner_tracks_recovery_with_legacy_context(temp_dir: Path) -> None: |
| 248 | async def assess_confidence(tool_name: str, tool_args: dict, context: str) -> ConfidenceAssessment: |
| 249 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 250 | |
| 251 | async def verify_action(tool_name: str, tool_args: dict, result: str, expected: str = "") -> ActionVerification: |
| 252 | raise AssertionError("Verification should not run for failed actions") |
| 253 | |
| 254 | context = build_context( |
| 255 | temp_dir=temp_dir, |
| 256 | messages=[], |
| 257 | safeguards=FakeSafeguards(), |
| 258 | assess_confidence=assess_confidence, |
| 259 | verify_action=verify_action, |
| 260 | auto_recover=True, |
| 261 | ) |
| 262 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 263 | tool_call = ToolCall(id="bash-1", name="bash", arguments={"command": "pytest"}) |
| 264 | executor = FakeExecutor([tool_outcome(tool_call=tool_call, output="command failed", is_error=True)]) |
| 265 | summary = TurnSummary(final_response="") |
| 266 | events: list[AgentEvent] = [] |
| 267 | |
| 268 | async def emit(event: AgentEvent) -> None: |
| 269 | events.append(event) |
| 270 | |
| 271 | await runner.execute_batch( |
| 272 | tool_calls=[tool_call], |
| 273 | tool_source="assistant", |
| 274 | pending_tool_calls_seen=set(), |
| 275 | emit=emit, |
| 276 | summary=summary, |
| 277 | dod=create_definition_of_done("Run tests"), |
| 278 | executor=executor, # type: ignore[arg-type] |
| 279 | on_confirmation=None, |
| 280 | on_user_question=None, |
| 281 | emit_confirmation=None, |
| 282 | consecutive_errors=0, |
| 283 | ) |
| 284 | |
| 285 | assert context.recovery_context is not None |
| 286 | assert summary.tool_result_messages |
| 287 | assert context.session.messages[-1] == summary.tool_result_messages[-1] |
| 288 | assert any(event.type == "recovery" for event in events) |
| 289 | |
| 290 | |
| 291 | @pytest.mark.asyncio |
| 292 | async def test_tool_batch_runner_emits_tool_metadata(temp_dir: Path) -> None: |
| 293 | async def assess_confidence(tool_name: str, tool_args: dict, context: str) -> ConfidenceAssessment: |
| 294 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 295 | |
| 296 | async def verify_action(tool_name: str, tool_args: dict, result: str, expected: str = "") -> ActionVerification: |
| 297 | raise AssertionError("Verification should not run for this scenario") |
| 298 | |
| 299 | context = build_context( |
| 300 | temp_dir=temp_dir, |
| 301 | messages=[], |
| 302 | safeguards=FakeSafeguards(), |
| 303 | assess_confidence=assess_confidence, |
| 304 | verify_action=verify_action, |
| 305 | auto_recover=False, |
| 306 | ) |
| 307 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 308 | tool_call = ToolCall( |
| 309 | id="bash-1", |
| 310 | name="bash", |
| 311 | arguments={"command": "python -m http.server 8000", "background": True}, |
| 312 | ) |
| 313 | metadata = { |
| 314 | "job_id": "bash-1", |
| 315 | "status": "running", |
| 316 | "background": True, |
| 317 | } |
| 318 | executor = FakeExecutor( |
| 319 | [ |
| 320 | tool_outcome( |
| 321 | tool_call=tool_call, |
| 322 | output="Started bash job bash-1", |
| 323 | is_error=False, |
| 324 | metadata=metadata, |
| 325 | ) |
| 326 | ] |
| 327 | ) |
| 328 | events: list[AgentEvent] = [] |
| 329 | |
| 330 | async def emit(event: AgentEvent) -> None: |
| 331 | events.append(event) |
| 332 | |
| 333 | await runner.execute_batch( |
| 334 | tool_calls=[tool_call], |
| 335 | tool_source="assistant", |
| 336 | pending_tool_calls_seen=set(), |
| 337 | emit=emit, |
| 338 | summary=TurnSummary(final_response=""), |
| 339 | dod=create_definition_of_done("Launch a preview server"), |
| 340 | executor=executor, # type: ignore[arg-type] |
| 341 | on_confirmation=None, |
| 342 | on_user_question=None, |
| 343 | emit_confirmation=None, |
| 344 | consecutive_errors=0, |
| 345 | ) |
| 346 | |
| 347 | tool_result = next(event for event in events if event.type == "tool_result") |
| 348 | assert tool_result.tool_metadata == metadata |
| 349 | |
| 350 | |
| 351 | @pytest.mark.asyncio |
| 352 | async def test_tool_batch_runner_verifies_with_context_services(temp_dir: Path) -> None: |
| 353 | verification_calls: list[str] = [] |
| 354 | |
| 355 | async def assess_confidence(tool_name: str, tool_args: dict, context: str) -> ConfidenceAssessment: |
| 356 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 357 | |
| 358 | async def verify_action(tool_name: str, tool_args: dict, result: str, expected: str = "") -> ActionVerification: |
| 359 | verification_calls.append(result) |
| 360 | return ActionVerification( |
| 361 | tool_name=tool_name, |
| 362 | tool_args=tool_args, |
| 363 | expected_outcome="Success", |
| 364 | actual_result=result, |
| 365 | verified=False, |
| 366 | discrepancies=["File contents did not match"], |
| 367 | needs_correction=True, |
| 368 | correction_suggestion="Read the file before editing again.", |
| 369 | ) |
| 370 | |
| 371 | existing_recovery = RecoveryContext( |
| 372 | original_tool="edit", |
| 373 | original_args={"file_path": "README.md"}, |
| 374 | ) |
| 375 | context = build_context( |
| 376 | temp_dir=temp_dir, |
| 377 | messages=[], |
| 378 | safeguards=FakeSafeguards(), |
| 379 | assess_confidence=assess_confidence, |
| 380 | verify_action=verify_action, |
| 381 | recovery_context=existing_recovery, |
| 382 | verification=True, |
| 383 | ) |
| 384 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 385 | tool_call = ToolCall(id="read-1", name="read", arguments={"file_path": "README.md"}) |
| 386 | executor = FakeExecutor([tool_outcome(tool_call=tool_call, output="file contents", is_error=False)]) |
| 387 | events: list[AgentEvent] = [] |
| 388 | |
| 389 | async def emit(event: AgentEvent) -> None: |
| 390 | events.append(event) |
| 391 | |
| 392 | await runner.execute_batch( |
| 393 | tool_calls=[tool_call], |
| 394 | tool_source="assistant", |
| 395 | pending_tool_calls_seen=set(), |
| 396 | emit=emit, |
| 397 | summary=TurnSummary(final_response=""), |
| 398 | dod=create_definition_of_done("Read the docs"), |
| 399 | executor=executor, # type: ignore[arg-type] |
| 400 | on_confirmation=None, |
| 401 | on_user_question=None, |
| 402 | emit_confirmation=None, |
| 403 | consecutive_errors=0, |
| 404 | ) |
| 405 | |
| 406 | assert verification_calls == ["file contents"] |
| 407 | assert context.recovery_context is existing_recovery |
| 408 | assert existing_recovery.successful_steps == [ |
| 409 | ("read", {"file_path": "README.md"}) |
| 410 | ] |
| 411 | assert context.session.messages[-1].role == Role.TOOL |
| 412 | assert context.session.messages[-1].content == "file contents" |
| 413 | assert any(event.type == "verification" for event in events) |
| 414 | |
| 415 | |
| 416 | @pytest.mark.asyncio |
| 417 | async def test_tool_batch_runner_preserves_recovery_context_across_diagnostic_success( |
| 418 | temp_dir: Path, |
| 419 | ) -> None: |
| 420 | async def assess_confidence( |
| 421 | tool_name: str, |
| 422 | tool_args: dict, |
| 423 | context: str, |
| 424 | ) -> ConfidenceAssessment: |
| 425 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 426 | |
| 427 | async def verify_action( |
| 428 | tool_name: str, |
| 429 | tool_args: dict, |
| 430 | result: str, |
| 431 | expected: str = "", |
| 432 | ) -> ActionVerification: |
| 433 | raise AssertionError("Verification should not run for this scenario") |
| 434 | |
| 435 | existing_recovery = RecoveryContext( |
| 436 | original_tool="read", |
| 437 | original_args={"file_path": "chapters/04-data-types.html"}, |
| 438 | ) |
| 439 | existing_recovery.add_attempt( |
| 440 | "read", |
| 441 | {"file_path": "chapters/04-data-types.html"}, |
| 442 | "File not found", |
| 443 | ) |
| 444 | context = build_context( |
| 445 | temp_dir=temp_dir, |
| 446 | messages=[], |
| 447 | safeguards=FakeSafeguards(), |
| 448 | assess_confidence=assess_confidence, |
| 449 | verify_action=verify_action, |
| 450 | recovery_context=existing_recovery, |
| 451 | auto_recover=False, |
| 452 | ) |
| 453 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 454 | tool_call = ToolCall( |
| 455 | id="bash-1", |
| 456 | name="bash", |
| 457 | arguments={"command": "ls chapters"}, |
| 458 | ) |
| 459 | executor = FakeExecutor( |
| 460 | [tool_outcome(tool_call=tool_call, output="01-introduction.html", is_error=False)] |
| 461 | ) |
| 462 | |
| 463 | summary = TurnSummary(final_response="") |
| 464 | await runner.execute_batch( |
| 465 | tool_calls=[tool_call], |
| 466 | tool_source="assistant", |
| 467 | pending_tool_calls_seen=set(), |
| 468 | emit=_noop_emit, |
| 469 | summary=summary, |
| 470 | dod=create_definition_of_done("Fix the chapter links"), |
| 471 | executor=executor, # type: ignore[arg-type] |
| 472 | on_confirmation=None, |
| 473 | on_user_question=None, |
| 474 | emit_confirmation=None, |
| 475 | consecutive_errors=0, |
| 476 | ) |
| 477 | |
| 478 | assert context.recovery_context is existing_recovery |
| 479 | assert existing_recovery.successful_steps == [ |
| 480 | ("bash", {"command": "ls chapters"}) |
| 481 | ] |
| 482 | |
| 483 | |
| 484 | @pytest.mark.asyncio |
| 485 | async def test_tool_batch_runner_clears_recovery_context_after_successful_mutation( |
| 486 | temp_dir: Path, |
| 487 | ) -> None: |
| 488 | async def assess_confidence( |
| 489 | tool_name: str, |
| 490 | tool_args: dict, |
| 491 | context: str, |
| 492 | ) -> ConfidenceAssessment: |
| 493 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 494 | |
| 495 | async def verify_action( |
| 496 | tool_name: str, |
| 497 | tool_args: dict, |
| 498 | result: str, |
| 499 | expected: str = "", |
| 500 | ) -> ActionVerification: |
| 501 | raise AssertionError("Verification should not run for this scenario") |
| 502 | |
| 503 | existing_recovery = RecoveryContext( |
| 504 | original_tool="read", |
| 505 | original_args={"file_path": "chapters/04-data-types.html"}, |
| 506 | ) |
| 507 | existing_recovery.add_attempt( |
| 508 | "read", |
| 509 | {"file_path": "chapters/04-data-types.html"}, |
| 510 | "File not found", |
| 511 | ) |
| 512 | context = build_context( |
| 513 | temp_dir=temp_dir, |
| 514 | messages=[], |
| 515 | safeguards=FakeSafeguards(), |
| 516 | assess_confidence=assess_confidence, |
| 517 | verify_action=verify_action, |
| 518 | recovery_context=existing_recovery, |
| 519 | auto_recover=False, |
| 520 | ) |
| 521 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 522 | tool_call = ToolCall( |
| 523 | id="patch-1", |
| 524 | name="patch", |
| 525 | arguments={ |
| 526 | "file_path": "index.html", |
| 527 | "hunks": [{"old_start": 1, "old_lines": 1, "new_start": 1, "new_lines": 1, "lines": ["-a", "+b"]}], |
| 528 | }, |
| 529 | ) |
| 530 | executor = FakeExecutor( |
| 531 | [tool_outcome(tool_call=tool_call, output="Patched index.html", is_error=False)] |
| 532 | ) |
| 533 | |
| 534 | summary = TurnSummary(final_response="") |
| 535 | await runner.execute_batch( |
| 536 | tool_calls=[tool_call], |
| 537 | tool_source="assistant", |
| 538 | pending_tool_calls_seen=set(), |
| 539 | emit=_noop_emit, |
| 540 | summary=summary, |
| 541 | dod=create_definition_of_done("Fix the chapter links"), |
| 542 | executor=executor, # type: ignore[arg-type] |
| 543 | on_confirmation=None, |
| 544 | on_user_question=None, |
| 545 | emit_confirmation=None, |
| 546 | consecutive_errors=0, |
| 547 | ) |
| 548 | |
| 549 | assert context.recovery_context is None |
| 550 | |
| 551 | |
| 552 | @pytest.mark.asyncio |
| 553 | async def test_tool_batch_runner_queues_duplicate_observation_nudge( |
| 554 | temp_dir: Path, |
| 555 | ) -> None: |
| 556 | async def assess_confidence( |
| 557 | tool_name: str, |
| 558 | tool_args: dict, |
| 559 | context: str, |
| 560 | ) -> ConfidenceAssessment: |
| 561 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 562 | |
| 563 | async def verify_action( |
| 564 | tool_name: str, |
| 565 | tool_args: dict, |
| 566 | result: str, |
| 567 | expected: str = "", |
| 568 | ) -> ActionVerification: |
| 569 | raise AssertionError("Verification should not run for this scenario") |
| 570 | |
| 571 | messages = [ |
| 572 | Message( |
| 573 | role=Role.TOOL, |
| 574 | content=( |
| 575 | "Observation [glob]: Result: " |
| 576 | f"{temp_dir}/chapters/01-introduction.html\n" |
| 577 | f"{temp_dir}/chapters/02-setup.html\n" |
| 578 | f"{temp_dir}/chapters/03-basics.html" |
| 579 | ), |
| 580 | tool_results=[], |
| 581 | ), |
| 582 | Message( |
| 583 | role=Role.ASSISTANT, |
| 584 | content="I already inspected the first chapter title.", |
| 585 | tool_calls=[ |
| 586 | ToolCall( |
| 587 | id="read-ch1", |
| 588 | name="read", |
| 589 | arguments={"file_path": str(temp_dir / 'chapters' / '01-introduction.html')}, |
| 590 | ) |
| 591 | ], |
| 592 | ), |
| 593 | Message.tool_result_message( |
| 594 | tool_call_id="read-ch1", |
| 595 | display_content="<h1>Chapter 1: Introduction to Fortran</h1>\n", |
| 596 | result_content="<h1>Chapter 1: Introduction to Fortran</h1>\n", |
| 597 | ), |
| 598 | Message( |
| 599 | role=Role.ASSISTANT, |
| 600 | content="I should update the index now.", |
| 601 | tool_calls=[ |
| 602 | ToolCall( |
| 603 | id="read-index", |
| 604 | name="read", |
| 605 | arguments={"file_path": str(temp_dir / 'index.html')}, |
| 606 | ) |
| 607 | ], |
| 608 | ), |
| 609 | ] |
| 610 | context = build_context( |
| 611 | temp_dir=temp_dir, |
| 612 | messages=messages, |
| 613 | safeguards=FakeSafeguards(), |
| 614 | assess_confidence=assess_confidence, |
| 615 | verify_action=verify_action, |
| 616 | auto_recover=False, |
| 617 | ) |
| 618 | (temp_dir / "chapters").mkdir() |
| 619 | (temp_dir / "index.html").write_text("<ul></ul>\n") |
| 620 | (temp_dir / "chapters" / "01-introduction.html").write_text("<h1>Intro</h1>\n") |
| 621 | (temp_dir / "chapters" / "02-setup.html").write_text("<h1>Setup</h1>\n") |
| 622 | (temp_dir / "chapters" / "03-basics.html").write_text("<h1>Basics</h1>\n") |
| 623 | implementation_plan = temp_dir / "implementation.md" |
| 624 | implementation_plan.write_text( |
| 625 | "\n".join( |
| 626 | [ |
| 627 | "# Implementation Plan", |
| 628 | "", |
| 629 | "## File Changes", |
| 630 | f"- `{temp_dir / 'index.html'}`", |
| 631 | f"- `{temp_dir / 'chapters' / '01-introduction.html'}`", |
| 632 | f"- `{temp_dir / 'chapters' / '02-setup.html'}`", |
| 633 | f"- `{temp_dir / 'chapters' / '03-basics.html'}`", |
| 634 | f"- `{temp_dir / 'chapters' / '04-variables.html'}`", |
| 635 | ] |
| 636 | ) |
| 637 | ) |
| 638 | context.session.current_task = ( |
| 639 | f"Update {temp_dir / 'index.html'} with the right chapter links." |
| 640 | ) |
| 641 | persistent_messages: list[str] = [] |
| 642 | ephemeral_messages: list[str] = [] |
| 643 | context.queue_steering_message_callback = persistent_messages.append |
| 644 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 645 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 646 | tool_call = ToolCall( |
| 647 | id="read-dup", |
| 648 | name="read", |
| 649 | arguments={"file_path": str(temp_dir / "index.html")}, |
| 650 | ) |
| 651 | duplicate_message = ( |
| 652 | "[Skipped - duplicate action: Already read " |
| 653 | f"{temp_dir / 'index.html'} recently without any intervening changes; " |
| 654 | "reuse the earlier read result instead of rereading]" |
| 655 | ) |
| 656 | executor = FakeExecutor( |
| 657 | [ |
| 658 | ToolExecutionOutcome( |
| 659 | tool_call=tool_call, |
| 660 | state=ToolExecutionState.DUPLICATE, |
| 661 | message=Message.tool_result_message( |
| 662 | tool_call_id=tool_call.id, |
| 663 | display_content=duplicate_message, |
| 664 | result_content=duplicate_message, |
| 665 | ), |
| 666 | event_content=duplicate_message, |
| 667 | is_error=False, |
| 668 | result_output=duplicate_message, |
| 669 | ) |
| 670 | ] |
| 671 | ) |
| 672 | |
| 673 | summary = TurnSummary(final_response="") |
| 674 | dod = create_definition_of_done("Fix the chapter links") |
| 675 | dod.implementation_plan = str(implementation_plan) |
| 676 | dod.pending_items.append("Create the remaining chapter files") |
| 677 | await runner.execute_batch( |
| 678 | tool_calls=[tool_call], |
| 679 | tool_source="assistant", |
| 680 | pending_tool_calls_seen=set(), |
| 681 | emit=_noop_emit, |
| 682 | summary=summary, |
| 683 | dod=dod, |
| 684 | executor=executor, # type: ignore[arg-type] |
| 685 | on_confirmation=None, |
| 686 | on_user_question=None, |
| 687 | emit_confirmation=None, |
| 688 | consecutive_errors=0, |
| 689 | ) |
| 690 | |
| 691 | assert len(persistent_messages) == 1 |
| 692 | assert "Reuse the earlier observation instead of repeating it." in persistent_messages[0] |
| 693 | assert "A declared output artifact is still missing." in persistent_messages[0] |
| 694 | assert "Resume by creating `04-variables.html` now." in persistent_messages[0] |
| 695 | assert ( |
| 696 | f"Prefer one `write` call for `{temp_dir / 'chapters' / '04-variables.html'}` instead of more rereads." |
| 697 | in persistent_messages[0] |
| 698 | ) |
| 699 | assert ephemeral_messages == [] |
| 700 | |
| 701 | |
| 702 | @pytest.mark.asyncio |
| 703 | async def test_tool_batch_runner_todo_write_does_not_regress_completed_file_todo( |
| 704 | temp_dir: Path, |
| 705 | ) -> None: |
| 706 | async def assess_confidence( |
| 707 | tool_name: str, |
| 708 | tool_args: dict, |
| 709 | context: str, |
| 710 | ) -> ConfidenceAssessment: |
| 711 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 712 | |
| 713 | async def verify_action( |
| 714 | tool_name: str, |
| 715 | tool_args: dict, |
| 716 | result: str, |
| 717 | expected: str = "", |
| 718 | ) -> ActionVerification: |
| 719 | raise AssertionError("Verification should not run for this scenario") |
| 720 | |
| 721 | context = build_context( |
| 722 | temp_dir=temp_dir, |
| 723 | messages=[], |
| 724 | safeguards=FakeSafeguards(), |
| 725 | assess_confidence=assess_confidence, |
| 726 | verify_action=verify_action, |
| 727 | auto_recover=False, |
| 728 | ) |
| 729 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 730 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 731 | sync_todos_to_definition_of_done( |
| 732 | dod, |
| 733 | [ |
| 734 | { |
| 735 | "content": "Create 03-first-website.html", |
| 736 | "active_form": "Creating 03-first-website.html", |
| 737 | "status": "pending", |
| 738 | }, |
| 739 | { |
| 740 | "content": "Create 04-configuration-basics.html", |
| 741 | "active_form": "Creating 04-configuration-basics.html", |
| 742 | "status": "pending", |
| 743 | }, |
| 744 | ], |
| 745 | ) |
| 746 | |
| 747 | chapter_path = temp_dir / "guides" / "nginx" / "chapters" / "03-first-website.html" |
| 748 | chapter_path.parent.mkdir(parents=True) |
| 749 | write_call = ToolCall( |
| 750 | id="write-ch3", |
| 751 | name="write", |
| 752 | arguments={"file_path": str(chapter_path), "content": "<html></html>\n"}, |
| 753 | ) |
| 754 | stale_todo_call = ToolCall( |
| 755 | id="todo-stale", |
| 756 | name="TodoWrite", |
| 757 | arguments={ |
| 758 | "todos": [ |
| 759 | { |
| 760 | "content": "Create 03-first-website.html", |
| 761 | "active_form": "Creating 03-first-website.html", |
| 762 | "status": "pending", |
| 763 | }, |
| 764 | { |
| 765 | "content": "Create 04-configuration-basics.html", |
| 766 | "active_form": "Creating 04-configuration-basics.html", |
| 767 | "status": "pending", |
| 768 | }, |
| 769 | ] |
| 770 | }, |
| 771 | ) |
| 772 | executor = FakeExecutor( |
| 773 | [ |
| 774 | tool_outcome( |
| 775 | tool_call=write_call, |
| 776 | output=f"Successfully wrote {chapter_path}", |
| 777 | is_error=False, |
| 778 | ), |
| 779 | tool_outcome( |
| 780 | tool_call=stale_todo_call, |
| 781 | output="Todos updated", |
| 782 | is_error=False, |
| 783 | metadata={ |
| 784 | "new_todos": [ |
| 785 | { |
| 786 | "content": "Create 03-first-website.html", |
| 787 | "active_form": "Creating 03-first-website.html", |
| 788 | "status": "pending", |
| 789 | }, |
| 790 | { |
| 791 | "content": "Create 04-configuration-basics.html", |
| 792 | "active_form": "Creating 04-configuration-basics.html", |
| 793 | "status": "pending", |
| 794 | }, |
| 795 | ] |
| 796 | }, |
| 797 | ), |
| 798 | ] |
| 799 | ) |
| 800 | |
| 801 | summary = TurnSummary(final_response="") |
| 802 | await runner.execute_batch( |
| 803 | tool_calls=[write_call, stale_todo_call], |
| 804 | tool_source="assistant", |
| 805 | pending_tool_calls_seen=set(), |
| 806 | emit=_noop_emit, |
| 807 | summary=summary, |
| 808 | dod=dod, |
| 809 | executor=executor, # type: ignore[arg-type] |
| 810 | on_confirmation=None, |
| 811 | on_user_question=None, |
| 812 | emit_confirmation=None, |
| 813 | consecutive_errors=0, |
| 814 | ) |
| 815 | |
| 816 | assert "Create 03-first-website.html" in dod.completed_items |
| 817 | assert "Create 03-first-website.html" not in dod.pending_items |
| 818 | assert "Create 04-configuration-basics.html" in dod.pending_items |
| 819 | |
| 820 | |
| 821 | @pytest.mark.asyncio |
| 822 | async def test_tool_batch_runner_proactively_queues_verified_html_inventory( |
| 823 | temp_dir: Path, |
| 824 | ) -> None: |
| 825 | async def assess_confidence( |
| 826 | tool_name: str, |
| 827 | tool_args: dict, |
| 828 | context: str, |
| 829 | ) -> ConfidenceAssessment: |
| 830 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 831 | |
| 832 | async def verify_action( |
| 833 | tool_name: str, |
| 834 | tool_args: dict, |
| 835 | result: str, |
| 836 | expected: str = "", |
| 837 | ) -> ActionVerification: |
| 838 | raise AssertionError("Verification should not run for this scenario") |
| 839 | |
| 840 | chapters = temp_dir / "chapters" |
| 841 | chapters.mkdir() |
| 842 | (chapters / "01-introduction.html").write_text( |
| 843 | "<h1>Chapter 1: Introduction to Fortran</h1>\n" |
| 844 | ) |
| 845 | (chapters / "02-setup.html").write_text( |
| 846 | "<h1>Chapter 2: Setting Up Your Environment</h1>\n" |
| 847 | ) |
| 848 | (temp_dir / "index.html").write_text("<ul></ul>\n") |
| 849 | |
| 850 | context = build_context( |
| 851 | temp_dir=temp_dir, |
| 852 | messages=[], |
| 853 | safeguards=FakeSafeguards(), |
| 854 | assess_confidence=assess_confidence, |
| 855 | verify_action=verify_action, |
| 856 | auto_recover=False, |
| 857 | ) |
| 858 | context.session.current_task = ( |
| 859 | f"Update {temp_dir / 'index.html'} so the chapter links match the sibling files." |
| 860 | ) |
| 861 | persistent_messages: list[str] = [] |
| 862 | ephemeral_messages: list[str] = [] |
| 863 | context.queue_steering_message_callback = persistent_messages.append |
| 864 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 865 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 866 | tool_call = ToolCall( |
| 867 | id="glob-1", |
| 868 | name="glob", |
| 869 | arguments={"path": str(chapters), "pattern": "*.html"}, |
| 870 | ) |
| 871 | executor = FakeExecutor( |
| 872 | [ |
| 873 | tool_outcome( |
| 874 | tool_call=tool_call, |
| 875 | output="\n".join( |
| 876 | [ |
| 877 | str(chapters / "01-introduction.html"), |
| 878 | str(chapters / "02-setup.html"), |
| 879 | ] |
| 880 | ), |
| 881 | is_error=False, |
| 882 | ) |
| 883 | ] |
| 884 | ) |
| 885 | |
| 886 | summary = TurnSummary(final_response="") |
| 887 | await runner.execute_batch( |
| 888 | tool_calls=[tool_call], |
| 889 | tool_source="assistant", |
| 890 | pending_tool_calls_seen=set(), |
| 891 | emit=_noop_emit, |
| 892 | summary=summary, |
| 893 | dod=create_definition_of_done("Fix the chapter links"), |
| 894 | executor=executor, # type: ignore[arg-type] |
| 895 | on_confirmation=None, |
| 896 | on_user_question=None, |
| 897 | emit_confirmation=None, |
| 898 | consecutive_errors=0, |
| 899 | ) |
| 900 | |
| 901 | assert persistent_messages == [] |
| 902 | assert ephemeral_messages == [] |
| 903 | assert len(summary.tool_result_messages) == 1 |
| 904 | assert "Verified chapter inventory:" not in summary.tool_result_messages[0].content |
| 905 | |
| 906 | |
| 907 | @pytest.mark.asyncio |
| 908 | async def test_tool_batch_runner_marks_validated_html_toc_completion_after_successful_edit( |
| 909 | temp_dir: Path, |
| 910 | ) -> None: |
| 911 | async def assess_confidence( |
| 912 | tool_name: str, |
| 913 | tool_args: dict, |
| 914 | context: str, |
| 915 | ) -> ConfidenceAssessment: |
| 916 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 917 | |
| 918 | async def verify_action( |
| 919 | tool_name: str, |
| 920 | tool_args: dict, |
| 921 | result: str, |
| 922 | expected: str = "", |
| 923 | ) -> ActionVerification: |
| 924 | raise AssertionError("Verification should not run for this scenario") |
| 925 | |
| 926 | chapters = temp_dir / "chapters" |
| 927 | chapters.mkdir() |
| 928 | (chapters / "01-introduction.html").write_text( |
| 929 | "<h1>Chapter 1: Introduction to Fortran</h1>\n" |
| 930 | ) |
| 931 | (chapters / "02-setup.html").write_text( |
| 932 | "<h1>Chapter 2: Setting Up Your Environment</h1>\n" |
| 933 | ) |
| 934 | index_path = temp_dir / "index.html" |
| 935 | old_block = ( |
| 936 | '<ul class="chapter-list">\n' |
| 937 | ' <li><a href="chapters/01-old.html">Chapter 1: Old</a></li>\n' |
| 938 | ' <li><a href="chapters/02-old.html">Chapter 2: Old</a></li>\n' |
| 939 | "</ul>\n" |
| 940 | ) |
| 941 | new_block = ( |
| 942 | '<ul class="chapter-list">\n' |
| 943 | ' <li><a href="chapters/01-introduction.html">Chapter 1: Introduction to Fortran</a></li>\n' |
| 944 | ' <li><a href="chapters/02-setup.html">Chapter 2: Setting Up Your Environment</a></li>\n' |
| 945 | "</ul>\n" |
| 946 | ) |
| 947 | index_path.write_text(new_block) |
| 948 | |
| 949 | context = build_context( |
| 950 | temp_dir=temp_dir, |
| 951 | messages=[], |
| 952 | safeguards=FakeSafeguards(), |
| 953 | assess_confidence=assess_confidence, |
| 954 | verify_action=verify_action, |
| 955 | auto_recover=False, |
| 956 | ) |
| 957 | context.session.current_task = ( |
| 958 | "Update index.html so every chapter link and title matches the real HTML files in chapters/." |
| 959 | ) |
| 960 | persistent_messages: list[str] = [] |
| 961 | ephemeral_messages: list[str] = [] |
| 962 | context.queue_steering_message_callback = persistent_messages.append |
| 963 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 964 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 965 | tool_call = ToolCall( |
| 966 | id="edit-1", |
| 967 | name="edit", |
| 968 | arguments={ |
| 969 | "file_path": str(index_path), |
| 970 | "old_string": old_block, |
| 971 | "new_string": new_block, |
| 972 | }, |
| 973 | ) |
| 974 | executor = FakeExecutor( |
| 975 | [ |
| 976 | tool_outcome( |
| 977 | tool_call=tool_call, |
| 978 | output=f"Successfully edited {index_path}", |
| 979 | is_error=False, |
| 980 | ) |
| 981 | ] |
| 982 | ) |
| 983 | |
| 984 | summary = TurnSummary(final_response="") |
| 985 | await runner.execute_batch( |
| 986 | tool_calls=[tool_call], |
| 987 | tool_source="assistant", |
| 988 | pending_tool_calls_seen=set(), |
| 989 | emit=_noop_emit, |
| 990 | summary=summary, |
| 991 | dod=create_definition_of_done( |
| 992 | "Update index.html so every chapter link and title matches the real HTML files in chapters/." |
| 993 | ), |
| 994 | executor=executor, # type: ignore[arg-type] |
| 995 | on_confirmation=None, |
| 996 | on_user_question=None, |
| 997 | emit_confirmation=None, |
| 998 | consecutive_errors=0, |
| 999 | ) |
| 1000 | |
| 1001 | assert all( |
| 1002 | "Semantic verification preview:" not in message.content |
| 1003 | for message in summary.tool_result_messages |
| 1004 | ) |
| 1005 | assert persistent_messages == [] |
| 1006 | assert ephemeral_messages == [] |
| 1007 | |
| 1008 | |
| 1009 | @pytest.mark.asyncio |
| 1010 | async def test_tool_batch_runner_does_not_apply_html_toc_handoff_to_reference_read( |
| 1011 | temp_dir: Path, |
| 1012 | ) -> None: |
| 1013 | async def assess_confidence( |
| 1014 | tool_name: str, |
| 1015 | tool_args: dict, |
| 1016 | context: str, |
| 1017 | ) -> ConfidenceAssessment: |
| 1018 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 1019 | |
| 1020 | async def verify_action( |
| 1021 | tool_name: str, |
| 1022 | tool_args: dict, |
| 1023 | result: str, |
| 1024 | expected: str = "", |
| 1025 | ) -> ActionVerification: |
| 1026 | raise AssertionError("Verification should not run for this scenario") |
| 1027 | |
| 1028 | chapters = temp_dir / "chapters" |
| 1029 | chapters.mkdir() |
| 1030 | (chapters / "01-introduction.html").write_text( |
| 1031 | "<h1>Chapter 1: Introduction to Fortran</h1>\n" |
| 1032 | ) |
| 1033 | (chapters / "02-setup.html").write_text( |
| 1034 | "<h1>Chapter 2: Setting Up Your Environment</h1>\n" |
| 1035 | ) |
| 1036 | index_path = temp_dir / "index.html" |
| 1037 | index_path.write_text( |
| 1038 | "<h2>Table of Contents</h2>\n" |
| 1039 | '<ul class="chapter-list">\n' |
| 1040 | ' <li><a href="chapters/01-introduction.html">Chapter 1: Introduction to Fortran</a></li>\n' |
| 1041 | ' <li><a href="chapters/02-setup.html">Chapter 2: Setting Up Your Environment</a></li>\n' |
| 1042 | "</ul>\n" |
| 1043 | ) |
| 1044 | |
| 1045 | prompt = ( |
| 1046 | "Have a look at ~/Loader/guides/fortran and chapters/ within. Get a feel " |
| 1047 | "for the structure and cadence of the guide. We are going to make an all " |
| 1048 | "new equally thorough guide on how to use the nginx tool." |
| 1049 | ) |
| 1050 | |
| 1051 | context = build_context( |
| 1052 | temp_dir=temp_dir, |
| 1053 | messages=[], |
| 1054 | safeguards=FakeSafeguards(), |
| 1055 | assess_confidence=assess_confidence, |
| 1056 | verify_action=verify_action, |
| 1057 | auto_recover=False, |
| 1058 | ) |
| 1059 | context.session.current_task = prompt # type: ignore[attr-defined] |
| 1060 | persistent_messages: list[str] = [] |
| 1061 | ephemeral_messages: list[str] = [] |
| 1062 | context.queue_steering_message_callback = persistent_messages.append |
| 1063 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 1064 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 1065 | tool_call = ToolCall( |
| 1066 | id="read-index", |
| 1067 | name="read", |
| 1068 | arguments={"file_path": str(index_path)}, |
| 1069 | ) |
| 1070 | executor = FakeExecutor( |
| 1071 | [ |
| 1072 | tool_outcome( |
| 1073 | tool_call=tool_call, |
| 1074 | output=index_path.read_text(), |
| 1075 | is_error=False, |
| 1076 | ) |
| 1077 | ] |
| 1078 | ) |
| 1079 | |
| 1080 | summary = TurnSummary(final_response="") |
| 1081 | await runner.execute_batch( |
| 1082 | tool_calls=[tool_call], |
| 1083 | tool_source="assistant", |
| 1084 | pending_tool_calls_seen=set(), |
| 1085 | emit=_noop_emit, |
| 1086 | summary=summary, |
| 1087 | dod=create_definition_of_done(prompt), |
| 1088 | executor=executor, # type: ignore[arg-type] |
| 1089 | on_confirmation=None, |
| 1090 | on_user_question=None, |
| 1091 | emit_confirmation=None, |
| 1092 | consecutive_errors=0, |
| 1093 | ) |
| 1094 | |
| 1095 | assert persistent_messages == [] |
| 1096 | assert ephemeral_messages == [] |
| 1097 | assert all( |
| 1098 | "Semantic verification preview:" not in message.content |
| 1099 | for message in summary.tool_result_messages |
| 1100 | ) |
| 1101 | |
| 1102 | |
| 1103 | @pytest.mark.asyncio |
| 1104 | async def test_tool_batch_runner_queues_next_pending_todo_after_discovery_progress( |
| 1105 | temp_dir: Path, |
| 1106 | ) -> None: |
| 1107 | async def assess_confidence( |
| 1108 | tool_name: str, |
| 1109 | tool_args: dict, |
| 1110 | context: str, |
| 1111 | ) -> ConfidenceAssessment: |
| 1112 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 1113 | |
| 1114 | async def verify_action( |
| 1115 | tool_name: str, |
| 1116 | tool_args: dict, |
| 1117 | result: str, |
| 1118 | expected: str = "", |
| 1119 | ) -> ActionVerification: |
| 1120 | raise AssertionError("Verification should not run for this scenario") |
| 1121 | |
| 1122 | reference = temp_dir / "fortran" / "chapters" / "01-introduction.html" |
| 1123 | reference.parent.mkdir(parents=True) |
| 1124 | reference.write_text("<h1>Introduction</h1>\n<p>Guide cadence.</p>\n") |
| 1125 | nginx_root = temp_dir / "Loader" / "guides" / "nginx" |
| 1126 | chapters = nginx_root / "chapters" |
| 1127 | implementation_plan = temp_dir / "implementation.md" |
| 1128 | implementation_plan.write_text( |
| 1129 | "\n".join( |
| 1130 | [ |
| 1131 | "# Implementation Plan", |
| 1132 | "", |
| 1133 | "## File Changes", |
| 1134 | f"- `{chapters}/`", |
| 1135 | f"- `{nginx_root / 'index.html'}`", |
| 1136 | "", |
| 1137 | ] |
| 1138 | ) |
| 1139 | ) |
| 1140 | |
| 1141 | context = build_context( |
| 1142 | temp_dir=temp_dir, |
| 1143 | messages=[], |
| 1144 | safeguards=FakeSafeguards(), |
| 1145 | assess_confidence=assess_confidence, |
| 1146 | verify_action=verify_action, |
| 1147 | auto_recover=False, |
| 1148 | ) |
| 1149 | persistent_messages: list[str] = [] |
| 1150 | ephemeral_messages: list[str] = [] |
| 1151 | context.queue_steering_message_callback = persistent_messages.append |
| 1152 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 1153 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 1154 | dod = create_definition_of_done("Create an equally thorough nginx guide.") |
| 1155 | dod.implementation_plan = str(implementation_plan) |
| 1156 | sync_todos_to_definition_of_done( |
| 1157 | dod, |
| 1158 | [ |
| 1159 | { |
| 1160 | "content": "Examine the existing Fortran guide structure to understand the cadence and format", |
| 1161 | "active_form": "Working on: Examine the existing Fortran guide structure to understand the cadence and format", |
| 1162 | "status": "pending", |
| 1163 | }, |
| 1164 | { |
| 1165 | "content": "Create the nginx directory structure", |
| 1166 | "active_form": "Working on: Create the nginx directory structure", |
| 1167 | "status": "pending", |
| 1168 | }, |
| 1169 | { |
| 1170 | "content": "Create the nginx index.html file", |
| 1171 | "active_form": "Working on: Create the nginx index.html file", |
| 1172 | "status": "pending", |
| 1173 | }, |
| 1174 | ], |
| 1175 | ) |
| 1176 | tool_call = ToolCall( |
| 1177 | id="read-reference", |
| 1178 | name="read", |
| 1179 | arguments={"file_path": str(reference)}, |
| 1180 | ) |
| 1181 | executor = FakeExecutor( |
| 1182 | [ |
| 1183 | tool_outcome( |
| 1184 | tool_call=tool_call, |
| 1185 | output="<h1>Introduction</h1>\n<p>Guide cadence.</p>\n", |
| 1186 | is_error=False, |
| 1187 | ) |
| 1188 | ] |
| 1189 | ) |
| 1190 | |
| 1191 | summary = TurnSummary(final_response="") |
| 1192 | await runner.execute_batch( |
| 1193 | tool_calls=[tool_call], |
| 1194 | tool_source="assistant", |
| 1195 | pending_tool_calls_seen=set(), |
| 1196 | emit=_noop_emit, |
| 1197 | summary=summary, |
| 1198 | dod=dod, |
| 1199 | executor=executor, # type: ignore[arg-type] |
| 1200 | on_confirmation=None, |
| 1201 | on_user_question=None, |
| 1202 | emit_confirmation=None, |
| 1203 | consecutive_errors=0, |
| 1204 | ) |
| 1205 | |
| 1206 | assert ( |
| 1207 | "Examine the existing Fortran guide structure to understand the cadence and format" |
| 1208 | in dod.completed_items |
| 1209 | ) |
| 1210 | assert any( |
| 1211 | "Continue with the next pending item: `Create the nginx directory structure`" |
| 1212 | in message |
| 1213 | for message in persistent_messages |
| 1214 | ) |
| 1215 | assert any( |
| 1216 | "Resume by creating `chapters/` now." in message |
| 1217 | for message in persistent_messages |
| 1218 | ) |
| 1219 | assert all("01-introduction.html" not in message for message in persistent_messages) |
| 1220 | assert ephemeral_messages == [] |
| 1221 | |
| 1222 | |
| 1223 | @pytest.mark.asyncio |
| 1224 | async def test_tool_batch_runner_queues_setup_directory_before_file_when_plan_lists_index_first( |
| 1225 | temp_dir: Path, |
| 1226 | ) -> None: |
| 1227 | async def assess_confidence( |
| 1228 | tool_name: str, |
| 1229 | tool_args: dict, |
| 1230 | context: str, |
| 1231 | ) -> ConfidenceAssessment: |
| 1232 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 1233 | |
| 1234 | async def verify_action( |
| 1235 | tool_name: str, |
| 1236 | tool_args: dict, |
| 1237 | result: str, |
| 1238 | expected: str = "", |
| 1239 | ) -> ActionVerification: |
| 1240 | raise AssertionError("Verification should not run for this scenario") |
| 1241 | |
| 1242 | reference = temp_dir / "fortran" / "chapters" / "01-introduction.html" |
| 1243 | reference.parent.mkdir(parents=True) |
| 1244 | reference.write_text("<h1>Introduction</h1>\n<p>Guide cadence.</p>\n") |
| 1245 | nginx_root = temp_dir / "Loader" / "guides" / "nginx" |
| 1246 | chapters = nginx_root / "chapters" |
| 1247 | implementation_plan = temp_dir / "implementation.md" |
| 1248 | implementation_plan.write_text( |
| 1249 | "\n".join( |
| 1250 | [ |
| 1251 | "# Implementation Plan", |
| 1252 | "", |
| 1253 | "## File Changes", |
| 1254 | f"- `{nginx_root / 'index.html'}`", |
| 1255 | f"- `{chapters}/`", |
| 1256 | "", |
| 1257 | ] |
| 1258 | ) |
| 1259 | ) |
| 1260 | |
| 1261 | context = build_context( |
| 1262 | temp_dir=temp_dir, |
| 1263 | messages=[], |
| 1264 | safeguards=FakeSafeguards(), |
| 1265 | assess_confidence=assess_confidence, |
| 1266 | verify_action=verify_action, |
| 1267 | auto_recover=False, |
| 1268 | ) |
| 1269 | persistent_messages: list[str] = [] |
| 1270 | ephemeral_messages: list[str] = [] |
| 1271 | context.queue_steering_message_callback = persistent_messages.append |
| 1272 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 1273 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 1274 | dod = create_definition_of_done("Create an equally thorough nginx guide.") |
| 1275 | dod.implementation_plan = str(implementation_plan) |
| 1276 | sync_todos_to_definition_of_done( |
| 1277 | dod, |
| 1278 | [ |
| 1279 | { |
| 1280 | "content": "Examine the existing Fortran guide structure to understand the cadence and format", |
| 1281 | "active_form": "Working on: Examine the existing Fortran guide structure to understand the cadence and format", |
| 1282 | "status": "pending", |
| 1283 | }, |
| 1284 | { |
| 1285 | "content": "Create the nginx directory structure", |
| 1286 | "active_form": "Working on: Create the nginx directory structure", |
| 1287 | "status": "pending", |
| 1288 | }, |
| 1289 | { |
| 1290 | "content": "Create the nginx index.html file", |
| 1291 | "active_form": "Working on: Create the nginx index.html file", |
| 1292 | "status": "pending", |
| 1293 | }, |
| 1294 | ], |
| 1295 | project_root=temp_dir, |
| 1296 | ) |
| 1297 | tool_call = ToolCall( |
| 1298 | id="read-reference-index-first", |
| 1299 | name="read", |
| 1300 | arguments={"file_path": str(reference)}, |
| 1301 | ) |
| 1302 | executor = FakeExecutor( |
| 1303 | [ |
| 1304 | tool_outcome( |
| 1305 | tool_call=tool_call, |
| 1306 | output="<h1>Introduction</h1>\n<p>Guide cadence.</p>\n", |
| 1307 | is_error=False, |
| 1308 | ) |
| 1309 | ] |
| 1310 | ) |
| 1311 | |
| 1312 | summary = TurnSummary(final_response="") |
| 1313 | await runner.execute_batch( |
| 1314 | tool_calls=[tool_call], |
| 1315 | tool_source="assistant", |
| 1316 | pending_tool_calls_seen=set(), |
| 1317 | emit=_noop_emit, |
| 1318 | summary=summary, |
| 1319 | dod=dod, |
| 1320 | executor=executor, # type: ignore[arg-type] |
| 1321 | on_confirmation=None, |
| 1322 | on_user_question=None, |
| 1323 | emit_confirmation=None, |
| 1324 | consecutive_errors=0, |
| 1325 | ) |
| 1326 | |
| 1327 | assert persistent_messages |
| 1328 | assert any( |
| 1329 | "Continue with the next pending item: `Create the nginx directory structure`" |
| 1330 | in message |
| 1331 | for message in persistent_messages |
| 1332 | ) |
| 1333 | assert any( |
| 1334 | "Resume by creating `chapters/` now." in message |
| 1335 | for message in persistent_messages |
| 1336 | ) |
| 1337 | assert all( |
| 1338 | "Next step: create `index.html`." not in message |
| 1339 | for message in persistent_messages |
| 1340 | ) |
| 1341 | assert ephemeral_messages == [] |
| 1342 | |
| 1343 | |
| 1344 | @pytest.mark.asyncio |
| 1345 | async def test_tool_batch_runner_duplicate_reference_read_prefers_next_pending_todo( |
| 1346 | temp_dir: Path, |
| 1347 | ) -> None: |
| 1348 | async def assess_confidence( |
| 1349 | tool_name: str, |
| 1350 | tool_args: dict, |
| 1351 | context: str, |
| 1352 | ) -> ConfidenceAssessment: |
| 1353 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 1354 | |
| 1355 | async def verify_action( |
| 1356 | tool_name: str, |
| 1357 | tool_args: dict, |
| 1358 | result: str, |
| 1359 | expected: str = "", |
| 1360 | ) -> ActionVerification: |
| 1361 | raise AssertionError("Verification should not run for this scenario") |
| 1362 | |
| 1363 | reference = temp_dir / "fortran" / "index.html" |
| 1364 | reference.parent.mkdir(parents=True) |
| 1365 | reference.write_text("<h1>Fortran Beginner's Guide</h1>\n") |
| 1366 | |
| 1367 | messages = [ |
| 1368 | Message( |
| 1369 | role=Role.TOOL, |
| 1370 | content=( |
| 1371 | "Observation [read]: Result: " |
| 1372 | "<h1>Fortran Beginner's Guide</h1>\n" |
| 1373 | ), |
| 1374 | ) |
| 1375 | ] |
| 1376 | context = build_context( |
| 1377 | temp_dir=temp_dir, |
| 1378 | messages=messages, |
| 1379 | safeguards=FakeSafeguards(), |
| 1380 | assess_confidence=assess_confidence, |
| 1381 | verify_action=verify_action, |
| 1382 | auto_recover=False, |
| 1383 | ) |
| 1384 | prompt = ( |
| 1385 | "Have a look at ~/Loader/guides/fortran and chapters/ within. Get a feel " |
| 1386 | "for the structure and cadence of the guide. We are going to make an all " |
| 1387 | "new equally thorough guide on how to use the nginx tool." |
| 1388 | ) |
| 1389 | context.session.current_task = prompt |
| 1390 | persistent_messages: list[str] = [] |
| 1391 | ephemeral_messages: list[str] = [] |
| 1392 | context.queue_steering_message_callback = persistent_messages.append |
| 1393 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 1394 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 1395 | dod = create_definition_of_done(prompt) |
| 1396 | sync_todos_to_definition_of_done( |
| 1397 | dod, |
| 1398 | [ |
| 1399 | { |
| 1400 | "content": "Examine the existing Fortran guide structure to understand the cadence and format", |
| 1401 | "active_form": "Working on: Examine the existing Fortran guide structure to understand the cadence and format", |
| 1402 | "status": "completed", |
| 1403 | }, |
| 1404 | { |
| 1405 | "content": "Create the nginx directory structure", |
| 1406 | "active_form": "Working on: Create the nginx directory structure", |
| 1407 | "status": "pending", |
| 1408 | }, |
| 1409 | { |
| 1410 | "content": "Create the nginx index.html file", |
| 1411 | "active_form": "Working on: Create the nginx index.html file", |
| 1412 | "status": "pending", |
| 1413 | }, |
| 1414 | ], |
| 1415 | ) |
| 1416 | tool_call = ToolCall( |
| 1417 | id="read-dup", |
| 1418 | name="read", |
| 1419 | arguments={"file_path": str(reference)}, |
| 1420 | ) |
| 1421 | duplicate_message = ( |
| 1422 | "[Skipped - duplicate action: Already read " |
| 1423 | f"{reference} recently without any intervening changes; " |
| 1424 | "reuse the earlier read result instead of rereading]" |
| 1425 | ) |
| 1426 | executor = FakeExecutor( |
| 1427 | [ |
| 1428 | ToolExecutionOutcome( |
| 1429 | tool_call=tool_call, |
| 1430 | state=ToolExecutionState.DUPLICATE, |
| 1431 | message=Message.tool_result_message( |
| 1432 | tool_call_id=tool_call.id, |
| 1433 | display_content=duplicate_message, |
| 1434 | result_content=duplicate_message, |
| 1435 | ), |
| 1436 | event_content=duplicate_message, |
| 1437 | is_error=False, |
| 1438 | result_output=duplicate_message, |
| 1439 | ) |
| 1440 | ] |
| 1441 | ) |
| 1442 | |
| 1443 | summary = TurnSummary(final_response="") |
| 1444 | await runner.execute_batch( |
| 1445 | tool_calls=[tool_call], |
| 1446 | tool_source="assistant", |
| 1447 | pending_tool_calls_seen=set(), |
| 1448 | emit=_noop_emit, |
| 1449 | summary=summary, |
| 1450 | dod=dod, |
| 1451 | executor=executor, # type: ignore[arg-type] |
| 1452 | on_confirmation=None, |
| 1453 | on_user_question=None, |
| 1454 | emit_confirmation=None, |
| 1455 | consecutive_errors=0, |
| 1456 | ) |
| 1457 | |
| 1458 | assert len(persistent_messages) == 1 |
| 1459 | assert "Reuse the earlier observation instead of repeating it." in persistent_messages[0] |
| 1460 | assert ( |
| 1461 | "Continue with the next pending item: `Create the nginx directory structure`" |
| 1462 | in persistent_messages[0] |
| 1463 | ) |
| 1464 | assert "Update `" not in persistent_messages[0] |
| 1465 | assert ephemeral_messages == [] |
| 1466 | |
| 1467 | |
| 1468 | @pytest.mark.asyncio |
| 1469 | async def test_tool_batch_runner_successful_reference_read_prioritizes_concrete_missing_artifact( |
| 1470 | temp_dir: Path, |
| 1471 | ) -> None: |
| 1472 | async def assess_confidence( |
| 1473 | tool_name: str, |
| 1474 | tool_args: dict, |
| 1475 | context: str, |
| 1476 | ) -> ConfidenceAssessment: |
| 1477 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 1478 | |
| 1479 | async def verify_action( |
| 1480 | tool_name: str, |
| 1481 | tool_args: dict, |
| 1482 | result: str, |
| 1483 | expected: str = "", |
| 1484 | ) -> ActionVerification: |
| 1485 | raise AssertionError("Verification should not run for this scenario") |
| 1486 | |
| 1487 | guide_root = temp_dir / "Loader" / "guides" / "nginx" |
| 1488 | chapters = guide_root / "chapters" |
| 1489 | chapters.mkdir(parents=True) |
| 1490 | chapter_one = chapters / "01-introduction.html" |
| 1491 | chapter_one.write_text("<html></html>\n") |
| 1492 | index_path = guide_root / "index.html" |
| 1493 | |
| 1494 | reference = temp_dir / "Loader" / "guides" / "fortran" / "chapters" / "01-introduction.html" |
| 1495 | reference.parent.mkdir(parents=True, exist_ok=True) |
| 1496 | reference.write_text("<h1>Introduction</h1>\n<p>Guide cadence.</p>\n") |
| 1497 | |
| 1498 | implementation_plan = temp_dir / "implementation.md" |
| 1499 | implementation_plan.write_text( |
| 1500 | "\n".join( |
| 1501 | [ |
| 1502 | "# Implementation Plan", |
| 1503 | "", |
| 1504 | "## File Changes", |
| 1505 | f"- `{guide_root}/`", |
| 1506 | f"- `{chapters}/`", |
| 1507 | f"- `{index_path}`", |
| 1508 | f"- `{chapter_one}`", |
| 1509 | f"- `{chapters / '02-installation.html'}`", |
| 1510 | "", |
| 1511 | ] |
| 1512 | ) |
| 1513 | ) |
| 1514 | |
| 1515 | context = build_context( |
| 1516 | temp_dir=temp_dir, |
| 1517 | messages=[], |
| 1518 | safeguards=FakeSafeguards(), |
| 1519 | assess_confidence=assess_confidence, |
| 1520 | verify_action=verify_action, |
| 1521 | auto_recover=False, |
| 1522 | ) |
| 1523 | persistent_messages: list[str] = [] |
| 1524 | ephemeral_messages: list[str] = [] |
| 1525 | context.queue_steering_message_callback = persistent_messages.append |
| 1526 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 1527 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 1528 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 1529 | dod.implementation_plan = str(implementation_plan) |
| 1530 | dod.touched_files.append(str(chapter_one)) |
| 1531 | sync_todos_to_definition_of_done( |
| 1532 | dod, |
| 1533 | [ |
| 1534 | { |
| 1535 | "content": "Examine the existing Fortran guide structure to understand the format and cadence", |
| 1536 | "active_form": "Working on: Examine the existing Fortran guide structure to understand the format and cadence", |
| 1537 | "status": "pending", |
| 1538 | }, |
| 1539 | { |
| 1540 | "content": "Create each chapter file with appropriate content", |
| 1541 | "active_form": "Working on: Create each chapter file with appropriate content", |
| 1542 | "status": "pending", |
| 1543 | }, |
| 1544 | { |
| 1545 | "content": "Ensure all files follow the same structure and style as the Fortran guide", |
| 1546 | "active_form": "Working on: Ensure all files follow the same structure and style as the Fortran guide", |
| 1547 | "status": "pending", |
| 1548 | }, |
| 1549 | ], |
| 1550 | ) |
| 1551 | tool_call = ToolCall( |
| 1552 | id="read-reference-chapter", |
| 1553 | name="read", |
| 1554 | arguments={"file_path": str(reference)}, |
| 1555 | ) |
| 1556 | read_output = "Observation [read]: Result: <h1>Introduction</h1>\n<p>Guide cadence.</p>\n" |
| 1557 | executor = FakeExecutor( |
| 1558 | [ |
| 1559 | ToolExecutionOutcome( |
| 1560 | tool_call=tool_call, |
| 1561 | state=ToolExecutionState.EXECUTED, |
| 1562 | message=Message.tool_result_message( |
| 1563 | tool_call_id=tool_call.id, |
| 1564 | display_content=read_output, |
| 1565 | result_content=read_output, |
| 1566 | ), |
| 1567 | event_content=read_output, |
| 1568 | is_error=False, |
| 1569 | result_output=read_output, |
| 1570 | ) |
| 1571 | ] |
| 1572 | ) |
| 1573 | |
| 1574 | summary = TurnSummary(final_response="") |
| 1575 | await runner.execute_batch( |
| 1576 | tool_calls=[tool_call], |
| 1577 | tool_source="assistant", |
| 1578 | pending_tool_calls_seen=set(), |
| 1579 | emit=_noop_emit, |
| 1580 | summary=summary, |
| 1581 | dod=dod, |
| 1582 | executor=executor, # type: ignore[arg-type] |
| 1583 | on_confirmation=None, |
| 1584 | on_user_question=None, |
| 1585 | emit_confirmation=None, |
| 1586 | consecutive_errors=0, |
| 1587 | ) |
| 1588 | |
| 1589 | assert persistent_messages |
| 1590 | assert any( |
| 1591 | "Confirmed progress: `Examine the existing Fortran guide structure to understand the format and cadence`" |
| 1592 | in message |
| 1593 | for message in persistent_messages |
| 1594 | ) |
| 1595 | assert any("Resume by creating `index.html` now." in message for message in persistent_messages) |
| 1596 | assert not any( |
| 1597 | "Continue with the next pending item: `Create each chapter file with appropriate content`" |
| 1598 | in message |
| 1599 | for message in persistent_messages |
| 1600 | ) |
| 1601 | assert ephemeral_messages == [] |
| 1602 | |
| 1603 | |
| 1604 | @pytest.mark.asyncio |
| 1605 | async def test_tool_batch_runner_duplicate_read_ignores_unplanned_expansion_after_plan_complete( |
| 1606 | temp_dir: Path, |
| 1607 | ) -> None: |
| 1608 | async def assess_confidence( |
| 1609 | tool_name: str, |
| 1610 | tool_args: dict, |
| 1611 | context: str, |
| 1612 | ) -> ConfidenceAssessment: |
| 1613 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 1614 | |
| 1615 | async def verify_action( |
| 1616 | tool_name: str, |
| 1617 | tool_args: dict, |
| 1618 | result: str, |
| 1619 | expected: str = "", |
| 1620 | ) -> ActionVerification: |
| 1621 | raise AssertionError("Verification should not run for this scenario") |
| 1622 | |
| 1623 | guide_root = temp_dir / "guides" / "nginx" |
| 1624 | chapters = guide_root / "chapters" |
| 1625 | guide_root.mkdir(parents=True) |
| 1626 | chapters.mkdir() |
| 1627 | index_path = guide_root / "index.html" |
| 1628 | chapter_one = chapters / "01-getting-started.html" |
| 1629 | chapter_two = chapters / "02-installation.html" |
| 1630 | index_path.write_text("<html></html>\n") |
| 1631 | chapter_one.write_text("<h1>One</h1>\n") |
| 1632 | chapter_two.write_text("<h1>Two</h1>\n") |
| 1633 | |
| 1634 | implementation_plan = temp_dir / "implementation.md" |
| 1635 | implementation_plan.write_text( |
| 1636 | "\n".join( |
| 1637 | [ |
| 1638 | "# Implementation Plan", |
| 1639 | "", |
| 1640 | "## File Changes", |
| 1641 | f"- `{guide_root}/`", |
| 1642 | f"- `{chapters}/`", |
| 1643 | f"- `{index_path}`", |
| 1644 | f"- `{chapter_one}`", |
| 1645 | f"- `{chapter_two}`", |
| 1646 | "", |
| 1647 | ] |
| 1648 | ) |
| 1649 | ) |
| 1650 | |
| 1651 | context = build_context( |
| 1652 | temp_dir=temp_dir, |
| 1653 | messages=[], |
| 1654 | safeguards=FakeSafeguards(), |
| 1655 | assess_confidence=assess_confidence, |
| 1656 | verify_action=verify_action, |
| 1657 | auto_recover=False, |
| 1658 | ) |
| 1659 | persistent_messages: list[str] = [] |
| 1660 | ephemeral_messages: list[str] = [] |
| 1661 | context.queue_steering_message_callback = persistent_messages.append |
| 1662 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 1663 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 1664 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 1665 | dod.implementation_plan = str(implementation_plan) |
| 1666 | dod.pending_items = [ |
| 1667 | "Create 07-performance-tuning.html", |
| 1668 | "Verify all guide files are linked and complete", |
| 1669 | "Complete the requested work", |
| 1670 | ] |
| 1671 | |
| 1672 | tool_call = ToolCall( |
| 1673 | id="read-dup", |
| 1674 | name="read", |
| 1675 | arguments={"file_path": str(chapter_one)}, |
| 1676 | ) |
| 1677 | duplicate_message = ( |
| 1678 | "[Skipped - duplicate action: Already read " |
| 1679 | f"{chapter_one} recently without any intervening changes; " |
| 1680 | "reuse the earlier read result instead of rereading]" |
| 1681 | ) |
| 1682 | executor = FakeExecutor( |
| 1683 | [ |
| 1684 | ToolExecutionOutcome( |
| 1685 | tool_call=tool_call, |
| 1686 | state=ToolExecutionState.DUPLICATE, |
| 1687 | message=Message.tool_result_message( |
| 1688 | tool_call_id=tool_call.id, |
| 1689 | display_content=duplicate_message, |
| 1690 | result_content=duplicate_message, |
| 1691 | ), |
| 1692 | event_content=duplicate_message, |
| 1693 | is_error=False, |
| 1694 | result_output=duplicate_message, |
| 1695 | ) |
| 1696 | ] |
| 1697 | ) |
| 1698 | |
| 1699 | summary = TurnSummary(final_response="") |
| 1700 | await runner.execute_batch( |
| 1701 | tool_calls=[tool_call], |
| 1702 | tool_source="assistant", |
| 1703 | pending_tool_calls_seen=set(), |
| 1704 | emit=_noop_emit, |
| 1705 | summary=summary, |
| 1706 | dod=dod, |
| 1707 | executor=executor, # type: ignore[arg-type] |
| 1708 | on_confirmation=None, |
| 1709 | on_user_question=None, |
| 1710 | emit_confirmation=None, |
| 1711 | consecutive_errors=0, |
| 1712 | ) |
| 1713 | |
| 1714 | assert len(persistent_messages) == 1 |
| 1715 | assert "Verify all guide files are linked and complete" in persistent_messages[0] |
| 1716 | assert "Create 07-performance-tuning.html" not in persistent_messages[0] |
| 1717 | assert ephemeral_messages == [] |
| 1718 | |
| 1719 | |
| 1720 | @pytest.mark.asyncio |
| 1721 | async def test_tool_batch_runner_duplicate_read_after_plan_complete_pushes_verification_handoff( |
| 1722 | temp_dir: Path, |
| 1723 | ) -> None: |
| 1724 | async def assess_confidence( |
| 1725 | tool_name: str, |
| 1726 | tool_args: dict, |
| 1727 | context: str, |
| 1728 | ) -> ConfidenceAssessment: |
| 1729 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 1730 | |
| 1731 | async def verify_action( |
| 1732 | tool_name: str, |
| 1733 | tool_args: dict, |
| 1734 | result: str, |
| 1735 | expected: str = "", |
| 1736 | ) -> ActionVerification: |
| 1737 | raise AssertionError("Verification should not run for this scenario") |
| 1738 | |
| 1739 | guide_root = temp_dir / "guides" / "nginx" |
| 1740 | chapters = guide_root / "chapters" |
| 1741 | guide_root.mkdir(parents=True) |
| 1742 | chapters.mkdir() |
| 1743 | index_path = guide_root / "index.html" |
| 1744 | chapter_one = chapters / "01-getting-started.html" |
| 1745 | chapter_two = chapters / "02-installation.html" |
| 1746 | index_path.write_text("<html></html>\n") |
| 1747 | chapter_one.write_text("<h1>One</h1>\n") |
| 1748 | chapter_two.write_text("<h1>Two</h1>\n") |
| 1749 | |
| 1750 | implementation_plan = temp_dir / "implementation.md" |
| 1751 | implementation_plan.write_text( |
| 1752 | "\n".join( |
| 1753 | [ |
| 1754 | "# Implementation Plan", |
| 1755 | "", |
| 1756 | "## File Changes", |
| 1757 | f"- `{guide_root}/`", |
| 1758 | f"- `{chapters}/`", |
| 1759 | f"- `{index_path}`", |
| 1760 | f"- `{chapter_one}`", |
| 1761 | f"- `{chapter_two}`", |
| 1762 | "", |
| 1763 | ] |
| 1764 | ) |
| 1765 | ) |
| 1766 | |
| 1767 | context = build_context( |
| 1768 | temp_dir=temp_dir, |
| 1769 | messages=[], |
| 1770 | safeguards=FakeSafeguards(), |
| 1771 | assess_confidence=assess_confidence, |
| 1772 | verify_action=verify_action, |
| 1773 | auto_recover=False, |
| 1774 | ) |
| 1775 | persistent_messages: list[str] = [] |
| 1776 | ephemeral_messages: list[str] = [] |
| 1777 | context.queue_steering_message_callback = persistent_messages.append |
| 1778 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 1779 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 1780 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 1781 | dod.implementation_plan = str(implementation_plan) |
| 1782 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 1783 | dod.pending_items = [ |
| 1784 | "Create 07-performance-tuning.html", |
| 1785 | "Complete the requested work", |
| 1786 | ] |
| 1787 | |
| 1788 | tool_call = ToolCall( |
| 1789 | id="read-dup", |
| 1790 | name="read", |
| 1791 | arguments={"file_path": str(chapter_one)}, |
| 1792 | ) |
| 1793 | duplicate_message = ( |
| 1794 | "[Skipped - duplicate action: Already read " |
| 1795 | f"{chapter_one} recently without any intervening changes; " |
| 1796 | "reuse the earlier read result instead of rereading]" |
| 1797 | ) |
| 1798 | executor = FakeExecutor( |
| 1799 | [ |
| 1800 | ToolExecutionOutcome( |
| 1801 | tool_call=tool_call, |
| 1802 | state=ToolExecutionState.DUPLICATE, |
| 1803 | message=Message.tool_result_message( |
| 1804 | tool_call_id=tool_call.id, |
| 1805 | display_content=duplicate_message, |
| 1806 | result_content=duplicate_message, |
| 1807 | ), |
| 1808 | event_content=duplicate_message, |
| 1809 | is_error=False, |
| 1810 | result_output=duplicate_message, |
| 1811 | ) |
| 1812 | ] |
| 1813 | ) |
| 1814 | |
| 1815 | summary = TurnSummary(final_response="") |
| 1816 | await runner.execute_batch( |
| 1817 | tool_calls=[tool_call], |
| 1818 | tool_source="assistant", |
| 1819 | pending_tool_calls_seen=set(), |
| 1820 | emit=_noop_emit, |
| 1821 | summary=summary, |
| 1822 | dod=dod, |
| 1823 | executor=executor, # type: ignore[arg-type] |
| 1824 | on_confirmation=None, |
| 1825 | on_user_question=None, |
| 1826 | emit_confirmation=None, |
| 1827 | consecutive_errors=0, |
| 1828 | ) |
| 1829 | |
| 1830 | assert len(persistent_messages) == 1 |
| 1831 | assert "All explicitly planned artifacts already exist." in persistent_messages[0] |
| 1832 | assert ( |
| 1833 | "Move to verification or final confirmation using the files already on disk." |
| 1834 | in persistent_messages[0] |
| 1835 | ) |
| 1836 | assert "Create 07-performance-tuning.html" not in persistent_messages[0] |
| 1837 | assert ephemeral_messages == [] |
| 1838 | |
| 1839 | |
| 1840 | @pytest.mark.asyncio |
| 1841 | async def test_tool_batch_runner_duplicate_read_after_plan_complete_ignores_stale_creation_todos( |
| 1842 | temp_dir: Path, |
| 1843 | ) -> None: |
| 1844 | async def assess_confidence( |
| 1845 | tool_name: str, |
| 1846 | tool_args: dict, |
| 1847 | context: str, |
| 1848 | ) -> ConfidenceAssessment: |
| 1849 | raise AssertionError("Confidence scoring should not run for this scenario") |
| 1850 | |
| 1851 | async def verify_action( |
| 1852 | tool_name: str, |
| 1853 | tool_args: dict, |
| 1854 | result: str, |
| 1855 | expected: str = "", |
| 1856 | ) -> ActionVerification: |
| 1857 | raise AssertionError("Verification should not run for this scenario") |
| 1858 | |
| 1859 | guide_root = temp_dir / "guides" / "nginx" |
| 1860 | chapters = guide_root / "chapters" |
| 1861 | guide_root.mkdir(parents=True) |
| 1862 | chapters.mkdir() |
| 1863 | index_path = guide_root / "index.html" |
| 1864 | chapter_one = chapters / "01-getting-started.html" |
| 1865 | chapter_two = chapters / "02-installation.html" |
| 1866 | index_path.write_text("<html></html>\n") |
| 1867 | chapter_one.write_text("<h1>One</h1>\n") |
| 1868 | chapter_two.write_text("<h1>Two</h1>\n") |
| 1869 | |
| 1870 | implementation_plan = temp_dir / "implementation.md" |
| 1871 | implementation_plan.write_text( |
| 1872 | "\n".join( |
| 1873 | [ |
| 1874 | "# Implementation Plan", |
| 1875 | "", |
| 1876 | "## File Changes", |
| 1877 | f"- `{guide_root}/`", |
| 1878 | f"- `{chapters}/`", |
| 1879 | f"- `{index_path}`", |
| 1880 | f"- `{chapter_one}`", |
| 1881 | f"- `{chapter_two}`", |
| 1882 | "", |
| 1883 | ] |
| 1884 | ) |
| 1885 | ) |
| 1886 | |
| 1887 | context = build_context( |
| 1888 | temp_dir=temp_dir, |
| 1889 | messages=[], |
| 1890 | safeguards=FakeSafeguards(), |
| 1891 | assess_confidence=assess_confidence, |
| 1892 | verify_action=verify_action, |
| 1893 | auto_recover=False, |
| 1894 | ) |
| 1895 | persistent_messages: list[str] = [] |
| 1896 | ephemeral_messages: list[str] = [] |
| 1897 | context.queue_steering_message_callback = persistent_messages.append |
| 1898 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 1899 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 1900 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 1901 | dod.implementation_plan = str(implementation_plan) |
| 1902 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 1903 | dod.pending_items = [ |
| 1904 | "Create 01-getting-started.html", |
| 1905 | "Creating 02-installation.html", |
| 1906 | "Complete the requested work", |
| 1907 | ] |
| 1908 | |
| 1909 | tool_call = ToolCall( |
| 1910 | id="read-dup-built-stale", |
| 1911 | name="read", |
| 1912 | arguments={"file_path": str(chapter_one)}, |
| 1913 | ) |
| 1914 | duplicate_message = ( |
| 1915 | "[Skipped - duplicate action: Already read " |
| 1916 | f"{chapter_one} recently without any intervening changes; " |
| 1917 | "reuse the earlier read result instead of rereading]" |
| 1918 | ) |
| 1919 | executor = FakeExecutor( |
| 1920 | [ |
| 1921 | ToolExecutionOutcome( |
| 1922 | tool_call=tool_call, |
| 1923 | state=ToolExecutionState.DUPLICATE, |
| 1924 | message=Message.tool_result_message( |
| 1925 | tool_call_id=tool_call.id, |
| 1926 | display_content=duplicate_message, |
| 1927 | result_content=duplicate_message, |
| 1928 | ), |
| 1929 | event_content=duplicate_message, |
| 1930 | is_error=False, |
| 1931 | result_output=duplicate_message, |
| 1932 | ) |
| 1933 | ] |
| 1934 | ) |
| 1935 | |
| 1936 | summary = TurnSummary(final_response="") |
| 1937 | await runner.execute_batch( |
| 1938 | tool_calls=[tool_call], |
| 1939 | tool_source="assistant", |
| 1940 | pending_tool_calls_seen=set(), |
| 1941 | emit=_noop_emit, |
| 1942 | summary=summary, |
| 1943 | dod=dod, |
| 1944 | executor=executor, # type: ignore[arg-type] |
| 1945 | on_confirmation=None, |
| 1946 | on_user_question=None, |
| 1947 | emit_confirmation=None, |
| 1948 | consecutive_errors=0, |
| 1949 | ) |
| 1950 | |
| 1951 | assert len(persistent_messages) == 1 |
| 1952 | assert "All explicitly planned artifacts already exist." in persistent_messages[0] |
| 1953 | assert ( |
| 1954 | "Move to verification or final confirmation using the files already on disk." |
| 1955 | in persistent_messages[0] |
| 1956 | ) |
| 1957 | assert "Create 01-getting-started.html" not in persistent_messages[0] |
| 1958 | assert "Creating 02-installation.html" not in persistent_messages[0] |
| 1959 | assert ephemeral_messages == [] |
| 1960 | |
| 1961 | |
| 1962 | @pytest.mark.asyncio |
| 1963 | async def test_tool_batch_runner_observation_handoff_pushes_mutation_step( |
| 1964 | temp_dir: Path, |
| 1965 | ) -> None: |
| 1966 | async def assess_confidence( |
| 1967 | tool_name: str, |
| 1968 | tool_args: dict, |
| 1969 | context: str, |
| 1970 | ) -> ConfidenceAssessment: |
| 1971 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 1972 | |
| 1973 | async def verify_action( |
| 1974 | tool_name: str, |
| 1975 | tool_args: dict, |
| 1976 | result: str, |
| 1977 | expected: str = "", |
| 1978 | ) -> ActionVerification: |
| 1979 | raise AssertionError("Verification should not run for this scenario") |
| 1980 | |
| 1981 | reference = temp_dir / "fortran" / "chapters" / "01-introduction.html" |
| 1982 | reference.parent.mkdir(parents=True) |
| 1983 | reference.write_text("<h1>Introduction</h1>\n<p>Guide cadence.</p>\n") |
| 1984 | |
| 1985 | context = build_context( |
| 1986 | temp_dir=temp_dir, |
| 1987 | messages=[], |
| 1988 | safeguards=FakeSafeguards(), |
| 1989 | assess_confidence=assess_confidence, |
| 1990 | verify_action=verify_action, |
| 1991 | auto_recover=False, |
| 1992 | ) |
| 1993 | persistent_messages: list[str] = [] |
| 1994 | ephemeral_messages: list[str] = [] |
| 1995 | context.queue_steering_message_callback = persistent_messages.append |
| 1996 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 1997 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 1998 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 1999 | sync_todos_to_definition_of_done( |
| 2000 | dod, |
| 2001 | [ |
| 2002 | { |
| 2003 | "content": "Examine the existing Fortran guide structure to understand the cadence and format", |
| 2004 | "active_form": "Working on: Examine the existing Fortran guide structure to understand the cadence and format", |
| 2005 | "status": "pending", |
| 2006 | }, |
| 2007 | { |
| 2008 | "content": "Create the nginx index.html file", |
| 2009 | "active_form": "Working on: Create the nginx index.html file", |
| 2010 | "status": "pending", |
| 2011 | }, |
| 2012 | ], |
| 2013 | ) |
| 2014 | tool_call = ToolCall( |
| 2015 | id="read-reference", |
| 2016 | name="read", |
| 2017 | arguments={"file_path": str(reference)}, |
| 2018 | ) |
| 2019 | executor = FakeExecutor( |
| 2020 | [ |
| 2021 | tool_outcome( |
| 2022 | tool_call=tool_call, |
| 2023 | output="<h1>Introduction</h1>\n<p>Guide cadence.</p>\n", |
| 2024 | is_error=False, |
| 2025 | ) |
| 2026 | ] |
| 2027 | ) |
| 2028 | |
| 2029 | summary = TurnSummary(final_response="") |
| 2030 | await runner.execute_batch( |
| 2031 | tool_calls=[tool_call], |
| 2032 | tool_source="assistant", |
| 2033 | pending_tool_calls_seen=set(), |
| 2034 | emit=_noop_emit, |
| 2035 | summary=summary, |
| 2036 | dod=dod, |
| 2037 | executor=executor, # type: ignore[arg-type] |
| 2038 | on_confirmation=None, |
| 2039 | on_user_question=None, |
| 2040 | emit_confirmation=None, |
| 2041 | consecutive_errors=0, |
| 2042 | ) |
| 2043 | |
| 2044 | assert any( |
| 2045 | "Continue with the next pending item: `Create the nginx index.html file`" |
| 2046 | in message |
| 2047 | for message in persistent_messages |
| 2048 | ) |
| 2049 | assert any( |
| 2050 | "stop gathering more reference material and perform the change now" in message |
| 2051 | for message in persistent_messages |
| 2052 | ) |
| 2053 | assert ephemeral_messages == [] |
| 2054 | |
| 2055 | |
| 2056 | @pytest.mark.asyncio |
| 2057 | async def test_tool_batch_runner_discovery_completion_handoff_stays_persistent( |
| 2058 | temp_dir: Path, |
| 2059 | ) -> None: |
| 2060 | async def assess_confidence( |
| 2061 | tool_name: str, |
| 2062 | tool_args: dict, |
| 2063 | context: str, |
| 2064 | ) -> ConfidenceAssessment: |
| 2065 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 2066 | |
| 2067 | async def verify_action( |
| 2068 | tool_name: str, |
| 2069 | tool_args: dict, |
| 2070 | result: str, |
| 2071 | expected: str = "", |
| 2072 | ) -> ActionVerification: |
| 2073 | raise AssertionError("Verification should not run for this scenario") |
| 2074 | |
| 2075 | reference = temp_dir / "fortran" / "chapters" / "01-introduction.html" |
| 2076 | reference.parent.mkdir(parents=True) |
| 2077 | reference.write_text("<h1>Introduction</h1>\n<p>Guide cadence.</p>\n") |
| 2078 | |
| 2079 | context = build_context( |
| 2080 | temp_dir=temp_dir, |
| 2081 | messages=[], |
| 2082 | safeguards=FakeSafeguards(), |
| 2083 | assess_confidence=assess_confidence, |
| 2084 | verify_action=verify_action, |
| 2085 | auto_recover=False, |
| 2086 | ) |
| 2087 | persistent_messages: list[str] = [] |
| 2088 | ephemeral_messages: list[str] = [] |
| 2089 | context.queue_steering_message_callback = persistent_messages.append |
| 2090 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 2091 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 2092 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 2093 | sync_todos_to_definition_of_done( |
| 2094 | dod, |
| 2095 | [ |
| 2096 | { |
| 2097 | "content": "First, examine the existing fortran guide structure and content", |
| 2098 | "active_form": "Working on: First, examine the existing fortran guide structure and content", |
| 2099 | "status": "pending", |
| 2100 | }, |
| 2101 | { |
| 2102 | "content": "Create the nginx directory structure", |
| 2103 | "active_form": "Working on: Create the nginx directory structure", |
| 2104 | "status": "pending", |
| 2105 | }, |
| 2106 | ], |
| 2107 | ) |
| 2108 | tool_call = ToolCall( |
| 2109 | id="read-reference", |
| 2110 | name="read", |
| 2111 | arguments={"file_path": str(reference)}, |
| 2112 | ) |
| 2113 | executor = FakeExecutor( |
| 2114 | [ |
| 2115 | tool_outcome( |
| 2116 | tool_call=tool_call, |
| 2117 | output="<h1>Introduction</h1>\n<p>Guide cadence.</p>\n", |
| 2118 | is_error=False, |
| 2119 | ) |
| 2120 | ] |
| 2121 | ) |
| 2122 | |
| 2123 | summary = TurnSummary(final_response="") |
| 2124 | await runner.execute_batch( |
| 2125 | tool_calls=[tool_call], |
| 2126 | tool_source="assistant", |
| 2127 | pending_tool_calls_seen=set(), |
| 2128 | emit=_noop_emit, |
| 2129 | summary=summary, |
| 2130 | dod=dod, |
| 2131 | executor=executor, # type: ignore[arg-type] |
| 2132 | on_confirmation=None, |
| 2133 | on_user_question=None, |
| 2134 | emit_confirmation=None, |
| 2135 | consecutive_errors=0, |
| 2136 | ) |
| 2137 | |
| 2138 | assert persistent_messages |
| 2139 | assert any( |
| 2140 | "Continue with the next pending item: `Create the nginx directory structure`" |
| 2141 | in message |
| 2142 | for message in persistent_messages |
| 2143 | ) |
| 2144 | assert ephemeral_messages == [] |
| 2145 | |
| 2146 | |
| 2147 | @pytest.mark.asyncio |
| 2148 | async def test_tool_batch_runner_missing_artifact_nudge_names_next_file_after_setup_mkdir( |
| 2149 | temp_dir: Path, |
| 2150 | ) -> None: |
| 2151 | async def assess_confidence( |
| 2152 | tool_name: str, |
| 2153 | tool_args: dict, |
| 2154 | context: str, |
| 2155 | ) -> ConfidenceAssessment: |
| 2156 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 2157 | |
| 2158 | async def verify_action( |
| 2159 | tool_name: str, |
| 2160 | tool_args: dict, |
| 2161 | result: str, |
| 2162 | expected: str = "", |
| 2163 | ) -> ActionVerification: |
| 2164 | raise AssertionError("Verification should not run for this scenario") |
| 2165 | |
| 2166 | nginx_root = temp_dir / "Loader" / "guides" / "nginx" |
| 2167 | chapters = nginx_root / "chapters" |
| 2168 | implementation_plan = temp_dir / "implementation.md" |
| 2169 | implementation_plan.write_text( |
| 2170 | "\n".join( |
| 2171 | [ |
| 2172 | "# Implementation Plan", |
| 2173 | "", |
| 2174 | "## File Changes", |
| 2175 | f"- `{chapters}/`", |
| 2176 | f"- `{nginx_root / 'index.html'}`", |
| 2177 | "", |
| 2178 | ] |
| 2179 | ) |
| 2180 | ) |
| 2181 | |
| 2182 | context = build_context( |
| 2183 | temp_dir=temp_dir, |
| 2184 | messages=[], |
| 2185 | safeguards=FakeSafeguards(), |
| 2186 | assess_confidence=assess_confidence, |
| 2187 | verify_action=verify_action, |
| 2188 | auto_recover=False, |
| 2189 | ) |
| 2190 | persistent_messages: list[str] = [] |
| 2191 | ephemeral_messages: list[str] = [] |
| 2192 | context.queue_steering_message_callback = persistent_messages.append |
| 2193 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 2194 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 2195 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 2196 | dod.implementation_plan = str(implementation_plan) |
| 2197 | sync_todos_to_definition_of_done( |
| 2198 | dod, |
| 2199 | [ |
| 2200 | { |
| 2201 | "content": "Create the nginx directory structure", |
| 2202 | "active_form": "Creating the nginx directory structure", |
| 2203 | "status": "pending", |
| 2204 | }, |
| 2205 | { |
| 2206 | "content": "Develop the main index.html file with proper structure", |
| 2207 | "active_form": "Developing the main index.html file with proper structure", |
| 2208 | "status": "pending", |
| 2209 | }, |
| 2210 | ], |
| 2211 | ) |
| 2212 | |
| 2213 | tool_call = ToolCall( |
| 2214 | id="mkdir-nginx", |
| 2215 | name="bash", |
| 2216 | arguments={"command": f"mkdir -p {chapters}"}, |
| 2217 | ) |
| 2218 | executor = FakeExecutor( |
| 2219 | [ |
| 2220 | tool_outcome( |
| 2221 | tool_call=tool_call, |
| 2222 | output="", |
| 2223 | is_error=False, |
| 2224 | ) |
| 2225 | ] |
| 2226 | ) |
| 2227 | |
| 2228 | summary = TurnSummary(final_response="") |
| 2229 | await runner.execute_batch( |
| 2230 | tool_calls=[tool_call], |
| 2231 | tool_source="assistant", |
| 2232 | pending_tool_calls_seen=set(), |
| 2233 | emit=_noop_emit, |
| 2234 | summary=summary, |
| 2235 | dod=dod, |
| 2236 | executor=executor, # type: ignore[arg-type] |
| 2237 | on_confirmation=None, |
| 2238 | on_user_question=None, |
| 2239 | emit_confirmation=None, |
| 2240 | consecutive_errors=0, |
| 2241 | ) |
| 2242 | |
| 2243 | assert persistent_messages |
| 2244 | message = persistent_messages[-1] |
| 2245 | assert "Directory setup is complete." in message |
| 2246 | assert "Continue with the next pending item: `Develop the main index.html file with proper structure`." in message |
| 2247 | assert "Resume by creating `index.html` now." in message |
| 2248 | assert ephemeral_messages == [] |
| 2249 | |
| 2250 | |
| 2251 | @pytest.mark.asyncio |
| 2252 | async def test_tool_batch_runner_first_chapter_handoff_becomes_ephemeral_after_first_file( |
| 2253 | temp_dir: Path, |
| 2254 | ) -> None: |
| 2255 | async def assess_confidence( |
| 2256 | tool_name: str, |
| 2257 | tool_args: dict, |
| 2258 | context: str, |
| 2259 | ) -> ConfidenceAssessment: |
| 2260 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 2261 | |
| 2262 | async def verify_action( |
| 2263 | tool_name: str, |
| 2264 | tool_args: dict, |
| 2265 | result: str, |
| 2266 | expected: str = "", |
| 2267 | ) -> ActionVerification: |
| 2268 | raise AssertionError("Verification should not run for this scenario") |
| 2269 | |
| 2270 | nginx_root = temp_dir / "guides" / "nginx" |
| 2271 | chapters = nginx_root / "chapters" |
| 2272 | chapters.mkdir(parents=True) |
| 2273 | index_path = nginx_root / "index.html" |
| 2274 | |
| 2275 | implementation_plan = temp_dir / "implementation.md" |
| 2276 | implementation_plan.write_text( |
| 2277 | "\n".join( |
| 2278 | [ |
| 2279 | "# Implementation Plan", |
| 2280 | "", |
| 2281 | "## File Changes", |
| 2282 | f"- `{chapters}/`", |
| 2283 | f"- `{index_path}`", |
| 2284 | f"- `{chapters / '01-introduction.html'}`", |
| 2285 | "", |
| 2286 | ] |
| 2287 | ) |
| 2288 | ) |
| 2289 | |
| 2290 | context = build_context( |
| 2291 | temp_dir=temp_dir, |
| 2292 | messages=[], |
| 2293 | safeguards=FakeSafeguards(), |
| 2294 | assess_confidence=assess_confidence, |
| 2295 | verify_action=verify_action, |
| 2296 | auto_recover=False, |
| 2297 | ) |
| 2298 | persistent_messages: list[str] = [] |
| 2299 | ephemeral_messages: list[str] = [] |
| 2300 | context.queue_steering_message_callback = persistent_messages.append |
| 2301 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 2302 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 2303 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 2304 | dod.implementation_plan = str(implementation_plan) |
| 2305 | sync_todos_to_definition_of_done( |
| 2306 | dod, |
| 2307 | [ |
| 2308 | { |
| 2309 | "content": "Create the main index.html file with proper structure", |
| 2310 | "active_form": "Creating the main index.html file with proper structure", |
| 2311 | "status": "pending", |
| 2312 | }, |
| 2313 | { |
| 2314 | "content": "Create each chapter file with appropriate content", |
| 2315 | "active_form": "Creating each chapter file with appropriate content", |
| 2316 | "status": "pending", |
| 2317 | }, |
| 2318 | ], |
| 2319 | ) |
| 2320 | |
| 2321 | tool_call = ToolCall( |
| 2322 | id="write-index", |
| 2323 | name="write", |
| 2324 | arguments={ |
| 2325 | "file_path": str(index_path), |
| 2326 | "content": "<html></html>\n", |
| 2327 | }, |
| 2328 | ) |
| 2329 | executor = FakeExecutor( |
| 2330 | [ |
| 2331 | tool_outcome( |
| 2332 | tool_call=tool_call, |
| 2333 | output=f"Successfully wrote 14 bytes to {index_path}", |
| 2334 | is_error=False, |
| 2335 | ) |
| 2336 | ] |
| 2337 | ) |
| 2338 | |
| 2339 | summary = TurnSummary(final_response="") |
| 2340 | await runner.execute_batch( |
| 2341 | tool_calls=[tool_call], |
| 2342 | tool_source="assistant", |
| 2343 | pending_tool_calls_seen=set(), |
| 2344 | emit=_noop_emit, |
| 2345 | summary=summary, |
| 2346 | dod=dod, |
| 2347 | executor=executor, # type: ignore[arg-type] |
| 2348 | on_confirmation=None, |
| 2349 | on_user_question=None, |
| 2350 | emit_confirmation=None, |
| 2351 | consecutive_errors=0, |
| 2352 | ) |
| 2353 | |
| 2354 | assert persistent_messages == [] |
| 2355 | assert ephemeral_messages |
| 2356 | message = ephemeral_messages[-1] |
| 2357 | assert "Confirmed progress:" in message |
| 2358 | assert "Next step: create `01-introduction.html`." in message |
| 2359 | assert ( |
| 2360 | f"Prefer one `write(file_path=..., content=...)` call for `{(chapters / '01-introduction.html').resolve(strict=False)}` now." |
| 2361 | in message |
| 2362 | ) |
| 2363 | assert "Do not reread reference material or spend the next turn on bookkeeping." in message |
| 2364 | |
| 2365 | |
| 2366 | @pytest.mark.asyncio |
| 2367 | async def test_tool_batch_runner_redirects_post_write_self_audit_to_next_missing_artifact( |
| 2368 | temp_dir: Path, |
| 2369 | ) -> None: |
| 2370 | async def assess_confidence( |
| 2371 | tool_name: str, |
| 2372 | tool_args: dict, |
| 2373 | context: str, |
| 2374 | ) -> ConfidenceAssessment: |
| 2375 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 2376 | |
| 2377 | async def verify_action( |
| 2378 | tool_name: str, |
| 2379 | tool_args: dict, |
| 2380 | result: str, |
| 2381 | expected: str = "", |
| 2382 | ) -> ActionVerification: |
| 2383 | raise AssertionError("Verification should not run in this scenario") |
| 2384 | |
| 2385 | nginx_root = temp_dir / "guides" / "nginx" |
| 2386 | chapters = nginx_root / "chapters" |
| 2387 | chapters.mkdir(parents=True) |
| 2388 | index_path = nginx_root / "index.html" |
| 2389 | index_path.write_text( |
| 2390 | "\n".join( |
| 2391 | [ |
| 2392 | "<html>", |
| 2393 | '<a href="chapters/01-introduction.html">Chapter 1: Introduction to Nginx</a>', |
| 2394 | '<a href="chapters/02-installation.html">Chapter 2: Installation and Setup</a>', |
| 2395 | "</html>", |
| 2396 | ] |
| 2397 | ) |
| 2398 | + "\n" |
| 2399 | ) |
| 2400 | |
| 2401 | implementation_plan = temp_dir / "implementation.md" |
| 2402 | implementation_plan.write_text( |
| 2403 | "\n".join( |
| 2404 | [ |
| 2405 | "# Implementation Plan", |
| 2406 | "", |
| 2407 | "## File Changes", |
| 2408 | f"- `{nginx_root}/`", |
| 2409 | f"- `{chapters}/`", |
| 2410 | f"- `{index_path}`", |
| 2411 | f"- `{chapters / '01-introduction.html'}`", |
| 2412 | "", |
| 2413 | ] |
| 2414 | ) |
| 2415 | ) |
| 2416 | |
| 2417 | context = build_context( |
| 2418 | temp_dir=temp_dir, |
| 2419 | messages=[], |
| 2420 | safeguards=FakeSafeguards(), |
| 2421 | assess_confidence=assess_confidence, |
| 2422 | verify_action=verify_action, |
| 2423 | auto_recover=False, |
| 2424 | ) |
| 2425 | persistent_messages: list[str] = [] |
| 2426 | ephemeral_messages: list[str] = [] |
| 2427 | context.queue_steering_message_callback = persistent_messages.append |
| 2428 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 2429 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 2430 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 2431 | dod.implementation_plan = str(implementation_plan) |
| 2432 | dod.touched_files.append(str(index_path)) |
| 2433 | dod.completed_items.append("Develop the main index.html file for the nginx guide") |
| 2434 | dod.pending_items.append("Create chapter files for the nginx guide") |
| 2435 | |
| 2436 | tool_call = ToolCall( |
| 2437 | id="read-index-self-audit", |
| 2438 | name="read", |
| 2439 | arguments={"file_path": str(index_path)}, |
| 2440 | ) |
| 2441 | executor = FakeExecutor( |
| 2442 | [ |
| 2443 | tool_outcome( |
| 2444 | tool_call=tool_call, |
| 2445 | output="1\t<html>\n", |
| 2446 | is_error=False, |
| 2447 | ) |
| 2448 | ] |
| 2449 | ) |
| 2450 | |
| 2451 | summary = TurnSummary(final_response="") |
| 2452 | await runner.execute_batch( |
| 2453 | tool_calls=[tool_call], |
| 2454 | tool_source="assistant", |
| 2455 | pending_tool_calls_seen=set(), |
| 2456 | emit=_noop_emit, |
| 2457 | summary=summary, |
| 2458 | dod=dod, |
| 2459 | executor=executor, # type: ignore[arg-type] |
| 2460 | on_confirmation=None, |
| 2461 | on_user_question=None, |
| 2462 | emit_confirmation=None, |
| 2463 | consecutive_errors=0, |
| 2464 | ) |
| 2465 | |
| 2466 | assert persistent_messages |
| 2467 | message = persistent_messages[-1] |
| 2468 | assert "You already have the current contents of `index.html` from the successful write." in message |
| 2469 | assert "Resume by creating `01-introduction.html` now." in message |
| 2470 | assert "Do not spend another turn rereading the file you just wrote or on TodoWrite alone." in message |
| 2471 | assert ephemeral_messages == [] |
| 2472 | |
| 2473 | |
| 2474 | @pytest.mark.asyncio |
| 2475 | async def test_tool_batch_runner_softens_first_file_handoff_after_recovery_prompt( |
| 2476 | temp_dir: Path, |
| 2477 | ) -> None: |
| 2478 | async def assess_confidence( |
| 2479 | tool_name: str, |
| 2480 | tool_args: dict, |
| 2481 | context: str, |
| 2482 | ) -> ConfidenceAssessment: |
| 2483 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 2484 | |
| 2485 | async def verify_action( |
| 2486 | tool_name: str, |
| 2487 | tool_args: dict, |
| 2488 | result: str, |
| 2489 | expected: str = "", |
| 2490 | ) -> ActionVerification: |
| 2491 | raise AssertionError("Verification should not run for this scenario") |
| 2492 | |
| 2493 | nginx_root = temp_dir / "guides" / "nginx" |
| 2494 | chapters = nginx_root / "chapters" |
| 2495 | chapters.mkdir(parents=True) |
| 2496 | index_path = nginx_root / "index.html" |
| 2497 | |
| 2498 | implementation_plan = temp_dir / "implementation.md" |
| 2499 | implementation_plan.write_text( |
| 2500 | "\n".join( |
| 2501 | [ |
| 2502 | "# Implementation Plan", |
| 2503 | "", |
| 2504 | "## File Changes", |
| 2505 | f"- `{chapters}/`", |
| 2506 | f"- `{index_path}`", |
| 2507 | f"- `{chapters / '01-introduction.html'}`", |
| 2508 | "", |
| 2509 | ] |
| 2510 | ) |
| 2511 | ) |
| 2512 | |
| 2513 | context = build_context( |
| 2514 | temp_dir=temp_dir, |
| 2515 | messages=[ |
| 2516 | Message( |
| 2517 | role=Role.USER, |
| 2518 | content=( |
| 2519 | "[EMPTY ASSISTANT RESPONSE]\n" |
| 2520 | "Respond with that concrete mutation tool call now. Do not return an empty response." |
| 2521 | ), |
| 2522 | ) |
| 2523 | ], |
| 2524 | safeguards=FakeSafeguards(), |
| 2525 | assess_confidence=assess_confidence, |
| 2526 | verify_action=verify_action, |
| 2527 | auto_recover=False, |
| 2528 | ) |
| 2529 | persistent_messages: list[str] = [] |
| 2530 | ephemeral_messages: list[str] = [] |
| 2531 | context.queue_steering_message_callback = persistent_messages.append |
| 2532 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 2533 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 2534 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 2535 | dod.implementation_plan = str(implementation_plan) |
| 2536 | sync_todos_to_definition_of_done( |
| 2537 | dod, |
| 2538 | [ |
| 2539 | { |
| 2540 | "content": "Create the main index.html file with proper structure", |
| 2541 | "active_form": "Creating the main index.html file with proper structure", |
| 2542 | "status": "pending", |
| 2543 | }, |
| 2544 | { |
| 2545 | "content": "Create each chapter file with appropriate content", |
| 2546 | "active_form": "Creating each chapter file with appropriate content", |
| 2547 | "status": "pending", |
| 2548 | }, |
| 2549 | ], |
| 2550 | ) |
| 2551 | |
| 2552 | tool_call = ToolCall( |
| 2553 | id="write-index-recovered", |
| 2554 | name="write", |
| 2555 | arguments={ |
| 2556 | "file_path": str(index_path), |
| 2557 | "content": "<html></html>\n", |
| 2558 | }, |
| 2559 | ) |
| 2560 | executor = FakeExecutor( |
| 2561 | [ |
| 2562 | tool_outcome( |
| 2563 | tool_call=tool_call, |
| 2564 | output=f"Successfully wrote 14 bytes to {index_path}", |
| 2565 | is_error=False, |
| 2566 | ) |
| 2567 | ] |
| 2568 | ) |
| 2569 | |
| 2570 | summary = TurnSummary(final_response="") |
| 2571 | await runner.execute_batch( |
| 2572 | tool_calls=[tool_call], |
| 2573 | tool_source="assistant", |
| 2574 | pending_tool_calls_seen=set(), |
| 2575 | emit=_noop_emit, |
| 2576 | summary=summary, |
| 2577 | dod=dod, |
| 2578 | executor=executor, # type: ignore[arg-type] |
| 2579 | on_confirmation=None, |
| 2580 | on_user_question=None, |
| 2581 | emit_confirmation=None, |
| 2582 | consecutive_errors=0, |
| 2583 | ) |
| 2584 | |
| 2585 | assert persistent_messages == [] |
| 2586 | assert ephemeral_messages |
| 2587 | message = ephemeral_messages[-1] |
| 2588 | assert "Next step: create `01-introduction.html`." in message |
| 2589 | |
| 2590 | |
| 2591 | @pytest.mark.asyncio |
| 2592 | async def test_tool_batch_runner_todowrite_uses_concrete_output_language_for_aggregate_chapter_step( |
| 2593 | temp_dir: Path, |
| 2594 | ) -> None: |
| 2595 | async def assess_confidence( |
| 2596 | tool_name: str, |
| 2597 | tool_args: dict, |
| 2598 | context: str, |
| 2599 | ) -> ConfidenceAssessment: |
| 2600 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 2601 | |
| 2602 | async def verify_action( |
| 2603 | tool_name: str, |
| 2604 | tool_args: dict, |
| 2605 | result: str, |
| 2606 | expected: str = "", |
| 2607 | ) -> ActionVerification: |
| 2608 | raise AssertionError("Verification should not run in this scenario") |
| 2609 | |
| 2610 | guide_root = temp_dir / "guides" / "nginx" |
| 2611 | chapters = guide_root / "chapters" |
| 2612 | chapters.mkdir(parents=True) |
| 2613 | index_path = guide_root / "index.html" |
| 2614 | index_path.write_text( |
| 2615 | "\n".join( |
| 2616 | [ |
| 2617 | "<html>", |
| 2618 | '<a href="chapters/01-introduction.html">Chapter 1: Introduction to Nginx</a>', |
| 2619 | '<a href="chapters/02-installation.html">Chapter 2: Installation and Setup</a>', |
| 2620 | "</html>", |
| 2621 | ] |
| 2622 | ) |
| 2623 | + "\n" |
| 2624 | ) |
| 2625 | |
| 2626 | implementation_plan = temp_dir / "implementation.md" |
| 2627 | implementation_plan.write_text( |
| 2628 | "\n".join( |
| 2629 | [ |
| 2630 | "# Implementation Plan", |
| 2631 | "", |
| 2632 | "## File Changes", |
| 2633 | f"- `{guide_root}/`", |
| 2634 | f"- `{chapters}/`", |
| 2635 | f"- `{index_path}`", |
| 2636 | "", |
| 2637 | ] |
| 2638 | ) |
| 2639 | ) |
| 2640 | |
| 2641 | context = build_context( |
| 2642 | temp_dir=temp_dir, |
| 2643 | messages=[], |
| 2644 | safeguards=FakeSafeguards(), |
| 2645 | assess_confidence=assess_confidence, |
| 2646 | verify_action=verify_action, |
| 2647 | ) |
| 2648 | queued_messages: list[str] = [] |
| 2649 | context.queue_steering_message_callback = queued_messages.append |
| 2650 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 2651 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 2652 | dod.implementation_plan = str(implementation_plan) |
| 2653 | dod.touched_files.append(str(index_path)) |
| 2654 | sync_todos_to_definition_of_done( |
| 2655 | dod, |
| 2656 | [ |
| 2657 | { |
| 2658 | "content": "Develop the main index.html file with proper structure", |
| 2659 | "active_form": "Developing the main index.html file with proper structure", |
| 2660 | "status": "completed", |
| 2661 | }, |
| 2662 | { |
| 2663 | "content": "Create chapter files with content and structure", |
| 2664 | "active_form": "Creating chapter files with content and structure", |
| 2665 | "status": "pending", |
| 2666 | }, |
| 2667 | ], |
| 2668 | ) |
| 2669 | |
| 2670 | todos = [ |
| 2671 | { |
| 2672 | "content": "Develop the main index.html file with proper structure", |
| 2673 | "active_form": "Developing the main index.html file with proper structure", |
| 2674 | "status": "completed", |
| 2675 | }, |
| 2676 | { |
| 2677 | "content": "Create chapter files with content and structure", |
| 2678 | "active_form": "Creating chapter files with content and structure", |
| 2679 | "status": "pending", |
| 2680 | }, |
| 2681 | ] |
| 2682 | tool_call = ToolCall( |
| 2683 | id="todo-aggregate", |
| 2684 | name="TodoWrite", |
| 2685 | arguments={"todos": todos}, |
| 2686 | ) |
| 2687 | executor = FakeExecutor( |
| 2688 | [ |
| 2689 | tool_outcome( |
| 2690 | tool_call=tool_call, |
| 2691 | output="Todos updated", |
| 2692 | is_error=False, |
| 2693 | metadata={"new_todos": todos}, |
| 2694 | ) |
| 2695 | ] |
| 2696 | ) |
| 2697 | |
| 2698 | summary = TurnSummary(final_response="") |
| 2699 | await runner.execute_batch( |
| 2700 | tool_calls=[tool_call], |
| 2701 | tool_source="assistant", |
| 2702 | pending_tool_calls_seen=set(), |
| 2703 | emit=_noop_emit, |
| 2704 | summary=summary, |
| 2705 | dod=dod, |
| 2706 | executor=executor, # type: ignore[arg-type] |
| 2707 | on_confirmation=None, |
| 2708 | on_user_question=None, |
| 2709 | emit_confirmation=None, |
| 2710 | consecutive_errors=0, |
| 2711 | ) |
| 2712 | |
| 2713 | assert queued_messages |
| 2714 | message = queued_messages[-1] |
| 2715 | assert "Continue with the next concrete output: `01-introduction.html`." in message |
| 2716 | assert "Resume by creating `01-introduction.html` now." in message |
| 2717 | assert ( |
| 2718 | "Continue with the next pending item: `Create chapter files with content and structure`." |
| 2719 | not in message |
| 2720 | ) |
| 2721 | |
| 2722 | |
| 2723 | @pytest.mark.asyncio |
| 2724 | async def test_duplicate_observation_nudge_prioritizes_missing_artifact_over_review( |
| 2725 | temp_dir: Path, |
| 2726 | ) -> None: |
| 2727 | async def assess_confidence( |
| 2728 | tool_name: str, |
| 2729 | tool_args: dict, |
| 2730 | context: str, |
| 2731 | ) -> ConfidenceAssessment: |
| 2732 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 2733 | |
| 2734 | async def verify_action( |
| 2735 | tool_name: str, |
| 2736 | tool_args: dict, |
| 2737 | result: str, |
| 2738 | expected: str = "", |
| 2739 | ) -> ActionVerification: |
| 2740 | raise AssertionError("Verification should not run for this scenario") |
| 2741 | |
| 2742 | guide_root = temp_dir / "guides" / "nginx" |
| 2743 | chapters = guide_root / "chapters" |
| 2744 | chapters.mkdir(parents=True) |
| 2745 | index_path = guide_root / "index.html" |
| 2746 | chapter_one = chapters / "01-getting-started.html" |
| 2747 | chapter_one.write_text("<h1>One</h1>\n") |
| 2748 | index_path.write_text("<a href=\"chapters/01-getting-started.html\">One</a>\n") |
| 2749 | |
| 2750 | implementation_plan = temp_dir / "implementation.md" |
| 2751 | implementation_plan.write_text( |
| 2752 | "\n".join( |
| 2753 | [ |
| 2754 | "# Implementation Plan", |
| 2755 | "", |
| 2756 | "## File Changes", |
| 2757 | f"- `{index_path}`", |
| 2758 | f"- `{chapter_one}`", |
| 2759 | f"- `{chapters / '06-ssl-configuration.html'}`", |
| 2760 | "", |
| 2761 | ] |
| 2762 | ) |
| 2763 | ) |
| 2764 | |
| 2765 | context = build_context( |
| 2766 | temp_dir=temp_dir, |
| 2767 | messages=[], |
| 2768 | safeguards=FakeSafeguards(), |
| 2769 | assess_confidence=assess_confidence, |
| 2770 | verify_action=verify_action, |
| 2771 | auto_recover=False, |
| 2772 | ) |
| 2773 | persistent_messages: list[str] = [] |
| 2774 | ephemeral_messages: list[str] = [] |
| 2775 | context.queue_steering_message_callback = persistent_messages.append |
| 2776 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 2777 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 2778 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 2779 | dod.implementation_plan = str(implementation_plan) |
| 2780 | sync_todos_to_definition_of_done( |
| 2781 | dod, |
| 2782 | [ |
| 2783 | { |
| 2784 | "content": "Ensure all files are properly linked and formatted consistently", |
| 2785 | "active_form": "Working on: Ensure all files are properly linked and formatted consistently", |
| 2786 | "status": "pending", |
| 2787 | }, |
| 2788 | { |
| 2789 | "content": "Create the final chapter (06-ssl-configuration.html)", |
| 2790 | "active_form": "Working on: Create the final chapter (06-ssl-configuration.html)", |
| 2791 | "status": "pending", |
| 2792 | }, |
| 2793 | ], |
| 2794 | ) |
| 2795 | assert tool_batches_should_prioritize_missing_artifact( |
| 2796 | dod=dod, |
| 2797 | next_pending=dod.pending_items[0], |
| 2798 | missing_artifact=(chapters / "06-ssl-configuration.html", False), |
| 2799 | project_root=temp_dir, |
| 2800 | ) |
| 2801 | |
| 2802 | tool_call = ToolCall( |
| 2803 | id="dup-read", |
| 2804 | name="read", |
| 2805 | arguments={"file_path": str(index_path)}, |
| 2806 | ) |
| 2807 | runner._queue_duplicate_observation_nudge(tool_call, dod=dod) # type: ignore[attr-defined] |
| 2808 | |
| 2809 | assert persistent_messages |
| 2810 | message = persistent_messages[-1] |
| 2811 | assert "06-ssl-configuration.html" in message |
| 2812 | assert "Do not switch into review or consistency-check mode" in message |
| 2813 | assert ( |
| 2814 | "Continue with the next pending item: `Ensure all files are properly linked and formatted consistently`" |
| 2815 | not in message |
| 2816 | ) |
| 2817 | |
| 2818 | |
| 2819 | @pytest.mark.asyncio |
| 2820 | async def test_tool_batch_runner_hands_off_to_verification_once_planned_artifacts_exist( |
| 2821 | temp_dir: Path, |
| 2822 | ) -> None: |
| 2823 | async def assess_confidence( |
| 2824 | tool_name: str, |
| 2825 | tool_args: dict, |
| 2826 | context: str, |
| 2827 | ) -> ConfidenceAssessment: |
| 2828 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 2829 | |
| 2830 | async def verify_action( |
| 2831 | tool_name: str, |
| 2832 | tool_args: dict, |
| 2833 | result: str, |
| 2834 | expected: str = "", |
| 2835 | ) -> ActionVerification: |
| 2836 | raise AssertionError("Verification should not run for this scenario") |
| 2837 | |
| 2838 | guide_root = temp_dir / "guides" / "nginx" |
| 2839 | chapters = guide_root / "chapters" |
| 2840 | chapters.mkdir(parents=True) |
| 2841 | index_path = guide_root / "index.html" |
| 2842 | chapter_one = chapters / "01-getting-started.html" |
| 2843 | chapter_two = chapters / "02-installation.html" |
| 2844 | index_path.write_text("<a href=\"chapters/01-getting-started.html\">One</a>\n") |
| 2845 | chapter_one.write_text("<h1>One</h1>\n") |
| 2846 | chapter_two.write_text("<h1>Two</h1>\n") |
| 2847 | |
| 2848 | implementation_plan = temp_dir / "implementation.md" |
| 2849 | implementation_plan.write_text( |
| 2850 | "\n".join( |
| 2851 | [ |
| 2852 | "# Implementation Plan", |
| 2853 | "", |
| 2854 | "## File Changes", |
| 2855 | f"- `{chapters}/`", |
| 2856 | f"- `{index_path}`", |
| 2857 | f"- `{chapter_one}`", |
| 2858 | f"- `{chapter_two}`", |
| 2859 | "", |
| 2860 | ] |
| 2861 | ) |
| 2862 | ) |
| 2863 | |
| 2864 | context = build_context( |
| 2865 | temp_dir=temp_dir, |
| 2866 | messages=[], |
| 2867 | safeguards=FakeSafeguards(), |
| 2868 | assess_confidence=assess_confidence, |
| 2869 | verify_action=verify_action, |
| 2870 | auto_recover=False, |
| 2871 | ) |
| 2872 | persistent_messages: list[str] = [] |
| 2873 | ephemeral_messages: list[str] = [] |
| 2874 | context.queue_steering_message_callback = persistent_messages.append |
| 2875 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 2876 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 2877 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 2878 | dod.implementation_plan = str(implementation_plan) |
| 2879 | sync_todos_to_definition_of_done( |
| 2880 | dod, |
| 2881 | [ |
| 2882 | { |
| 2883 | "content": "Create the guide files", |
| 2884 | "active_form": "Working on: Create the guide files", |
| 2885 | "status": "completed", |
| 2886 | }, |
| 2887 | { |
| 2888 | "content": "Ensure all files are properly linked and formatted consistently", |
| 2889 | "active_form": "Working on: Ensure all files are properly linked and formatted consistently", |
| 2890 | "status": "pending", |
| 2891 | }, |
| 2892 | ], |
| 2893 | ) |
| 2894 | tool_call = ToolCall( |
| 2895 | id="write-final", |
| 2896 | name="write", |
| 2897 | arguments={ |
| 2898 | "file_path": str(chapter_two), |
| 2899 | "content": "<h1>Two</h1>\n", |
| 2900 | }, |
| 2901 | ) |
| 2902 | executor = FakeExecutor( |
| 2903 | [ |
| 2904 | tool_outcome( |
| 2905 | tool_call=tool_call, |
| 2906 | output=f"Successfully wrote {chapter_two}", |
| 2907 | is_error=False, |
| 2908 | ) |
| 2909 | ] |
| 2910 | ) |
| 2911 | |
| 2912 | summary = TurnSummary(final_response="") |
| 2913 | await runner.execute_batch( |
| 2914 | tool_calls=[tool_call], |
| 2915 | tool_source="assistant", |
| 2916 | pending_tool_calls_seen=set(), |
| 2917 | emit=_noop_emit, |
| 2918 | summary=summary, |
| 2919 | dod=dod, |
| 2920 | executor=executor, # type: ignore[arg-type] |
| 2921 | on_confirmation=None, |
| 2922 | on_user_question=None, |
| 2923 | emit_confirmation=None, |
| 2924 | consecutive_errors=0, |
| 2925 | ) |
| 2926 | |
| 2927 | assert any( |
| 2928 | "All explicitly planned artifacts now exist." in message |
| 2929 | for message in persistent_messages |
| 2930 | ) |
| 2931 | assert any( |
| 2932 | "Ensure all files are properly linked and formatted consistently" in message |
| 2933 | for message in persistent_messages |
| 2934 | ) |
| 2935 | assert any( |
| 2936 | "Move to verification once no specific mismatch remains." in message |
| 2937 | for message in persistent_messages |
| 2938 | ) |
| 2939 | |
| 2940 | |
| 2941 | @pytest.mark.asyncio |
| 2942 | async def test_tool_batch_runner_mutation_handoff_points_at_next_missing_artifact( |
| 2943 | temp_dir: Path, |
| 2944 | ) -> None: |
| 2945 | async def assess_confidence( |
| 2946 | tool_name: str, |
| 2947 | tool_args: dict, |
| 2948 | context: str, |
| 2949 | ) -> ConfidenceAssessment: |
| 2950 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 2951 | |
| 2952 | async def verify_action( |
| 2953 | tool_name: str, |
| 2954 | tool_args: dict, |
| 2955 | result: str, |
| 2956 | expected: str = "", |
| 2957 | ) -> ActionVerification: |
| 2958 | raise AssertionError("Verification should not run in this scenario") |
| 2959 | |
| 2960 | guide_root = temp_dir / "guides" / "nginx" |
| 2961 | chapters = guide_root / "chapters" |
| 2962 | guide_root.mkdir(parents=True) |
| 2963 | chapters.mkdir() |
| 2964 | index_path = guide_root / "index.html" |
| 2965 | index_path.write_text("<html></html>\n") |
| 2966 | chapter_one = chapters / "01-getting-started.html" |
| 2967 | chapter_two = chapters / "02-installation.html" |
| 2968 | implementation_plan = temp_dir / "implementation.md" |
| 2969 | implementation_plan.write_text( |
| 2970 | "\n".join( |
| 2971 | [ |
| 2972 | "# Implementation Plan", |
| 2973 | "", |
| 2974 | "## File Changes", |
| 2975 | f"- `{guide_root}/`", |
| 2976 | f"- `{index_path}`", |
| 2977 | f"- `{chapter_one}`", |
| 2978 | f"- `{chapter_two}`", |
| 2979 | "", |
| 2980 | ] |
| 2981 | ) |
| 2982 | ) |
| 2983 | |
| 2984 | context = build_context( |
| 2985 | temp_dir=temp_dir, |
| 2986 | messages=[], |
| 2987 | safeguards=FakeSafeguards(), |
| 2988 | assess_confidence=assess_confidence, |
| 2989 | verify_action=verify_action, |
| 2990 | auto_recover=False, |
| 2991 | ) |
| 2992 | persistent_messages: list[str] = [] |
| 2993 | ephemeral_messages: list[str] = [] |
| 2994 | context.queue_steering_message_callback = persistent_messages.append |
| 2995 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 2996 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 2997 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 2998 | dod.implementation_plan = str(implementation_plan) |
| 2999 | sync_todos_to_definition_of_done( |
| 3000 | dod, |
| 3001 | [ |
| 3002 | { |
| 3003 | "content": "Create the main index.html file with proper structure", |
| 3004 | "active_form": "Working on: Create the main index.html file with proper structure", |
| 3005 | "status": "pending", |
| 3006 | }, |
| 3007 | { |
| 3008 | "content": "Create each chapter file in sequence, following the established pattern", |
| 3009 | "active_form": "Working on: Create each chapter file in sequence, following the established pattern", |
| 3010 | "status": "pending", |
| 3011 | }, |
| 3012 | { |
| 3013 | "content": "Ensure all files are properly linked and formatted consistently", |
| 3014 | "active_form": "Working on: Ensure all files are properly linked and formatted consistently", |
| 3015 | "status": "pending", |
| 3016 | }, |
| 3017 | ], |
| 3018 | ) |
| 3019 | tool_call = ToolCall( |
| 3020 | id="write-index", |
| 3021 | name="write", |
| 3022 | arguments={"file_path": str(index_path), "content": "<html></html>\n"}, |
| 3023 | ) |
| 3024 | executor = FakeExecutor( |
| 3025 | [tool_outcome(tool_call=tool_call, output=f"Successfully wrote {index_path}", is_error=False)] |
| 3026 | ) |
| 3027 | |
| 3028 | summary = TurnSummary(final_response="") |
| 3029 | await runner.execute_batch( |
| 3030 | tool_calls=[tool_call], |
| 3031 | tool_source="assistant", |
| 3032 | pending_tool_calls_seen=set(), |
| 3033 | emit=_noop_emit, |
| 3034 | summary=summary, |
| 3035 | dod=dod, |
| 3036 | executor=executor, # type: ignore[arg-type] |
| 3037 | on_confirmation=None, |
| 3038 | on_user_question=None, |
| 3039 | emit_confirmation=None, |
| 3040 | consecutive_errors=0, |
| 3041 | ) |
| 3042 | |
| 3043 | assert persistent_messages == [] |
| 3044 | assert ephemeral_messages |
| 3045 | message = ephemeral_messages[-1] |
| 3046 | assert "Next step: create `01-getting-started.html`." in message |
| 3047 | assert "refresh `TodoWrite`" not in message |
| 3048 | assert "Do not reread reference material or spend the next turn on bookkeeping." in message |
| 3049 | |
| 3050 | |
| 3051 | @pytest.mark.asyncio |
| 3052 | async def test_tool_batch_runner_large_plan_does_not_claim_completion_early( |
| 3053 | temp_dir: Path, |
| 3054 | ) -> None: |
| 3055 | async def assess_confidence( |
| 3056 | tool_name: str, |
| 3057 | tool_args: dict, |
| 3058 | context: str, |
| 3059 | ) -> ConfidenceAssessment: |
| 3060 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 3061 | |
| 3062 | async def verify_action( |
| 3063 | tool_name: str, |
| 3064 | tool_args: dict, |
| 3065 | result: str, |
| 3066 | expected: str = "", |
| 3067 | ) -> ActionVerification: |
| 3068 | raise AssertionError("Verification should not run in this scenario") |
| 3069 | |
| 3070 | guide_root = temp_dir / "guides" / "nginx" |
| 3071 | chapters = guide_root / "chapters" |
| 3072 | guide_root.mkdir(parents=True) |
| 3073 | chapters.mkdir() |
| 3074 | index_path = guide_root / "index.html" |
| 3075 | index_path.write_text("<html></html>\n") |
| 3076 | |
| 3077 | chapter_paths = [ |
| 3078 | chapters / "01-getting-started.html", |
| 3079 | chapters / "02-installation.html", |
| 3080 | chapters / "03-first-website.html", |
| 3081 | chapters / "04-configuration-basics.html", |
| 3082 | chapters / "05-advanced-configurations.html", |
| 3083 | chapters / "06-performance-tuning.html", |
| 3084 | chapters / "07-security-best-practices.html", |
| 3085 | ] |
| 3086 | for chapter in chapter_paths[:4]: |
| 3087 | chapter.write_text(f"<h1>{chapter.stem}</h1>\n") |
| 3088 | chapter_paths[4].write_text("<h1>Advanced configurations</h1>\n") |
| 3089 | |
| 3090 | implementation_plan = temp_dir / "implementation.md" |
| 3091 | implementation_plan.write_text( |
| 3092 | "\n".join( |
| 3093 | [ |
| 3094 | "# Implementation Plan", |
| 3095 | "", |
| 3096 | "## File Changes", |
| 3097 | f"- `{guide_root}/`", |
| 3098 | f"- `{chapters}/`", |
| 3099 | f"- `{index_path}`", |
| 3100 | *[f"- `{path}`" for path in chapter_paths], |
| 3101 | "", |
| 3102 | ] |
| 3103 | ) |
| 3104 | ) |
| 3105 | |
| 3106 | context = build_context( |
| 3107 | temp_dir=temp_dir, |
| 3108 | messages=[], |
| 3109 | safeguards=FakeSafeguards(), |
| 3110 | assess_confidence=assess_confidence, |
| 3111 | verify_action=verify_action, |
| 3112 | auto_recover=False, |
| 3113 | ) |
| 3114 | persistent_messages: list[str] = [] |
| 3115 | ephemeral_messages: list[str] = [] |
| 3116 | context.queue_steering_message_callback = persistent_messages.append |
| 3117 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 3118 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 3119 | dod = create_definition_of_done("Create a thorough nginx guide.") |
| 3120 | dod.implementation_plan = str(implementation_plan) |
| 3121 | sync_todos_to_definition_of_done( |
| 3122 | dod, |
| 3123 | [ |
| 3124 | { |
| 3125 | "content": "Create the nginx guide artifacts", |
| 3126 | "active_form": "Creating nginx guide artifacts", |
| 3127 | "status": "pending", |
| 3128 | }, |
| 3129 | { |
| 3130 | "content": "Verify all guide files are linked and complete", |
| 3131 | "active_form": "Verifying guide linkage and completeness", |
| 3132 | "status": "pending", |
| 3133 | }, |
| 3134 | ], |
| 3135 | ) |
| 3136 | tool_call = ToolCall( |
| 3137 | id="write-chapter-05", |
| 3138 | name="write", |
| 3139 | arguments={ |
| 3140 | "file_path": str(chapter_paths[4]), |
| 3141 | "content": "<h1>Advanced configurations</h1>\n", |
| 3142 | }, |
| 3143 | ) |
| 3144 | executor = FakeExecutor( |
| 3145 | [ |
| 3146 | tool_outcome( |
| 3147 | tool_call=tool_call, |
| 3148 | output=f"Successfully wrote {chapter_paths[4]}", |
| 3149 | is_error=False, |
| 3150 | ) |
| 3151 | ] |
| 3152 | ) |
| 3153 | |
| 3154 | summary = TurnSummary(final_response="") |
| 3155 | await runner.execute_batch( |
| 3156 | tool_calls=[tool_call], |
| 3157 | tool_source="assistant", |
| 3158 | pending_tool_calls_seen=set(), |
| 3159 | emit=_noop_emit, |
| 3160 | summary=summary, |
| 3161 | dod=dod, |
| 3162 | executor=executor, # type: ignore[arg-type] |
| 3163 | on_confirmation=None, |
| 3164 | on_user_question=None, |
| 3165 | emit_confirmation=None, |
| 3166 | consecutive_errors=0, |
| 3167 | ) |
| 3168 | |
| 3169 | assert any( |
| 3170 | "Next step: create `06-performance-tuning.html`." in message |
| 3171 | for message in ephemeral_messages |
| 3172 | ) |
| 3173 | assert not any( |
| 3174 | "All explicitly planned artifacts now exist." in message |
| 3175 | for message in ephemeral_messages |
| 3176 | ) |
| 3177 | |
| 3178 | |
| 3179 | @pytest.mark.asyncio |
| 3180 | async def test_tool_batch_runner_uses_compact_missing_artifact_nudge_after_substantial_progress( |
| 3181 | temp_dir: Path, |
| 3182 | ) -> None: |
| 3183 | async def assess_confidence( |
| 3184 | tool_name: str, |
| 3185 | tool_args: dict, |
| 3186 | context: str, |
| 3187 | ) -> ConfidenceAssessment: |
| 3188 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 3189 | |
| 3190 | async def verify_action( |
| 3191 | tool_name: str, |
| 3192 | tool_args: dict, |
| 3193 | result: str, |
| 3194 | expected: str = "", |
| 3195 | ) -> ActionVerification: |
| 3196 | raise AssertionError("Verification should not run in this scenario") |
| 3197 | |
| 3198 | guide_root = temp_dir / "guides" / "nginx" |
| 3199 | chapters = guide_root / "chapters" |
| 3200 | guide_root.mkdir(parents=True) |
| 3201 | chapters.mkdir() |
| 3202 | index_path = guide_root / "index.html" |
| 3203 | chapter_paths = [ |
| 3204 | chapters / "01-introduction.html", |
| 3205 | chapters / "02-installation.html", |
| 3206 | chapters / "03-configuration.html", |
| 3207 | chapters / "04-basic-usage.html", |
| 3208 | chapters / "05-advanced-features.html", |
| 3209 | ] |
| 3210 | for path in (index_path, *chapter_paths[:4]): |
| 3211 | path.write_text("<html></html>\n") |
| 3212 | |
| 3213 | implementation_plan = temp_dir / "implementation.md" |
| 3214 | implementation_plan.write_text( |
| 3215 | "\n".join( |
| 3216 | [ |
| 3217 | "# Implementation Plan", |
| 3218 | "", |
| 3219 | "## File Changes", |
| 3220 | f"- `{guide_root}/`", |
| 3221 | f"- `{chapters}/`", |
| 3222 | f"- `{index_path}`", |
| 3223 | *[f"- `{path}`" for path in chapter_paths], |
| 3224 | "", |
| 3225 | ] |
| 3226 | ) |
| 3227 | ) |
| 3228 | |
| 3229 | context = build_context( |
| 3230 | temp_dir=temp_dir, |
| 3231 | messages=[], |
| 3232 | safeguards=FakeSafeguards(), |
| 3233 | assess_confidence=assess_confidence, |
| 3234 | verify_action=verify_action, |
| 3235 | auto_recover=False, |
| 3236 | ) |
| 3237 | persistent_messages: list[str] = [] |
| 3238 | ephemeral_messages: list[str] = [] |
| 3239 | context.queue_steering_message_callback = persistent_messages.append |
| 3240 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 3241 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 3242 | dod = create_definition_of_done("Create a thorough nginx guide.") |
| 3243 | dod.implementation_plan = str(implementation_plan) |
| 3244 | dod.touched_files.extend(str(path) for path in (index_path, *chapter_paths[:4])) |
| 3245 | dod.completed_items.extend( |
| 3246 | [ |
| 3247 | "Create the nginx directory structure", |
| 3248 | "Create the main index.html file with proper structure", |
| 3249 | ] |
| 3250 | ) |
| 3251 | sync_todos_to_definition_of_done( |
| 3252 | dod, |
| 3253 | [ |
| 3254 | { |
| 3255 | "content": "Create each chapter file with appropriate content", |
| 3256 | "active_form": "Creating each chapter file with appropriate content", |
| 3257 | "status": "pending", |
| 3258 | } |
| 3259 | ], |
| 3260 | ) |
| 3261 | tool_call = ToolCall( |
| 3262 | id="write-chapter-04", |
| 3263 | name="write", |
| 3264 | arguments={ |
| 3265 | "file_path": str(chapter_paths[3]), |
| 3266 | "content": "<html>updated</html>\n", |
| 3267 | }, |
| 3268 | ) |
| 3269 | executor = FakeExecutor( |
| 3270 | [ |
| 3271 | tool_outcome( |
| 3272 | tool_call=tool_call, |
| 3273 | output=f"Successfully wrote {chapter_paths[3]}", |
| 3274 | is_error=False, |
| 3275 | ) |
| 3276 | ] |
| 3277 | ) |
| 3278 | |
| 3279 | summary = TurnSummary(final_response="") |
| 3280 | await runner.execute_batch( |
| 3281 | tool_calls=[tool_call], |
| 3282 | tool_source="assistant", |
| 3283 | pending_tool_calls_seen=set(), |
| 3284 | emit=_noop_emit, |
| 3285 | summary=summary, |
| 3286 | dod=dod, |
| 3287 | executor=executor, # type: ignore[arg-type] |
| 3288 | on_confirmation=None, |
| 3289 | on_user_question=None, |
| 3290 | emit_confirmation=None, |
| 3291 | consecutive_errors=0, |
| 3292 | ) |
| 3293 | |
| 3294 | assert ephemeral_messages |
| 3295 | message = ephemeral_messages[-1] |
| 3296 | assert "Next step: create `05-advanced-features.html`." in message |
| 3297 | assert "Do not reread reference material or spend the next turn on bookkeeping." in message |
| 3298 | assert "refresh `TodoWrite`" not in message |
| 3299 | |
| 3300 | |
| 3301 | @pytest.mark.asyncio |
| 3302 | async def test_tool_batch_runner_todowrite_with_missing_artifact_requeues_exact_resume_step( |
| 3303 | temp_dir: Path, |
| 3304 | ) -> None: |
| 3305 | async def assess_confidence( |
| 3306 | tool_name: str, |
| 3307 | tool_args: dict, |
| 3308 | context: str, |
| 3309 | ) -> ConfidenceAssessment: |
| 3310 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 3311 | |
| 3312 | async def verify_action( |
| 3313 | tool_name: str, |
| 3314 | tool_args: dict, |
| 3315 | result: str, |
| 3316 | expected: str = "", |
| 3317 | ) -> ActionVerification: |
| 3318 | raise AssertionError("Verification should not run in this scenario") |
| 3319 | |
| 3320 | guide_root = temp_dir / "guides" / "nginx" |
| 3321 | chapters = guide_root / "chapters" |
| 3322 | guide_root.mkdir(parents=True) |
| 3323 | chapters.mkdir() |
| 3324 | index_path = guide_root / "index.html" |
| 3325 | index_path.write_text("<html></html>\n") |
| 3326 | chapter_one = chapters / "01-getting-started.html" |
| 3327 | chapter_two = chapters / "02-installation.html" |
| 3328 | chapter_one.write_text("<h1>One</h1>\n") |
| 3329 | |
| 3330 | implementation_plan = temp_dir / "implementation.md" |
| 3331 | implementation_plan.write_text( |
| 3332 | "\n".join( |
| 3333 | [ |
| 3334 | "# Implementation Plan", |
| 3335 | "", |
| 3336 | "## File Changes", |
| 3337 | f"- `{guide_root}/`", |
| 3338 | f"- `{chapters}/`", |
| 3339 | f"- `{index_path}`", |
| 3340 | f"- `{chapter_one}`", |
| 3341 | f"- `{chapter_two}`", |
| 3342 | "", |
| 3343 | ] |
| 3344 | ) |
| 3345 | ) |
| 3346 | |
| 3347 | context = build_context( |
| 3348 | temp_dir=temp_dir, |
| 3349 | messages=[], |
| 3350 | safeguards=FakeSafeguards(), |
| 3351 | assess_confidence=assess_confidence, |
| 3352 | verify_action=verify_action, |
| 3353 | auto_recover=False, |
| 3354 | ) |
| 3355 | persistent_messages: list[str] = [] |
| 3356 | ephemeral_messages: list[str] = [] |
| 3357 | context.queue_steering_message_callback = persistent_messages.append |
| 3358 | context.queue_ephemeral_steering_message_callback = ephemeral_messages.append |
| 3359 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 3360 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 3361 | dod.implementation_plan = str(implementation_plan) |
| 3362 | sync_todos_to_definition_of_done( |
| 3363 | dod, |
| 3364 | [ |
| 3365 | { |
| 3366 | "content": "Create 01-getting-started.html", |
| 3367 | "active_form": "Creating 01-getting-started.html", |
| 3368 | "status": "completed", |
| 3369 | }, |
| 3370 | { |
| 3371 | "content": "Create 02-installation.html", |
| 3372 | "active_form": "Creating 02-installation.html", |
| 3373 | "status": "pending", |
| 3374 | }, |
| 3375 | ], |
| 3376 | ) |
| 3377 | dod.touched_files.extend([str(index_path), str(chapter_one)]) |
| 3378 | |
| 3379 | tool_call = ToolCall( |
| 3380 | id="todo-only", |
| 3381 | name="TodoWrite", |
| 3382 | arguments={ |
| 3383 | "todos": [ |
| 3384 | { |
| 3385 | "content": "Create 01-getting-started.html", |
| 3386 | "active_form": "Creating 01-getting-started.html", |
| 3387 | "status": "completed", |
| 3388 | }, |
| 3389 | { |
| 3390 | "content": "Create 02-installation.html", |
| 3391 | "active_form": "Creating 02-installation.html", |
| 3392 | "status": "pending", |
| 3393 | }, |
| 3394 | ] |
| 3395 | }, |
| 3396 | ) |
| 3397 | executor = FakeExecutor( |
| 3398 | [ |
| 3399 | tool_outcome( |
| 3400 | tool_call=tool_call, |
| 3401 | output="Todos updated", |
| 3402 | is_error=False, |
| 3403 | metadata={ |
| 3404 | "new_todos": [ |
| 3405 | { |
| 3406 | "content": "Create 01-getting-started.html", |
| 3407 | "active_form": "Creating 01-getting-started.html", |
| 3408 | "status": "completed", |
| 3409 | }, |
| 3410 | { |
| 3411 | "content": "Create 02-installation.html", |
| 3412 | "active_form": "Creating 02-installation.html", |
| 3413 | "status": "pending", |
| 3414 | }, |
| 3415 | ] |
| 3416 | }, |
| 3417 | ) |
| 3418 | ] |
| 3419 | ) |
| 3420 | |
| 3421 | summary = TurnSummary(final_response="") |
| 3422 | await runner.execute_batch( |
| 3423 | tool_calls=[tool_call], |
| 3424 | tool_source="assistant", |
| 3425 | pending_tool_calls_seen=set(), |
| 3426 | emit=_noop_emit, |
| 3427 | summary=summary, |
| 3428 | dod=dod, |
| 3429 | executor=executor, # type: ignore[arg-type] |
| 3430 | on_confirmation=None, |
| 3431 | on_user_question=None, |
| 3432 | emit_confirmation=None, |
| 3433 | consecutive_errors=0, |
| 3434 | ) |
| 3435 | |
| 3436 | assert persistent_messages |
| 3437 | message = persistent_messages[-1] |
| 3438 | assert "Todo tracking is updated. A declared output artifact is still missing." in message |
| 3439 | assert "Resume by creating `02-installation.html` now." in message |
| 3440 | assert "refresh `TodoWrite`" in message |
| 3441 | assert "Do not spend the next turn on TodoWrite alone" in message |
| 3442 | assert ephemeral_messages == [] |
| 3443 | |
| 3444 | |
| 3445 | @pytest.mark.asyncio |
| 3446 | async def test_tool_batch_runner_todowrite_after_artifacts_exist_pushes_verification_handoff( |
| 3447 | temp_dir: Path, |
| 3448 | ) -> None: |
| 3449 | async def assess_confidence( |
| 3450 | tool_name: str, |
| 3451 | tool_args: dict, |
| 3452 | context: str, |
| 3453 | ) -> ConfidenceAssessment: |
| 3454 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 3455 | |
| 3456 | async def verify_action( |
| 3457 | tool_name: str, |
| 3458 | tool_args: dict, |
| 3459 | result: str, |
| 3460 | expected: str = "", |
| 3461 | ) -> ActionVerification: |
| 3462 | raise AssertionError("Verification should not run in this scenario") |
| 3463 | |
| 3464 | guide_root = temp_dir / "guides" / "nginx" |
| 3465 | chapters = guide_root / "chapters" |
| 3466 | guide_root.mkdir(parents=True) |
| 3467 | chapters.mkdir() |
| 3468 | index_path = guide_root / "index.html" |
| 3469 | chapter_one = chapters / "01-getting-started.html" |
| 3470 | chapter_two = chapters / "02-installation.html" |
| 3471 | index_path.write_text("<html></html>\n") |
| 3472 | chapter_one.write_text("<h1>One</h1>\n") |
| 3473 | chapter_two.write_text("<h1>Two</h1>\n") |
| 3474 | |
| 3475 | implementation_plan = temp_dir / "implementation.md" |
| 3476 | implementation_plan.write_text( |
| 3477 | "\n".join( |
| 3478 | [ |
| 3479 | "# Implementation Plan", |
| 3480 | "", |
| 3481 | "## File Changes", |
| 3482 | f"- `{guide_root}/`", |
| 3483 | f"- `{chapters}/`", |
| 3484 | f"- `{index_path}`", |
| 3485 | f"- `{chapter_one}`", |
| 3486 | f"- `{chapter_two}`", |
| 3487 | "", |
| 3488 | ] |
| 3489 | ) |
| 3490 | ) |
| 3491 | |
| 3492 | context = build_context( |
| 3493 | temp_dir=temp_dir, |
| 3494 | messages=[], |
| 3495 | safeguards=FakeSafeguards(), |
| 3496 | assess_confidence=assess_confidence, |
| 3497 | verify_action=verify_action, |
| 3498 | auto_recover=False, |
| 3499 | ) |
| 3500 | queued_messages: list[str] = [] |
| 3501 | context.queue_steering_message_callback = queued_messages.append |
| 3502 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 3503 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 3504 | dod.implementation_plan = str(implementation_plan) |
| 3505 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 3506 | sync_todos_to_definition_of_done( |
| 3507 | dod, |
| 3508 | [ |
| 3509 | { |
| 3510 | "content": "First, examine the existing Fortran guide structure to understand the format and content organization", |
| 3511 | "active_form": "Working on: First, examine the existing Fortran guide structure to understand the format and content organization", |
| 3512 | "status": "pending", |
| 3513 | }, |
| 3514 | { |
| 3515 | "content": "Verify all guide files are linked and complete", |
| 3516 | "active_form": "Working on: Verify all guide files are linked and complete", |
| 3517 | "status": "pending", |
| 3518 | }, |
| 3519 | ], |
| 3520 | project_root=temp_dir, |
| 3521 | ) |
| 3522 | |
| 3523 | tool_call = ToolCall( |
| 3524 | id="todo-only", |
| 3525 | name="TodoWrite", |
| 3526 | arguments={ |
| 3527 | "todos": [ |
| 3528 | { |
| 3529 | "content": "First, examine the existing Fortran guide structure to understand the format and content organization", |
| 3530 | "active_form": "Working on: First, examine the existing Fortran guide structure to understand the format and content organization", |
| 3531 | "status": "pending", |
| 3532 | }, |
| 3533 | { |
| 3534 | "content": "Verify all guide files are linked and complete", |
| 3535 | "active_form": "Working on: Verify all guide files are linked and complete", |
| 3536 | "status": "pending", |
| 3537 | }, |
| 3538 | ] |
| 3539 | }, |
| 3540 | ) |
| 3541 | executor = FakeExecutor( |
| 3542 | [ |
| 3543 | tool_outcome( |
| 3544 | tool_call=tool_call, |
| 3545 | output="Todos updated", |
| 3546 | is_error=False, |
| 3547 | metadata={ |
| 3548 | "new_todos": [ |
| 3549 | { |
| 3550 | "content": "First, examine the existing Fortran guide structure to understand the format and content organization", |
| 3551 | "active_form": "Working on: First, examine the existing Fortran guide structure to understand the format and content organization", |
| 3552 | "status": "pending", |
| 3553 | }, |
| 3554 | { |
| 3555 | "content": "Verify all guide files are linked and complete", |
| 3556 | "active_form": "Working on: Verify all guide files are linked and complete", |
| 3557 | "status": "pending", |
| 3558 | }, |
| 3559 | ] |
| 3560 | }, |
| 3561 | ) |
| 3562 | ] |
| 3563 | ) |
| 3564 | |
| 3565 | summary = TurnSummary(final_response="") |
| 3566 | await runner.execute_batch( |
| 3567 | tool_calls=[tool_call], |
| 3568 | tool_source="assistant", |
| 3569 | pending_tool_calls_seen=set(), |
| 3570 | emit=_noop_emit, |
| 3571 | summary=summary, |
| 3572 | dod=dod, |
| 3573 | executor=executor, # type: ignore[arg-type] |
| 3574 | on_confirmation=None, |
| 3575 | on_user_question=None, |
| 3576 | emit_confirmation=None, |
| 3577 | consecutive_errors=0, |
| 3578 | ) |
| 3579 | |
| 3580 | assert queued_messages |
| 3581 | message = queued_messages[-1] |
| 3582 | assert "Todo tracking is updated. All explicitly planned artifacts now exist." in message |
| 3583 | assert "Verify all guide files are linked and complete" in message |
| 3584 | assert "Move to verification once no specific mismatch remains." in message |
| 3585 | assert "reopen reference materials" in message |
| 3586 | assert "Fortran guide structure" not in message |
| 3587 | |
| 3588 | |
| 3589 | @pytest.mark.asyncio |
| 3590 | async def test_tool_batch_runner_todowrite_with_existing_output_roots_requeues_next_mutation( |
| 3591 | temp_dir: Path, |
| 3592 | ) -> None: |
| 3593 | async def assess_confidence( |
| 3594 | tool_name: str, |
| 3595 | tool_args: dict, |
| 3596 | context: str, |
| 3597 | ) -> ConfidenceAssessment: |
| 3598 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 3599 | |
| 3600 | async def verify_action( |
| 3601 | tool_name: str, |
| 3602 | tool_args: dict, |
| 3603 | result: str, |
| 3604 | expected: str = "", |
| 3605 | ) -> ActionVerification: |
| 3606 | raise AssertionError("Verification should not run in this scenario") |
| 3607 | |
| 3608 | guide_root = temp_dir / "guides" / "nginx" |
| 3609 | chapters = guide_root / "chapters" |
| 3610 | guide_root.mkdir(parents=True) |
| 3611 | chapters.mkdir() |
| 3612 | index_path = guide_root / "index.html" |
| 3613 | index_path.write_text( |
| 3614 | "\n".join( |
| 3615 | [ |
| 3616 | "<!DOCTYPE html>", |
| 3617 | "<html>", |
| 3618 | "<body>", |
| 3619 | '<a href="chapters/01-introduction.html">Introduction</a>', |
| 3620 | "</body>", |
| 3621 | "</html>", |
| 3622 | "", |
| 3623 | ] |
| 3624 | ) |
| 3625 | ) |
| 3626 | |
| 3627 | implementation_plan = temp_dir / "implementation.md" |
| 3628 | implementation_plan.write_text( |
| 3629 | "\n".join( |
| 3630 | [ |
| 3631 | "# Implementation Plan", |
| 3632 | "", |
| 3633 | "## File Changes", |
| 3634 | f"- `{guide_root}/`", |
| 3635 | f"- `{chapters}/`", |
| 3636 | f"- `{index_path}`", |
| 3637 | "", |
| 3638 | ] |
| 3639 | ) |
| 3640 | ) |
| 3641 | |
| 3642 | context = build_context( |
| 3643 | temp_dir=temp_dir, |
| 3644 | messages=[], |
| 3645 | safeguards=FakeSafeguards(), |
| 3646 | assess_confidence=assess_confidence, |
| 3647 | verify_action=verify_action, |
| 3648 | auto_recover=False, |
| 3649 | ) |
| 3650 | queued_messages: list[str] = [] |
| 3651 | context.queue_steering_message_callback = queued_messages.append |
| 3652 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 3653 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 3654 | dod.implementation_plan = str(implementation_plan) |
| 3655 | dod.touched_files.append(str(index_path)) |
| 3656 | sync_todos_to_definition_of_done( |
| 3657 | dod, |
| 3658 | [ |
| 3659 | { |
| 3660 | "content": "Examine the existing Fortran guide structure", |
| 3661 | "active_form": "Examining the existing Fortran guide structure", |
| 3662 | "status": "completed", |
| 3663 | }, |
| 3664 | { |
| 3665 | "content": "Create the nginx directory structure", |
| 3666 | "active_form": "Creating the nginx directory structure", |
| 3667 | "status": "completed", |
| 3668 | }, |
| 3669 | { |
| 3670 | "content": "Write the introduction chapter", |
| 3671 | "active_form": "Writing the introduction chapter", |
| 3672 | "status": "pending", |
| 3673 | }, |
| 3674 | ], |
| 3675 | project_root=temp_dir, |
| 3676 | ) |
| 3677 | |
| 3678 | tool_call = ToolCall( |
| 3679 | id="todo-next-mutation", |
| 3680 | name="TodoWrite", |
| 3681 | arguments={ |
| 3682 | "todos": [ |
| 3683 | { |
| 3684 | "content": "Examine the existing Fortran guide structure", |
| 3685 | "active_form": "Examining the existing Fortran guide structure", |
| 3686 | "status": "completed", |
| 3687 | }, |
| 3688 | { |
| 3689 | "content": "Create the nginx directory structure", |
| 3690 | "active_form": "Creating the nginx directory structure", |
| 3691 | "status": "completed", |
| 3692 | }, |
| 3693 | { |
| 3694 | "content": "Write the introduction chapter", |
| 3695 | "active_form": "Writing the introduction chapter", |
| 3696 | "status": "pending", |
| 3697 | }, |
| 3698 | ] |
| 3699 | }, |
| 3700 | ) |
| 3701 | executor = FakeExecutor( |
| 3702 | [ |
| 3703 | tool_outcome( |
| 3704 | tool_call=tool_call, |
| 3705 | output="Todos updated", |
| 3706 | is_error=False, |
| 3707 | metadata={ |
| 3708 | "new_todos": [ |
| 3709 | { |
| 3710 | "content": "Examine the existing Fortran guide structure", |
| 3711 | "active_form": "Examining the existing Fortran guide structure", |
| 3712 | "status": "completed", |
| 3713 | }, |
| 3714 | { |
| 3715 | "content": "Create the nginx directory structure", |
| 3716 | "active_form": "Creating the nginx directory structure", |
| 3717 | "status": "completed", |
| 3718 | }, |
| 3719 | { |
| 3720 | "content": "Write the introduction chapter", |
| 3721 | "active_form": "Writing the introduction chapter", |
| 3722 | "status": "pending", |
| 3723 | }, |
| 3724 | ] |
| 3725 | }, |
| 3726 | ) |
| 3727 | ] |
| 3728 | ) |
| 3729 | |
| 3730 | summary = TurnSummary(final_response="") |
| 3731 | await runner.execute_batch( |
| 3732 | tool_calls=[tool_call], |
| 3733 | tool_source="assistant", |
| 3734 | pending_tool_calls_seen=set(), |
| 3735 | emit=_noop_emit, |
| 3736 | summary=summary, |
| 3737 | dod=dod, |
| 3738 | executor=executor, # type: ignore[arg-type] |
| 3739 | on_confirmation=None, |
| 3740 | on_user_question=None, |
| 3741 | emit_confirmation=None, |
| 3742 | consecutive_errors=0, |
| 3743 | ) |
| 3744 | |
| 3745 | assert queued_messages |
| 3746 | message = queued_messages[-1] |
| 3747 | assert "Todo tracking is updated. A declared output artifact is still missing." in message |
| 3748 | assert "Continue with the next pending item: `Write the introduction chapter`." in message |
| 3749 | assert "Resume by creating `01-introduction.html` now." in message |
| 3750 | assert "Prefer one `write` call for `" in message |
| 3751 | assert "01-introduction.html` instead of more rereads." in message |
| 3752 | assert "Do not spend the next turn on TodoWrite alone" in message |
| 3753 | |
| 3754 | |
| 3755 | @pytest.mark.asyncio |
| 3756 | async def test_tool_batch_runner_todowrite_prefers_pending_index_over_empty_output_directory( |
| 3757 | temp_dir: Path, |
| 3758 | ) -> None: |
| 3759 | async def assess_confidence( |
| 3760 | tool_name: str, |
| 3761 | tool_args: dict, |
| 3762 | context: str, |
| 3763 | ) -> ConfidenceAssessment: |
| 3764 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 3765 | |
| 3766 | async def verify_action( |
| 3767 | tool_name: str, |
| 3768 | tool_args: dict, |
| 3769 | result: str, |
| 3770 | expected: str = "", |
| 3771 | ) -> ActionVerification: |
| 3772 | raise AssertionError("Verification should not run in this scenario") |
| 3773 | |
| 3774 | guide_root = temp_dir / "Loader" / "guides" / "nginx" |
| 3775 | chapters = guide_root / "chapters" |
| 3776 | chapters.mkdir(parents=True) |
| 3777 | index_path = guide_root / "index.html" |
| 3778 | implementation_plan = temp_dir / "implementation.md" |
| 3779 | implementation_plan.write_text( |
| 3780 | "\n".join( |
| 3781 | [ |
| 3782 | "# Implementation Plan", |
| 3783 | "", |
| 3784 | "## File Changes", |
| 3785 | f"- `{chapters}/`", |
| 3786 | f"- `{index_path}`", |
| 3787 | "", |
| 3788 | ] |
| 3789 | ) |
| 3790 | ) |
| 3791 | |
| 3792 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 3793 | dod.implementation_plan = str(implementation_plan) |
| 3794 | sync_todos_to_definition_of_done( |
| 3795 | dod, |
| 3796 | [ |
| 3797 | { |
| 3798 | "content": "Examine the existing Fortran guide structure to understand the format and depth", |
| 3799 | "active_form": "Examining the existing Fortran guide structure", |
| 3800 | "status": "completed", |
| 3801 | }, |
| 3802 | { |
| 3803 | "content": "Create the new nginx guide directory structure", |
| 3804 | "active_form": "Creating the new nginx guide directory structure", |
| 3805 | "status": "completed", |
| 3806 | }, |
| 3807 | { |
| 3808 | "content": "Create a new index.html for the nginx guide", |
| 3809 | "active_form": "Creating a new index.html for the nginx guide", |
| 3810 | "status": "pending", |
| 3811 | }, |
| 3812 | { |
| 3813 | "content": "Create the first chapter for the nginx guide", |
| 3814 | "active_form": "Creating the first chapter for the nginx guide", |
| 3815 | "status": "pending", |
| 3816 | }, |
| 3817 | ], |
| 3818 | project_root=temp_dir, |
| 3819 | ) |
| 3820 | |
| 3821 | queued_messages: list[str] = [] |
| 3822 | context = build_context( |
| 3823 | temp_dir=temp_dir, |
| 3824 | messages=[], |
| 3825 | safeguards=FakeSafeguards(), |
| 3826 | assess_confidence=assess_confidence, |
| 3827 | verify_action=verify_action, |
| 3828 | auto_recover=False, |
| 3829 | ) |
| 3830 | context.queue_steering_message_callback = queued_messages.append |
| 3831 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 3832 | |
| 3833 | todos = [ |
| 3834 | { |
| 3835 | "content": "Examine the existing Fortran guide structure to understand the format and depth", |
| 3836 | "active_form": "Examining the existing Fortran guide structure", |
| 3837 | "status": "completed", |
| 3838 | }, |
| 3839 | { |
| 3840 | "content": "Create the new nginx guide directory structure", |
| 3841 | "active_form": "Creating the new nginx guide directory structure", |
| 3842 | "status": "completed", |
| 3843 | }, |
| 3844 | { |
| 3845 | "content": "Create a new index.html for the nginx guide", |
| 3846 | "active_form": "Creating a new index.html for the nginx guide", |
| 3847 | "status": "pending", |
| 3848 | }, |
| 3849 | { |
| 3850 | "content": "Create the first chapter for the nginx guide", |
| 3851 | "active_form": "Creating the first chapter for the nginx guide", |
| 3852 | "status": "pending", |
| 3853 | }, |
| 3854 | ] |
| 3855 | tool_call = ToolCall( |
| 3856 | id="todo-index-before-chapter", |
| 3857 | name="TodoWrite", |
| 3858 | arguments={"todos": todos}, |
| 3859 | ) |
| 3860 | executor = FakeExecutor( |
| 3861 | [ |
| 3862 | tool_outcome( |
| 3863 | tool_call=tool_call, |
| 3864 | output="Todos updated", |
| 3865 | is_error=False, |
| 3866 | metadata={"new_todos": todos}, |
| 3867 | ) |
| 3868 | ] |
| 3869 | ) |
| 3870 | |
| 3871 | summary = TurnSummary(final_response="") |
| 3872 | await runner.execute_batch( |
| 3873 | tool_calls=[tool_call], |
| 3874 | tool_source="assistant", |
| 3875 | pending_tool_calls_seen=set(), |
| 3876 | emit=_noop_emit, |
| 3877 | summary=summary, |
| 3878 | dod=dod, |
| 3879 | executor=executor, # type: ignore[arg-type] |
| 3880 | on_confirmation=None, |
| 3881 | on_user_question=None, |
| 3882 | emit_confirmation=None, |
| 3883 | consecutive_errors=0, |
| 3884 | ) |
| 3885 | |
| 3886 | assert queued_messages |
| 3887 | message = queued_messages[-1] |
| 3888 | assert "Continue with the next pending item: `Create a new index.html for the nginx guide`." in message |
| 3889 | assert "Resume by creating `index.html` now." in message |
| 3890 | assert f"Prefer one `write` call for `{index_path.resolve(strict=False)}`" in message |
| 3891 | assert "01-introduction.html" not in message |
| 3892 | |
| 3893 | |
| 3894 | @pytest.mark.asyncio |
| 3895 | async def test_tool_batch_runner_todowrite_with_declared_child_targets_names_next_missing_file( |
| 3896 | temp_dir: Path, |
| 3897 | ) -> None: |
| 3898 | async def assess_confidence( |
| 3899 | tool_name: str, |
| 3900 | tool_args: dict, |
| 3901 | context: str, |
| 3902 | ) -> ConfidenceAssessment: |
| 3903 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 3904 | |
| 3905 | async def verify_action( |
| 3906 | tool_name: str, |
| 3907 | tool_args: dict, |
| 3908 | result: str, |
| 3909 | expected: str = "", |
| 3910 | ) -> ActionVerification: |
| 3911 | raise AssertionError("Verification should not run in this scenario") |
| 3912 | |
| 3913 | guide_root = temp_dir / "guides" / "nginx" |
| 3914 | chapters = guide_root / "chapters" |
| 3915 | guide_root.mkdir(parents=True) |
| 3916 | chapters.mkdir() |
| 3917 | index_path = guide_root / "index.html" |
| 3918 | index_path.write_text( |
| 3919 | "\n".join( |
| 3920 | [ |
| 3921 | "<html>", |
| 3922 | '<a href="chapters/introduction.html">Introduction</a>', |
| 3923 | '<a href="chapters/installation.html">Installation</a>', |
| 3924 | "</html>", |
| 3925 | ] |
| 3926 | ) |
| 3927 | + "\n" |
| 3928 | ) |
| 3929 | |
| 3930 | implementation_plan = temp_dir / "implementation.md" |
| 3931 | implementation_plan.write_text( |
| 3932 | "\n".join( |
| 3933 | [ |
| 3934 | "# Implementation Plan", |
| 3935 | "", |
| 3936 | "## File Changes", |
| 3937 | f"- `{guide_root}/`", |
| 3938 | f"- `{chapters}/`", |
| 3939 | f"- `{index_path}`", |
| 3940 | "", |
| 3941 | ] |
| 3942 | ) |
| 3943 | ) |
| 3944 | |
| 3945 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 3946 | dod.implementation_plan = str(implementation_plan) |
| 3947 | dod.pending_items = [ |
| 3948 | "Write the introduction chapter", |
| 3949 | "Complete the requested work", |
| 3950 | ] |
| 3951 | dod.touched_files.append(str(index_path)) |
| 3952 | |
| 3953 | queued_messages: list[str] = [] |
| 3954 | context = build_context( |
| 3955 | temp_dir=temp_dir, |
| 3956 | messages=[], |
| 3957 | safeguards=FakeSafeguards(), |
| 3958 | assess_confidence=assess_confidence, |
| 3959 | verify_action=verify_action, |
| 3960 | auto_recover=False, |
| 3961 | ) |
| 3962 | context.queue_steering_message_callback = queued_messages.append |
| 3963 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 3964 | |
| 3965 | tool_call = ToolCall( |
| 3966 | id="todo-1", |
| 3967 | name="TodoWrite", |
| 3968 | arguments={ |
| 3969 | "todos": [ |
| 3970 | { |
| 3971 | "content": "Write the introduction chapter", |
| 3972 | "activeForm": "Writing the introduction chapter", |
| 3973 | "status": "pending", |
| 3974 | } |
| 3975 | ] |
| 3976 | }, |
| 3977 | ) |
| 3978 | executor = FakeExecutor( |
| 3979 | [ |
| 3980 | tool_outcome( |
| 3981 | tool_call=tool_call, |
| 3982 | output="Todos updated", |
| 3983 | is_error=False, |
| 3984 | metadata={ |
| 3985 | "new_todos": [ |
| 3986 | { |
| 3987 | "content": "Write the introduction chapter", |
| 3988 | "active_form": "Writing the introduction chapter", |
| 3989 | "status": "pending", |
| 3990 | } |
| 3991 | ] |
| 3992 | }, |
| 3993 | ) |
| 3994 | ] |
| 3995 | ) |
| 3996 | |
| 3997 | summary = TurnSummary(final_response="") |
| 3998 | await runner.execute_batch( |
| 3999 | tool_calls=[tool_call], |
| 4000 | tool_source="assistant", |
| 4001 | pending_tool_calls_seen=set(), |
| 4002 | emit=_noop_emit, |
| 4003 | summary=summary, |
| 4004 | dod=dod, |
| 4005 | executor=executor, # type: ignore[arg-type] |
| 4006 | on_confirmation=None, |
| 4007 | on_user_question=None, |
| 4008 | emit_confirmation=None, |
| 4009 | consecutive_errors=0, |
| 4010 | ) |
| 4011 | |
| 4012 | assert queued_messages |
| 4013 | message = queued_messages[-1] |
| 4014 | assert "Todo tracking is updated. A declared output artifact is still missing." in message |
| 4015 | assert "Continue with the next pending item: `Write the introduction chapter`." in message |
| 4016 | assert "Resume by creating `introduction.html` now." in message |
| 4017 | assert "Prefer one `write` call for `" in message |
| 4018 | assert "introduction.html` instead of more rereads." in message |
| 4019 | assert "Do not spend the next turn on TodoWrite alone" in message |
| 4020 | |
| 4021 | |
| 4022 | @pytest.mark.asyncio |
| 4023 | async def test_tool_batch_runner_todowrite_names_concrete_pending_file_after_artifacts_exist( |
| 4024 | temp_dir: Path, |
| 4025 | ) -> None: |
| 4026 | async def assess_confidence( |
| 4027 | tool_name: str, |
| 4028 | tool_args: dict, |
| 4029 | context: str, |
| 4030 | ) -> ConfidenceAssessment: |
| 4031 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 4032 | |
| 4033 | async def verify_action( |
| 4034 | tool_name: str, |
| 4035 | tool_args: dict, |
| 4036 | result: str, |
| 4037 | expected: str = "", |
| 4038 | ) -> ActionVerification: |
| 4039 | raise AssertionError("Verification should not run in this scenario") |
| 4040 | |
| 4041 | guide_root = temp_dir / "guides" / "nginx" |
| 4042 | chapters = guide_root / "chapters" |
| 4043 | guide_root.mkdir(parents=True) |
| 4044 | chapters.mkdir() |
| 4045 | index_path = guide_root / "index.html" |
| 4046 | chapter_one = chapters / "01-introduction.html" |
| 4047 | index_path.write_text( |
| 4048 | "\n".join( |
| 4049 | [ |
| 4050 | "<html>", |
| 4051 | '<a href="chapters/01-introduction.html">Chapter 1: Introduction to NGINX Tool</a>', |
| 4052 | '<a href="chapters/02-installation.html">Chapter 2: Installation and Setup</a>', |
| 4053 | "</html>", |
| 4054 | ] |
| 4055 | ) |
| 4056 | + "\n" |
| 4057 | ) |
| 4058 | chapter_one.write_text("<html></html>\n") |
| 4059 | |
| 4060 | implementation_plan = temp_dir / "implementation.md" |
| 4061 | implementation_plan.write_text( |
| 4062 | "\n".join( |
| 4063 | [ |
| 4064 | "# Implementation Plan", |
| 4065 | "", |
| 4066 | "## File Changes", |
| 4067 | f"- `{guide_root}/`", |
| 4068 | f"- `{chapters}/`", |
| 4069 | f"- `{index_path}`", |
| 4070 | "", |
| 4071 | ] |
| 4072 | ) |
| 4073 | ) |
| 4074 | |
| 4075 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 4076 | dod.implementation_plan = str(implementation_plan) |
| 4077 | dod.pending_items = [ |
| 4078 | "Creating Chapter 2: Installation and Setup", |
| 4079 | "Complete the requested work", |
| 4080 | ] |
| 4081 | dod.touched_files.extend([str(index_path), str(chapter_one)]) |
| 4082 | |
| 4083 | queued_messages: list[str] = [] |
| 4084 | context = build_context( |
| 4085 | temp_dir=temp_dir, |
| 4086 | messages=[], |
| 4087 | safeguards=FakeSafeguards(), |
| 4088 | assess_confidence=assess_confidence, |
| 4089 | verify_action=verify_action, |
| 4090 | auto_recover=False, |
| 4091 | ) |
| 4092 | context.queue_steering_message_callback = queued_messages.append |
| 4093 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 4094 | |
| 4095 | tool_call = ToolCall( |
| 4096 | id="todo-1", |
| 4097 | name="TodoWrite", |
| 4098 | arguments={ |
| 4099 | "todos": [ |
| 4100 | { |
| 4101 | "content": "Creating Chapter 2: Installation and Setup", |
| 4102 | "activeForm": "Creating Chapter 2: Installation and Setup", |
| 4103 | "status": "pending", |
| 4104 | } |
| 4105 | ] |
| 4106 | }, |
| 4107 | ) |
| 4108 | executor = FakeExecutor( |
| 4109 | [ |
| 4110 | tool_outcome( |
| 4111 | tool_call=tool_call, |
| 4112 | output="Todos updated", |
| 4113 | is_error=False, |
| 4114 | metadata={ |
| 4115 | "new_todos": [ |
| 4116 | { |
| 4117 | "content": "Creating Chapter 2: Installation and Setup", |
| 4118 | "active_form": "Creating Chapter 2: Installation and Setup", |
| 4119 | "status": "pending", |
| 4120 | } |
| 4121 | ] |
| 4122 | }, |
| 4123 | ) |
| 4124 | ] |
| 4125 | ) |
| 4126 | |
| 4127 | summary = TurnSummary(final_response="") |
| 4128 | await runner.execute_batch( |
| 4129 | tool_calls=[tool_call], |
| 4130 | tool_source="assistant", |
| 4131 | pending_tool_calls_seen=set(), |
| 4132 | emit=_noop_emit, |
| 4133 | summary=summary, |
| 4134 | dod=dod, |
| 4135 | executor=executor, # type: ignore[arg-type] |
| 4136 | on_confirmation=None, |
| 4137 | on_user_question=None, |
| 4138 | emit_confirmation=None, |
| 4139 | consecutive_errors=0, |
| 4140 | ) |
| 4141 | |
| 4142 | assert queued_messages |
| 4143 | message = queued_messages[-1] |
| 4144 | assert "Todo tracking is updated. A declared output artifact is still missing." in message |
| 4145 | assert "Continue with the next pending item: `Creating Chapter 2: Installation and Setup`." in message |
| 4146 | assert "Resume by creating `02-installation.html` now." in message |
| 4147 | assert ( |
| 4148 | f"Prefer one `write` call for `{(chapters / '02-installation.html').resolve(strict=False)}` " |
| 4149 | "instead of more rereads." |
| 4150 | in message |
| 4151 | ) |
| 4152 | assert "Make your next response the concrete mutation tool call itself" in message |
| 4153 | |
| 4154 | |
| 4155 | @pytest.mark.asyncio |
| 4156 | async def test_tool_batch_runner_todowrite_uses_observed_sibling_pattern_for_next_file( |
| 4157 | temp_dir: Path, |
| 4158 | ) -> None: |
| 4159 | async def assess_confidence( |
| 4160 | tool_name: str, |
| 4161 | tool_args: dict, |
| 4162 | context: str, |
| 4163 | ) -> ConfidenceAssessment: |
| 4164 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 4165 | |
| 4166 | async def verify_action( |
| 4167 | tool_name: str, |
| 4168 | tool_args: dict, |
| 4169 | result: str, |
| 4170 | expected: str = "", |
| 4171 | ) -> ActionVerification: |
| 4172 | raise AssertionError("Verification should not run in this scenario") |
| 4173 | |
| 4174 | reference_chapters = temp_dir / "fortran" / "chapters" |
| 4175 | reference_chapters.mkdir(parents=True) |
| 4176 | (reference_chapters / "01-introduction.html").write_text("<h1>Introduction</h1>\n") |
| 4177 | |
| 4178 | guide_root = temp_dir / "guides" / "nginx" |
| 4179 | chapters = guide_root / "chapters" |
| 4180 | guide_root.mkdir(parents=True) |
| 4181 | chapters.mkdir() |
| 4182 | index_path = guide_root / "index.html" |
| 4183 | index_path.write_text("<html></html>\n") |
| 4184 | |
| 4185 | implementation_plan = temp_dir / "implementation.md" |
| 4186 | implementation_plan.write_text( |
| 4187 | "\n".join( |
| 4188 | [ |
| 4189 | "# Implementation Plan", |
| 4190 | "", |
| 4191 | "## File Changes", |
| 4192 | f"- `{guide_root}/`", |
| 4193 | f"- `{chapters}/`", |
| 4194 | f"- `{index_path}`", |
| 4195 | "", |
| 4196 | ] |
| 4197 | ) |
| 4198 | ) |
| 4199 | |
| 4200 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 4201 | dod.implementation_plan = str(implementation_plan) |
| 4202 | dod.pending_items = [ |
| 4203 | "Write the introduction chapter", |
| 4204 | "Complete the requested work", |
| 4205 | ] |
| 4206 | dod.touched_files.append(str(index_path)) |
| 4207 | |
| 4208 | queued_messages: list[str] = [] |
| 4209 | context = build_context( |
| 4210 | temp_dir=temp_dir, |
| 4211 | messages=[ |
| 4212 | Message( |
| 4213 | role=Role.ASSISTANT, |
| 4214 | content="", |
| 4215 | tool_calls=[ |
| 4216 | ToolCall( |
| 4217 | id="read-ref-1", |
| 4218 | name="read", |
| 4219 | arguments={"file_path": str(reference_chapters / "01-introduction.html")}, |
| 4220 | ) |
| 4221 | ], |
| 4222 | ) |
| 4223 | ], |
| 4224 | safeguards=FakeSafeguards(), |
| 4225 | assess_confidence=assess_confidence, |
| 4226 | verify_action=verify_action, |
| 4227 | auto_recover=False, |
| 4228 | ) |
| 4229 | context.queue_steering_message_callback = queued_messages.append |
| 4230 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 4231 | |
| 4232 | tool_call = ToolCall( |
| 4233 | id="todo-observed-1", |
| 4234 | name="TodoWrite", |
| 4235 | arguments={ |
| 4236 | "todos": [ |
| 4237 | { |
| 4238 | "content": "Write the introduction chapter", |
| 4239 | "activeForm": "Writing the introduction chapter", |
| 4240 | "status": "pending", |
| 4241 | } |
| 4242 | ] |
| 4243 | }, |
| 4244 | ) |
| 4245 | executor = FakeExecutor( |
| 4246 | [ |
| 4247 | tool_outcome( |
| 4248 | tool_call=tool_call, |
| 4249 | output="Todos updated", |
| 4250 | is_error=False, |
| 4251 | metadata={ |
| 4252 | "new_todos": [ |
| 4253 | { |
| 4254 | "content": "Write the introduction chapter", |
| 4255 | "active_form": "Writing the introduction chapter", |
| 4256 | "status": "pending", |
| 4257 | } |
| 4258 | ] |
| 4259 | }, |
| 4260 | ) |
| 4261 | ] |
| 4262 | ) |
| 4263 | |
| 4264 | summary = TurnSummary(final_response="") |
| 4265 | await runner.execute_batch( |
| 4266 | tool_calls=[tool_call], |
| 4267 | tool_source="assistant", |
| 4268 | pending_tool_calls_seen=set(), |
| 4269 | emit=_noop_emit, |
| 4270 | summary=summary, |
| 4271 | dod=dod, |
| 4272 | executor=executor, # type: ignore[arg-type] |
| 4273 | on_confirmation=None, |
| 4274 | on_user_question=None, |
| 4275 | emit_confirmation=None, |
| 4276 | consecutive_errors=0, |
| 4277 | ) |
| 4278 | |
| 4279 | assert queued_messages |
| 4280 | message = queued_messages[-1] |
| 4281 | assert "Todo tracking is updated. A declared output artifact is still missing." in message |
| 4282 | assert "Continue with the next pending item: `Write the introduction chapter`." in message |
| 4283 | assert "Resume by creating `01-introduction.html` now." in message |
| 4284 | assert ( |
| 4285 | "It mirrors the observed filename pattern from another `chapters/` directory " |
| 4286 | "you already inspected." |
| 4287 | in message |
| 4288 | ) |
| 4289 | assert "01-introduction.html` instead of more rereads." in message |
| 4290 | |
| 4291 | |
| 4292 | @pytest.mark.asyncio |
| 4293 | async def test_tool_batch_runner_bookkeeping_note_with_missing_artifact_requeues_resume_step( |
| 4294 | temp_dir: Path, |
| 4295 | ) -> None: |
| 4296 | async def assess_confidence( |
| 4297 | tool_name: str, |
| 4298 | tool_args: dict, |
| 4299 | context: str, |
| 4300 | ) -> ConfidenceAssessment: |
| 4301 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 4302 | |
| 4303 | async def verify_action( |
| 4304 | tool_name: str, |
| 4305 | tool_args: dict, |
| 4306 | result: str, |
| 4307 | expected: str = "", |
| 4308 | ) -> ActionVerification: |
| 4309 | raise AssertionError("Verification should not run in this scenario") |
| 4310 | |
| 4311 | guide_root = temp_dir / "guides" / "nginx" |
| 4312 | chapters = guide_root / "chapters" |
| 4313 | guide_root.mkdir(parents=True) |
| 4314 | chapters.mkdir() |
| 4315 | index_path = guide_root / "index.html" |
| 4316 | chapter_one = chapters / "01-getting-started.html" |
| 4317 | chapter_two = chapters / "02-installation.html" |
| 4318 | index_path.write_text("<html></html>\n") |
| 4319 | chapter_one.write_text("<h1>One</h1>\n") |
| 4320 | |
| 4321 | implementation_plan = temp_dir / "implementation.md" |
| 4322 | implementation_plan.write_text( |
| 4323 | "\n".join( |
| 4324 | [ |
| 4325 | "# Implementation Plan", |
| 4326 | "", |
| 4327 | "## File Changes", |
| 4328 | f"- `{guide_root}/`", |
| 4329 | f"- `{chapters}/`", |
| 4330 | f"- `{index_path}`", |
| 4331 | f"- `{chapter_one}`", |
| 4332 | f"- `{chapter_two}`", |
| 4333 | "", |
| 4334 | ] |
| 4335 | ) |
| 4336 | ) |
| 4337 | |
| 4338 | context = build_context( |
| 4339 | temp_dir=temp_dir, |
| 4340 | messages=[], |
| 4341 | safeguards=FakeSafeguards(), |
| 4342 | assess_confidence=assess_confidence, |
| 4343 | verify_action=verify_action, |
| 4344 | auto_recover=False, |
| 4345 | ) |
| 4346 | queued_messages: list[str] = [] |
| 4347 | context.queue_steering_message_callback = queued_messages.append |
| 4348 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 4349 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 4350 | dod.implementation_plan = str(implementation_plan) |
| 4351 | sync_todos_to_definition_of_done( |
| 4352 | dod, |
| 4353 | [ |
| 4354 | { |
| 4355 | "content": "Create 01-getting-started.html", |
| 4356 | "active_form": "Creating 01-getting-started.html", |
| 4357 | "status": "completed", |
| 4358 | }, |
| 4359 | { |
| 4360 | "content": "Create 02-installation.html", |
| 4361 | "active_form": "Creating 02-installation.html", |
| 4362 | "status": "pending", |
| 4363 | }, |
| 4364 | ], |
| 4365 | project_root=temp_dir, |
| 4366 | ) |
| 4367 | dod.touched_files.extend([str(index_path), str(chapter_one)]) |
| 4368 | |
| 4369 | tool_call = ToolCall( |
| 4370 | id="working-note", |
| 4371 | name="notepad_write_working", |
| 4372 | arguments={"content": "Creating the second chapter file: Installation"}, |
| 4373 | ) |
| 4374 | executor = FakeExecutor( |
| 4375 | [ |
| 4376 | tool_outcome( |
| 4377 | tool_call=tool_call, |
| 4378 | output="Working note recorded", |
| 4379 | is_error=False, |
| 4380 | ) |
| 4381 | ] |
| 4382 | ) |
| 4383 | |
| 4384 | summary = TurnSummary(final_response="") |
| 4385 | await runner.execute_batch( |
| 4386 | tool_calls=[tool_call], |
| 4387 | tool_source="assistant", |
| 4388 | pending_tool_calls_seen=set(), |
| 4389 | emit=_noop_emit, |
| 4390 | summary=summary, |
| 4391 | dod=dod, |
| 4392 | executor=executor, # type: ignore[arg-type] |
| 4393 | on_confirmation=None, |
| 4394 | on_user_question=None, |
| 4395 | emit_confirmation=None, |
| 4396 | consecutive_errors=0, |
| 4397 | ) |
| 4398 | |
| 4399 | assert queued_messages |
| 4400 | message = queued_messages[-1] |
| 4401 | assert "Bookkeeping note is recorded. A declared output artifact is still missing." in message |
| 4402 | assert "Resume by creating `02-installation.html` now." in message |
| 4403 | assert "Make your next response the concrete mutation tool call itself" in message |
| 4404 | assert "refresh `TodoWrite`" in message |
| 4405 | assert "Do not spend the next turn on additional notes, rediscovery, verification, or final confirmation" in message |
| 4406 | |
| 4407 | |
| 4408 | @pytest.mark.asyncio |
| 4409 | async def test_tool_batch_runner_working_note_respects_discovery_first_pending_step( |
| 4410 | temp_dir: Path, |
| 4411 | ) -> None: |
| 4412 | async def assess_confidence( |
| 4413 | tool_name: str, |
| 4414 | tool_args: dict, |
| 4415 | context: str, |
| 4416 | ) -> ConfidenceAssessment: |
| 4417 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 4418 | |
| 4419 | async def verify_action( |
| 4420 | tool_name: str, |
| 4421 | tool_args: dict, |
| 4422 | result: str, |
| 4423 | expected: str = "", |
| 4424 | ) -> ActionVerification: |
| 4425 | raise AssertionError("Verification should not run in this scenario") |
| 4426 | |
| 4427 | implementation_plan = temp_dir / "implementation.md" |
| 4428 | implementation_plan.write_text( |
| 4429 | "\n".join( |
| 4430 | [ |
| 4431 | "# Implementation Plan", |
| 4432 | "", |
| 4433 | "## File Changes", |
| 4434 | f"- `{temp_dir / 'guides' / 'nginx' / 'index.html'}`", |
| 4435 | f"- `{temp_dir / 'guides' / 'nginx' / 'chapters'}`", |
| 4436 | "", |
| 4437 | ] |
| 4438 | ) |
| 4439 | ) |
| 4440 | |
| 4441 | context = build_context( |
| 4442 | temp_dir=temp_dir, |
| 4443 | messages=[], |
| 4444 | safeguards=FakeSafeguards(), |
| 4445 | assess_confidence=assess_confidence, |
| 4446 | verify_action=verify_action, |
| 4447 | auto_recover=False, |
| 4448 | ) |
| 4449 | queued_messages: list[str] = [] |
| 4450 | context.queue_steering_message_callback = queued_messages.append |
| 4451 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 4452 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 4453 | dod.implementation_plan = str(implementation_plan) |
| 4454 | dod.pending_items.extend( |
| 4455 | [ |
| 4456 | "First, examine the existing fortran guide structure and content to understand the format", |
| 4457 | "Create the nginx directory structure", |
| 4458 | "Develop the main index.html file for the nginx guide", |
| 4459 | ] |
| 4460 | ) |
| 4461 | |
| 4462 | tool_call = ToolCall( |
| 4463 | id="working-note", |
| 4464 | name="notepad_write_working", |
| 4465 | arguments={"content": "Analyzing the fortran guide structure before creating nginx guide"}, |
| 4466 | ) |
| 4467 | executor = FakeExecutor( |
| 4468 | [ |
| 4469 | tool_outcome( |
| 4470 | tool_call=tool_call, |
| 4471 | output="Working note recorded", |
| 4472 | is_error=False, |
| 4473 | ) |
| 4474 | ] |
| 4475 | ) |
| 4476 | |
| 4477 | summary = TurnSummary(final_response="") |
| 4478 | await runner.execute_batch( |
| 4479 | tool_calls=[tool_call], |
| 4480 | tool_source="assistant", |
| 4481 | pending_tool_calls_seen=set(), |
| 4482 | emit=_noop_emit, |
| 4483 | summary=summary, |
| 4484 | dod=dod, |
| 4485 | executor=executor, # type: ignore[arg-type] |
| 4486 | on_confirmation=None, |
| 4487 | on_user_question=None, |
| 4488 | emit_confirmation=None, |
| 4489 | consecutive_errors=0, |
| 4490 | ) |
| 4491 | |
| 4492 | assert queued_messages |
| 4493 | message = queued_messages[-1] |
| 4494 | assert ( |
| 4495 | "Continue with the next pending item: `First, examine the existing fortran guide structure and content to understand the format`." |
| 4496 | in message |
| 4497 | ) |
| 4498 | assert "one concrete evidence-gathering tool call" in message |
| 4499 | assert "Resume by creating `index.html` now." not in message |
| 4500 | |
| 4501 | |
| 4502 | @pytest.mark.asyncio |
| 4503 | async def test_tool_batch_runner_working_note_prefers_declared_output_gap_over_stale_discovery( |
| 4504 | temp_dir: Path, |
| 4505 | ) -> None: |
| 4506 | async def assess_confidence( |
| 4507 | tool_name: str, |
| 4508 | tool_args: dict, |
| 4509 | context: str, |
| 4510 | ) -> ConfidenceAssessment: |
| 4511 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 4512 | |
| 4513 | async def verify_action( |
| 4514 | tool_name: str, |
| 4515 | tool_args: dict, |
| 4516 | result: str, |
| 4517 | expected: str = "", |
| 4518 | ) -> ActionVerification: |
| 4519 | raise AssertionError("Verification should not run in this scenario") |
| 4520 | |
| 4521 | guide_root = temp_dir / "guides" / "nginx" |
| 4522 | chapters_dir = guide_root / "chapters" |
| 4523 | chapters_dir.mkdir(parents=True) |
| 4524 | index_path = guide_root / "index.html" |
| 4525 | first_chapter = chapters_dir / "01-introduction.html" |
| 4526 | index_path.write_text( |
| 4527 | "\n".join( |
| 4528 | [ |
| 4529 | '<a href="chapters/01-introduction.html">Introduction</a>', |
| 4530 | '<a href="chapters/02-installation.html">Installation</a>', |
| 4531 | '<a href="chapters/03-configuration.html">Configuration</a>', |
| 4532 | ] |
| 4533 | ) |
| 4534 | ) |
| 4535 | first_chapter.write_text("<h1>Introduction</h1>\n") |
| 4536 | |
| 4537 | implementation_plan = temp_dir / "implementation.md" |
| 4538 | implementation_plan.write_text( |
| 4539 | "\n".join( |
| 4540 | [ |
| 4541 | "# Implementation Plan", |
| 4542 | "", |
| 4543 | "## File Changes", |
| 4544 | f"- `{guide_root / 'index.html'}`", |
| 4545 | f"- `{chapters_dir}/`", |
| 4546 | "", |
| 4547 | ] |
| 4548 | ) |
| 4549 | ) |
| 4550 | |
| 4551 | context = build_context( |
| 4552 | temp_dir=temp_dir, |
| 4553 | messages=[], |
| 4554 | safeguards=FakeSafeguards(), |
| 4555 | assess_confidence=assess_confidence, |
| 4556 | verify_action=verify_action, |
| 4557 | auto_recover=False, |
| 4558 | ) |
| 4559 | queued_messages: list[str] = [] |
| 4560 | context.queue_steering_message_callback = queued_messages.append |
| 4561 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 4562 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 4563 | dod.implementation_plan = str(implementation_plan) |
| 4564 | dod.pending_items.extend( |
| 4565 | [ |
| 4566 | "First, examine the existing fortran guide structure and content to understand the format", |
| 4567 | "Create chapter files following the established pattern", |
| 4568 | ] |
| 4569 | ) |
| 4570 | dod.touched_files.extend([str(index_path), str(first_chapter)]) |
| 4571 | |
| 4572 | tool_call = ToolCall( |
| 4573 | id="working-note", |
| 4574 | name="notepad_write_working", |
| 4575 | arguments={"content": "Created index and first chapter; next is chapter 2"}, |
| 4576 | ) |
| 4577 | executor = FakeExecutor( |
| 4578 | [ |
| 4579 | tool_outcome( |
| 4580 | tool_call=tool_call, |
| 4581 | output="Working note recorded", |
| 4582 | is_error=False, |
| 4583 | ) |
| 4584 | ] |
| 4585 | ) |
| 4586 | |
| 4587 | summary = TurnSummary(final_response="") |
| 4588 | await runner.execute_batch( |
| 4589 | tool_calls=[tool_call], |
| 4590 | tool_source="assistant", |
| 4591 | pending_tool_calls_seen=set(), |
| 4592 | emit=_noop_emit, |
| 4593 | summary=summary, |
| 4594 | dod=dod, |
| 4595 | executor=executor, # type: ignore[arg-type] |
| 4596 | on_confirmation=None, |
| 4597 | on_user_question=None, |
| 4598 | emit_confirmation=None, |
| 4599 | consecutive_errors=0, |
| 4600 | ) |
| 4601 | |
| 4602 | assert queued_messages |
| 4603 | message = queued_messages[-1] |
| 4604 | assert "Bookkeeping note is recorded. A declared output artifact is still missing." in message |
| 4605 | assert "Resume by creating `02-installation.html` now." in message |
| 4606 | assert "Continue with the next pending item: `First, examine the existing fortran guide structure" not in message |
| 4607 | |
| 4608 | |
| 4609 | @pytest.mark.asyncio |
| 4610 | async def test_tool_batch_runner_shallow_glob_does_not_handoff_before_content_read( |
| 4611 | temp_dir: Path, |
| 4612 | ) -> None: |
| 4613 | async def assess_confidence( |
| 4614 | tool_name: str, |
| 4615 | tool_args: dict, |
| 4616 | context: str, |
| 4617 | ) -> ConfidenceAssessment: |
| 4618 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 4619 | |
| 4620 | async def verify_action( |
| 4621 | tool_name: str, |
| 4622 | tool_args: dict, |
| 4623 | result: str, |
| 4624 | expected: str = "", |
| 4625 | ) -> ActionVerification: |
| 4626 | raise AssertionError("Verification should not run in this scenario") |
| 4627 | |
| 4628 | fortran_root = temp_dir / "Loader" / "guides" / "fortran" |
| 4629 | chapters_dir = fortran_root / "chapters" |
| 4630 | chapters_dir.mkdir(parents=True) |
| 4631 | |
| 4632 | implementation_plan = temp_dir / "implementation.md" |
| 4633 | implementation_plan.write_text( |
| 4634 | "\n".join( |
| 4635 | [ |
| 4636 | "# Implementation Plan", |
| 4637 | "", |
| 4638 | "## File Changes", |
| 4639 | f"- `{temp_dir / 'Loader' / 'guides' / 'nginx' / 'index.html'}`", |
| 4640 | f"- `{temp_dir / 'Loader' / 'guides' / 'nginx' / 'chapters'}`", |
| 4641 | "", |
| 4642 | ] |
| 4643 | ) |
| 4644 | ) |
| 4645 | |
| 4646 | context = build_context( |
| 4647 | temp_dir=temp_dir, |
| 4648 | messages=[], |
| 4649 | safeguards=FakeSafeguards(), |
| 4650 | assess_confidence=assess_confidence, |
| 4651 | verify_action=verify_action, |
| 4652 | auto_recover=False, |
| 4653 | ) |
| 4654 | queued_messages: list[str] = [] |
| 4655 | context.queue_steering_message_callback = queued_messages.append |
| 4656 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 4657 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 4658 | dod.implementation_plan = str(implementation_plan) |
| 4659 | dod.pending_items.extend( |
| 4660 | [ |
| 4661 | "First, examine the existing fortran guide structure and content", |
| 4662 | "Create the nginx directory structure", |
| 4663 | "Develop the main index.html file for nginx guide", |
| 4664 | ] |
| 4665 | ) |
| 4666 | |
| 4667 | tool_call = ToolCall( |
| 4668 | id="glob-1", |
| 4669 | name="glob", |
| 4670 | arguments={"pattern": "**", "path": str(fortran_root)}, |
| 4671 | ) |
| 4672 | executor = FakeExecutor( |
| 4673 | [ |
| 4674 | tool_outcome( |
| 4675 | tool_call=tool_call, |
| 4676 | output=f"{fortran_root}\n{chapters_dir}", |
| 4677 | is_error=False, |
| 4678 | ) |
| 4679 | ] |
| 4680 | ) |
| 4681 | |
| 4682 | summary = TurnSummary(final_response="") |
| 4683 | await runner.execute_batch( |
| 4684 | tool_calls=[tool_call], |
| 4685 | tool_source="assistant", |
| 4686 | pending_tool_calls_seen=set(), |
| 4687 | emit=_noop_emit, |
| 4688 | summary=summary, |
| 4689 | dod=dod, |
| 4690 | executor=executor, # type: ignore[arg-type] |
| 4691 | on_confirmation=None, |
| 4692 | on_user_question=None, |
| 4693 | emit_confirmation=None, |
| 4694 | consecutive_errors=0, |
| 4695 | ) |
| 4696 | |
| 4697 | assert queued_messages == [] |
| 4698 | |
| 4699 | |
| 4700 | @pytest.mark.asyncio |
| 4701 | async def test_tool_batch_runner_hands_off_noop_toc_edit_when_file_is_already_valid( |
| 4702 | temp_dir: Path, |
| 4703 | ) -> None: |
| 4704 | async def assess_confidence( |
| 4705 | tool_name: str, |
| 4706 | tool_args: dict, |
| 4707 | context: str, |
| 4708 | ) -> ConfidenceAssessment: |
| 4709 | raise AssertionError("Confidence scoring should not run in this scenario") |
| 4710 | |
| 4711 | async def verify_action( |
| 4712 | tool_name: str, |
| 4713 | tool_args: dict, |
| 4714 | result: str, |
| 4715 | expected: str = "", |
| 4716 | ) -> ActionVerification: |
| 4717 | raise AssertionError("Verification should not run in this scenario") |
| 4718 | |
| 4719 | prompt = ( |
| 4720 | "Have a look at ~/Loader/guides/fortran/index.html, then " |
| 4721 | "~/Loader/guides/fortran/chapters. The table of contents links in " |
| 4722 | "index.html are inaccurate and the href’s are wrong. Let’s update the " |
| 4723 | "links and their link texts to be correct." |
| 4724 | ) |
| 4725 | chapters = temp_dir / "chapters" |
| 4726 | chapters.mkdir() |
| 4727 | (chapters / "01-introduction.html").write_text( |
| 4728 | "<h1>Chapter 1: Introduction to Fortran</h1>\n" |
| 4729 | ) |
| 4730 | (chapters / "02-setup.html").write_text( |
| 4731 | "<h1>Chapter 2: Setting Up Your Environment</h1>\n" |
| 4732 | ) |
| 4733 | current_block = ( |
| 4734 | "<h2>Table of Contents</h2>\n" |
| 4735 | ' <ul class="chapter-list">\n' |
| 4736 | ' <li><a href="chapters/01-introduction.html">Chapter 1: Introduction to Fortran</a></li>\n' |
| 4737 | ' <li><a href="chapters/02-setup.html">Chapter 2: Setting Up Your Environment</a></li>\n' |
| 4738 | " </ul>\n" |
| 4739 | ) |
| 4740 | index_path = temp_dir / "index.html" |
| 4741 | index_path.write_text(current_block) |
| 4742 | |
| 4743 | context = build_context( |
| 4744 | temp_dir=temp_dir, |
| 4745 | messages=[], |
| 4746 | safeguards=FakeSafeguards(), |
| 4747 | assess_confidence=assess_confidence, |
| 4748 | verify_action=verify_action, |
| 4749 | auto_recover=False, |
| 4750 | ) |
| 4751 | context.session.current_task = prompt # type: ignore[attr-defined] |
| 4752 | queued_messages: list[str] = [] |
| 4753 | context.queue_steering_message_callback = queued_messages.append |
| 4754 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 4755 | tool_call = ToolCall( |
| 4756 | id="edit-1", |
| 4757 | name="edit", |
| 4758 | arguments={ |
| 4759 | "file_path": str(index_path), |
| 4760 | "old_string": current_block, |
| 4761 | "new_string": current_block, |
| 4762 | }, |
| 4763 | ) |
| 4764 | executor = FakeExecutor( |
| 4765 | [ |
| 4766 | tool_outcome( |
| 4767 | tool_call=tool_call, |
| 4768 | output=( |
| 4769 | "[Blocked - old_string and new_string are identical - no change " |
| 4770 | "would occur] Suggestion: Provide different old and new strings" |
| 4771 | ), |
| 4772 | is_error=True, |
| 4773 | state=ToolExecutionState.BLOCKED, |
| 4774 | ) |
| 4775 | ] |
| 4776 | ) |
| 4777 | |
| 4778 | await runner.execute_batch( |
| 4779 | tool_calls=[tool_call], |
| 4780 | tool_source="assistant", |
| 4781 | pending_tool_calls_seen=set(), |
| 4782 | emit=_noop_emit, |
| 4783 | summary=TurnSummary(final_response=""), |
| 4784 | dod=create_definition_of_done(prompt), |
| 4785 | executor=executor, # type: ignore[arg-type] |
| 4786 | on_confirmation=None, |
| 4787 | on_user_question=None, |
| 4788 | emit_confirmation=None, |
| 4789 | consecutive_errors=0, |
| 4790 | ) |
| 4791 | |
| 4792 | assert queued_messages == [] |
| 4793 | |
| 4794 | |
| 4795 | def test_tool_batch_runner_blocked_noop_edit_nudge_stays_on_active_repair_target( |
| 4796 | temp_dir: Path, |
| 4797 | ) -> None: |
| 4798 | async def assess_confidence( |
| 4799 | tool_name: str, |
| 4800 | tool_args: dict, |
| 4801 | context: str, |
| 4802 | ) -> ConfidenceAssessment: |
| 4803 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 4804 | |
| 4805 | async def verify_action( |
| 4806 | tool_name: str, |
| 4807 | tool_args: dict, |
| 4808 | result: str, |
| 4809 | expected: str = "", |
| 4810 | ) -> ActionVerification: |
| 4811 | raise AssertionError("Verification should not run in this scenario") |
| 4812 | |
| 4813 | repair_target = temp_dir / "guide" / "chapters" / "04-basic-usage.html" |
| 4814 | context = build_context( |
| 4815 | temp_dir=temp_dir, |
| 4816 | messages=[ |
| 4817 | Message( |
| 4818 | role=Role.ASSISTANT, |
| 4819 | content=( |
| 4820 | "Repair focus:\n" |
| 4821 | f"- Fix the broken local reference `05-advanced-topics.html` in `{repair_target}`.\n" |
| 4822 | f"- Immediate next step: edit `{repair_target}`.\n" |
| 4823 | 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" |
| 4824 | ), |
| 4825 | ) |
| 4826 | ], |
| 4827 | safeguards=FakeSafeguards(), |
| 4828 | assess_confidence=assess_confidence, |
| 4829 | verify_action=verify_action, |
| 4830 | ) |
| 4831 | queued: list[str] = [] |
| 4832 | context.queue_steering_message_callback = queued.append |
| 4833 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 4834 | |
| 4835 | runner._queue_blocked_html_edit_nudge( |
| 4836 | ToolCall( |
| 4837 | id="edit-1", |
| 4838 | name="edit", |
| 4839 | arguments={ |
| 4840 | "file_path": str(repair_target), |
| 4841 | "old_string": "same", |
| 4842 | "new_string": "same", |
| 4843 | }, |
| 4844 | ), |
| 4845 | "[Blocked - old_string and new_string are identical - no change would occur] Suggestion: Provide different old and new strings", |
| 4846 | ) |
| 4847 | |
| 4848 | assert queued |
| 4849 | assert str(repair_target) in queued[0] |
| 4850 | assert "no on-disk change" in queued[0] |
| 4851 | assert "replace the surrounding block" in queued[0] |
| 4852 | assert "Do not reopen unrelated reference materials" in queued[0] |
| 4853 | |
| 4854 | |
| 4855 | async def _noop_emit(event: AgentEvent) -> None: |
| 4856 | return None |
| 4857 | |
| 4858 | |
| 4859 | @pytest.mark.asyncio |
| 4860 | async def test_tool_batch_runner_marks_verification_planned_after_new_mutation( |
| 4861 | temp_dir: Path, |
| 4862 | ) -> None: |
| 4863 | async def assess_confidence( |
| 4864 | tool_name: str, |
| 4865 | tool_args: dict, |
| 4866 | context: str, |
| 4867 | ) -> ConfidenceAssessment: |
| 4868 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 4869 | |
| 4870 | async def verify_action( |
| 4871 | tool_name: str, |
| 4872 | tool_args: dict, |
| 4873 | result: str, |
| 4874 | expected: str = "", |
| 4875 | ) -> ActionVerification: |
| 4876 | raise AssertionError("Verification should not run for this scenario") |
| 4877 | |
| 4878 | context = build_context( |
| 4879 | temp_dir=temp_dir, |
| 4880 | messages=[], |
| 4881 | safeguards=FakeSafeguards(), |
| 4882 | assess_confidence=assess_confidence, |
| 4883 | verify_action=verify_action, |
| 4884 | ) |
| 4885 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 4886 | tool_call = ToolCall( |
| 4887 | id="write-1", |
| 4888 | name="write", |
| 4889 | arguments={"file_path": str(temp_dir / "README.md"), "content": "updated\n"}, |
| 4890 | ) |
| 4891 | executor = FakeExecutor( |
| 4892 | [tool_outcome(tool_call=tool_call, output="wrote file", is_error=False)] |
| 4893 | ) |
| 4894 | summary = TurnSummary(final_response="") |
| 4895 | dod = create_definition_of_done("Update README and verify it still works.") |
| 4896 | events: list[AgentEvent] = [] |
| 4897 | |
| 4898 | async def emit(event: AgentEvent) -> None: |
| 4899 | events.append(event) |
| 4900 | |
| 4901 | await runner.execute_batch( |
| 4902 | tool_calls=[tool_call], |
| 4903 | tool_source="assistant", |
| 4904 | pending_tool_calls_seen=set(), |
| 4905 | emit=emit, |
| 4906 | summary=summary, |
| 4907 | dod=dod, |
| 4908 | executor=executor, # type: ignore[arg-type] |
| 4909 | on_confirmation=None, |
| 4910 | on_user_question=None, |
| 4911 | emit_confirmation=None, |
| 4912 | consecutive_errors=0, |
| 4913 | ) |
| 4914 | |
| 4915 | assert dod.last_verification_result == "planned" |
| 4916 | assert dod.verification_commands |
| 4917 | assert "Collect verification evidence" in dod.pending_items |
| 4918 | assert dod.active_verification_attempt_id == "verification-attempt-1" |
| 4919 | assert dod.active_verification_attempt_number == 1 |
| 4920 | assert summary.workflow_timeline[-1].reason_code == "verification_planned" |
| 4921 | assert summary.workflow_timeline[-1].policy_outcome == "planned" |
| 4922 | assert summary.workflow_timeline[-1].verification_observations[0].status == "planned" |
| 4923 | assert ( |
| 4924 | summary.workflow_timeline[-1].verification_observations[0].attempt_id |
| 4925 | == "verification-attempt-1" |
| 4926 | ) |
| 4927 | assert ( |
| 4928 | summary.workflow_timeline[-1].verification_observations[0].attempt_number == 1 |
| 4929 | ) |
| 4930 | |
| 4931 | |
| 4932 | @pytest.mark.asyncio |
| 4933 | async def test_tool_batch_runner_does_not_mark_verification_planned_after_setup_only_mkdir( |
| 4934 | temp_dir: Path, |
| 4935 | ) -> None: |
| 4936 | async def assess_confidence( |
| 4937 | tool_name: str, |
| 4938 | tool_args: dict, |
| 4939 | context: str, |
| 4940 | ) -> ConfidenceAssessment: |
| 4941 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 4942 | |
| 4943 | async def verify_action( |
| 4944 | tool_name: str, |
| 4945 | tool_args: dict, |
| 4946 | result: str, |
| 4947 | expected: str = "", |
| 4948 | ) -> ActionVerification: |
| 4949 | raise AssertionError("Verification should not run in this scenario") |
| 4950 | |
| 4951 | context = build_context( |
| 4952 | temp_dir=temp_dir, |
| 4953 | messages=[], |
| 4954 | safeguards=FakeSafeguards(), |
| 4955 | assess_confidence=assess_confidence, |
| 4956 | verify_action=verify_action, |
| 4957 | ) |
| 4958 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 4959 | nginx_root = temp_dir / "Loader" / "guides" / "nginx" |
| 4960 | chapters = nginx_root / "chapters" |
| 4961 | implementation_plan = temp_dir / "implementation.md" |
| 4962 | implementation_plan.write_text( |
| 4963 | "\n".join( |
| 4964 | [ |
| 4965 | "# Implementation Plan", |
| 4966 | "", |
| 4967 | "## File Changes", |
| 4968 | f"- `{chapters}/`", |
| 4969 | f"- `{nginx_root / 'index.html'}`", |
| 4970 | "", |
| 4971 | ] |
| 4972 | ) |
| 4973 | ) |
| 4974 | |
| 4975 | tool_call = ToolCall( |
| 4976 | id="mkdir-1", |
| 4977 | name="bash", |
| 4978 | arguments={"command": f"mkdir -p {chapters}"}, |
| 4979 | ) |
| 4980 | executor = FakeExecutor( |
| 4981 | [tool_outcome(tool_call=tool_call, output="", is_error=False)] |
| 4982 | ) |
| 4983 | summary = TurnSummary(final_response="") |
| 4984 | dod = create_definition_of_done("Create an equally thorough nginx guide with chapters.") |
| 4985 | dod.implementation_plan = str(implementation_plan) |
| 4986 | events: list[AgentEvent] = [] |
| 4987 | |
| 4988 | async def emit(event: AgentEvent) -> None: |
| 4989 | events.append(event) |
| 4990 | |
| 4991 | await runner.execute_batch( |
| 4992 | tool_calls=[tool_call], |
| 4993 | tool_source="assistant", |
| 4994 | pending_tool_calls_seen=set(), |
| 4995 | emit=emit, |
| 4996 | summary=summary, |
| 4997 | dod=dod, |
| 4998 | executor=executor, # type: ignore[arg-type] |
| 4999 | on_confirmation=None, |
| 5000 | on_user_question=None, |
| 5001 | emit_confirmation=None, |
| 5002 | consecutive_errors=0, |
| 5003 | ) |
| 5004 | |
| 5005 | assert dod.last_verification_result is None |
| 5006 | assert "Collect verification evidence" not in dod.pending_items |
| 5007 | assert not any( |
| 5008 | entry.reason_code == "verification_planned" for entry in summary.workflow_timeline |
| 5009 | ) |
| 5010 | |
| 5011 | |
| 5012 | @pytest.mark.asyncio |
| 5013 | async def test_tool_batch_runner_marks_passed_verification_stale_after_new_mutation( |
| 5014 | temp_dir: Path, |
| 5015 | ) -> None: |
| 5016 | async def assess_confidence( |
| 5017 | tool_name: str, |
| 5018 | tool_args: dict, |
| 5019 | context: str, |
| 5020 | ) -> ConfidenceAssessment: |
| 5021 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 5022 | |
| 5023 | async def verify_action( |
| 5024 | tool_name: str, |
| 5025 | tool_args: dict, |
| 5026 | result: str, |
| 5027 | expected: str = "", |
| 5028 | ) -> ActionVerification: |
| 5029 | raise AssertionError("Verification should not run for this scenario") |
| 5030 | |
| 5031 | context = build_context( |
| 5032 | temp_dir=temp_dir, |
| 5033 | messages=[], |
| 5034 | safeguards=FakeSafeguards(), |
| 5035 | assess_confidence=assess_confidence, |
| 5036 | verify_action=verify_action, |
| 5037 | ) |
| 5038 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 5039 | tool_call = ToolCall( |
| 5040 | id="write-1", |
| 5041 | name="write", |
| 5042 | arguments={"file_path": str(temp_dir / "README.md"), "content": "updated\n"}, |
| 5043 | ) |
| 5044 | executor = FakeExecutor( |
| 5045 | [tool_outcome(tool_call=tool_call, output="wrote file", is_error=False)] |
| 5046 | ) |
| 5047 | summary = TurnSummary(final_response="") |
| 5048 | dod = create_definition_of_done("Update README and verify it still works.") |
| 5049 | dod.verification_commands = ["uv run pytest -q"] |
| 5050 | dod.last_verification_result = "passed" |
| 5051 | dod.verification_attempt_counter = 1 |
| 5052 | dod.active_verification_attempt_id = "verification-attempt-1" |
| 5053 | dod.active_verification_attempt_number = 1 |
| 5054 | dod.evidence = [ |
| 5055 | VerificationEvidence( |
| 5056 | command="uv run pytest -q", |
| 5057 | passed=True, |
| 5058 | stdout="401 passed", |
| 5059 | kind="test", |
| 5060 | ) |
| 5061 | ] |
| 5062 | dod.completed_items.append("Collect verification evidence") |
| 5063 | events: list[AgentEvent] = [] |
| 5064 | |
| 5065 | async def emit(event: AgentEvent) -> None: |
| 5066 | events.append(event) |
| 5067 | |
| 5068 | await runner.execute_batch( |
| 5069 | tool_calls=[tool_call], |
| 5070 | tool_source="assistant", |
| 5071 | pending_tool_calls_seen=set(), |
| 5072 | emit=emit, |
| 5073 | summary=summary, |
| 5074 | dod=dod, |
| 5075 | executor=executor, # type: ignore[arg-type] |
| 5076 | on_confirmation=None, |
| 5077 | on_user_question=None, |
| 5078 | emit_confirmation=None, |
| 5079 | consecutive_errors=0, |
| 5080 | ) |
| 5081 | |
| 5082 | assert dod.last_verification_result == "stale" |
| 5083 | assert dod.evidence == [] |
| 5084 | assert "Collect verification evidence" in dod.pending_items |
| 5085 | assert "Collect verification evidence" not in dod.completed_items |
| 5086 | assert dod.active_verification_attempt_id == "verification-attempt-2" |
| 5087 | assert dod.active_verification_attempt_number == 2 |
| 5088 | assert summary.workflow_timeline[-1].reason_code == "verification_stale" |
| 5089 | assert summary.workflow_timeline[-1].policy_outcome == "stale" |
| 5090 | assert summary.workflow_timeline[-1].verification_observations[0].status == "stale" |
| 5091 | assert ( |
| 5092 | summary.workflow_timeline[-1].verification_observations[0].attempt_id |
| 5093 | == "verification-attempt-1" |
| 5094 | ) |
| 5095 | assert ( |
| 5096 | summary.workflow_timeline[-1].verification_observations[0].attempt_number == 1 |
| 5097 | ) |
| 5098 | assert ( |
| 5099 | summary.workflow_timeline[-1].verification_observations[0].supersedes_attempt_id |
| 5100 | == "verification-attempt-2" |
| 5101 | ) |
| 5102 | assert ( |
| 5103 | summary.workflow_timeline[-1].verification_observations[0].command |
| 5104 | == "uv run pytest -q" |
| 5105 | ) |
| 5106 | |
| 5107 | |
| 5108 | def test_tool_batch_runner_blocked_active_repair_nudge_uses_repair_scope(temp_dir: Path) -> None: |
| 5109 | async def assess_confidence( |
| 5110 | tool_name: str, |
| 5111 | tool_args: dict, |
| 5112 | context: str, |
| 5113 | ) -> ConfidenceAssessment: |
| 5114 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 5115 | |
| 5116 | async def verify_action( |
| 5117 | tool_name: str, |
| 5118 | tool_args: dict, |
| 5119 | result: str, |
| 5120 | expected: str = "", |
| 5121 | ) -> ActionVerification: |
| 5122 | raise AssertionError("Verification should not run in this scenario") |
| 5123 | |
| 5124 | repair_target = temp_dir / "guide" / "index.html" |
| 5125 | context = build_context( |
| 5126 | temp_dir=temp_dir, |
| 5127 | messages=[ |
| 5128 | Message( |
| 5129 | role=Role.ASSISTANT, |
| 5130 | content=( |
| 5131 | "Repair focus:\n" |
| 5132 | f"- Fix the broken local reference `chapters/01-getting-started.html` in `{repair_target}`.\n" |
| 5133 | f"- Immediate next step: edit `{repair_target}`.\n" |
| 5134 | 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" |
| 5135 | ), |
| 5136 | ) |
| 5137 | ], |
| 5138 | safeguards=FakeSafeguards(), |
| 5139 | assess_confidence=assess_confidence, |
| 5140 | verify_action=verify_action, |
| 5141 | ) |
| 5142 | queued: list[str] = [] |
| 5143 | context.queue_steering_message_callback = queued.append |
| 5144 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 5145 | |
| 5146 | runner._queue_blocked_active_repair_nudge( |
| 5147 | "[Blocked - active repair scope: verification already identified the repair target.]" |
| 5148 | ) |
| 5149 | |
| 5150 | assert queued |
| 5151 | assert str(repair_target) in queued[0] |
| 5152 | assert str(temp_dir / "guide" / "chapters" / "01-getting-started.html") in queued[0] |
| 5153 | assert "Do not reopen unrelated reference materials" in queued[0] |
| 5154 | |
| 5155 | |
| 5156 | def test_tool_batch_runner_blocked_active_repair_mutation_nudge_uses_allowed_paths( |
| 5157 | temp_dir: Path, |
| 5158 | ) -> None: |
| 5159 | async def assess_confidence( |
| 5160 | tool_name: str, |
| 5161 | tool_args: dict, |
| 5162 | context: str, |
| 5163 | ) -> ConfidenceAssessment: |
| 5164 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 5165 | |
| 5166 | async def verify_action( |
| 5167 | tool_name: str, |
| 5168 | tool_args: dict, |
| 5169 | result: str, |
| 5170 | expected: str = "", |
| 5171 | ) -> ActionVerification: |
| 5172 | raise AssertionError("Verification should not run in this scenario") |
| 5173 | |
| 5174 | repair_target = temp_dir / "guide" / "chapters" / "05-advanced-configurations.html" |
| 5175 | stylesheet = temp_dir / "guide" / "styles.css" |
| 5176 | context = build_context( |
| 5177 | temp_dir=temp_dir, |
| 5178 | messages=[ |
| 5179 | Message( |
| 5180 | role=Role.ASSISTANT, |
| 5181 | content=( |
| 5182 | "Repair focus:\n" |
| 5183 | f"- Fix the broken local reference `../styles.css` in `{repair_target}`.\n" |
| 5184 | f"- Immediate next step: edit `{repair_target}`.\n" |
| 5185 | f"- If the broken reference should remain, create `{stylesheet}`; otherwise remove or replace `../styles.css`.\n" |
| 5186 | ), |
| 5187 | ) |
| 5188 | ], |
| 5189 | safeguards=FakeSafeguards(), |
| 5190 | assess_confidence=assess_confidence, |
| 5191 | verify_action=verify_action, |
| 5192 | ) |
| 5193 | queued: list[str] = [] |
| 5194 | context.queue_steering_message_callback = queued.append |
| 5195 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 5196 | |
| 5197 | runner._queue_blocked_active_repair_mutation_nudge( |
| 5198 | "[Blocked - active repair mutation scope: verification already identified the repair target.]" |
| 5199 | ) |
| 5200 | |
| 5201 | assert queued |
| 5202 | assert str(repair_target) in queued[0] |
| 5203 | assert str(stylesheet) in queued[0] |
| 5204 | assert "before widening the change set" in queued[0] |
| 5205 | |
| 5206 | |
| 5207 | def test_tool_batch_runner_blocked_late_reference_drift_nudge_points_to_missing_artifact( |
| 5208 | temp_dir: Path, |
| 5209 | ) -> None: |
| 5210 | async def assess_confidence( |
| 5211 | tool_name: str, |
| 5212 | tool_args: dict, |
| 5213 | context: str, |
| 5214 | ) -> ConfidenceAssessment: |
| 5215 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 5216 | |
| 5217 | async def verify_action( |
| 5218 | tool_name: str, |
| 5219 | tool_args: dict, |
| 5220 | result: str, |
| 5221 | expected: str = "", |
| 5222 | ) -> ActionVerification: |
| 5223 | raise AssertionError("Verification should not run in this scenario") |
| 5224 | |
| 5225 | context = build_context( |
| 5226 | temp_dir=temp_dir, |
| 5227 | messages=[], |
| 5228 | safeguards=FakeSafeguards(), |
| 5229 | assess_confidence=assess_confidence, |
| 5230 | verify_action=verify_action, |
| 5231 | ) |
| 5232 | queued: list[str] = [] |
| 5233 | context.queue_steering_message_callback = queued.append |
| 5234 | store = DefinitionOfDoneStore(temp_dir) |
| 5235 | dod = create_definition_of_done("Create a multi-file guide from a reference") |
| 5236 | plan_path = temp_dir / "implementation.md" |
| 5237 | plan_path.write_text( |
| 5238 | "# File Changes\n" |
| 5239 | "- `guide/index.html`\n" |
| 5240 | "- `guide/chapters/01-getting-started.html`\n" |
| 5241 | "- `guide/chapters/02-installation.html`\n" |
| 5242 | "- `guide/chapters/03-first-website.html`\n" |
| 5243 | ) |
| 5244 | dod.implementation_plan = str(plan_path) |
| 5245 | (temp_dir / "guide" / "chapters").mkdir(parents=True, exist_ok=True) |
| 5246 | (temp_dir / "guide" / "index.html").write_text("index") |
| 5247 | (temp_dir / "guide" / "chapters" / "01-getting-started.html").write_text("one") |
| 5248 | (temp_dir / "guide" / "chapters" / "02-installation.html").write_text("two") |
| 5249 | runner = ToolBatchRunner(context, store) |
| 5250 | |
| 5251 | runner._queue_blocked_late_reference_drift_nudge( |
| 5252 | "[Blocked - late reference drift: several planned artifacts already exist.]", |
| 5253 | dod=dod, |
| 5254 | ) |
| 5255 | |
| 5256 | assert queued |
| 5257 | assert "03-first-website.html" in queued[0] |
| 5258 | assert "older reference materials" in queued[0] |
| 5259 | |
| 5260 | |
| 5261 | def test_tool_batch_runner_blocked_completed_artifact_scope_nudge_prefers_verification( |
| 5262 | temp_dir: Path, |
| 5263 | ) -> None: |
| 5264 | async def assess_confidence( |
| 5265 | tool_name: str, |
| 5266 | tool_args: dict, |
| 5267 | context: str, |
| 5268 | ) -> ConfidenceAssessment: |
| 5269 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 5270 | |
| 5271 | async def verify_action( |
| 5272 | tool_name: str, |
| 5273 | tool_args: dict, |
| 5274 | result: str, |
| 5275 | expected: str = "", |
| 5276 | ) -> ActionVerification: |
| 5277 | raise AssertionError("Verification should not run in this scenario") |
| 5278 | |
| 5279 | guide_root = temp_dir / "guide" |
| 5280 | chapters = guide_root / "chapters" |
| 5281 | guide_root.mkdir(parents=True) |
| 5282 | chapters.mkdir() |
| 5283 | index_path = guide_root / "index.html" |
| 5284 | chapter_one = chapters / "01-getting-started.html" |
| 5285 | chapter_two = chapters / "02-installation.html" |
| 5286 | index_path.write_text("index") |
| 5287 | chapter_one.write_text("one") |
| 5288 | chapter_two.write_text("two") |
| 5289 | |
| 5290 | implementation_plan = temp_dir / "implementation.md" |
| 5291 | implementation_plan.write_text( |
| 5292 | "\n".join( |
| 5293 | [ |
| 5294 | "# Implementation Plan", |
| 5295 | "", |
| 5296 | "## File Changes", |
| 5297 | f"- `{guide_root}`", |
| 5298 | f"- `{chapters}`", |
| 5299 | f"- `{index_path}`", |
| 5300 | f"- `{chapter_one}`", |
| 5301 | f"- `{chapter_two}`", |
| 5302 | "", |
| 5303 | ] |
| 5304 | ) |
| 5305 | ) |
| 5306 | |
| 5307 | context = build_context( |
| 5308 | temp_dir=temp_dir, |
| 5309 | messages=[], |
| 5310 | safeguards=FakeSafeguards(), |
| 5311 | assess_confidence=assess_confidence, |
| 5312 | verify_action=verify_action, |
| 5313 | ) |
| 5314 | queued: list[str] = [] |
| 5315 | context.queue_steering_message_callback = queued.append |
| 5316 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 5317 | dod = create_definition_of_done("Create a multi-file guide from a reference") |
| 5318 | dod.implementation_plan = str(implementation_plan) |
| 5319 | dod.verification_commands = [f"ls -la {guide_root}"] |
| 5320 | sync_todos_to_definition_of_done( |
| 5321 | dod, |
| 5322 | [ |
| 5323 | { |
| 5324 | "content": "Verify all guide files are linked and complete", |
| 5325 | "active_form": "Working on: Verify all guide files are linked and complete", |
| 5326 | "status": "pending", |
| 5327 | } |
| 5328 | ], |
| 5329 | project_root=temp_dir, |
| 5330 | ) |
| 5331 | |
| 5332 | runner._queue_blocked_completed_artifact_scope_nudge( |
| 5333 | "[Blocked - completed artifact set scope: all explicitly planned artifacts already exist.]", |
| 5334 | dod=dod, |
| 5335 | ) |
| 5336 | |
| 5337 | assert queued |
| 5338 | assert "All explicitly planned artifacts already exist." in queued[0] |
| 5339 | assert "Verify all guide files are linked and complete" in queued[0] |
| 5340 | assert "Do not reopen earlier reference materials." in queued[0] |
| 5341 | |
| 5342 | |
| 5343 | def test_tool_batch_runner_blocked_html_declared_target_nudge_uses_closest_declared_target( |
| 5344 | temp_dir: Path, |
| 5345 | ) -> None: |
| 5346 | async def assess_confidence( |
| 5347 | tool_name: str, |
| 5348 | tool_args: dict, |
| 5349 | context: str, |
| 5350 | ) -> ConfidenceAssessment: |
| 5351 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 5352 | |
| 5353 | async def verify_action( |
| 5354 | tool_name: str, |
| 5355 | tool_args: dict, |
| 5356 | result: str, |
| 5357 | expected: str = "", |
| 5358 | ) -> ActionVerification: |
| 5359 | raise AssertionError("Verification should not run in this scenario") |
| 5360 | |
| 5361 | context = build_context( |
| 5362 | temp_dir=temp_dir, |
| 5363 | messages=[], |
| 5364 | safeguards=FakeSafeguards(), |
| 5365 | assess_confidence=assess_confidence, |
| 5366 | verify_action=verify_action, |
| 5367 | ) |
| 5368 | queued: list[str] = [] |
| 5369 | context.queue_steering_message_callback = queued.append |
| 5370 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 5371 | |
| 5372 | runner._queue_blocked_html_declared_target_nudge( |
| 5373 | ToolCall( |
| 5374 | id="write-ch1", |
| 5375 | name="write", |
| 5376 | arguments={"file_path": str(temp_dir / "guide" / "chapters" / "01-introduction.html")}, |
| 5377 | ), |
| 5378 | ( |
| 5379 | "[Blocked - HTML page introduces new local targets outside the current declared artifact set] " |
| 5380 | "Suggestion: Keep non-root HTML pages within the root-declared local-link set and avoid " |
| 5381 | "introducing new sibling targets that the guide root does not declare, for example fix: 02-setup.html. " |
| 5382 | "Already-declared local targets include: chapters/01-introduction.html, chapters/02-installation.html, " |
| 5383 | "chapters/03-configuration.html. Closest declared local targets include: chapters/02-installation.html" |
| 5384 | ), |
| 5385 | ) |
| 5386 | |
| 5387 | assert queued |
| 5388 | assert str(temp_dir / "guide" / "chapters" / "01-introduction.html") in queued[0] |
| 5389 | assert "`chapters/02-installation.html`" in queued[0] |
| 5390 | assert "same file now" in queued[0] |
| 5391 | |
| 5392 | |
| 5393 | @pytest.mark.asyncio |
| 5394 | async def test_tool_batch_runner_blocked_empty_file_path_nudges_concrete_next_artifact( |
| 5395 | temp_dir: Path, |
| 5396 | ) -> None: |
| 5397 | async def assess_confidence( |
| 5398 | tool_name: str, |
| 5399 | tool_args: dict, |
| 5400 | context: str, |
| 5401 | ) -> ConfidenceAssessment: |
| 5402 | raise AssertionError("Confidence scoring should be disabled in this scenario") |
| 5403 | |
| 5404 | async def verify_action( |
| 5405 | tool_name: str, |
| 5406 | tool_args: dict, |
| 5407 | result: str, |
| 5408 | expected: str = "", |
| 5409 | ) -> ActionVerification: |
| 5410 | raise AssertionError("Verification should not run in this scenario") |
| 5411 | |
| 5412 | guide_root = temp_dir / "guides" / "nginx" |
| 5413 | chapters = guide_root / "chapters" |
| 5414 | chapters.mkdir(parents=True) |
| 5415 | index_path = guide_root / "index.html" |
| 5416 | chapter_one = chapters / "01-introduction.html" |
| 5417 | chapter_two = chapters / "02-installation.html" |
| 5418 | index_path.write_text("<html></html>\n") |
| 5419 | chapter_one.write_text("<h1>Intro</h1>\n") |
| 5420 | |
| 5421 | implementation_plan = temp_dir / "implementation.md" |
| 5422 | implementation_plan.write_text( |
| 5423 | "\n".join( |
| 5424 | [ |
| 5425 | "# Implementation Plan", |
| 5426 | "", |
| 5427 | "## File Changes", |
| 5428 | f"- `{index_path}`", |
| 5429 | f"- `{chapter_one}`", |
| 5430 | f"- `{chapter_two}`", |
| 5431 | "", |
| 5432 | ] |
| 5433 | ) |
| 5434 | ) |
| 5435 | |
| 5436 | context = build_context( |
| 5437 | temp_dir=temp_dir, |
| 5438 | messages=[], |
| 5439 | safeguards=FakeSafeguards(), |
| 5440 | assess_confidence=assess_confidence, |
| 5441 | verify_action=verify_action, |
| 5442 | auto_recover=False, |
| 5443 | ) |
| 5444 | queued: list[str] = [] |
| 5445 | context.queue_steering_message_callback = queued.append |
| 5446 | runner = ToolBatchRunner(context, DefinitionOfDoneStore(temp_dir)) |
| 5447 | tool_call = ToolCall( |
| 5448 | id="write-2", |
| 5449 | name="write", |
| 5450 | arguments={"file_path": "", "content": "<html></html>\n"}, |
| 5451 | ) |
| 5452 | blocked_message = "[Blocked - Empty file path] Suggestion: Provide a valid file path" |
| 5453 | executor = FakeExecutor( |
| 5454 | [ |
| 5455 | ToolExecutionOutcome( |
| 5456 | tool_call=tool_call, |
| 5457 | state=ToolExecutionState.BLOCKED, |
| 5458 | message=Message.tool_result_message( |
| 5459 | tool_call_id=tool_call.id, |
| 5460 | display_content=blocked_message, |
| 5461 | result_content=blocked_message, |
| 5462 | is_error=True, |
| 5463 | ), |
| 5464 | event_content=blocked_message, |
| 5465 | is_error=True, |
| 5466 | result_output=blocked_message, |
| 5467 | ) |
| 5468 | ] |
| 5469 | ) |
| 5470 | dod = create_definition_of_done("Create a multi-file nginx guide.") |
| 5471 | dod.implementation_plan = str(implementation_plan) |
| 5472 | dod.touched_files.extend([str(index_path), str(chapter_one)]) |
| 5473 | dod.pending_items.append("Creating Chapter 2: Installation and Setup") |
| 5474 | |
| 5475 | await runner.execute_batch( |
| 5476 | tool_calls=[tool_call], |
| 5477 | tool_source="assistant", |
| 5478 | pending_tool_calls_seen=set(), |
| 5479 | emit=_noop_emit, |
| 5480 | summary=TurnSummary(final_response=""), |
| 5481 | dod=dod, |
| 5482 | executor=executor, # type: ignore[arg-type] |
| 5483 | on_confirmation=None, |
| 5484 | on_user_question=None, |
| 5485 | emit_confirmation=None, |
| 5486 | consecutive_errors=0, |
| 5487 | ) |
| 5488 | |
| 5489 | assert queued |
| 5490 | assert "did not provide a valid `file_path`" in queued[0] |
| 5491 | assert "Resume by creating `02-installation.html` now." in queued[0] |
| 5492 | assert ( |
| 5493 | f"Prefer one `write` call for `{chapter_two}` instead of more rereads." |
| 5494 | in queued[0] |
| 5495 | ) |
| 5496 | assert context.recovery_context is not None |
| 5497 | assert context.recovery_context.attempts[-1].error == blocked_message |