Repository navigation
Expand file tree
/
Copy pathquery.ts
More file actions
1729 lines (1612 loc) · 67.1 KB
/
Copy pathquery.ts
File metadata and controls
1729 lines (1612 loc) · 67.1 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
9
8000
09
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
// biome-ignore-all assist/source/organizeImports: ANT-ONLY import markers must not be reordered
import type {
ToolResultBlockParam,
ToolUseBlock,
} from '@anthropic-ai/sdk/resources/index.mjs'
import type { CanUseToolFn } from './hooks/useCanUseTool.js'
import { FallbackTriggeredError } from './services/api/withRetry.js'
import {
calculateTokenWarningState,
isAutoCompactEnabled,
type AutoCompactTrackingState,
} from './services/compact/autoCompact.js'
import { buildPostCompactMessages } from './services/compact/compact.js'
/* eslint-disable @typescript-eslint/no-require-imports */
const reactiveCompact = feature('REACTIVE_COMPACT')
? (require('./services/compact/reactiveCompact.js') as typeof import('./services/compact/reactiveCompact.js'))
: null
const contextCollapse = feature('CONTEXT_COLLAPSE')
? (require('./services/contextCollapse/index.js') as typeof import('./services/contextCollapse/index.js'))
: null
/* eslint-enable @typescript-eslint/no-require-imports */
import {
logEvent,
type AnalyticsMetadata_I_VERIFIED_THIS_IS_NOT_CODE_OR_FILEPATHS,
} from 'src/services/analytics/index.js'
import { ImageSizeError } from './utils/imageValidation.js'
import { ImageResizeError } from './utils/imageResizer.js'
import { findToolByName, type ToolUseContext } from './Tool.js'
import { asSystemPrompt, type SystemPrompt } from './utils/systemPromptType.js'
import type {
AssistantMessage,
AttachmentMessage,
Message,
RequestStartEvent,
StreamEvent,
ToolUseSummaryMessage,
UserMessage,
TombstoneMessage,
} from './types/message.js'
import { logError } from './utils/log.js'
import {
PROMPT_TOO_LONG_ERROR_MESSAGE,
isPromptTooLongMessage,
} from './services/api/errors.js'
import { logAntError, logForDebugging } from './utils/debug.js'
import {
createUserMessage,
createUserInterruptionMessage,
normalizeMessagesForAPI,
createSystemMessage,
createAssistantAPIErrorMessage,
getMessagesAfterCompactBoundary,
createToolUseSummaryMessage,
createMicrocompactBoundaryMessage,
stripSignatureBlocks,
} from './utils/messages.js'
import { generateToolUseSummary } from './services/toolUseSummary/toolUseSummaryGenerator.js'
import { prependUserContext, appendSystemContext } from './utils/api.js'
import {
createAttachmentMessage,
filterDuplicateMemoryAttachments,
getAttachmentMessages,
startRelevantMemoryPrefetch,
} from './utils/attachments.js'
/* eslint-disable @typescript-eslint/no-require-imports */
const skillPrefetch = feature('EXPERIMENTAL_SKILL_SEARCH')
? (require('./services/skillSearch/prefetch.js') as typeof import('./services/skillSearch/prefetch.js'))
: null
const jobClassifier = feature('TEMPLATES')
? (require('./jobs/classifier.js') as typeof import('./jobs/classifier.js'))
: null
/* eslint-enable @typescript-eslint/no-require-imports */
import {
remove as removeFromQueue,
getCommandsByMaxPriority,
isSlashCommand,
} from './utils/messageQueueManager.js'
import { notifyCommandLifecycle } from './utils/commandLifecycle.js'
import { headlessProfilerCheckpoint } from './utils/headlessProfiler.js'
import {
getRuntimeMainLoopModel,
renderModelName,
} from './utils/model/model.js'
import {
doesMostRecentAssistantMessageExceed200k,
finalContextTokensFromLastResponse,
tokenCountWithEstimation,
} from './utils/tokens.js'
import { ESCALATED_MAX_TOKENS } from './utils/context.js'
import { getFeatureValue_CACHED_MAY_BE_STALE } from './services/analytics/growthbook.js'
import { SLEEP_TOOL_NAME } from './tools/SleepTool/prompt.js'
import { executePostSamplingHooks } from './utils/hooks/postSamplingHooks.js'
import { executeStopFailureHooks } from './utils/hooks.js'
import type { QuerySource } from './constants/querySource.js'
import { createDumpPromptsFetch } from './services/api/dumpPrompts.js'
import { StreamingToolExecutor } from './services/tools/StreamingToolExecutor.js'
import { queryCheckpoint } from './utils/queryProfiler.js'
import { runTools } from './services/tools/toolOrchestration.js'
import { applyToolResultBudget } from './utils/toolResultStorage.js'
import { recordContentReplacement } from './utils/sessionStorage.js'
import { handleStopHooks } from './query/stopHooks.js'
import { buildQueryConfig } from './query/config.js'
import { productionDeps, type QueryDeps } from './query/deps.js'
import type { Terminal, Continue } from './query/transitions.js'
import { feature } from 'bun:bundle'
import {
getCurrentTurnTokenBudget,
getTurnOutputTokens,
incrementBudgetContinuationCount,
} from './bootstrap/state.js'
import { createBudgetTracker, checkTokenBudget } from './query/tokenBudget.js'
import { count } from './utils/array.js'
/* eslint-disable @typescript-eslint/no-require-imports */
const snipModule = feature('HISTORY_SNIP')
? (require('./services/compact/snipCompact.js') as typeof import('./services/compact/snipCompact.js'))
: null
const taskSummaryModule = feature('BG_SESSIONS')
? (require('./utils/taskSummary.js') as typeof import('./utils/taskSummary.js'))
: null
/* eslint-enable @typescript-eslint/no-require-imports */
function* yieldMissingToolResultBlocks(
assistantMessages: AssistantMessage[],
errorMessage: string,
) {
for (const assistantMessage of assistantMessages) {
// Extract all tool use blocks from this assistant message
const toolUseBlocks = assistantMessage.message.content.filter(
content => content.type === 'tool_use',
) as ToolUseBlock[]
// Emit an interruption message for each tool use
for (const toolUse of toolUseBlocks) {
yield createUserMessage({
content: [
{
type: 'tool_result',
content: errorMessage,
is_error: true,
tool_use_id: toolUse.id,
},
],
toolUseResult: errorMessage,
sourceToolAssistantUUID: assistantMessage.uuid,
})
}
}
}
/**
* The rules of thinking are lengthy and fortuitous. They require plenty of thinking
* of most long duration and deep meditation for a wizard to wrap one's noggin around.
*
* The rules follow:
* 1. A message that contains a thinking or redacted_thinking block must be part of a query whose max_thinking_length > 0
* 2. A thinking block may not be the last message in a block
* 3. Thinking blocks must be preserved for the duration of an assistant trajectory (a single turn, or if that turn includes a tool_use block then also its subsequent tool_result and the following assistant message)
*
* Heed these rules well, young wizard. For they are the rules of thinking, and
* the rules of thinking are the rules of the universe. If ye does not heed these
* rules, ye will be punished with an entire day of debugging and hair pulling.
*/
const MAX_OUTPUT_TOKENS_RECOVERY_LIMIT = 3
/**
* Is this a max_output_tokens error message? If so, the streaming loop should
* withhold it from SDK callers until we know whether the recovery loop can
* continue. Yielding early leaks an intermediate error to SDK callers (e.g.
* cowork/desktop) that terminate the session on any `error` field — the
* recovery loop keeps running but nobody is listening.
*
* Mirrors reactiveCompact.isWithheldPromptTooLong.
*/
function isWithheldMaxOutputTokens(
msg: Message | StreamEvent | undefined,
): msg is AssistantMessage {
return msg?.type === 'assistant' && msg.apiError === 'max_output_tokens'
}
export type QueryParams = {
messages: Message[]
systemPrompt: SystemPrompt
userContext: { [k: string]: string }
systemContext: { [k: string]: string }
canUseTool: CanUseToolFn
toolUseContext: ToolUseContext
fallbackModel?: string
querySource: QuerySource
maxOutputTokensOverride?: number
maxTurns?: number
skipCacheWrite?: boolean
// API task_budget (output_config.task_budget, beta task-budgets-2026-03-13).
// Distinct from the tokenBudget +500k auto-continue feature. `total` is the
// budget for the whole agentic turn; `remaining` is computed per iteration
// from cumulative API usage. See configureTaskBudgetParams in claude.ts.
taskBudget?: { total: number }
deps?: QueryDeps
}
// -- query loop state
// Mutable state carried between loop iterations
type State = {
messages: Message[]
toolUseContext: ToolUseContext
autoCompactTracking: AutoCompactTrackingState | undefined
maxOutputTokensRecoveryCount: number
hasAttemptedReactiveCompact: boolean
maxOutputTokensOverride: number | undefined
pendingToolUseSummary: Promise<ToolUseSummaryMessage | null> | undefined
stopHookActive: boolean | undefined
turnCount: number
// Why the previous iteration continued. Undefined on first iteration.
// Lets tests assert recovery paths fired without inspecting message contents.
transition: Continue | undefined
}
export async function* query(
params: QueryParams,
): AsyncGenerator<
| StreamEvent
| RequestStartEvent
| Message
| TombstoneMessage
| ToolUseSummaryMessage,
Terminal
> {
const consumedCommandUuids: string[] = []
const terminal = yield* queryLoop(params, consumedCommandUuids)
// Only reached if queryLoop returned normally. Skipped on throw (error
// propagates through yield*) and on .return() (Return completion closes
// both generators). This gives the same asymmetric started-without-completed
// signal as print.ts's drainCommandQueue when the turn fails.
for (const uuid of consumedCommandUuids) {
notifyCommandLifecycle(uuid, 'completed')
}
return terminal
}
async function* queryLoop(
params: QueryParams,
consumedCommandUuids: string[],
): AsyncGenerator<
| StreamEvent
| RequestStartEvent
| Message
| TombstoneMessage
| ToolUseSummaryMessage,
Terminal
> {
// Immutable params — never reassigned during the query loop.
const {
systemPrompt,
userContext,
systemContext,
canUseTool,
fallbackModel,
querySource,
maxTurns,
skipCacheWrite,
} = params
const deps = params.deps ?? productionDeps()
// Mutable cross-iteration state. The loop body destructures this at the top
// of each iteration so reads stay bare-name (`messages`, `toolUseContext`).
// Continue sites write `state = { ... }` instead of 9 separate assignments.
let state: State = {
messages: params.messages,
toolUseContext: params.toolUseContext,
maxOutputTokensOverride: params.maxOutputTokensOverride,
autoCompactTracking: undefined,
stopHookActive: undefined,
maxOutputTokensRecoveryCount: 0,
hasAttemptedReactiveCompact: false,
turnCount: 1,
pendingToolUseSummary: undefined,
transition: undefined,
}
const budgetTracker = feature('TOKEN_BUDGET') ? createBudgetTracker() : null
// task_budget.remaining tracking across compaction boundaries. Undefined
// until first compact fires — while context is uncompacted the server can
// see the full history and handles the countdown from {total} itself (see
// api/api/sampling/prompt/renderer.py:292). After a compact, the server sees
// only the summary and would under-count spend; remaining tells it the
// pre-compact final window that got summarized away. Cumulative across
// multiple compacts: each subtracts the final context at that compact's
// trigger point. Loop-local (not on State) to avoid touching the 7 continue
// sites.
let taskBudgetRemaining: number | undefined = undefined
// Snapshot immutable env/statsig/session state once at entry. See QueryConfig
// for what's included and why feature() gates are intentionally excluded.
const config = buildQueryConfig()
// Fired once per user turn — the prompt is invariant across loop iterations,
// so per-iteration firing would ask sideQuery the same question N times.
// Consume point polls settledAt (never blocks). `using` disposes on all
// generator exit paths — see MemoryPrefetch for dispose/telemetry semantics.
using pendingMemoryPrefetch = startRelevantMemoryPrefetch(
state.messages,
state.toolUseContext,
)
// eslint-disable-next-line no-constant-condition
while (true) {
// Destructure state at the top of each iteration. toolUseContext alone
// is reassigned within an iteration (queryTracking, messages updates);
// the rest are read-only between continue sites.
let { toolUseContext } = state
const {
messages,
autoCompactTracking,
maxOutputTokensRecoveryCount,
hasAttemptedReactiveCompact,
maxOutputTokensOverride,
pendingToolUseSummary,
stopHookActive,
turnCount,
} = state
// Skill discovery prefetch — per-iteration (uses findWritePivot guard
// that returns early on non-write iterations). Discovery runs while the
// model streams and tools execute; awaited post-tools alongside the
// memory prefetch consume. Replaces the blocking assistant_turn path
// that ran inside getAttachmentMessages (97% of those calls found
// nothing in prod). Turn-0 user-input discovery still blocks in
// userInputAttachments — that's the one signal where there's no prior
// work to hide under.
const pendingSkillPrefetch = skillPrefetch?.startSkillDiscoveryPrefetch(
null,
messages,
toolUseContext,
)
yield { type: 'stream_request_start' }
queryCheckpoint('query_fn_entry')
// Record query start for headless latency tracking (skip for subagents)
if (!toolUseContext.agentId) {
headlessProfilerCheckpoint('query_started')
}
// Initialize or increment query chain tracking
const queryTracking = toolUseContext.queryTracking
? {
chai
8
nId: too
10BC8
lUseContext.queryTracking.chainId,
depth: toolUseContext.queryTracking.depth + 1,
}
: {
chainId: deps.uuid(),
depth: 0,
}
const queryChainIdForAnalytics =
queryTracking.chainId as AnalyticsMetadata_I_VERIFIED_THIS_IS_NOT_CODE_OR_FILEPATHS
toolUseContext = {
...toolUseContext,
queryTracking,
}
let messagesForQuery = [...getMessagesAfterCompactBoundary(messages)]
let tracking = autoCompactTracking
// Enforce per-message budget on aggregate tool result size. Runs BEFORE
// microcompact — cached MC operates purely by tool_use_id (never inspects
// content), so content replacement is invisible to it and the two compose
// cleanly. No-ops when contentReplacementState is undefined (feature off).
// Persist only for querySources that read records back on resume: agentId
// routes to sidechain file (AgentTool resume) or session file (/resume).
// Ephemeral runForkedAgent callers (agent_summary etc.) don't persist.
const persistReplacements =
querySource.startsWith('agent:') ||
querySource.startsWith('repl_main_thread')
messagesForQuery = await applyToolResultBudget(
messagesForQuery,
toolUseContext.contentReplacementState,
persistReplacements
? records =>
void recordContentReplacement(
records,
toolUseContext.agentId,
).catch(logError)
: undefined,
new Set(
toolUseContext.options.tools
.filter(t => !Number.isFinite(t.maxResultSizeChars))
.map(t => t.name),
),
)
// Apply snip before microcompact (both may run — they are not mutually exclusive).
// snipTokensFreed is plumbed to autocompact so its threshold check reflects
// what snip removed; tokenCountWithEstimation alone can't see it (reads usage
// from the protected-tail assistant, which survives snip unchanged).
let snipTokensFreed = 0
if (feature('HISTORY_SNIP')) {
queryCheckpoint('query_snip_start')
const snipResult = snipModule!.snipCompactIfNeeded(messagesForQuery)
messagesForQuery = snipResult.messages
snipTokensFreed = snipResult.tokensFreed
if (snipResult.boundaryMessage) {
yield snipResult.boundaryMessage
}
queryCheckpoint('query_snip_end')
}
// Apply microcompact before autocompact
queryCheckpoint('query_microcompact_start')
const microcompactResult = await deps.microcompact(
messagesForQuery,
toolUseContext,
querySource,
)
messagesForQuery = microcompactResult.messages
// For cached microcompact (cache editing), defer boundary message until after
// the API response so we can use actual cache_deleted_input_tokens.
// Gated behind feature() so the string is eliminated from external builds.
const pendingCacheEdits = feature('CACHED_MICROCOMPACT')
? microcompactResult.compactionInfo?.pendingCacheEdits
: undefined
queryCheckpoint('query_microcompact_end')
// Project the collapsed context view and maybe commit more collapses.
// Runs BEFORE autocompact so that if collapse gets us under the
// autocompact threshold, autocompact is a no-op and we keep granular
// context instead of a single summary.
//
// Nothing is yielded — the collapsed view is a read-time projection
// over the REPL's full history. Summary messages live in the collapse
// store, not the REPL array. This is what makes collapses persist
// across turns: projectView() replays the commit log on every entry.
// Within a turn, the view flows forward via state.messages at the
// continue site (query.ts:1192), and the next projectView() no-ops
// because the archived messages are already gone from its input.
if (feature('CONTEXT_COLLAPSE') && contextCollapse) {
const collapseResult = await contextCollapse.applyCollapsesIfNeeded(
messagesForQuery,
toolUseContext,
querySource,
)
messagesForQuery = collapseResult.messages
}
const fullSystemPrompt = asSystemPrompt(
appendSystemContext(systemPrompt, systemContext),
)
queryCheckpoint('query_autocompact_start')
const { compactionResult, consecutiveFailures } = await deps.autocompact(
messagesForQuery,
toolUseContext,
{
systemPrompt,
userContext,
systemContext,
toolUseContext,
forkContextMessages: messagesForQuery,
},
querySource,
tracking,
snipTokensFreed,
)
queryCheckpoint('query_autocompact_end')
if (compactionResult) {
const {
preCompactTokenCount,
postCompactTokenCount,
truePostCompactTokenCount,
compactionUsage,
} = compactionResult
logEvent('tengu_auto_compact_succeeded', {
originalMessageCount: messages.length,
compactedMessageCount:
compactionResult.summaryMessages.length +
compactionResult.attachments.length +
compactionResult.hookResults.length,
preCompactTokenCount,
postCompactTokenCount,
truePostCompactTokenCount,
compactionInputTokens: compactionUsage?.input_tokens,
compactionOutputTokens: compactionUsage?.output_tokens,
compactionCacheReadTokens:
compactionUsage?.cache_read_input_tokens ?? 0,
compactionCacheCreationTokens:
compactionUsage?.cache_creation_input_tokens ?? 0,
compactionTotalTokens: compactionUsage
? compactionUsage.input_tokens +
(compactionUsage.cache_creation_input_tokens ?? 0) +
(compactionUsage.cache_read_input_tokens ?? 0) +
compactionUsage.output_tokens
: 0,
queryChainId: queryChainIdForAnalytics,
queryDepth: queryTracking.depth,
})
// task_budget: capture pre-compact final context window before
// messagesForQuery is replaced with postCompactMessages below.
// iterations[-1] is the authoritative final window (post server tool
// loops); see #304930.
if (params.taskBudget) {
const preCompactContext =
finalContextTokensFromLastResponse(messagesForQuery)
taskBudgetRemaining = Math.max(
0,
(taskBudgetRemaining ?? params.taskBudget.total) - preCompactContext,
)
}
// Reset on every compact so turnCounter/turnId reflect the MOST RECENT
// compact. recompactionInfo (autoCompact.ts:190) already captured the
// old values for turnsSincePreviousCompact/previousCompactTurnId before
// the call, so this reset doesn't lose those.
tracking = {
compacted: true,
turnId: deps.uuid(),
turnCounter: 0,
consecutiveFailures: 0,
}
const postCompactMessages = buildPostCompactMessages(compactionResult)
for (const message of postCompactMessages) {
yield message
}
// Continue on with the current query call using the post compact messages
messagesForQuery = postCompactMessages
} else if (consecutiveFailures !== undefined) {
// Autocompact failed — propagate failure count so the circuit breaker
// can stop retrying on the next iteration.
tracking = {
...(tracking ?? { compacted: false, turnId: '', turnCounter: 0 }),
consecutiveFailures,
}
}
//TODO: no need to set toolUseContext.messages during set-up since it is updated here
toolUseContext = {
...toolUseContext,
messages: messagesForQuery,
}
const assistantMessages: AssistantMessage[] = []
const toolResults: (UserMessage | AttachmentMessage)[] = []
// @see https://docs.claude.com/en/docs/build-with-claude/tool-use
// Note: stop_reason === 'tool_use' is unreliable -- it's not always set correctly.
// Set during streaming whenever a tool_use block arrives — the sole
// loop-exit signal. If false after streaming, we're done (modulo stop-hook retry).
const toolUseBlocks: ToolUseBlock[] = []
let needsFollowUp = false
queryCheckpoint('query_setup_start')
const useStreamingToolExecution = config.gates.streamingToolExecution
let streamingToolExecutor = useStreamingToolExecution
? new StreamingToolExecutor(
toolUseContext.options.tools,
canUseTool,
toolUseContext,
)
: null
const appState = toolUseContext.getAppState()
const permissionMode = appState.toolPermissionContext.mode
let currentModel = getRuntimeMainLoopModel({
permissionMode,
mainLoopModel: toolUseContext.options.mainLoopModel,
exceeds200kTokens:
permissionMode === 'plan' &&
doesMostRecentAssistantMessageExceed200k(messagesForQuery),
})
queryCheckpoint('query_setup_end')
// Create fetch wrapper once per query session to avoid memory retention.
// Each call to createDumpPromptsFetch creates a closure that captures the request body.
// Creating it once means only the latest request body is retained (~700KB),
// instead of all request bodies from the session (~500MB for long sessions).
// Note: agentId is effectively constant during a query() call - it only changes
// between queries (e.g., /clear command or session resume).
const dumpPromptsFetch = config.gates.isAnt
? createDumpPromptsFetch(toolUseContext.agentId ?? config.sessionId)
: undefined
// Block if we've hit the hard blocking limit (only applies when auto-compact is OFF)
// This reserves space so users can still run /compact manually
// Skip this check if compaction just happened - the compaction result is already
// validated to be under the threshold, and tokenCountWithEstimation would use
// stale input_tokens from kept messages that reflect pre-compaction context size.
// Same staleness applies to snip: subtract snipTokensFreed (otherwise we'd
// falsely block in the window where snip brought us under autocompact threshold
// but the stale usage is still above blocking limit — before this PR that
// window never existed because autocompact always fired on the stale count).
// Also skip for compact/session_memory queries — these are forked agents that
// inherit the full conversation and would deadlock if blocked here (the compact
// agent needs to run to REDUCE the token count).
// Also skip when reactive compact is enabled and automatic compaction is
// allowed — the preempt's synthetic error returns before the API call,
// so reactive compact would never see a prompt-too-long to react to.
// Widened to walrus so RC can act as fallback when proactive fails.
//
// Same skip for context-collapse: its recoverFromOverflow drains
// staged collapses on a REAL API 413, then falls through to
// reactiveCompact. A synthetic preempt here would return before the
// API call and starve both recovery paths. The isAutoCompactEnabled()
// conjunct preserves the user's explicit "no automatic anything"
// config — if they set DISABLE_AUTO_COMPACT, they get the preempt.
let collapseOwnsIt = false
if (feature('CONTEXT_COLLAPSE')) {
collapseOwnsIt =
(contextCollapse?.isContextCollapseEnabled() ?? false) &&
isAutoCompactEnabled()
}
// Hoist media-recovery gate once per turn. Withholding (inside the
// stream loop) and recovery (after) must agree; CACHED_MAY_BE_STALE can
// flip during the 5-30s stream, and withhold-without-recover would eat
// the message. PTL doesn't hoist because its withholding is ungated —
// it predates the experiment and is already the control-arm baseline.
const mediaRecoveryEnabled =
reactiveCompact?.isReactiveCompactEnabled() ?? false
if (
!compactionResult &&
querySource !== 'compact' &&
querySource !== 'session_memory' &&
!(
reactiveCompact?.isReactiveCompactEnabled() && isAutoCompactEnabled()
) &&
!collapseOwnsIt
) {
const { isAtBlockingLimit } = calculateTokenWarningState(
tokenCountWithEstimation(messagesForQuery) - snipTokensFreed,
toolUseContext.options.mainLoopModel,
)
if (isAtBlockingLimit) {
yield createAssistantAPIErrorMessage({
content: PROMPT_TOO_LONG_ERROR_MESSAGE,
error: 'invalid_request',
})
return { reason: 'blocking_limit' }
}
}
let attemptWithFallback = true
queryCheckpoint('query_api_loop_start')
try {
while (attemptWithFallback) {
attemptWithFallback = false
try {
let streamingFallbackOccured = false
queryCheckpoint('query_api_streaming_start')
for await (const message of deps.callModel({
messages: prependUserContext(messagesForQuery, userContext),
systemPrompt: fullSystemPrompt,
thinkingConfig: toolUseContext.options.thinkingConfig,
tools: toolUseContext.options.tools,
signal: toolUseContext.abortController.signal,
options: {
async getToolPermissionContext() {
const appState = toolUseContext.getAppState()
return appState.toolPermissionContext
},
model: currentModel,
...(config.gates.fastModeEnabled && {
fastMode: appState.fastMode,
}),
toolChoice: undefined,
isNonInteractiveSession:
toolUseContext.options.isNonInteractiveSession,
fallbackModel,
onStreamingFallback: () => {
streamingFallbackOccured = true
},
querySource,
agents: toolUseContext.options.agentDefinitions.activeAgents,
allowedAgentTypes:
toolUseContext.options.agentDefinitions.allowedAgentTypes,
hasAppendSystemPrompt:
!!toolUseContext.options.appendSystemPrompt,
maxOutputTokensOverride,
fetchOverride: dumpPromptsFetch,
mcpTools: appState.mcp.tools,
hasPendingMcpServers: appState.mcp.clients.some(
c => c.type === 'pending',
),
queryTracking,
effortValue: appState.effortValue,
advisorModel: appState.advisorModel,
skipCacheWrite,
agentId: toolUseContext.agentId,
addNotification: toolUseContext.addNotification,
...(params.taskBudget && {
taskBudget: {
total: params.taskBudget.total,
...(taskBudgetRemaining !== undefined && {
remaining: taskBudgetRemaining,
}),
},
}),
},
})) {
// We won't use the tool_calls from the first attempt
// We could.. but then we'd have to merge assistant messages
// with different ids and double up on full the tool_results
if (streamingFallbackOccured) {
// Yield tombstones for orphaned messages so they're removed from UI and transcript.
// These partial messages (especially thinking blocks) have invalid signatures
// that would cause "thinking blocks cannot be modified" API errors.
for (const msg of assistantMessages) {
yield { type: 'tombstone' as const, message: msg }
}
logEvent('tengu_orphaned_messages_tombstoned', {
orphanedMessageCount: assistantMessages.length,
queryChainId: queryChainIdForAnalytics,
queryDepth: queryTracking.depth,
})
assistantMessages.length = 0
toolResults.length = 0
toolUseBlocks.length = 0
needsFollowUp = false
// Discard pending results from the failed streaming attempt and create
// a fresh executor. This prevents orphan tool_results (with old tool_use_ids)
// from being yielded after the fallback response arrives.
if (streamingToolExecutor) {
streamingToolExecutor.discard()
streamingToolExecutor = new StreamingToolExecutor(
toolUseContext.options.tools,
canUseTool,
toolUseContext,
)
}
}
// Backfill tool_use inputs on a cloned message before yield so
// SDK stream output and transcript serialization see legacy/derived
// fields. The original `message` is left untouched for
// assistantMessages.push below — it flows back to the API and
// mutating it would break prompt caching (byte mismatch).
let yieldMessage: typeof message = message
if (message.type === 'assistant') {
let clonedContent: typeof message.message.content | undefined
for (let i = 0; i < message.message.content.length; i++) {
const block = message.message.content[i]!
if (
block.type === 'tool_use' &&
typeof block.input === 'object' &&
block.input !== null
) {
const tool = findToolByName(
toolUseContext.options.tools,
block.name,
)
if (tool?.backfillObservableInput) {
const originalInput = block.input as Record<string, unknown>
const inputCopy = { ...originalInput }
tool.backfillObservableInput(inputCopy)
// Only yield a clone when backfill ADDED fields; skip if
// it only OVERWROTE existing ones (e.g. file tools
// expanding file_path). Overwrites change the serialized
// transcript and break VCR fixture hashes on resume,
// while adding nothing the SDK stream needs — hooks get
// the expanded path via toolExecution.ts separately.
const addedFields = Object.keys(inputCopy).some(
k => !(k in originalInput),
)
if (addedFields) {
clonedContent ??= [...message.message.content]
clonedContent[i] = { ...block, input: inputCopy }
}
}
}
}
if (clonedContent) {
yieldMessage = {
...message,
message: { ...message.message, content: clonedContent },
}
}
}
// Withhold recoverable errors (prompt-too-long, max-output-tokens)
// until we know whether recovery (collapse drain / reactive
// compact / truncation retry) can succeed. Still pushed to
// assistantMessages so the recovery checks below find them.
// Either subsystem's withhold is sufficient — they're
// independent so turning one off doesn't break the other's
// recovery path.
//
// feature() only works in if/ternary conditions (bun:bundle
// tree-shaking constraint), so the collapse check is nested
// rather than composed.
let withheld = false
if (feature('CONTEXT_COLLAPSE')) {
if (
contextCollapse?.isWithheldPromptTooLong(
message,
isPromptTooLongMessage,
querySource,
)
) {
withheld = true
}
}
if (reactiveCompact?.isWithheldPromptTooLong(message)) {
withheld = true
}
if (
mediaRecoveryEnabled &&
reactiveCompact?.isWithheldMediaSizeError(message)
) {
withheld = true
}
if (isWithheldMaxOutputTokens(message)) {
withheld = true
}
if (!withheld) {
yield yieldMessage
}
if (message.type === 'assistant') {
assistantMessages.push(message)
const msgToolUseBlocks = message.message.content.filter(
content => content.type === 'tool_use',
) as ToolUseBlock[]
if (msgToolUseBlocks.length > 0) {
toolUseBlocks.push(...msgToolUseBlocks)
needsFollowUp = true
}
if (
streamingToolExecutor &&
!toolUseContext.abortController.signal.aborted
) {
for (const toolBlock of msgToolUseBlocks) {
streamingToolExecutor.addTool(toolBlock, message)
}
}
}
if (
streamingToolExecutor &&
!toolUseContext.abortController.signal.aborted
) {
for (const result of streamingToolExecutor.getCompletedResults()) {
if (result.message) {
yield result.message
toolResults.push(
...normalizeMessagesForAPI(
[result.message],
toolUseContext.options.tools,
).filter(_ => _.type === 'user'),
)
}
}
}
}
queryCheckpoint('query_api_streaming_end')
// Yield deferred microcompact boundary message using actual API-reported
// token deletion count instead of client-side estimates.
// Entire block gated behind feature() so the excluded string
// is eliminated from external builds.
if (feature('CACHED_MICROCOMPACT') && pendingCacheEdits) {
const lastAssistant = assistantMessages.at(-1)
// The API field is cumulative/sticky across requests, so we
// subtract the baseline captured before this request to get the delta.
const usage = lastAssistant?.message.usage
const cumulativeDeleted = usage
? ((usage as unknown as Record<string, number>)
.cache_deleted_input_tokens ?? 0)
: 0
const deletedTokens = Math.max(
0,
cumulativeDeleted - pendingCacheEdits.baselineCacheDeletedTokens,
)
if (deletedTokens > 0) {
yield createMicrocompactBoundaryMessage(
pendingCacheEdits.trigger,
0,
deletedTokens,
pendingCacheEdits.deletedToolIds,
[],
)
}
}
} catch (innerError) {
if (innerError instanceof FallbackTriggeredError && fallbackModel) {
// Fallback was triggered - switch model and retry
currentModel = fallbackModel
attemptWithFallback = true
// Clear assistant messages since we'll retry the entire request
yield* yieldMissingToolResultBlocks(
assistantMessages,
'Model fallback triggered',
)
assistantMessages.length = 0
toolResults.length = 0
toolUseBlocks.length = 0
needsFollowUp = false
// Discard pending results from the failed attempt and create a
// fresh executor. This prevents orphan tool_results (with old
// tool_use_ids) from leaking into the retry.
if (streamingToolExecutor) {
streamingToolExecutor.discard()
streamingToolExecutor = new StreamingToolExecutor(
toolUseContext.options.tools,
canUseTool,
toolUseContext,
)
}
// Update tool use context with new model
toolUseContext.options.mainLoopModel = fallbackModel
// Thinking signatures are model-bound: replaying a protected-thinking
// block (e.g. capybara) to an unprotected fallback (e.g. opus) 400s.
// Strip before retry so the fallback model gets clean history.
if (process.env.USER_TYPE === 'ant') {
messagesForQuery = stripSignatureBlocks(messagesForQuery)
}
// Log the fallback event
logEvent('tengu_model_fallback_triggered', {
original_model:
innerError.originalModel as AnalyticsMetadata_I_VERIFIED_THIS_IS_NOT_CODE_OR_FILEPATHS,
fallback_model:
fallbackModel as AnalyticsMetadata_I_VERIFIED_THIS_IS_NOT_CODE_OR_FILEPATHS,
entrypoint:
'cli' as AnalyticsMetadata_I_VERIFIED_THIS_IS_NOT_CODE_OR_FILEPATHS,
queryChainId: queryChainIdForAnalytics,
queryDepth: queryTracking.depth,
})
// Yield system message about fallback — use 'warning' level so
// users see the notification without needing verbose mode
yield createSystemMessage(
`Switched to ${renderModelName(innerError.fallbackModel)} due to high demand for ${renderModelName(innerError.originalModel)}`,
'warning',
)
continue
}
throw innerError
}
}
} catch (error) {
logError(error)
const errorMessage =
error instanceof Error ? error.message : String(error)
logEvent('tengu_query_error', {
assistantMessages: assistantMessages.length,
toolUses: assistantMessages.flatMap(_ =>
_.message.content.filter(content => content.type === 'tool_use'),
).length,
queryChainId: queryChainIdForAnalytics,
queryDepth: queryTracking.depth,
})
// Handle image size/resize errors with user-friendly messages
if (
error instanceof ImageSizeError ||
error instanceof ImageResizeError
) {
yield createAssistantAPIErrorMessage({
content: error.message,
})
return { reason: 'image_error' }
}
// Generally queryModelWithStreaming should not throw errors but instead
// yield them as synthetic assistant messages. However if it does throw
// due to a bug, we may end up in a state where we have already emitted
// a tool_use block but will stop before emitting the tool_result.
yield* yieldMissingToolResultBlocks(assistantMessages, errorMessage)
// Surface the real error instead of a misleading "[Request interrupted
// by user]" — this path is a model/runtime failure, not a user action.
// SDK consumers were seeing phantom interrupts on e.g. Node 18's missing
// Array.prototype.with(), masking the actual cause.
yield createAssistantAPIErrorMessage({
content: errorMessage,
})
// To help track down bugs, log loudly for ants
logAntError('Query error', error)
return { reason: 'model_error', error }
}
// Execute post-sampling hooks after model response is complete
if (assistantMessages.length > 0) {