Repository navigation
Expand file tree
/
Copy pathagent_loop.py
More file actions
1716 lines (1500 loc) · 62.8 KB
/
Copy pathagent_loop.py
File metadata and controls
1716 lines (1500 loc) · 62.8 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
from __future__ import annotations
import asyncio
from collections.abc import AsyncGenerator, Callable, Generator
import contextlib
import copy
from enum import StrEnum, auto
from functools import wraps
from http import HTTPStatus
import inspect
import os
from pathlib import Path
import threading
from threading import Thread
import time
from typing import TYPE_CHECKING, Any, Literal
from uuid import uuid4
from opentelemetry import trace
from pydantic import BaseModel
from vibe.cli.terminal_detect import detect_terminal
from vibe.core.agents.manager import AgentManager
from vibe.core.agents.models import AgentProfile, BuiltinAgentName
from vibe.core.config import ModelConfig, ProviderConfig, VibeConfig
from vibe.core.hooks.manager import HooksManager
from vibe.core.hooks.models import HookConfigResult, HookType, HookUserMessage
from vibe.core.llm.backend.factory import BACKEND_FACTORY
from vibe.core.llm.exceptions import BackendError
from vibe.core.llm.format import (
APIToolFormatHandler,
FailedToolCall,
ResolvedMessage,
ResolvedToolCall,
)
from vibe.core.llm.types import BackendLike
from vibe.core.middleware import (
CHAT_AGENT_EXIT,
CHAT_AGENT_REMINDER,
PLAN_AGENT_EXIT,
AutoCompactMiddleware,
ContextWarningMiddleware,
ConversationContext,
MiddlewareAction,
MiddlewarePipeline,
MiddlewareResult,
PriceLimitMiddleware,
ReadOnlyAgentMiddleware,
ResetReason,
TurnLimitMiddleware,
make_plan_agent_reminder,
)
from vibe.core.plan_session import PlanSession
from vibe.core.prompts import UtilityPrompt
from vibe.core.rewind import RewindManager
from vibe.core.scratchpad import init_scratchpad
from vibe.core.session.session_id import extract_suffix, generate_session_id
from vibe.core.session.session_logger import SessionLogger
from vibe.core.session.session_migration import migrate_sessions_entrypoint
from vibe.core.skills.manager import SkillManager
from vibe.core.system_prompt import get_universal_system_prompt
from vibe.core.telemetry.build_metadata import build_request_metadata
from vibe.core.telemetry.send import TelemetryClient
from vibe.core.telemetry.types import (
EntrypointMetadata,
TelemetryCallType,
TelemetryRequestMetadata,
)
from vibe.core.teleport.errors import ServiceTeleportError
from vibe.core.teleport.telemetry import TeleportTelemetryTracker
from vibe.core.teleport.types import TeleportCompleteEvent
from vibe.core.tools.base import (
BaseTool,
InvokeContext,
ToolError,
ToolPermission,
ToolPermissionError,
)
from vibe.core.tools.connectors import ConnectorRegistry, connectors_enabled
from vibe.core.tools.manager import ToolManager
from vibe.core.tools.mcp import MCPRegistry
from vibe.core.tools.mcp_sampling import MCPSamplingHandler
from vibe.core.tools.permissions import (
ApprovedRule,
PermissionContext,
RequiredPermission,
)
from vibe.core.tools.utils import wildcard_match
from vibe.core.tracing import agent_span, set_tool_result, tool_span
from vibe.core.trusted_folders import has_agents_md_file
from vibe.core.types import (
AgentProfileChangedEvent,
AgentStats,
ApprovalCallback,
ApprovalResponse,
AssistantEvent,
BaseEvent,
CompactEndEvent,
CompactStartEvent,
ContextTooLongError,
LLMChunk,
LLMMessage,
LLMUsage,
MessageList,
RateLimitError,
ReasoningEvent,
Role,
ToolCall,
ToolCallEvent,
ToolResultEvent,
ToolStreamEvent,
UserInputCallback,
UserMessageEvent,
)
from vibe.core.utils import (
CANCELLATION_TAG,
TOOL_ERROR_TAG,
VIBE_STOP_EVENT_TAG,
CancellationReason,
get_server_url_from_api_base,
get_user_agent,
get_user_cancellation_message,
is_user_cancellation_event,
)
try:
from vibe.core.teleport.teleport import TeleportService as _TeleportService
_TELEPORT_AVAILABLE = True
except ImportError:
_TELEPORT_AVAILABLE = False
_TeleportService = None
if TYPE_CHECKING:
from vibe.core.teleport.teleport import TeleportService
from vibe.core.teleport.types import TeleportPushResponseEvent, TeleportYieldEvent
class ToolExecutionResponse(StrEnum):
SKIP = auto()
EXECUTE = auto()
class ToolDecision(BaseModel):
verdict: ToolExecutionResponse
approval_type: ToolPermission
feedback: str | None = None
class AgentLoopError(Exception):
"""Base exception for AgentLoop errors."""
class AgentLoopStateError(AgentLoopError):
"""Raised when agent loop is in an invalid state."""
class AgentLoopLLMResponseError(AgentLoopError):
"""Raised when LLM response is malformed or missing expected data."""
class TeleportError(AgentLoopError):
"""Raised when teleport to Vibe Code fails."""
def _should_raise_rate_limit_error(e: Exception) -> bool:
return isinstance(e, BackendError) and e.status == HTTPStatus.TOO_MANY_REQUESTS
def _is_context_too_long_error(e: Exception) -> bool:
if isinstance(e, BackendError):
return e.is_context_too_long
if isinstance(e, RuntimeError) and isinstance(e.__cause__, BackendError):
return e.__cause__.is_context_too_long
return False
def _is_non_retryable_error(e: BaseException) -> bool:
# Detect Temporal-style ``non_retryable`` flag without importing temporalio.
# Walks ``__cause__`` so an ``ActivityError`` whose cause is a non-retryable
# ``ApplicationError`` is detected too — that's what callers driving the
# agent loop from a Temporal activity will see when a sub-activity has
# already failed terminally.
seen: set[int] = set()
current: BaseException | None = e
while current is not None and id(current) not in seen:
if getattr(current, "non_retryable", False):
return True
seen.add(id(current))
current = current.__cause__
return False
def requires_init(fn: Callable[..., Any]) -> Callable[..., Any]:
"""Decorator that awaits deferred initialization before executing the method."""
if inspect.isasyncgenfunction(fn):
@wraps(fn)
async def gen_wrapper(self: AgentLoop, *args: Any, **kwargs: Any) -> Any:
await self.wait_until_ready()
agen = fn(self, *args, **kwargs)
sent: Any = None
try:
while True:
sent = yield await agen.asend(sent)
except StopAsyncIteration:
return
finally:
await agen.aclose()
return gen_wrapper
@wraps(fn)
async def wrapper(self: AgentLoop, *args: Any, **kwargs: Any) -> Any:
await self.wait_until_ready()
return await fn(self, *args, **kwargs)
return wrapper
class AgentLoop:
def __init__( # noqa: PLR0913, PLR0915
self,
config: VibeConfig,
*,
agent_name: str = BuiltinAgentName.DEFAULT,
message_observer: Callable[[LLMMessage], None] | None = None,
max_turns: int | None = None,
max_price: float | None = None,
backend: BackendLike | None = None,
enable_streaming: bool = False,
entrypoint_metadata: EntrypointMetadata | None = None,
is_subagent: bool = False,
defer_heavy_init: bool = False,
headless: bool = False,
hook_config_result: HookConfigResult | None = None,
) -> None:
self._base_config = config
self._headless = headless
self._defer_heavy_init = defer_heavy_init
self._deferred_init_thread: threading.Thread | None = None
self._deferred_init_lock = threading.Lock()
self._init_error: Exception | None = None
self._init_start_time = time.monotonic()
self._init_duration_ms: int | None = None
self.mcp_registry = MCPRegistry()
self.connector_registry = self._create_connector_registry()
self.agent_manager = AgentManager(
lambda: self._base_config,
initial_agent=agent_name,
allow_subagent=is_subagent,
)
self.tool_manager = ToolManager(
lambda: self.config,
mcp_registry=self.mcp_registry,
connector_registry=self.connector_registry,
defer_mcp=defer_heavy_init,
)
self.skill_manager = SkillManager(lambda: self.config)
self.message_observer = message_observer
self._max_turns = max_turns
self._max_price = max_price
self._plan_session = PlanSession()
self.format_handler = APIToolFormatHandler()
self.backend_factory = lambda: backend or self._select_backend()
self.backend = self.backend_factory()
self._sampling_handler = MCPSamplingHandler(
backend_getter=lambda: self.backend,
config_getter=lambda: self.config,
metadata_getter=lambda: self._build_backend_metadata(
call_type="secondary_call"
).model_dump(exclude_none=True),
extra_headers_getter=self._get_extra_headers,
)
self.enable_streaming = enable_streaming
self.middleware_pipeline = MiddlewarePipeline()
self._setup_middleware()
self.session_id = generate_session_id()
self.parent_session_id: str | None = None
self.scratchpad_dir = (
init_scratchpad(self.session_id) if not is_subagent else None
)
system_prompt = get_universal_system_prompt(
self.tool_manager,
self.config,
self.skill_manager,
self.agent_manager,
include_git_status=not defer_heavy_init,
scratchpad_dir=self.scratchpad_dir,
headless=self._headless,
)
system_message = LLMMessage(role=Role.system, content=system_prompt)
self.messages = MessageList(initial=[system_message], observer=message_observer)
self.stats = AgentStats()
self.approval_callback: ApprovalCallback | None = None
self.user_input_callback: UserInputCallback | None = None
self.entrypoint_metadata = entrypoint_metadata
try:
active_model = config.get_active_model()
self.stats.input_price_per_million = active_model.input_price
self.stats.output_price_per_million = active_model.output_price
except ValueError:
pass
self._current_user_message_id: str | None = None
self._is_user_prompt_call: bool = False
self._session_rules: list[ApprovedRule] = []
self._approval_lock = asyncio.Lock()
self.telemetry_client = TelemetryClient(
config_getter=lambda: self.config,
session_id_getter=lambda: self.session_id,
parent_session_id_getter=lambda: self.parent_session_id,
entrypoint_metadata_getter=lambda: self.entrypoint_metadata,
)
self.session_logger = SessionLogger(config.session_logging, self.session_id)
self._hook_config_result = hook_config_result
self._hooks_manager = (
HooksManager(hook_config_result.hooks) if hook_config_result else None
)
self.hook_config_issues = (
hook_config_result.issues if hook_config_result else []
)
self.rewind_manager = RewindManager(
messages=self.messages,
save_messages=self._save_messages,
reset_session=self._reset_session,
)
self._teleport_service: TeleportService | None = None
Thread(
target=migrate_sessions_entrypoint,
args=(config.session_logging,),
daemon=True,
name="migrate_sessions",
).start()
if defer_heavy_init:
self._start_deferred_init()
def _start_deferred_init(self) -> threading.Thread:
"""Spawn a daemon thread that finishes deferred heavy I/O once."""
with self._deferred_init_lock:
if self._deferred_init_thread is not None:
return self._deferred_init_thread
thread = threading.Thread(
target=self._complete_init, daemon=True, name="agent_loop_init"
)
self._deferred_init_thread = thread
thread.start()
return thread
@property
def is_initialized(self) -> bool:
"""Whether deferred initialization has completed (successfully or not)."""
if not self._defer_heavy_init:
return True
thread = self._deferred_init_thread
return thread is not None and not thread.is_alive()
def _complete_init(self) -> None:
"""Run deferred heavy I/O: MCP and connector discovery.
Intended to be called from a background thread when
``defer_heavy_init=True`` was passed to ``__init__``.
"""
try:
self.tool_manager.integrate_all(raise_on_mcp_failure=True)
system_prompt = get_universal_system_prompt(
self.tool_manager,
self.config,
self.skill_manager,
self.agent_manager,
scratchpad_dir=self.scratchpad_dir,
headless=self._headless,
)
self.messages.update_system_prompt(system_prompt)
self._init_duration_ms = int(
(time.monotonic() - self._init_start_time) * 1000
)
except Exception as exc:
self._init_error = exc
async def wait_until_ready(self) -> None:
"""Await deferred initialization from an async context."""
if not self._defer_heavy_init:
return
thread = self._start_deferred_init()
await asyncio.to_thread(thread.join)
if err := self._init_error:
raise copy.copy(err).with_traceback(err.__traceback__)
if self._init_duration_ms is not None:
duration, self._init_duration_ms = self._init_duration_ms, None
self.emit_ready_telemetry(duration)
@property
def agent_profile(self) -> AgentProfile:
return self.agent_manager.active_profile
@property
def base_config(self) -> VibeConfig:
return self._base_config
@property
def config(self) -> VibeConfig:
return self.agent_manager.config
@property
def bypass_tool_permissions(self) -> bool:
return self.config.bypass_tool_permissions
def refresh_config(self) -> None:
self._base_config = VibeConfig.load()
self.agent_manager.invalidate_config()
def set_approval_callback(self, callback: ApprovalCallback) -> None:
self.approval_callback = callback
def set_user_input_callback(self, callback: UserInputCallback) -> None:
self.user_input_callback = callback
def set_tool_permission(
self, tool_name: str, permission: ToolPermission, save_permanently: bool = False
) -> None:
if save_permanently:
VibeConfig.save_updates({
"tools": {tool_name: {"permission": permission.value}}
})
if tool_name not in self.config.tools:
self.config.tools[tool_name] = {}
self.config.tools[tool_name]["permission"] = permission.value
def _add_session_rule(self, rule: ApprovedRule) -> None:
self._session_rules.append(rule)
def _is_permission_covered(self, tool_name: str, rp: RequiredPermission) -> bool:
return any(
rule.tool_name == tool_name
and rule.scope == rp.scope
and wildcard_match(rp.invocation_pattern, rule.session_pattern)
for rule in self._session_rules
)
def approve_always(
self,
tool_name: str,
required_permissions: list[RequiredPermission] | None,
save_permanently: bool = False,
) -> None:
"""Handle 'Allow Always' approval: add session rules or set tool-level permission."""
if required_permissions:
for rp in required_permissions:
self._add_session_rule(
ApprovedRule(
tool_name=tool_name,
scope=rp.scope,
session_pattern=rp.session_pattern,
)
)
if save_permanently:
self.config.add_tool_allowlist_patterns(
tool_name, [rp.session_pattern for rp in required_permissions]
)
else:
self.set_tool_permission(
tool_name, ToolPermission.ALWAYS, save_permanently=save_permanently
)
def emit_new_session_telemetry(self) -> None:
entrypoint = (
self.entrypoint_metadata.agent_entrypoint
if self.entrypoint_metadata
else "unknown"
)
client_name = (
self.entrypoint_metadata.client_name if self.entrypoint_metadata else None
)
client_version = (
self.entrypoint_metadata.client_version
if self.entrypoint_metadata
else None
)
has_agents_md = has_agents_md_file(Path.cwd())
nb_skills = len(self.skill_manager.available_skills)
nb_mcp_servers = len(self.config.mcp_servers)
nb_models = len(self.config.models)
terminal_emulator = None
if entrypoint == "cli":
terminal_emulator = detect_terminal().value
self.telemetry_client.send_new_session(
has_agents_md=has_agents_md,
nb_skills=nb_skills,
nb_mcp_servers=nb_mcp_servers,
nb_models=nb_models,
entrypoint=entrypoint,
client_name=client_name,
client_version=client_version,
terminal_emulator=terminal_emulator,
)
def emit_ready_telemetry(self, init_duration_ms: int) -> None:
self.telemetry_client.send_ready(init_duration_ms=init_duration_ms)
def emit_session_closed_telemetry(self) -> None:
self.telemetry_client.send_session_closed()
async def aclose(self) -> None:
with contextlib.suppress(Exception):
await self.backend.__aexit__(None, None, None)
def _create_connector_registry(self) -> ConnectorRegistry | None:
if not connectors_enabled():
return None
provider = self._base_config.get_mistral_provider()
if provider is None:
return None
api_key_env = provider.api_key_env_var or "MISTRAL_API_KEY"
api_key = os.getenv(api_key_env, "")
if not api_key:
return None
server_url = get_server_url_from_api_base(provider.api_base)
return ConnectorRegistry(api_key=api_key, server_url=server_url)
@requires_init
async def refresh_system_prompt(self) -> None:
"""Rebuild and replace the system prompt with current tool/skill state."""
system_prompt = get_universal_system_prompt(
self.tool_manager,
self.config,
self.skill_manager,
self.agent_manager,
headless=self._headless,
)
self.messages.update_system_prompt(system_prompt)
def _select_backend(self) -> BackendLike:
provider = self.config.get_active_provider()
timeout = self.config.api_timeout
return BACKEND_FACTORY[provider.backend](provider=provider, timeout=timeout)
async def _save_messages(self) -> None:
await self.session_logger.save_interaction(
self.messages,
self.stats,
self._base_config,
self.tool_manager,
self.agent_profile,
)
@requires_init
async def inject_user_context(self, content: str) -> None:
self.messages.append(LLMMessage(role=Role.user, content=content, injected=True))
await self._save_messages()
@requires_init
async def act(
self, msg: str, client_message_id: str | None = None
) -> AsyncGenerator[BaseEvent, None]:
self._clean_message_history()
self.rewind_manager.create_checkpoint()
try:
model_name = self.config.get_active_model().name
except ValueError:
model_name = None
async with agent_span(model=model_name, session_id=self.session_id):
async for event in self._conversation_loop(
msg, client_message_id=client_message_id
):
yield event
@property
def teleport_service(self) -> TeleportService:
if not _TELEPORT_AVAILABLE:
raise TeleportError(
"Teleport requires git to be installed. "
"Please install git and try again."
)
if self._teleport_service is None:
if _TeleportService is None:
raise TeleportError("_TeleportService is unexpectedly None")
self._teleport_service = _TeleportService(
session_logger=self.session_logger,
vibe_code_base_url=self.config.vibe_code_base_url,
vibe_code_workflow_id=self.config.vibe_code_workflow_id,
vibe_code_api_key=self.config.vibe_code_api_key,
vibe_code_task_queue=self.config.vibe_code_task_queue,
vibe_config=self._base_config,
)
return self._teleport_service
@requires_init
async def teleport_to_vibe_code(
self, prompt: str | None
) -> AsyncGenerator[TeleportYieldEvent, TeleportPushResponseEvent | None]:
from vibe.core.teleport.nuage import TeleportSession
session_messages = [
msg.model_dump(exclude_none=True) for msg in self.messages[1:]
]
telemetry_tracker = TeleportTelemetryTracker(
telemetry_client=self.telemetry_client,
nb_session_messages=len(session_messages),
stage="no_history"
if prompt is None and not session_messages
else "git_check",
)
session = TeleportSession(
metadata={
"agent": self.agent_profile.name,
"model": self.config.active_model,
"stats": self.stats.model_dump(),
},
messages=session_messages,
)
try:
async with self.teleport_service:
gen = self.teleport_service.execute(prompt=prompt, session=session)
response: TeleportPushResponseEvent | None = None
while True:
try:
event = await gen.asend(response)
telemetry_tracker.record_event(event)
if isinstance(event, TeleportCompleteEvent):
telemetry_tracker.send_success()
response = yield event
except StopAsyncIteration:
break
except ServiceTeleportError as e:
telemetry_tracker.record_service_error(e)
raise TeleportError(str(e)) from e
except (asyncio.CancelledError, GeneratorExit):
telemetry_tracker.record_cancelled()
raise
except Exception as e:
telemetry_tracker.record_unexpected_error(e)
raise
finally:
telemetry_tracker.send_failure_if_needed()
self._teleport_service = None
def _setup_middleware(self) -> None:
"""Configure middleware pipeline for this conversation."""
self.middleware_pipeline.clear()
if self._max_turns is not None:
self.middleware_pipeline.add(TurnLimitMiddleware(self._max_turns))
if self._max_price is not None:
self.middleware_pipeline.add(PriceLimitMiddleware(self._max_price))
self.middleware_pipeline.add(AutoCompactMiddleware())
if self.config.context_warnings:
self.middleware_pipeline.add(ContextWarningMiddleware(0.5))
self.middleware_pipeline.add(
ReadOnlyAgentMiddleware(
lambda: self.agent_profile,
BuiltinAgentName.PLAN,
lambda: make_plan_agent_reminder(
self._plan_session.plan_file_path_str,
has_ask_user_question="ask_user_question"
in self.tool_manager.available_tools,
has_exit_plan_mode="exit_plan_mode"
in self.tool_manager.available_tools,
),
PLAN_AGENT_EXIT,
)
)
self.middleware_pipeline.add(
ReadOnlyAgentMiddleware(
lambda: self.agent_profile,
BuiltinAgentName.CHAT,
CHAT_AGENT_REMINDER,
CHAT_AGENT_EXIT,
)
)
async def _handle_middleware_result(
self, result: MiddlewareResult
) -> AsyncGenerator[BaseEvent]:
match result.action:
case MiddlewareAction.STOP:
yield AssistantEvent(
content=f"<{VIBE_STOP_EVENT_TAG}>{result.reason}</{VIBE_STOP_EVENT_TAG}>",
stopped_by_middleware=True,
)
case MiddlewareAction.INJECT_MESSAGE:
if result.message:
injected_message = LLMMessage(
role=Role.user, content=result.message, injected=True
)
self.messages.append(injected_message)
case MiddlewareAction.COMPACT:
old_tokens = result.metadata.get(
"old_tokens", self.stats.context_tokens
)
threshold = result.metadata.get(
"threshold", self.config.get_active_model().auto_compact_threshold
)
old_session_id = self.session_id
old_parent_session_id = self.parent_session_id
tool_call_id = str(uuid4())
yield CompactStartEvent(
tool_call_id=tool_call_id,
current_context_tokens=old_tokens,
threshold=threshold,
)
compact_status: Literal["success", "failure", "cancelled"] = "success"
new_tokens = self.stats.context_tokens
try:
summary = await self.compact()
except asyncio.CancelledError:
compact_status = "cancelled"
raise
except Exception:
compact_status = "failure"
raise
finally:
new_tokens = self.stats.context_tokens
self.telemetry_client.send_auto_compact_triggered(
nb_context_tokens_before=old_tokens,
nb_context_tokens_after=new_tokens,
auto_compact_threshold=threshold,
status=compact_status,
session_id=old_session_id,
parent_session_id=old_parent_session_id,
)
yield CompactEndEvent(
tool_call_id=tool_call_id,
old_context_tokens=old_tokens,
new_context_tokens=new_tokens,
summary_length=len(summary),
old_session_id=old_session_id,
new_session_id=self.session_id,
)
case MiddlewareAction.CONTINUE:
pass
def _get_context(self) -> ConversationContext:
return ConversationContext(
messages=self.messages, stats=self.stats, config=self.config
)
def _build_backend_metadata(
self, call_type: TelemetryCallType | None = None
) -> TelemetryRequestMetadata:
return build_request_metadata(
entrypoint_metadata=self.entrypoint_metadata,
session_id=self.session_id,
parent_session_id=self.parent_session_id,
call_type=(
call_type
if call_type is not None
else ("main_call" if self._is_user_prompt_call else "secondary_call")
),
message_id=self._current_user_message_id,
)
def _get_extra_headers(
self, provider: ProviderConfig | None = None
) -> dict[str, str]:
provider = self.config.get_active_provider() if provider is None else provider
headers: dict[str, str] = {**provider.extra_headers}
headers["user-agent"] = get_user_agent(provider.backend)
headers["x-affinity"] = self.session_id
return headers
async def _conversation_loop(
self, user_msg: str, client_message_id: str | None = None
) -> AsyncGenerator[BaseEvent]:
user_message = LLMMessage(
role=Role.user, content=user_msg, message_id=client_message_id
)
self.messages.append(user_message)
self.stats.steps += 1
self._current_user_message_id = user_message.message_id
if user_message.message_id is None:
raise AgentLoopError("User message must have a message_id")
yield UserMessageEvent(content=user_msg, message_id=user_message.message_id)
if self._hooks_manager:
self._hooks_manager.reset_retry_count()
try:
should_break_loop = False
first_llm_turn = True
while not should_break_loop:
self._is_user_prompt_call = False
result = await self.middleware_pipeline.run_before_turn(
self._get_context()
)
async for event in self._handle_middleware_result(result):
yield event
if result.action == MiddlewareAction.STOP:
return
self.stats.steps += 1
user_cancelled = False
if first_llm_turn:
self._is_user_prompt_call = True
first_llm_turn = False
async for event in self._perform_llm_turn():
if is_user_cancellation_event(event):
user_cancelled = True
yield event
await self._save_messages()
self._is_user_prompt_call = False
last_message = self.messages[-1]
should_break_loop = last_message.role != Role.tool
if user_cancelled:
return
if should_break_loop and self._hooks_manager:
hook_retry: HookUserMessage | None = None
async for hook_event in self._hooks_manager.run(
HookType.POST_AGENT_TURN, self.session_id, self.session_logger
):
if isinstance(hook_event, HookUserMessage):
hook_retry = hook_event
else:
yield hook_event
if hook_retry is not None:
self.messages.append(
LLMMessage(
role=Role.user,
content=hook_retry.content,
injected=True,
)
)
should_break_loop = False
finally:
await self._save_messages()
async def _perform_llm_turn(self) -> AsyncGenerator[BaseEvent, None]:
if self.enable_streaming:
async for event in self._stream_assistant_events():
yield event
else:
assistant_event = await self._get_assistant_event()
if assistant_event.content:
yield assistant_event
last_message = self.messages[-1]
parsed = self.format_handler.parse_message(last_message)
resolved = self.format_handler.resolve_tool_calls(parsed, self.tool_manager)
if not resolved.tool_calls and not resolved.failed_calls:
return
profile_before = self.agent_profile.name
async for event in self._handle_tool_calls(resolved):
yield event
if self.agent_profile.name != profile_before:
yield AgentProfileChangedEvent(agent_name=self.agent_profile.name)
def _build_tool_call_events(
self, tool_calls: list[ToolCall] | None, emitted_ids: set[str]
) -> Generator[ToolCallEvent, None, None]:
for tc in tool_calls or []:
if tc.id is None or not tc.function.name:
continue
if tc.id in emitted_ids:
continue
tool_class = self.tool_manager.available_tools.get(tc.function.name)
if tool_class is None:
continue
yield ToolCallEvent(
tool_call_id=tc.id,
tool_call_index=tc.index,
tool_name=tc.function.name,
tool_class=tool_class,
)
async def _stream_assistant_events(
self,
) -> AsyncGenerator[AssistantEvent | ReasoningEvent | ToolCallEvent]:
message_id: str | None = None
reasoning_message_id: str | None = None
emitted_tool_call_ids = set[str]()
async for chunk in self._chat_streaming():
if message_id is None:
message_id = chunk.message.message_id
if reasoning_message_id is None:
reasoning_message_id = chunk.message.reasoning_message_id
for event in self._build_tool_call_events(
chunk.message.tool_calls, emitted_tool_call_ids
):
emitted_tool_call_ids.add(event.tool_call_id)
yield event
if chunk.message.reasoning_content:
yield ReasoningEvent(
content=chunk.message.reasoning_content,
message_id=reasoning_message_id,
)
if chunk.message.content:
yield AssistantEvent(
content=chunk.message.content, message_id=message_id
)
async def _get_assistant_event(self) -> AssistantEvent:
llm_result = await self._chat()
return AssistantEvent(
content=llm_result.message.content or "",
message_id=llm_result.message.message_id,
)
async def _emit_failed_tool_events(
self, failed_calls: list[FailedToolCall]
) -> AsyncGenerator[ToolResultEvent]:
for failed in failed_calls:
error_msg = f"<{TOOL_ERROR_TAG}>{failed.tool_name}: {failed.error}</{TOOL_ERROR_TAG}>"
yield ToolResultEvent(
tool_name=failed.tool_name,
tool_class=None,
error=error_msg,
tool_call_id=failed.call_id,
)
self.stats.tool_calls_failed += 1
self.messages.append(
self.format_handler.create_failed_tool_response_message(
failed, error_msg
)
)
async def _process_one_tool_call(
self, tool_call: ResolvedToolCall
) -> AsyncGenerator[ToolResultEvent | ToolStreamEvent]:
async with tool_span(
tool_name=tool_call.tool_name,
call_id=tool_call.call_id,
arguments=tool_call.validated_args.model_dump_json(),
) as span:
async for event in self._execute_tool_call(span, tool_call):
yield event
async def _execute_tool_call(
self, span: trace.Span, tool_call: ResolvedToolCall
) -> AsyncGenerator[ToolResultEvent | ToolStreamEvent]:
try:
tool_instance = self.tool_manager.get(tool_call.tool_name)
except Exception as exc:
error_msg = f"Error getting tool '{tool_call.tool_name}': {exc}"
yield self._tool_failure_event(tool_call, error_msg, span=span)
return
decision: ToolDecision | None = None
try:
decision = await self._should_execute_tool(
tool_instance, tool_call.validated_args, tool_call.call_id
)
if decision.verdict == ToolExecutionResponse.SKIP:
self.stats.tool_calls_rejected += 1
skip_reason = decision.feedback or str(
get_user_cancellation_message(
CancellationReason.TOOL_SKIPPED, tool_call.tool_name
)
)
yield ToolResultEvent(
tool_name=tool_call.tool_name,
tool_class=tool_call.tool_class,
skipped=True,