| 1 | """Tests for no-tool text completion orchestration.""" |
| 2 | |
| 3 | from __future__ import annotations |
| 4 | |
| 5 | from pathlib import Path |
| 6 | |
| 7 | import pytest |
| 8 | |
| 9 | from loader.agent.loop import Agent, AgentConfig |
| 10 | from loader.llm.base import Message, Role |
| 11 | from loader.runtime.conversation import ConversationRuntime |
| 12 | from loader.runtime.dod import VerificationEvidence |
| 13 | from loader.runtime.phases import TurnPhase |
| 14 | from loader.runtime.turn_completion import TurnCompletionAction |
| 15 | from loader.runtime.verification_observations import VerificationObservationStatus |
| 16 | from tests.helpers.runtime_harness import ScriptedBackend |
| 17 | |
| 18 | |
| 19 | def non_streaming_config() -> AgentConfig: |
| 20 | """Shared config for direct turn-completion tests.""" |
| 21 | |
| 22 | return AgentConfig(auto_context=False, stream=False, max_iterations=8) |
| 23 | |
| 24 | |
| 25 | @pytest.mark.asyncio |
| 26 | async def test_turn_completion_requests_continuation_for_premature_text_response( |
| 27 | temp_dir: Path, |
| 28 | ) -> None: |
| 29 | backend = ScriptedBackend() |
| 30 | agent = Agent( |
| 31 | backend=backend, |
| 32 | config=non_streaming_config(), |
| 33 | project_root=temp_dir, |
| 34 | ) |
| 35 | runtime = ConversationRuntime(agent) |
| 36 | events = [] |
| 37 | |
| 38 | async def capture(event) -> None: |
| 39 | events.append(event) |
| 40 | |
| 41 | prepared = await runtime.turn_preparation.prepare( |
| 42 | task="Fix the README heading.", |
| 43 | emit=capture, |
| 44 | requested_mode="execute", |
| 45 | original_task=None, |
| 46 | on_user_question=None, |
| 47 | ) |
| 48 | await runtime.phase_tracker.enter( |
| 49 | TurnPhase.ASSISTANT, |
| 50 | capture, |
| 51 | detail="Requesting assistant response", |
| 52 | reason_code="request_assistant_response", |
| 53 | ) |
| 54 | |
| 55 | decision = await runtime.turn_completion.handle_text_response( |
| 56 | content="I looked into it.", |
| 57 | response_content="I looked into it.", |
| 58 | task=prepared.task, |
| 59 | effective_task=prepared.effective_task, |
| 60 | iterations=1, |
| 61 | max_iterations=agent.config.max_iterations, |
| 62 | actions_taken=[], |
| 63 | continuation_count=0, |
| 64 | dod=prepared.definition_of_done, |
| 65 | emit=capture, |
| 66 | summary=prepared.summary, |
| 67 | executor=prepared.executor, |
| 68 | rollback_plan=prepared.rollback_plan, |
| 69 | ) |
| 70 | |
| 71 | assert decision.action == TurnCompletionAction.CONTINUE |
| 72 | assert decision.continuation_count == 1 |
| 73 | assert prepared.summary.completion_decision_code == "premature_completion_nudge" |
| 74 | assert prepared.summary.completion_decision_summary == ( |
| 75 | "requested one continuation because the non-mutating response looked incomplete" |
| 76 | ) |
| 77 | assert agent.session.last_completion_decision_code == "premature_completion_nudge" |
| 78 | assert [ |
| 79 | entry.decision_code for entry in prepared.summary.completion_trace |
| 80 | ] == ["premature_completion_nudge"] |
| 81 | assert prepared.summary.completion_trace[0].stage == "continuation_check" |
| 82 | assert [entry.kind for entry in prepared.summary.workflow_timeline[-1:]] == [ |
| 83 | "completion_continue" |
| 84 | ] |
| 85 | assert prepared.summary.workflow_timeline[-1].policy_stage == "continuation_check" |
| 86 | assert prepared.summary.workflow_timeline[-1].policy_outcome == "continue" |
| 87 | assert agent.session.messages[-1].role.value == "user" |
| 88 | assert "concrete evidence" in agent.session.messages[-1].content |
| 89 | assert "Carry out the requested change or command now" in agent.session.messages[-1].content |
| 90 | assert any(event.type == "completion_check" for event in events) |
| 91 | |
| 92 | |
| 93 | @pytest.mark.asyncio |
| 94 | async def test_turn_completion_marks_non_mutating_response_done( |
| 95 | temp_dir: Path, |
| 96 | ) -> None: |
| 97 | backend = ScriptedBackend() |
| 98 | agent = Agent( |
| 99 | backend=backend, |
| 100 | config=non_streaming_config(), |
| 101 | project_root=temp_dir, |
| 102 | ) |
| 103 | runtime = ConversationRuntime(agent) |
| 104 | events = [] |
| 105 | |
| 106 | async def capture(event) -> None: |
| 107 | events.append(event) |
| 108 | |
| 109 | prepared = await runtime.turn_preparation.prepare( |
| 110 | task="Explain Loader's clarify loop.", |
| 111 | emit=capture, |
| 112 | requested_mode="execute", |
| 113 | original_task=None, |
| 114 | on_user_question=None, |
| 115 | ) |
| 116 | await runtime.phase_tracker.enter( |
| 117 | TurnPhase.ASSISTANT, |
| 118 | capture, |
| 119 | detail="Requesting assistant response", |
| 120 | reason_code="request_assistant_response", |
| 121 | ) |
| 122 | |
| 123 | decision = await runtime.turn_completion.handle_text_response( |
| 124 | content="Loader uses a bounded clarify loop before execution.", |
| 125 | response_content="Loader uses a bounded clarify loop before execution.", |
| 126 | task=prepared.task, |
| 127 | effective_task=prepared.effective_task, |
| 128 | iterations=1, |
| 129 | max_iterations=agent.config.max_iterations, |
| 130 | actions_taken=[], |
| 131 | continuation_count=0, |
| 132 | dod=prepared.definition_of_done, |
| 133 | emit=capture, |
| 134 | summary=prepared.summary, |
| 135 | executor=prepared.executor, |
| 136 | rollback_plan=prepared.rollback_plan, |
| 137 | ) |
| 138 | |
| 139 | assert decision.action == TurnCompletionAction.COMPLETE |
| 140 | assert prepared.summary.final_response == ( |
| 141 | "Loader uses a bounded clarify loop before execution." |
| 142 | ) |
| 143 | assert prepared.summary.completion_decision_code == "non_mutating_response_accepted" |
| 144 | assert prepared.summary.completion_decision_summary == ( |
| 145 | "accepted the response because no mutating work required verification" |
| 146 | ) |
| 147 | assert agent.session.last_completion_decision_code == ( |
| 148 | "non_mutating_response_accepted" |
| 149 | ) |
| 150 | assert [ |
| 151 | entry.decision_code for entry in prepared.summary.completion_trace |
| 152 | ] == [ |
| 153 | "completion_response_accepted", |
| 154 | "non_mutating_response_accepted", |
| 155 | ] |
| 156 | policy_entries = [ |
| 157 | entry |
| 158 | for entry in prepared.summary.workflow_timeline |
| 159 | if entry.kind.startswith("completion_") |
| 160 | ] |
| 161 | assert [entry.kind for entry in policy_entries] == [ |
| 162 | "completion_check", |
| 163 | "completion_complete", |
| 164 | ] |
| 165 | assert policy_entries[0].policy_stage == "continuation_check" |
| 166 | assert policy_entries[-1].policy_stage == "definition_of_done" |
| 167 | assert [item.summary for item in prepared.summary.completion_trace[-1].evidence_provenance] == [ |
| 168 | "verification was skipped because no mutating work required checks" |
| 169 | ] |
| 170 | assert [ |
| 171 | item.status |
| 172 | for item in prepared.summary.completion_trace[-1].verification_observations |
| 173 | ] == [VerificationObservationStatus.SKIPPED.value] |
| 174 | assert [ |
| 175 | item.summary |
| 176 | for item in prepared.summary.completion_trace[-1].verification_observations |
| 177 | ] == ["verification was skipped because no mutating work required checks"] |
| 178 | assert [item.status for item in policy_entries[-1].verification_observations] == [ |
| 179 | VerificationObservationStatus.SKIPPED.value |
| 180 | ] |
| 181 | assert prepared.definition_of_done.status == "done" |
| 182 | assert prepared.definition_of_done.last_verification_result == "skipped" |
| 183 | assert any(event.type == "response" for event in events) |
| 184 | assert any( |
| 185 | event.type == "dod_status" and event.dod_status == "done" |
| 186 | for event in events |
| 187 | ) |
| 188 | |
| 189 | |
| 190 | @pytest.mark.asyncio |
| 191 | async def test_turn_completion_blocks_false_completion_without_preserving_it( |
| 192 | temp_dir: Path, |
| 193 | ) -> None: |
| 194 | backend = ScriptedBackend() |
| 195 | agent = Agent( |
| 196 | backend=backend, |
| 197 | config=non_streaming_config(), |
| 198 | project_root=temp_dir, |
| 199 | ) |
| 200 | runtime = ConversationRuntime(agent) |
| 201 | events = [] |
| 202 | |
| 203 | async def capture(event) -> None: |
| 204 | events.append(event) |
| 205 | |
| 206 | prepared = await runtime.turn_preparation.prepare( |
| 207 | task=( |
| 208 | "Create a multi-file nginx guide under ~/Loader/guides/nginx " |
| 209 | "with an index and chapter files." |
| 210 | ), |
| 211 | emit=capture, |
| 212 | requested_mode="execute", |
| 213 | original_task=None, |
| 214 | on_user_question=None, |
| 215 | ) |
| 216 | await runtime.phase_tracker.enter( |
| 217 | TurnPhase.ASSISTANT, |
| 218 | capture, |
| 219 | detail="Requesting assistant response", |
| 220 | reason_code="request_assistant_response", |
| 221 | ) |
| 222 | |
| 223 | implementation_plan = temp_dir / "implementation.md" |
| 224 | implementation_plan.write_text( |
| 225 | "# Implementation Plan\n\n" |
| 226 | "## File Changes\n\n" |
| 227 | "1. Create main index.html file:\n" |
| 228 | " - `index.html`\n\n" |
| 229 | "2. Create chapter files:\n" |
| 230 | " - `chapters/01-getting-started.html`\n" |
| 231 | " - `chapters/06-troubleshooting.html`\n" |
| 232 | ) |
| 233 | chapters_dir = temp_dir / "chapters" |
| 234 | chapters_dir.mkdir() |
| 235 | (chapters_dir / "01-getting-started.html").write_text("<h1>Getting Started</h1>\n") |
| 236 | (temp_dir / "index.html").write_text("<h1>NGINX Guide</h1>\n") |
| 237 | |
| 238 | prepared.definition_of_done.implementation_plan = str(implementation_plan) |
| 239 | prepared.definition_of_done.mutating_actions.append("write") |
| 240 | prepared.definition_of_done.touched_files.extend( |
| 241 | [ |
| 242 | str(temp_dir / "index.html"), |
| 243 | str(chapters_dir / "01-getting-started.html"), |
| 244 | ] |
| 245 | ) |
| 246 | |
| 247 | queued_messages: list[str] = [] |
| 248 | runtime.context.queue_steering_message_callback = queued_messages.append |
| 249 | |
| 250 | completion_claim = ( |
| 251 | "I've successfully completed the NGINX guide with all planned files " |
| 252 | "and verified everything is done." |
| 253 | ) |
| 254 | decision = await runtime.turn_completion.handle_text_response( |
| 255 | content=completion_claim, |
| 256 | response_content=completion_claim, |
| 257 | task=prepared.task, |
| 258 | effective_task=prepared.effective_task, |
| 259 | iterations=1, |
| 260 | max_iterations=agent.config.max_iterations, |
| 261 | actions_taken=[], |
| 262 | continuation_count=0, |
| 263 | dod=prepared.definition_of_done, |
| 264 | emit=capture, |
| 265 | summary=prepared.summary, |
| 266 | executor=prepared.executor, |
| 267 | rollback_plan=prepared.rollback_plan, |
| 268 | ) |
| 269 | |
| 270 | assert decision.action == TurnCompletionAction.CONTINUE |
| 271 | assert prepared.summary.assistant_messages == [] |
| 272 | assert not any( |
| 273 | message.role.value == "assistant" and message.content == completion_claim |
| 274 | for message in agent.session.messages |
| 275 | ) |
| 276 | assert agent.session.messages[-1].role.value == "user" |
| 277 | assert agent.session.messages[-1].content.startswith( |
| 278 | "[PLANNED ARTIFACTS STILL MISSING]" |
| 279 | ) |
| 280 | assert "`06-troubleshooting.html`" in agent.session.messages[-1].content |
| 281 | assert queued_messages |
| 282 | assert "06-troubleshooting.html" in queued_messages[-1] |
| 283 | assert "Do not summarize, mark completion, or write bookkeeping notes yet" in queued_messages[-1] |
| 284 | assert not any(event.type == "response" for event in events) |
| 285 | |
| 286 | |
| 287 | @pytest.mark.asyncio |
| 288 | async def test_turn_completion_interrupts_progress_intent_once_output_files_exist( |
| 289 | temp_dir: Path, |
| 290 | ) -> None: |
| 291 | backend = ScriptedBackend() |
| 292 | agent = Agent( |
| 293 | backend=backend, |
| 294 | config=non_streaming_config(), |
| 295 | project_root=temp_dir, |
| 296 | ) |
| 297 | runtime = ConversationRuntime(agent) |
| 298 | events = [] |
| 299 | |
| 300 | async def capture(event) -> None: |
| 301 | events.append(event) |
| 302 | |
| 303 | prepared = await runtime.turn_preparation.prepare( |
| 304 | task=( |
| 305 | "Create a multi-file nginx guide under ~/Loader/guides/nginx " |
| 306 | "with an index and chapter files." |
| 307 | ), |
| 308 | emit=capture, |
| 309 | requested_mode="execute", |
| 310 | original_task=None, |
| 311 | on_user_question=None, |
| 312 | ) |
| 313 | await runtime.phase_tracker.enter( |
| 314 | TurnPhase.ASSISTANT, |
| 315 | capture, |
| 316 | detail="Requesting assistant response", |
| 317 | reason_code="request_assistant_response", |
| 318 | ) |
| 319 | |
| 320 | implementation_plan = temp_dir / "implementation.md" |
| 321 | implementation_plan.write_text( |
| 322 | "# Implementation Plan\n\n" |
| 323 | "## File Changes\n\n" |
| 324 | "1. Create main index.html file:\n" |
| 325 | f" - `{temp_dir / 'index.html'}`\n\n" |
| 326 | "2. Create chapter files:\n" |
| 327 | f" - `{temp_dir / 'chapters' / '01-introduction.html'}`\n" |
| 328 | f" - `{temp_dir / 'chapters' / '02-installation.html'}`\n" |
| 329 | ) |
| 330 | chapters_dir = temp_dir / "chapters" |
| 331 | chapters_dir.mkdir() |
| 332 | (temp_dir / "index.html").write_text("<h1>NGINX Guide</h1>\n") |
| 333 | (chapters_dir / "01-introduction.html").write_text("<h1>Intro</h1>\n") |
| 334 | |
| 335 | prepared.definition_of_done.implementation_plan = str(implementation_plan) |
| 336 | prepared.definition_of_done.mutating_actions.append("write") |
| 337 | prepared.definition_of_done.touched_files.extend( |
| 338 | [ |
| 339 | str(temp_dir / "index.html"), |
| 340 | str(chapters_dir / "01-introduction.html"), |
| 341 | ] |
| 342 | ) |
| 343 | prepared.definition_of_done.pending_items.append("Create chapter files for nginx guide") |
| 344 | |
| 345 | content = "Now I'll create the second chapter file for the nginx guide." |
| 346 | decision = await runtime.turn_completion.handle_text_response( |
| 347 | content=content, |
| 348 | response_content=content, |
| 349 | task=prepared.task, |
| 350 | effective_task=prepared.effective_task, |
| 351 | iterations=1, |
| 352 | max_iterations=agent.config.max_iterations, |
| 353 | actions_taken=[], |
| 354 | continuation_count=0, |
| 355 | dod=prepared.definition_of_done, |
| 356 | emit=capture, |
| 357 | summary=prepared.summary, |
| 358 | executor=prepared.executor, |
| 359 | rollback_plan=prepared.rollback_plan, |
| 360 | ) |
| 361 | |
| 362 | assert decision.action == TurnCompletionAction.CONTINUE |
| 363 | assert decision.continuation_count == 1 |
| 364 | assert prepared.summary.completion_decision_code == "in_progress_transition_continue" |
| 365 | assert prepared.summary.assistant_messages[-1].content == content |
| 366 | assert agent.session.messages[-1].role.value == "user" |
| 367 | assert agent.session.messages[-1].content.startswith("[CONTINUE CURRENT STEP]") |
| 368 | assert "02-installation.html" in agent.session.messages[-1].content |
| 369 | assert not any( |
| 370 | message.role.value == "user" |
| 371 | and message.content.startswith("[PLANNED ARTIFACTS STILL MISSING]") |
| 372 | for message in agent.session.messages |
| 373 | ) |
| 374 | |
| 375 | |
| 376 | @pytest.mark.asyncio |
| 377 | async def test_turn_completion_allows_first_progress_narration_before_any_output_exists( |
| 378 | temp_dir: Path, |
| 379 | ) -> None: |
| 380 | backend = ScriptedBackend() |
| 381 | config = non_streaming_config() |
| 382 | config.reasoning.completion_check = False |
| 383 | agent = Agent( |
| 384 | backend=backend, |
| 385 | config=config, |
| 386 | project_root=temp_dir, |
| 387 | ) |
| 388 | runtime = ConversationRuntime(agent) |
| 389 | events = [] |
| 390 | |
| 391 | async def capture(event) -> None: |
| 392 | events.append(event) |
| 393 | |
| 394 | prepared = await runtime.turn_preparation.prepare( |
| 395 | task=( |
| 396 | "Create a multi-file nginx guide under ~/Loader/guides/nginx " |
| 397 | "with an index and chapter files." |
| 398 | ), |
| 399 | emit=capture, |
| 400 | requested_mode="execute", |
| 401 | original_task=None, |
| 402 | on_user_question=None, |
| 403 | ) |
| 404 | await runtime.phase_tracker.enter( |
| 405 | TurnPhase.ASSISTANT, |
| 406 | capture, |
| 407 | detail="Requesting assistant response", |
| 408 | reason_code="request_assistant_response", |
| 409 | ) |
| 410 | |
| 411 | implementation_plan = temp_dir / "implementation.md" |
| 412 | implementation_plan.write_text( |
| 413 | "# Implementation Plan\n\n" |
| 414 | "## File Changes\n\n" |
| 415 | f"- `{temp_dir / 'index.html'}`\n" |
| 416 | f"- `{temp_dir / 'chapters' / '01-introduction.html'}`\n" |
| 417 | ) |
| 418 | |
| 419 | prepared.definition_of_done.implementation_plan = str(implementation_plan) |
| 420 | prepared.definition_of_done.pending_items.append( |
| 421 | "Develop the main index.html file for nginx guide" |
| 422 | ) |
| 423 | |
| 424 | content = "Now I'll create the main index.html file for the nginx guide." |
| 425 | decision = await runtime.turn_completion.handle_text_response( |
| 426 | content=content, |
| 427 | response_content=content, |
| 428 | task=prepared.task, |
| 429 | effective_task=prepared.effective_task, |
| 430 | iterations=1, |
| 431 | max_iterations=agent.config.max_iterations, |
| 432 | actions_taken=[], |
| 433 | continuation_count=0, |
| 434 | dod=prepared.definition_of_done, |
| 435 | emit=capture, |
| 436 | summary=prepared.summary, |
| 437 | executor=prepared.executor, |
| 438 | rollback_plan=prepared.rollback_plan, |
| 439 | ) |
| 440 | |
| 441 | assert decision.action == TurnCompletionAction.CONTINUE |
| 442 | assert decision.continuation_count == 1 |
| 443 | assert prepared.summary.assistant_messages[-1].content == content |
| 444 | assert agent.session.messages[-1].role.value == "assistant" |
| 445 | |
| 446 | |
| 447 | @pytest.mark.asyncio |
| 448 | async def test_turn_completion_interrupts_repeated_concrete_progress_narration( |
| 449 | temp_dir: Path, |
| 450 | ) -> None: |
| 451 | backend = ScriptedBackend() |
| 452 | config = non_streaming_config() |
| 453 | config.reasoning.completion_check = False |
| 454 | agent = Agent( |
| 455 | backend=backend, |
| 456 | config=config, |
| 457 | project_root=temp_dir, |
| 458 | ) |
| 459 | runtime = ConversationRuntime(agent) |
| 460 | events = [] |
| 461 | |
| 462 | async def capture(event) -> None: |
| 463 | events.append(event) |
| 464 | |
| 465 | prepared = await runtime.turn_preparation.prepare( |
| 466 | task=( |
| 467 | "Create a multi-file nginx guide under ~/Loader/guides/nginx " |
| 468 | "with an index and chapter files." |
| 469 | ), |
| 470 | emit=capture, |
| 471 | requested_mode="execute", |
| 472 | original_task=None, |
| 473 | on_user_question=None, |
| 474 | ) |
| 475 | await runtime.phase_tracker.enter( |
| 476 | TurnPhase.ASSISTANT, |
| 477 | capture, |
| 478 | detail="Requesting assistant response", |
| 479 | reason_code="request_assistant_response", |
| 480 | ) |
| 481 | |
| 482 | implementation_plan = temp_dir / "implementation.md" |
| 483 | implementation_plan.write_text( |
| 484 | "# Implementation Plan\n\n" |
| 485 | "## File Changes\n\n" |
| 486 | "1. Create main index.html file:\n" |
| 487 | f" - `{temp_dir / 'index.html'}`\n\n" |
| 488 | "2. Create chapter files:\n" |
| 489 | f" - `{temp_dir / 'chapters' / '01-introduction.html'}`\n" |
| 490 | f" - `{temp_dir / 'chapters' / '02-installation.html'}`\n" |
| 491 | ) |
| 492 | chapters_dir = temp_dir / "chapters" |
| 493 | chapters_dir.mkdir() |
| 494 | (temp_dir / "index.html").write_text("<h1>NGINX Guide</h1>\n") |
| 495 | (chapters_dir / "01-introduction.html").write_text("<h1>Intro</h1>\n") |
| 496 | |
| 497 | prepared.definition_of_done.implementation_plan = str(implementation_plan) |
| 498 | prepared.definition_of_done.mutating_actions.append("write") |
| 499 | prepared.definition_of_done.touched_files.extend( |
| 500 | [ |
| 501 | str(temp_dir / "index.html"), |
| 502 | str(chapters_dir / "01-introduction.html"), |
| 503 | ] |
| 504 | ) |
| 505 | prepared.definition_of_done.pending_items.append("Create chapter files for nginx guide") |
| 506 | |
| 507 | content = "Now I'll create the second chapter file for the nginx guide." |
| 508 | decision = await runtime.turn_completion.handle_text_response( |
| 509 | content=content, |
| 510 | response_content=content, |
| 511 | task=prepared.task, |
| 512 | effective_task=prepared.effective_task, |
| 513 | iterations=1, |
| 514 | max_iterations=agent.config.max_iterations, |
| 515 | actions_taken=[], |
| 516 | continuation_count=1, |
| 517 | dod=prepared.definition_of_done, |
| 518 | emit=capture, |
| 519 | summary=prepared.summary, |
| 520 | executor=prepared.executor, |
| 521 | rollback_plan=prepared.rollback_plan, |
| 522 | ) |
| 523 | |
| 524 | assert decision.action == TurnCompletionAction.CONTINUE |
| 525 | assert decision.continuation_count == 2 |
| 526 | assert agent.session.messages[-1].role.value == "user" |
| 527 | assert agent.session.messages[-1].content.startswith("[CONTINUE CURRENT STEP]") |
| 528 | assert "02-installation.html" in agent.session.messages[-1].content |
| 529 | |
| 530 | |
| 531 | @pytest.mark.asyncio |
| 532 | async def test_turn_completion_interrupts_first_narration_after_concrete_target_prompt( |
| 533 | temp_dir: Path, |
| 534 | ) -> None: |
| 535 | backend = ScriptedBackend() |
| 536 | config = non_streaming_config() |
| 537 | config.reasoning.completion_check = False |
| 538 | agent = Agent( |
| 539 | backend=backend, |
| 540 | config=config, |
| 541 | project_root=temp_dir, |
| 542 | ) |
| 543 | runtime = ConversationRuntime(agent) |
| 544 | events = [] |
| 545 | |
| 546 | async def capture(event) -> None: |
| 547 | events.append(event) |
| 548 | |
| 549 | prepared = await runtime.turn_preparation.prepare( |
| 550 | task=( |
| 551 | "Create a multi-file nginx guide under ~/Loader/guides/nginx " |
| 552 | "with an index and chapter files." |
| 553 | ), |
| 554 | emit=capture, |
| 555 | requested_mode="execute", |
| 556 | original_task=None, |
| 557 | on_user_question=None, |
| 558 | ) |
| 559 | await runtime.phase_tracker.enter( |
| 560 | TurnPhase.ASSISTANT, |
| 561 | capture, |
| 562 | detail="Requesting assistant response", |
| 563 | reason_code="request_assistant_response", |
| 564 | ) |
| 565 | |
| 566 | implementation_plan = temp_dir / "implementation.md" |
| 567 | implementation_plan.write_text( |
| 568 | "# Implementation Plan\n\n" |
| 569 | "## File Changes\n\n" |
| 570 | f"- `{temp_dir / 'index.html'}`\n" |
| 571 | f"- `{temp_dir / 'chapters' / '01-introduction.html'}`\n" |
| 572 | ) |
| 573 | chapters_dir = temp_dir / "chapters" |
| 574 | chapters_dir.mkdir() |
| 575 | |
| 576 | prepared.definition_of_done.implementation_plan = str(implementation_plan) |
| 577 | prepared.definition_of_done.pending_items.append( |
| 578 | "Develop the main index.html file for nginx guide" |
| 579 | ) |
| 580 | |
| 581 | agent.session.append( |
| 582 | Message( |
| 583 | role=Role.USER, |
| 584 | content=( |
| 585 | "[USER INTERRUPTION]: Directory setup is complete. Continue with the next pending item: " |
| 586 | "`Develop the main index.html file for nginx guide`. Resume by creating `index.html` now. " |
| 587 | f"Prefer one `write` call for `{(temp_dir / 'index.html').resolve(strict=False)}` instead of more rereads. " |
| 588 | "Make your next response the concrete mutation tool call itself, not another bookkeeping-only turn." |
| 589 | ), |
| 590 | ) |
| 591 | ) |
| 592 | |
| 593 | content = "Now I'll create the main index.html file for the nginx guide." |
| 594 | decision = await runtime.turn_completion.handle_text_response( |
| 595 | content=content, |
| 596 | response_content=content, |
| 597 | task=prepared.task, |
| 598 | effective_task=prepared.effective_task, |
| 599 | iterations=1, |
| 600 | max_iterations=agent.config.max_iterations, |
| 601 | actions_taken=[], |
| 602 | continuation_count=0, |
| 603 | dod=prepared.definition_of_done, |
| 604 | emit=capture, |
| 605 | summary=prepared.summary, |
| 606 | executor=prepared.executor, |
| 607 | rollback_plan=prepared.rollback_plan, |
| 608 | ) |
| 609 | |
| 610 | assert decision.action == TurnCompletionAction.CONTINUE |
| 611 | assert decision.continuation_count == 1 |
| 612 | assert prepared.summary.assistant_messages[-1].content == content |
| 613 | assert agent.session.messages[-1].role.value == "user" |
| 614 | assert agent.session.messages[-1].content.startswith("[CONTINUE CURRENT STEP]") |
| 615 | assert "index.html" in agent.session.messages[-1].content |
| 616 | |
| 617 | |
| 618 | @pytest.mark.asyncio |
| 619 | async def test_turn_completion_handles_fake_tool_narration_without_reroute( |
| 620 | temp_dir: Path, |
| 621 | ) -> None: |
| 622 | backend = ScriptedBackend() |
| 623 | config = non_streaming_config() |
| 624 | config.reasoning.completion_check = False |
| 625 | agent = Agent( |
| 626 | backend=backend, |
| 627 | config=config, |
| 628 | project_root=temp_dir, |
| 629 | ) |
| 630 | runtime = ConversationRuntime(agent) |
| 631 | events = [] |
| 632 | |
| 633 | async def capture(event) -> None: |
| 634 | events.append(event) |
| 635 | |
| 636 | prepared = await runtime.turn_preparation.prepare( |
| 637 | task="Summarize the current test status.", |
| 638 | emit=capture, |
| 639 | requested_mode="execute", |
| 640 | original_task=None, |
| 641 | on_user_question=None, |
| 642 | ) |
| 643 | await runtime.phase_tracker.enter( |
| 644 | TurnPhase.ASSISTANT, |
| 645 | capture, |
| 646 | detail="Requesting assistant response", |
| 647 | reason_code="request_assistant_response", |
| 648 | ) |
| 649 | |
| 650 | narrated = "Used bash tool with command `pytest -q` and everything passed." |
| 651 | decision = await runtime.turn_completion.handle_text_response( |
| 652 | content=narrated, |
| 653 | response_content=narrated, |
| 654 | task=prepared.task, |
| 655 | effective_task=prepared.effective_task, |
| 656 | iterations=1, |
| 657 | max_iterations=agent.config.max_iterations, |
| 658 | actions_taken=[], |
| 659 | continuation_count=0, |
| 660 | dod=prepared.definition_of_done, |
| 661 | emit=capture, |
| 662 | summary=prepared.summary, |
| 663 | executor=prepared.executor, |
| 664 | rollback_plan=prepared.rollback_plan, |
| 665 | ) |
| 666 | |
| 667 | assert decision.action == TurnCompletionAction.COMPLETE |
| 668 | assert prepared.summary.final_response == narrated |
| 669 | assert prepared.summary.completion_decision_code == "non_mutating_response_accepted" |
| 670 | assert prepared.summary.completion_trace[-1].decision_code == ( |
| 671 | "non_mutating_response_accepted" |
| 672 | ) |
| 673 | assert not any( |
| 674 | "PRETENDING to use tools" in message.content |
| 675 | for message in agent.session.messages |
| 676 | ) |
| 677 | assert any(event.type == "response" and event.content == narrated for event in events) |
| 678 | |
| 679 | |
| 680 | @pytest.mark.asyncio |
| 681 | async def test_turn_completion_handles_deflection_text_without_repair_prompt( |
| 682 | temp_dir: Path, |
| 683 | ) -> None: |
| 684 | backend = ScriptedBackend() |
| 685 | config = non_streaming_config() |
| 686 | config.reasoning.completion_check = False |
| 687 | agent = Agent( |
| 688 | backend=backend, |
| 689 | config=config, |
| 690 | project_root=temp_dir, |
| 691 | ) |
| 692 | runtime = ConversationRuntime(agent) |
| 693 | events = [] |
| 694 | |
| 695 | async def capture(event) -> None: |
| 696 | events.append(event) |
| 697 | |
| 698 | prepared = await runtime.turn_preparation.prepare( |
| 699 | task="What should I verify next?", |
| 700 | emit=capture, |
| 701 | requested_mode="execute", |
| 702 | original_task=None, |
| 703 | on_user_question=None, |
| 704 | ) |
| 705 | await runtime.phase_tracker.enter( |
| 706 | TurnPhase.ASSISTANT, |
| 707 | capture, |
| 708 | detail="Requesting assistant response", |
| 709 | reason_code="request_assistant_response", |
| 710 | ) |
| 711 | |
| 712 | deflection = "You can run pytest -q to verify the current state." |
| 713 | decision = await runtime.turn_completion.handle_text_response( |
| 714 | content=deflection, |
| 715 | response_content=deflection, |
| 716 | task=prepared.task, |
| 717 | effective_task=prepared.effective_task, |
| 718 | iterations=1, |
| 719 | max_iterations=agent.config.max_iterations, |
| 720 | actions_taken=[], |
| 721 | continuation_count=0, |
| 722 | dod=prepared.definition_of_done, |
| 723 | emit=capture, |
| 724 | summary=prepared.summary, |
| 725 | executor=prepared.executor, |
| 726 | rollback_plan=prepared.rollback_plan, |
| 727 | ) |
| 728 | |
| 729 | assert decision.action == TurnCompletionAction.COMPLETE |
| 730 | assert prepared.summary.final_response == deflection |
| 731 | assert prepared.summary.completion_decision_code == "non_mutating_response_accepted" |
| 732 | assert prepared.summary.completion_trace[-1].decision_code == ( |
| 733 | "non_mutating_response_accepted" |
| 734 | ) |
| 735 | assert not any( |
| 736 | "Please use your tools to execute the task" in message.content |
| 737 | for message in agent.session.messages |
| 738 | ) |
| 739 | assert any(event.type == "response" and event.content == deflection for event in events) |
| 740 | |
| 741 | |
| 742 | @pytest.mark.asyncio |
| 743 | async def test_turn_completion_skips_self_critique_reroute( |
| 744 | temp_dir: Path, |
| 745 | ) -> None: |
| 746 | backend = ScriptedBackend() |
| 747 | config = non_streaming_config() |
| 748 | config.reasoning.completion_check = False |
| 749 | config.reasoning.self_critique = True |
| 750 | agent = Agent( |
| 751 | backend=backend, |
| 752 | config=config, |
| 753 | project_root=temp_dir, |
| 754 | ) |
| 755 | runtime = ConversationRuntime(agent) |
| 756 | events = [] |
| 757 | |
| 758 | async def capture(event) -> None: |
| 759 | events.append(event) |
| 760 | |
| 761 | prepared = await runtime.turn_preparation.prepare( |
| 762 | task="Explain Loader's clarify loop.", |
| 763 | emit=capture, |
| 764 | requested_mode="execute", |
| 765 | original_task=None, |
| 766 | on_user_question=None, |
| 767 | ) |
| 768 | await runtime.phase_tracker.enter( |
| 769 | TurnPhase.ASSISTANT, |
| 770 | capture, |
| 771 | detail="Requesting assistant response", |
| 772 | reason_code="request_assistant_response", |
| 773 | ) |
| 774 | |
| 775 | detailed = ( |
| 776 | "Loader might begin with a bounded clarify pass, perhaps asking follow-up " |
| 777 | "questions when the task leaves touchpoints or decision boundaries unclear. " |
| 778 | "It then shifts into execution once the workflow policy is satisfied." |
| 779 | ) |
| 780 | decision = await runtime.turn_completion.handle_text_response( |
| 781 | content=detailed, |
| 782 | response_content=detailed, |
| 783 | task=prepared.task, |
| 784 | effective_task=prepared.effective_task, |
| 785 | iterations=1, |
| 786 | max_iterations=agent.config.max_iterations, |
| 787 | actions_taken=[], |
| 788 | continuation_count=0, |
| 789 | dod=prepared.definition_of_done, |
| 790 | emit=capture, |
| 791 | summary=prepared.summary, |
| 792 | executor=prepared.executor, |
| 793 | rollback_plan=prepared.rollback_plan, |
| 794 | ) |
| 795 | |
| 796 | assert decision.action == TurnCompletionAction.COMPLETE |
| 797 | assert prepared.summary.final_response == detailed |
| 798 | assert prepared.summary.completion_decision_code == "non_mutating_response_accepted" |
| 799 | assert prepared.summary.completion_trace[-1].decision_code == ( |
| 800 | "non_mutating_response_accepted" |
| 801 | ) |
| 802 | assert not any("[SELF-CRITIQUE]" in message.content for message in agent.session.messages) |
| 803 | assert not any(event.type == "critique" for event in events) |
| 804 | |
| 805 | |
| 806 | @pytest.mark.asyncio |
| 807 | async def test_turn_completion_finalizes_when_follow_through_budget_is_exhausted( |
| 808 | temp_dir: Path, |
| 809 | ) -> None: |
| 810 | backend = ScriptedBackend() |
| 811 | agent = Agent( |
| 812 | backend=backend, |
| 813 | config=non_streaming_config(), |
| 814 | project_root=temp_dir, |
| 815 | ) |
| 816 | runtime = ConversationRuntime(agent) |
| 817 | events = [] |
| 818 | |
| 819 | async def capture(event) -> None: |
| 820 | events.append(event) |
| 821 | |
| 822 | prepared = await runtime.turn_preparation.prepare( |
| 823 | task="Fix the README heading.", |
| 824 | emit=capture, |
| 825 | requested_mode="execute", |
| 826 | original_task=None, |
| 827 | on_user_question=None, |
| 828 | ) |
| 829 | await runtime.phase_tracker.enter( |
| 830 | TurnPhase.ASSISTANT, |
| 831 | capture, |
| 832 | detail="Requesting assistant response", |
| 833 | reason_code="request_assistant_response", |
| 834 | ) |
| 835 | |
| 836 | decision = await runtime.turn_completion.handle_text_response( |
| 837 | content="I looked into it.", |
| 838 | response_content="I looked into it.", |
| 839 | task=prepared.task, |
| 840 | effective_task=prepared.effective_task, |
| 841 | iterations=1, |
| 842 | max_iterations=agent.config.max_iterations, |
| 843 | actions_taken=[], |
| 844 | continuation_count=agent.config.reasoning.max_continuation_prompts, |
| 845 | dod=prepared.definition_of_done, |
| 846 | emit=capture, |
| 847 | summary=prepared.summary, |
| 848 | executor=prepared.executor, |
| 849 | rollback_plan=prepared.rollback_plan, |
| 850 | ) |
| 851 | |
| 852 | assert decision.action == TurnCompletionAction.FINALIZE |
| 853 | assert decision.finalize_reason_code == "continuation_budget_exhausted" |
| 854 | assert prepared.summary.final_response.startswith( |
| 855 | "I stopped because I still could not show enough evidence" |
| 856 | ) |
| 857 | assert prepared.summary.completion_decision_code == "continuation_budget_exhausted" |
| 858 | assert prepared.summary.failures == [ |
| 859 | "missing follow-through evidence after continuation budget exhaustion" |
| 860 | ] |
| 861 | assert prepared.summary.completion_trace[-1].outcome == "finalize" |
| 862 | assert prepared.summary.completion_trace[-1].decision_code == ( |
| 863 | "continuation_budget_exhausted" |
| 864 | ) |
| 865 | assert prepared.summary.completion_trace[-1].evidence_summary == [ |
| 866 | "showing the requested work was actually carried out" |
| 867 | ] |
| 868 | assert [item.status for item in prepared.summary.completion_trace[-1].evidence_provenance] == [ |
| 869 | "missing" |
| 870 | ] |
| 871 | assert prepared.summary.workflow_timeline[-1].kind == "completion_finalize" |
| 872 | assert prepared.summary.workflow_timeline[-1].evidence_summary == [ |
| 873 | "showing the requested work was actually carried out" |
| 874 | ] |
| 875 | assert [event.type for event in events[-3:]] == [ |
| 876 | "completion_check", |
| 877 | "error", |
| 878 | "response", |
| 879 | ] |
| 880 | |
| 881 | |
| 882 | @pytest.mark.asyncio |
| 883 | async def test_turn_completion_uses_observed_verification_for_budget_exhaustion( |
| 884 | temp_dir: Path, |
| 885 | ) -> None: |
| 886 | backend = ScriptedBackend() |
| 887 | agent = Agent( |
| 888 | backend=backend, |
| 889 | config=non_streaming_config(), |
| 890 | project_root=temp_dir, |
| 891 | ) |
| 892 | runtime = ConversationRuntime(agent) |
| 893 | events = [] |
| 894 | |
| 895 | async def capture(event) -> None: |
| 896 | events.append(event) |
| 897 | |
| 898 | prepared = await runtime.turn_preparation.prepare( |
| 899 | task="Run pytest -q and make sure it works.", |
| 900 | emit=capture, |
| 901 | requested_mode="execute", |
| 902 | original_task=None, |
| 903 | on_user_question=None, |
| 904 | ) |
| 905 | prepared.definition_of_done.verification_commands = ["pytest -q"] |
| 906 | prepared.definition_of_done.evidence = [ |
| 907 | VerificationEvidence( |
| 908 | command="pytest -q", |
| 909 | passed=False, |
| 910 | stderr="1 failed", |
| 911 | kind="test", |
| 912 | ) |
| 913 | ] |
| 914 | prepared.definition_of_done.last_verification_result = "failed" |
| 915 | await runtime.phase_tracker.enter( |
| 916 | TurnPhase.ASSISTANT, |
| 917 | capture, |
| 918 | detail="Requesting assistant response", |
| 919 | reason_code="request_assistant_response", |
| 920 | ) |
| 921 | |
| 922 | decision = await runtime.turn_completion.handle_text_response( |
| 923 | content="The tests are done.", |
| 924 | response_content="The tests are done.", |
| 925 | task=prepared.task, |
| 926 | effective_task=prepared.effective_task, |
| 927 | iterations=1, |
| 928 | max_iterations=agent.config.max_iterations, |
| 929 | actions_taken=[], |
| 930 | continuation_count=agent.config.reasoning.max_continuation_prompts, |
| 931 | dod=prepared.definition_of_done, |
| 932 | emit=capture, |
| 933 | summary=prepared.summary, |
| 934 | executor=prepared.executor, |
| 935 | rollback_plan=prepared.rollback_plan, |
| 936 | ) |
| 937 | |
| 938 | assert decision.action == TurnCompletionAction.FINALIZE |
| 939 | assert decision.finalize_reason_code == "continuation_budget_exhausted" |
| 940 | assert prepared.summary.final_response == ( |
| 941 | "I stopped because the continuation budget was exhausted and observed " |
| 942 | "verification still showed: verification failed for `pytest -q` [1 failed]." |
| 943 | ) |
| 944 | assert prepared.summary.completion_trace[-1].decision_code == ( |
| 945 | "continuation_budget_exhausted" |
| 946 | ) |
| 947 | assert [ |
| 948 | item.status |
| 949 | for item in prepared.summary.completion_trace[-1].verification_observations |
| 950 | ] == [VerificationObservationStatus.FAILED.value] |
| 951 | assert [ |
| 952 | item.summary |
| 953 | for item in prepared.summary.completion_trace[-1].verification_observations |
| 954 | ] == ["verification failed for `pytest -q`"] |
| 955 | assert [ |
| 956 | item.status |
| 957 | for item in prepared.summary.workflow_timeline[-1].verification_observations |
| 958 | ] == [VerificationObservationStatus.FAILED.value] |