@adhdev/daemon-core 1.0.28-rc.15 → 1.0.28-rc.17
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/index.d.ts +1 -1
- package/dist/index.js +258 -32
- package/dist/index.js.map +1 -1
- package/dist/index.mjs +257 -32
- package/dist/index.mjs.map +1 -1
- package/dist/mesh/mesh-events-stale.d.ts +59 -0
- package/dist/mesh/mesh-queue-assignment.d.ts +1 -0
- package/dist/providers/cli-provider-instance.d.ts +34 -0
- package/package.json +3 -3
- package/src/cli-adapters/provider-cli-adapter.ts +7 -1
- package/src/commands/med-family/cli-agent.ts +32 -13
- package/src/index.ts +1 -0
- package/src/mesh/mesh-completion-synthesis.ts +35 -1
- package/src/mesh/mesh-event-forwarding.ts +129 -1
- package/src/mesh/mesh-events-stale.ts +109 -0
- package/src/mesh/mesh-queue-assignment.ts +19 -1
- package/src/mesh/mesh-reconcile-loop.ts +59 -12
- package/src/providers/cli-provider-instance.ts +135 -7
|
@@ -1,4 +1,52 @@
|
|
|
1
1
|
import type { MeshLedgerKind } from './mesh-ledger.js';
|
|
2
|
+
export interface TranscriptSynthCausalAdmission {
|
|
3
|
+
/**
|
|
4
|
+
* Synchronous re-check of a resolvable LOCAL live instance's CURRENT turn state — the
|
|
5
|
+
* same hasLiveTurnPendingEvidence() discriminator the 3a48f660 mid-turn gate uses on the
|
|
6
|
+
* native-event path. A pending verdict VETOES an eager transcript synth. Fail-open by
|
|
7
|
+
* construction: absent (remote/unknown session, separate-process caller), not a function,
|
|
8
|
+
* or throwing → no veto; the caller's bounded transcript evidence/settle behavior remains
|
|
9
|
+
* the operative net, so a missing transcript source never wedges.
|
|
10
|
+
*/
|
|
11
|
+
liveTurnPendingEvidence?: () => boolean;
|
|
12
|
+
/**
|
|
13
|
+
* The caller is a BOUNDED last-resort backstop firing at its max-wait (the acked-hold
|
|
14
|
+
* death deadline, the delivered-no-turn redrive deadline). A bounded backstop overrides
|
|
15
|
+
* the live-pending veto — preserving the genuine-final fail-open / max-wait semantics —
|
|
16
|
+
* but NEVER overrides the trailing-tool veto (a stale weak interim summary must not
|
|
17
|
+
* become final merely because a timeout fired while newer transcript/tool activity
|
|
18
|
+
* exists).
|
|
19
|
+
*/
|
|
20
|
+
boundedBackstop?: boolean;
|
|
21
|
+
/**
|
|
22
|
+
* The caller's transcript read shows tool/terminal activity AFTER the latest final-looking
|
|
23
|
+
* assistant bubble (hasTrailingToolActivityAfterFinalAssistant) — the bubble is interim
|
|
24
|
+
* narration ("Let me explore…"), the turn is still executing. Absolute veto: the worker is
|
|
25
|
+
* demonstrably alive and mid-turn, so even a bounded backstop keeps holding (the hold is
|
|
26
|
+
* still released by the session-death read-failure backstop and the stranded-reclaim /
|
|
27
|
+
* orphan-prune nets, so this cannot wedge forever).
|
|
28
|
+
*/
|
|
29
|
+
trailingToolActivityAfterFinalAssistant?: boolean;
|
|
30
|
+
}
|
|
31
|
+
export type TranscriptSynthAdmissionVerdict = {
|
|
32
|
+
admitted: true;
|
|
33
|
+
} | {
|
|
34
|
+
admitted: false;
|
|
35
|
+
reason: 'live_turn_pending_evidence' | 'trailing_tool_activity_after_final_assistant';
|
|
36
|
+
};
|
|
37
|
+
export declare function evaluateTranscriptSynthAdmission(admission?: TranscriptSynthCausalAdmission): TranscriptSynthAdmissionVerdict;
|
|
38
|
+
/**
|
|
39
|
+
* Resolve the liveTurnPendingEvidence probe for `sessionId` from a components-shaped object
|
|
40
|
+
* (anything exposing instanceManager.getInstance). Returns undefined when no live instance
|
|
41
|
+
* resolves or the instance does not expose hasLiveTurnPendingEvidence — the fail-open
|
|
42
|
+
* remote/unknown case. Shared by every transcript-synth caller so the probe is resolved ONE
|
|
43
|
+
* way (UNIFY: same discriminator as the 3a48f660 native-event gate).
|
|
44
|
+
*/
|
|
45
|
+
export declare function resolveLiveTurnPendingEvidence(components: {
|
|
46
|
+
instanceManager?: {
|
|
47
|
+
getInstance?: (sessionId: string) => unknown;
|
|
48
|
+
};
|
|
49
|
+
} | undefined, sessionId: string): (() => boolean) | undefined;
|
|
2
50
|
export declare function findRecentTerminalLedgerEvidence(args: {
|
|
3
51
|
meshId: string;
|
|
4
52
|
sessionId?: string;
|
|
@@ -64,6 +112,17 @@ export declare function reconcileDirectDispatchCompletionFromTranscript(args: {
|
|
|
64
112
|
* Direct-dispatch callers omit it → the gates apply exactly as before (no regression).
|
|
65
113
|
*/
|
|
66
114
|
preValidatedTranscriptEvidence?: boolean;
|
|
115
|
+
/**
|
|
116
|
+
* MID-TURN-CAUSAL-ADMISSION (rc.16): causal admission evidence for the unified guard —
|
|
117
|
+
* see TranscriptSynthCausalAdmission. Every transcript-synth caller (PHASE 4 reconcile,
|
|
118
|
+
* watchdog transcript propagation, MCP mesh_status poll) passes what it knows; the choke
|
|
119
|
+
* point evaluates it BEFORE writing the terminal ledger or queuing the coordinator
|
|
120
|
+
* completion. A veto returns { reconciled: false, reason } — callers treat that exactly
|
|
121
|
+
* like their other not-yet-proven outcomes (hold and re-evaluate next tick), so the held
|
|
122
|
+
* completion re-evaluates the causal evidence every pass and releases exactly once after
|
|
123
|
+
* the state clears (the terminal-ledger dedup above makes the release idempotent).
|
|
124
|
+
*/
|
|
125
|
+
causalAdmission?: TranscriptSynthCausalAdmission;
|
|
67
126
|
source?: string;
|
|
68
127
|
}): {
|
|
69
128
|
reconciled: boolean;
|
|
@@ -31,6 +31,7 @@ interface AwaitClaimBackoffState {
|
|
|
31
31
|
}
|
|
32
32
|
export declare function __resetAutoLaunchAwaitClaimBackoffForTests(): void;
|
|
33
33
|
export declare function __seedAutoLaunchAwaitClaimBackoffForTests(meshId: string, taskId: string, state: AwaitClaimBackoffState): void;
|
|
34
|
+
export declare function isIdleSessionState(state: any): boolean;
|
|
34
35
|
/** Active assignments that hold the one-active-per-node / global-parallel invariant
|
|
35
36
|
* (everything except read-only diagnoses, which run unbounded by the write cap). */
|
|
36
37
|
export declare function activeWriteAssignedCount(meshId: string): number;
|
|
@@ -355,6 +355,40 @@ export declare class CliProviderInstance implements ProviderInstance {
|
|
|
355
355
|
private isTransientToolConsent;
|
|
356
356
|
/** True when this session is parked on a modal awaiting a human answer. */
|
|
357
357
|
isModalParked(): boolean;
|
|
358
|
+
/**
|
|
359
|
+
* MID-TURN-LIVE-STATE-GATE (broader false-idle RCA, mid-turn follow-up): a live,
|
|
360
|
+
* synchronous re-check of whether this session's CURRENT turn genuinely still has
|
|
361
|
+
* unresolved work. Public wrapper so the coordinator (mesh-event-forwarding) can
|
|
362
|
+
* independently re-verify an incoming agent:generating_completed for a LOCAL session
|
|
363
|
+
* before trusting it — defense-in-depth against a race where the completion emit and the
|
|
364
|
+
* coordinator's receipt straddle a state change (screen-redraw parse artifact, decoupled-
|
|
365
|
+
* immediate emit). Reuses the EXACT same discriminators this instance's own finalization
|
|
366
|
+
* gate (getCompletedFinalizationBlock) uses — hasAdapterPendingResponse() (adapter
|
|
367
|
+
* isWaitingForResponse / currentTurnScope / isProcessing() / a non-empty partial response)
|
|
368
|
+
* OR isModalParked() (a live approval/choice modal) — so a session this method reports
|
|
369
|
+
* pending is, by construction, one the local finalization gate would also refuse to
|
|
370
|
+
* finalize right now.
|
|
371
|
+
*
|
|
372
|
+
* NATIVE-TRAILING-TOOL-GATE (rc.16 follow-up): the three adapter-state discriminators
|
|
373
|
+
* above are blind to a background shell tool that keeps a turn alive without the adapter
|
|
374
|
+
* reporting a pending response — e.g. a backgrounded `sleep 40 &` between narration
|
|
375
|
+
* bubbles on a write-lag native-source provider (claude-cli), which is deliberately
|
|
376
|
+
* un-floored (noExternalTranscriptSource omitted, mission f2f6da1b owner decision) so its
|
|
377
|
+
* transcript write-lag emits promptly. That un-flooring is correct for the ordinary
|
|
378
|
+
* fraction-of-a-second trail, but it also means the growth-hold / busy-lease protections
|
|
379
|
+
* in getCompletedFinalizationBlock never engage for claude-cli, so an interim final-
|
|
380
|
+
* LOOKING bubble followed by continuing tool calls can still finalize as a completion
|
|
381
|
+
* while the transcript demonstrably shows the turn still executing. Close that gap here,
|
|
382
|
+
* narrowly: for a native-source provider only, reuse the SAME bounded transcript read the
|
|
383
|
+
* class's own completion judgment already performs (probeNativeTranscriptSignals) and
|
|
384
|
+
* apply the SAME trailing-tool-activity veto the transcript-synth admission choke point
|
|
385
|
+
* uses (hasTrailingToolActivityAfterFinalAssistant) — the latest final-looking assistant
|
|
386
|
+
* bubble followed by tool/terminal activity is interim narration, not a turn end. Fail-open
|
|
387
|
+
* by construction: a non-native-source class (probe returns null), an unresolved
|
|
388
|
+
* transcript, or a read error never reaches the veto, so a missing/unavailable transcript
|
|
389
|
+
* can never wedge a session as "pending" forever.
|
|
390
|
+
*/
|
|
391
|
+
hasLiveTurnPendingEvidence(): boolean;
|
|
358
392
|
/**
|
|
359
393
|
* PTY-OVERTRUST-DRAIN (Defect B). The deliverability/drain status the mesh
|
|
360
394
|
* reconcile loop must consult — the RAW adapter turn-state, with the
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@adhdev/daemon-core",
|
|
3
|
-
"version": "1.0.28-rc.
|
|
3
|
+
"version": "1.0.28-rc.17",
|
|
4
4
|
"description": "ADHDev daemon core — CDP, IDE detection, providers, command execution",
|
|
5
5
|
"main": "dist/index.js",
|
|
6
6
|
"types": "dist/index.d.ts",
|
|
@@ -49,8 +49,8 @@
|
|
|
49
49
|
"author": "vilmire",
|
|
50
50
|
"license": "AGPL-3.0-or-later",
|
|
51
51
|
"dependencies": {
|
|
52
|
-
"@adhdev/mesh-shared": "1.0.28-rc.
|
|
53
|
-
"@adhdev/session-host-core": "1.0.28-rc.
|
|
52
|
+
"@adhdev/mesh-shared": "1.0.28-rc.17",
|
|
53
|
+
"@adhdev/session-host-core": "1.0.28-rc.17",
|
|
54
54
|
"@agentclientprotocol/sdk": "^0.16.1",
|
|
55
55
|
"ajv": "^8.20.0",
|
|
56
56
|
"ajv-formats": "^3.0.1",
|
|
@@ -1161,7 +1161,13 @@ export class ProviderCliAdapter implements CliAdapter {
|
|
|
1161
1161
|
: 1;
|
|
1162
1162
|
this.staticIdlePollStreak += 1;
|
|
1163
1163
|
if (this.staticIdlePollStreak >= requiredStreak) {
|
|
1164
|
-
this.engine.confirmPollStaticIdle('poll_static_idle')
|
|
1164
|
+
if (this.engine.confirmPollStaticIdle('poll_static_idle')) {
|
|
1165
|
+
// This transition originates from the read-only status poll rather
|
|
1166
|
+
// than PTY output. Propagate it through the same instance callback as
|
|
1167
|
+
// output-driven FSM transitions so completion bookkeeping observes
|
|
1168
|
+
// generating→idle and can arm/finalize the pending completion.
|
|
1169
|
+
this.onStatusChange?.();
|
|
1170
|
+
}
|
|
1165
1171
|
this.staticIdlePollStreak = 0;
|
|
1166
1172
|
}
|
|
1167
1173
|
} else {
|
|
@@ -15,6 +15,7 @@ import { getRecentActivity } from '../../config/recent-activity.js';
|
|
|
15
15
|
import { getSavedProviderSessions } from '../../config/saved-sessions.js';
|
|
16
16
|
import { listProviderHistorySessions } from '../../config/chat-history.js';
|
|
17
17
|
import { buildMeshWorkerRelayStamp, readNonEmptyString } from '../../mesh/mesh-events-utils.js';
|
|
18
|
+
import { isIdleSessionState } from '../../mesh/mesh-queue-assignment.js';
|
|
18
19
|
import { LOG } from '../../logging/logger.js';
|
|
19
20
|
import { readStringValue } from '../router.js';
|
|
20
21
|
import type { MedFamilyContext, MedFamilyHandler } from './types.js';
|
|
@@ -231,8 +232,8 @@ export const cliAgentHandlers: Record<string, MedFamilyHandler> = {
|
|
|
231
232
|
// completion event sits in the pending queue until a read_chat reconcile drains
|
|
232
233
|
// it. Stamping here makes a reused/relaunched remote session relay-safe at
|
|
233
234
|
// dispatch time even when it was not launched via mesh_launch_session.
|
|
235
|
+
const dispatchSessionId = readStringValue(args?.targetSessionId, (args as any)?.sessionId, (args as any)?.instanceId);
|
|
234
236
|
{
|
|
235
|
-
const dispatchSessionId = readStringValue(args?.targetSessionId, (args as any)?.sessionId, (args as any)?.instanceId);
|
|
236
237
|
const dispatchMeshContext = args?.meshContext as Record<string, unknown> | undefined;
|
|
237
238
|
if (dispatchSessionId && dispatchMeshContext) {
|
|
238
239
|
try {
|
|
@@ -289,18 +290,36 @@ export const cliAgentHandlers: Record<string, MedFamilyHandler> = {
|
|
|
289
290
|
// by the local queue-claim gate too so remote and local dispatch agree on when to defer.
|
|
290
291
|
const { shouldDeferDispatchForBootstrap } = await import('../../mesh/worktree-bootstrap-config.js');
|
|
291
292
|
if (shouldDeferDispatchForBootstrap(nodeObj as any)) {
|
|
292
|
-
|
|
293
|
-
|
|
294
|
-
|
|
295
|
-
|
|
296
|
-
|
|
297
|
-
|
|
298
|
-
|
|
299
|
-
|
|
300
|
-
|
|
301
|
-
|
|
302
|
-
|
|
303
|
-
|
|
293
|
+
// BOOTSTRAP-POLICY-CONSISTENCY (Fix B, rc.15 orchestration RCA): the node-level
|
|
294
|
+
// worktreeBootstrap.status flag is COARSE — it can read stale 'running' (the
|
|
295
|
+
// detached async persist chain lags the inline stamp) even though the explicit
|
|
296
|
+
// mesh_send_task caller already independently confirmed, via its own live
|
|
297
|
+
// get_status_metadata probe, that the PINNED target session is genuinely ready.
|
|
298
|
+
// Refusing that dispatch anyway is a false block. Narrowest safe override: only
|
|
299
|
+
// when the caller named a specific target session AND that session's live status
|
|
300
|
+
// — re-probed here via the same isIdleSessionState liveness check the queue
|
|
301
|
+
// assignment gate uses — is independently confirmed idle right now. A session
|
|
302
|
+
// that is starting / generating / waiting_approval / any other non-idle status
|
|
303
|
+
// never qualifies, so this can never mask a genuinely half-built worktree.
|
|
304
|
+
const dispatchSessionState = dispatchSessionId
|
|
305
|
+
? ctx.deps.instanceManager?.getInstance?.(dispatchSessionId)?.getState?.()
|
|
306
|
+
: undefined;
|
|
307
|
+
const sessionConfirmedReady = !!dispatchSessionState && isIdleSessionState(dispatchSessionState);
|
|
308
|
+
if (!sessionConfirmedReady) {
|
|
309
|
+
return {
|
|
310
|
+
success: false,
|
|
311
|
+
recoverable: true,
|
|
312
|
+
dispatched: false,
|
|
313
|
+
code: 'mesh_node_bootstrap_pending',
|
|
314
|
+
reason: 'bootstrap_still_running',
|
|
315
|
+
nodeId: dispatchNodeId,
|
|
316
|
+
meshId: dispatchMeshId,
|
|
317
|
+
...(readStringValue(meshCtx?.taskId) ? { taskId: readStringValue(meshCtx?.taskId) } : {}),
|
|
318
|
+
error: `Node '${dispatchNodeId}' worktree bootstrap is still running; a task injected now would land in the session input buffer before the provider is ready to consume it and be silently lost. Dispatch deferred.`,
|
|
319
|
+
nextAction: 'Wait for the worktree_bootstrap_complete event (or poll mesh_status until the node session is ready), then re-send the task with mesh_send_task. Alternatively use mesh_enqueue_task so the queue auto-assigns it once a ready session is available.',
|
|
320
|
+
};
|
|
321
|
+
}
|
|
322
|
+
LOG.info('MeshQueue', `Overriding stale worktreeBootstrap 'running' flag for node ${dispatchNodeId}: explicit target session ${dispatchSessionId} is independently confirmed idle/ready — dispatching mesh_send_task directly.`);
|
|
304
323
|
}
|
|
305
324
|
} catch { /* best-effort — if the bootstrap probe fails, fall through and dispatch */ }
|
|
306
325
|
}
|
package/src/index.ts
CHANGED
|
@@ -25,7 +25,7 @@ import { daemonIdListIncludes } from './mesh-reconcile-identity.js';
|
|
|
25
25
|
import { getActiveDirectDispatches, getQueue } from './mesh-work-queue.js';
|
|
26
26
|
import { readLedgerEntries } from './mesh-ledger.js';
|
|
27
27
|
import { pruneStaleDirectDispatches } from './mesh-active-work.js';
|
|
28
|
-
import { reconcileDirectDispatchCompletionFromTranscript } from './mesh-events-stale.js';
|
|
28
|
+
import { reconcileDirectDispatchCompletionFromTranscript, resolveLiveTurnPendingEvidence } from './mesh-events-stale.js';
|
|
29
29
|
import { extractFinalAssistantSummaryEvidence, hasTrailingToolActivityAfterFinalAssistant, readChatMessageTimestampMs } from '../providers/chat-message-normalization.js';
|
|
30
30
|
import type { ChatMessage } from '../types.js';
|
|
31
31
|
import {
|
|
@@ -230,6 +230,29 @@ export async function reconcileUnterminatedDirectDispatches(
|
|
|
230
230
|
const messages = Array.isArray(payload.messages) ? payload.messages as ChatMessage[] : [];
|
|
231
231
|
const evidence = extractFinalAssistantSummaryEvidence(messages);
|
|
232
232
|
|
|
233
|
+
// MID-TURN-CAUSAL-ADMISSION (rc.16, trailing-tool veto): the latest final-LOOKING
|
|
234
|
+
// assistant bubble is followed by tool/terminal activity — the worker emitted interim
|
|
235
|
+
// narration ("Let me find…") and is still running tools; the bubble is a preamble,
|
|
236
|
+
// not a turn end. This is the exact false-completion shape of the verified incidents
|
|
237
|
+
// (weak synth after an interim progress sentence while tools continued), and PHASE 4
|
|
238
|
+
// previously never ran the veto pollAssignedTaskTerminalEvidence has. The veto applies
|
|
239
|
+
// to EVERY synth path below — never-acked first-idle, acked fast-track, AND the death
|
|
240
|
+
// deadline: a stale weak interim summary must not become final merely because a
|
|
241
|
+
// timeout fired while newer transcript/tool activity exists. The fast-track streak is
|
|
242
|
+
// reset so the continuous idle-with-final-assistant run must restart from the genuine
|
|
243
|
+
// final bubble. This cannot wedge: the next tick re-reads and re-evaluates (a genuine
|
|
244
|
+
// final assistant lands AFTER the last tool call, completing one tick later), and the
|
|
245
|
+
// hold is still released by the session-death read-failure backstop / stranded-reclaim
|
|
246
|
+
// / orphan-prune nets if the worker truly goes away.
|
|
247
|
+
if (hasTrailingToolActivityAfterFinalAssistant(messages)) {
|
|
248
|
+
setHoldState(synthKey, mesh.id, { liveConfirmedSinceAck: true, consecutiveReadFailures: 0 });
|
|
249
|
+
LOG.info('MeshReconcile', `Mid-turn causal admission: task ${taskId} on node ${nodeId} (mesh ${mesh.id}) — the latest final-looking assistant bubble is followed by trailing tool/terminal activity (interim narration; the turn is still executing); holding the transcript synth`);
|
|
250
|
+
traceMeshEventDrop('reconcile_synth_veto_trailing_tool_activity', {
|
|
251
|
+
taskId, sessionId, nodeId, meshId: mesh.id, event: 'agent:generating_completed',
|
|
252
|
+
});
|
|
253
|
+
continue;
|
|
254
|
+
}
|
|
255
|
+
|
|
233
256
|
if (isAcked) {
|
|
234
257
|
const ackedAtMs = Date.parse(readNonEmptyString(dispatch.updatedAt));
|
|
235
258
|
const sinceAckMs = Number.isFinite(ackedAtMs) ? nowMs - ackedAtMs : Number.POSITIVE_INFINITY;
|
|
@@ -357,6 +380,17 @@ export async function reconcileUnterminatedDirectDispatches(
|
|
|
357
380
|
// The ledger-recovered dispatching-coordinator daemon (inside the reconcile fn)
|
|
358
381
|
// takes PRIORITY over this arg; this remains the best-available fallback.
|
|
359
382
|
...(coordinatorDaemonId ? { targetCoordinatorDaemonId: coordinatorDaemonId } : {}),
|
|
383
|
+
// MID-TURN-CAUSAL-ADMISSION (rc.16): pass the LOCAL live adapter's synchronous
|
|
384
|
+
// turn-state probe into the unified choke point — a pending verdict vetoes this
|
|
385
|
+
// eager synth (never-acked first-idle / acked fast-track). The death-deadline
|
|
386
|
+
// backstop is the bounded max-wait net and overrides the veto (genuine-final
|
|
387
|
+
// fail-open preserved); a remote/missing live source resolves to undefined and
|
|
388
|
+
// fails open onto the bounded transcript evidence above. The trailing-tool veto
|
|
389
|
+
// already ran at the top of this tick.
|
|
390
|
+
causalAdmission: {
|
|
391
|
+
liveTurnPendingEvidence: resolveLiveTurnPendingEvidence(components, sessionId),
|
|
392
|
+
boundedBackstop: backstopKind === 'ackedHoldDeathDeadlineFired',
|
|
393
|
+
},
|
|
360
394
|
source: 'daemon_reconcile_transcript_completion',
|
|
361
395
|
});
|
|
362
396
|
if (result.reconciled) {
|
|
@@ -505,6 +505,50 @@ function resolveActiveDirectDispatchTaskId(meshId: string, sessionId: string): s
|
|
|
505
505
|
}
|
|
506
506
|
}
|
|
507
507
|
|
|
508
|
+
// ---------------------------------------------------------------------------
|
|
509
|
+
// CAUSAL-COMPLETION-GATE (Fix A, rc.15 orchestration RCA): auto-launch sessions can emit a
|
|
510
|
+
// spurious agent:generating_completed before the task was ever delivered to / consumed by the
|
|
511
|
+
// session — a freshly spawned worker's boot/idle-prompt output can be misread by the provider's
|
|
512
|
+
// own completion detector as a finished turn (totalMessages=0: the worker never actually
|
|
513
|
+
// processed the task). The provider's weak/false-idle self-report is not guaranteed to catch a
|
|
514
|
+
// pure boot artifact, so this is an INDEPENDENT, coordinator-side check scoped to the exact race:
|
|
515
|
+
// a task still 'pending' whose in-window autoLaunch record names THIS session. For that narrow
|
|
516
|
+
// case we require causal evidence — the delivery was consumed (taskDeliveryConsumed) or a
|
|
517
|
+
// task_dispatched ledger entry proves the turn actually started — before letting the completion
|
|
518
|
+
// through. An already-claimed/dispatched task (the overwhelming majority of completions) never
|
|
519
|
+
// matches (its status is no longer 'pending'), so this never broadens into general suppression.
|
|
520
|
+
// ---------------------------------------------------------------------------
|
|
521
|
+
|
|
522
|
+
// The in-window, not-yet-claimed queue task (if any) whose autoLaunch record names this exact
|
|
523
|
+
// session. Bounded to AUTO_LAUNCH_AWAIT_CLAIM_MS — the same window the claim-churn/respawn guards
|
|
524
|
+
// use — so a task whose autoLaunch attempt is long expired (and has since moved on to backoff /
|
|
525
|
+
// direct-dispatch fallback, or a fresh auto-launch attempt) is never matched here.
|
|
526
|
+
function findInWindowUnclaimedAutoLaunchTask(meshId: string, sessionId: string, nowMs: number): MeshWorkQueueEntry | undefined {
|
|
527
|
+
return getQueue(meshId, { status: ['pending'] }).find(task => {
|
|
528
|
+
const al = task.autoLaunch;
|
|
529
|
+
if (!al || al.status !== 'completed' || !al.sessionId) return false;
|
|
530
|
+
if (!sessionIdsEquivalent(al.sessionId, sessionId)) return false;
|
|
531
|
+
const launchedAtMs = Date.parse(al.updatedAt);
|
|
532
|
+
return Number.isFinite(launchedAtMs) && nowMs - launchedAtMs < AUTO_LAUNCH_AWAIT_CLAIM_MS;
|
|
533
|
+
});
|
|
534
|
+
}
|
|
535
|
+
|
|
536
|
+
// Turn-start evidence distinct from taskDeliveryConsumed: a task_dispatched ledger entry naming
|
|
537
|
+
// BOTH this taskId and this sessionId proves the coordinator itself recorded a genuine dispatch
|
|
538
|
+
// of this task onto this session. A task still 'pending' (the only case this gate applies to)
|
|
539
|
+
// normally has no such entry — task_dispatched is written when the claim transitions the row to
|
|
540
|
+
// 'assigned' — so this is a defensive alternate signal, not the expected path.
|
|
541
|
+
function hasMatchingTaskDispatchedLedgerEntry(meshId: string, taskId: string, sessionId: string): boolean {
|
|
542
|
+
const entries = readLedgerEntries(meshId, { tail: 200 });
|
|
543
|
+
for (let i = entries.length - 1; i >= 0; i--) {
|
|
544
|
+
const entry = entries[i];
|
|
545
|
+
if (entry.kind !== 'task_dispatched') continue;
|
|
546
|
+
if (!sessionIdsEquivalent(entry.sessionId, sessionId)) continue;
|
|
547
|
+
if (readNonEmptyString(entry.payload?.taskId) === taskId) return true;
|
|
548
|
+
}
|
|
549
|
+
return false;
|
|
550
|
+
}
|
|
551
|
+
|
|
508
552
|
// Coordinator-side suppression/reconcile gate for an incoming mesh event. Each clause is a
|
|
509
553
|
// closed dedup/suppression concern that only inspects the event + already-resolved context and
|
|
510
554
|
// either (a) returns a `suppress` result the caller forwards verbatim, (b) returns a `reconcile`
|
|
@@ -528,12 +572,13 @@ function evaluateMeshEventSuppression(
|
|
|
528
572
|
eventNodeId: string;
|
|
529
573
|
eventTimestamp: number | null;
|
|
530
574
|
workerCoordinatorDaemonId: string | undefined;
|
|
575
|
+
components: DaemonComponents;
|
|
531
576
|
},
|
|
532
577
|
):
|
|
533
578
|
| { kind: 'suppress'; result: { success: true; forwarded: 0; suppressed: true; [extra: string]: unknown } }
|
|
534
579
|
| { kind: 'reconcile'; metadataEvent: Record<string, unknown> }
|
|
535
580
|
| null {
|
|
536
|
-
const { traceCtx, eventSessionId, eventNodeId, eventTimestamp, workerCoordinatorDaemonId } = ctx;
|
|
581
|
+
const { traceCtx, eventSessionId, eventNodeId, eventTimestamp, workerCoordinatorDaemonId, components } = ctx;
|
|
537
582
|
|
|
538
583
|
const intentionalCleanupStop = shouldSuppressIntentionalCleanupStop({
|
|
539
584
|
event: args.event,
|
|
@@ -603,6 +648,88 @@ function evaluateMeshEventSuppression(
|
|
|
603
648
|
}
|
|
604
649
|
}
|
|
605
650
|
if (args.event === 'agent:generating_completed' && eventSessionId) {
|
|
651
|
+
// MID-TURN-LIVE-STATE-GATE (broader false-idle RCA, mid-turn follow-up): before trusting
|
|
652
|
+
// ANY incoming completion, independently re-verify a resolvable LOCAL live instance's
|
|
653
|
+
// CURRENT turn state — not just the event's self-reported diagnostic — via the exact same
|
|
654
|
+
// discriminators the instance's OWN emission gate (getCompletedFinalizationBlock) uses
|
|
655
|
+
// (adapter pending-response: isWaitingForResponse / currentTurnScope / isProcessing() /
|
|
656
|
+
// a non-empty partial response; or a live approval/choice modal). A screen-redraw parse
|
|
657
|
+
// artifact, a decoupled-immediate emit, or a TOCTOU between emit and receipt can let a
|
|
658
|
+
// genuinely mid-turn session's completion reach here; this is a stateless, synchronous
|
|
659
|
+
// safety net for that race.
|
|
660
|
+
//
|
|
661
|
+
// Scope: ONLY when components.instanceManager.getInstance(sessionId) resolves a live
|
|
662
|
+
// instance exposing hasLiveTurnPendingEvidence() — i.e. a co-located worker/coordinator or
|
|
663
|
+
// single-daemon (standalone) topology. A REMOTE session (no live instance on this daemon)
|
|
664
|
+
// is untouched by this gate — absence of a resolvable instance is NOT treated as pending
|
|
665
|
+
// evidence, so a remote/unknown session is never fail-closed here; its existing async
|
|
666
|
+
// reconcile-loop protections (evaluateEarlyIdleTranscriptArm, transcript trailing-tool-
|
|
667
|
+
// activity checks) remain the operative safety net, unchanged.
|
|
668
|
+
//
|
|
669
|
+
// Bounded / never wedges: this SUPPRESSES only the ONE premature event and persists no new
|
|
670
|
+
// hold state — it relies on the adapter's OWN bounded finalization retry (which already
|
|
671
|
+
// owns re-emitting once genuinely idle, COMPLETED_FINALIZATION_MAX_WAIT_MS) plus the
|
|
672
|
+
// existing reconcile-loop completion nets (PHASE 4 transcript synth, assigned-stranded
|
|
673
|
+
// deadline, acked-hold death backstop) as fail-open backstops if the adapter's re-emit is
|
|
674
|
+
// itself ever lost. A later genuine completion (pending evidence cleared) is NOT touched by
|
|
675
|
+
// this gate and is processed normally.
|
|
676
|
+
const liveInstance = components.instanceManager?.getInstance?.(eventSessionId) as
|
|
677
|
+
{ hasLiveTurnPendingEvidence?: () => boolean } | undefined;
|
|
678
|
+
if (typeof liveInstance?.hasLiveTurnPendingEvidence === 'function') {
|
|
679
|
+
let midTurnPending = false;
|
|
680
|
+
try { midTurnPending = liveInstance.hasLiveTurnPendingEvidence() === true; } catch { /* fail open — never block on a diagnostic throw */ }
|
|
681
|
+
if (midTurnPending) {
|
|
682
|
+
LOG.info('MeshEvents', `Suppressed agent:generating_completed for session ${eventSessionId} (mesh ${args.meshId}): live adapter re-check shows the turn is still pending (mid-tool / streaming / modal) — treating as a premature/redraw-race emit; the adapter's own finalization retry or the reconcile loop's transcript evidence will surface the real completion`);
|
|
683
|
+
traceMeshEventDrop('mid_turn_live_state_pending', traceCtx);
|
|
684
|
+
return {
|
|
685
|
+
kind: 'suppress',
|
|
686
|
+
result: { success: true, forwarded: 0, suppressed: true, midTurnLiveStatePending: true },
|
|
687
|
+
};
|
|
688
|
+
}
|
|
689
|
+
}
|
|
690
|
+
// NO-DISPATCH-NATIVE-COMPLETION-GATE (rc.16 follow-up): a session that was launched
|
|
691
|
+
// but never given ANY task can still emit its own native agent:generating_completed off
|
|
692
|
+
// a startup/greeting artifact (e.g. cursor-cli's "→ Plan, search, build anything" idle
|
|
693
|
+
// prompt right after boot) — the WARMUPGAP note on markSessionTerminal below already
|
|
694
|
+
// recognizes this shape (no echoed taskId, no active assignment) but only skips the
|
|
695
|
+
// dispatch-row side effect; the event itself still becomes a coordinator-visible
|
|
696
|
+
// task_completed. Suppress it outright here, narrowly: the event carries no taskId, the
|
|
697
|
+
// session holds no active assignment (queue OR direct-dispatch), AND the session has NO
|
|
698
|
+
// terminal ledger history at all (a session with a PRIOR terminal has been dispatched at
|
|
699
|
+
// least once before and falls through to the existing terminal/dedup logic below
|
|
700
|
+
// unchanged). Structural/causal only — no message-content inspection. A completion that
|
|
701
|
+
// is NOT weak (a real final summary / worker result, not a bare false-idle status flag)
|
|
702
|
+
// still passes through: a session that was somehow handed a genuine answer without a
|
|
703
|
+
// tracked dispatch must not have that answer silently dropped.
|
|
704
|
+
if (!readNonEmptyString(args.metadataEvent.taskId)
|
|
705
|
+
&& !sessionHasActiveAssignment(args.meshId, eventSessionId)
|
|
706
|
+
&& !findRecentTerminalLedgerEvidence({ meshId: args.meshId, sessionId: eventSessionId, nodeId: eventNodeId || undefined })
|
|
707
|
+
&& isWeakCompletionEvidence(args.metadataEvent)) {
|
|
708
|
+
LOG.info('MeshEvents', `Suppressed agent:generating_completed for session ${eventSessionId} (mesh ${args.meshId}): no taskId, no active assignment, and no prior terminal ledger evidence — a pre-dispatch startup/greeting artifact, not a real task completion`);
|
|
709
|
+
traceMeshEventDrop('no_dispatch_native_completion', traceCtx);
|
|
710
|
+
return {
|
|
711
|
+
kind: 'suppress',
|
|
712
|
+
result: { success: true, forwarded: 0, suppressed: true, noDispatchNativeCompletion: true },
|
|
713
|
+
};
|
|
714
|
+
}
|
|
715
|
+
// CAUSAL-COMPLETION-GATE (Fix A): fail closed ONLY for the scoped auto-launch race —
|
|
716
|
+
// an in-window, still-'pending' task whose autoLaunch record names this exact session,
|
|
717
|
+
// with neither a consumed delivery nor a matching task_dispatched ledger entry. Every
|
|
718
|
+
// other completion (already-claimed tasks, direct dispatches, expired auto-launch
|
|
719
|
+
// windows) falls through unchanged to the existing dedup/terminal-ledger logic below.
|
|
720
|
+
const inWindowTask = findInWindowUnclaimedAutoLaunchTask(args.meshId, eventSessionId, Date.now());
|
|
721
|
+
if (inWindowTask) {
|
|
722
|
+
const causalEvidence = MeshRuntimeStore.getInstance().taskDeliveryConsumed(args.meshId, inWindowTask.id)
|
|
723
|
+
|| hasMatchingTaskDispatchedLedgerEntry(args.meshId, inWindowTask.id, eventSessionId);
|
|
724
|
+
if (!causalEvidence) {
|
|
725
|
+
LOG.warn('MeshEvents', `Suppressed premature agent:generating_completed for auto-launch session ${eventSessionId} (mesh ${args.meshId}): task ${inWindowTask.id} delivery not yet consumed and no turn-start evidence — likely a boot artifact before the task was ever dispatched`);
|
|
726
|
+
traceMeshEventDrop('autolaunch_completion_before_causal_evidence', traceCtx, `taskId=${inWindowTask.id}`);
|
|
727
|
+
return {
|
|
728
|
+
kind: 'suppress',
|
|
729
|
+
result: { success: true, forwarded: 0, suppressed: true, autoLaunchCausalGateFailed: true, taskId: inWindowTask.id },
|
|
730
|
+
};
|
|
731
|
+
}
|
|
732
|
+
}
|
|
606
733
|
const terminal = findRecentTerminalLedgerEvidence({
|
|
607
734
|
meshId: args.meshId,
|
|
608
735
|
sessionId: eventSessionId,
|
|
@@ -1056,6 +1183,7 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
|
|
|
1056
1183
|
eventNodeId,
|
|
1057
1184
|
eventTimestamp,
|
|
1058
1185
|
workerCoordinatorDaemonId,
|
|
1186
|
+
components,
|
|
1059
1187
|
});
|
|
1060
1188
|
if (suppression) {
|
|
1061
1189
|
if (suppression.kind === 'reconcile') {
|
|
@@ -5,6 +5,7 @@ import { markSessionDeliveriesTerminal } from './mesh-delivery-policy.js';
|
|
|
5
5
|
import { queuePendingMeshCoordinatorEvent } from './mesh-events-pending.js';
|
|
6
6
|
import { readNonEmptyString, readRecord, resolveEventSessionId, readWorkerResultMetadata, isWeakCompletionEvidence, buildMeshSystemMessage } from './mesh-events-utils.js';
|
|
7
7
|
import { recordDebugTrace } from '../logging/debug-trace.js';
|
|
8
|
+
import { LOG } from '../logging/logger.js';
|
|
8
9
|
import { meshNodeIdMatches, sessionIdsEquivalent, type MeshNodeIdentified } from '@adhdev/mesh-shared';
|
|
9
10
|
|
|
10
11
|
// EARLYNOTIFY-GATEBYPASS (d): every completed-emit producer that bypasses the CLI-provider
|
|
@@ -40,6 +41,88 @@ function recordSynthCompletionGateTrace(stage: string, payload: Record<string, u
|
|
|
40
41
|
const DIRECT_DISPATCH_RECONCILE_GRACE_MS = 60_000;
|
|
41
42
|
const DIRECT_DISPATCH_IDLE_SESSION_RECONCILE_GRACE_MS = 120_000;
|
|
42
43
|
|
|
44
|
+
// ---------------------------------------------------------------------------
|
|
45
|
+
// MID-TURN-CAUSAL-ADMISSION (rc.16 unified completion ingress)
|
|
46
|
+
// ---------------------------------------------------------------------------
|
|
47
|
+
// Verified incidents: the coordinator received a WEAK synthesized completion mid-turn —
|
|
48
|
+
// after an interim narration bubble ("Let me find…") — while the worker's raw PTY still
|
|
49
|
+
// showed `Working (43s)` and further tool calls continued. Every transcript-synth ingress
|
|
50
|
+
// (reconcile-loop PHASE 4, the assigned-stranded watchdog's transcript propagation, the MCP
|
|
51
|
+
// mesh_status poll) funnels through reconcileDirectDispatchCompletionFromTranscript, which
|
|
52
|
+
// queues the coordinator completion DIRECTLY — bypassing the injectMeshSystemMessage →
|
|
53
|
+
// evaluateMeshEventSuppression → hasLiveTurnPendingEvidence gate (3a48f660) that protects
|
|
54
|
+
// the native provider-event path. evaluateTranscriptSynthAdmission is the ONE reusable
|
|
55
|
+
// causal admission point closing that bypass: every synth caller passes its causal evidence
|
|
56
|
+
// through the `causalAdmission` arg and the choke point enforces it uniformly.
|
|
57
|
+
export interface TranscriptSynthCausalAdmission {
|
|
58
|
+
/**
|
|
59
|
+
* Synchronous re-check of a resolvable LOCAL live instance's CURRENT turn state — the
|
|
60
|
+
* same hasLiveTurnPendingEvidence() discriminator the 3a48f660 mid-turn gate uses on the
|
|
61
|
+
* native-event path. A pending verdict VETOES an eager transcript synth. Fail-open by
|
|
62
|
+
* construction: absent (remote/unknown session, separate-process caller), not a function,
|
|
63
|
+
* or throwing → no veto; the caller's bounded transcript evidence/settle behavior remains
|
|
64
|
+
* the operative net, so a missing transcript source never wedges.
|
|
65
|
+
*/
|
|
66
|
+
liveTurnPendingEvidence?: () => boolean;
|
|
67
|
+
/**
|
|
68
|
+
* The caller is a BOUNDED last-resort backstop firing at its max-wait (the acked-hold
|
|
69
|
+
* death deadline, the delivered-no-turn redrive deadline). A bounded backstop overrides
|
|
70
|
+
* the live-pending veto — preserving the genuine-final fail-open / max-wait semantics —
|
|
71
|
+
* but NEVER overrides the trailing-tool veto (a stale weak interim summary must not
|
|
72
|
+
* become final merely because a timeout fired while newer transcript/tool activity
|
|
73
|
+
* exists).
|
|
74
|
+
*/
|
|
75
|
+
boundedBackstop?: boolean;
|
|
76
|
+
/**
|
|
77
|
+
* The caller's transcript read shows tool/terminal activity AFTER the latest final-looking
|
|
78
|
+
* assistant bubble (hasTrailingToolActivityAfterFinalAssistant) — the bubble is interim
|
|
79
|
+
* narration ("Let me explore…"), the turn is still executing. Absolute veto: the worker is
|
|
80
|
+
* demonstrably alive and mid-turn, so even a bounded backstop keeps holding (the hold is
|
|
81
|
+
* still released by the session-death read-failure backstop and the stranded-reclaim /
|
|
82
|
+
* orphan-prune nets, so this cannot wedge forever).
|
|
83
|
+
*/
|
|
84
|
+
trailingToolActivityAfterFinalAssistant?: boolean;
|
|
85
|
+
}
|
|
86
|
+
|
|
87
|
+
export type TranscriptSynthAdmissionVerdict =
|
|
88
|
+
| { admitted: true }
|
|
89
|
+
| { admitted: false; reason: 'live_turn_pending_evidence' | 'trailing_tool_activity_after_final_assistant' };
|
|
90
|
+
|
|
91
|
+
export function evaluateTranscriptSynthAdmission(admission?: TranscriptSynthCausalAdmission): TranscriptSynthAdmissionVerdict {
|
|
92
|
+
if (!admission) return { admitted: true };
|
|
93
|
+
if (admission.trailingToolActivityAfterFinalAssistant === true) {
|
|
94
|
+
return { admitted: false, reason: 'trailing_tool_activity_after_final_assistant' };
|
|
95
|
+
}
|
|
96
|
+
if (!admission.boundedBackstop && typeof admission.liveTurnPendingEvidence === 'function') {
|
|
97
|
+
let pending = false;
|
|
98
|
+
try { pending = admission.liveTurnPendingEvidence() === true; } catch { /* fail open — never block on a diagnostic throw */ }
|
|
99
|
+
if (pending) return { admitted: false, reason: 'live_turn_pending_evidence' };
|
|
100
|
+
}
|
|
101
|
+
return { admitted: true };
|
|
102
|
+
}
|
|
103
|
+
|
|
104
|
+
/**
|
|
105
|
+
* Resolve the liveTurnPendingEvidence probe for `sessionId` from a components-shaped object
|
|
106
|
+
* (anything exposing instanceManager.getInstance). Returns undefined when no live instance
|
|
107
|
+
* resolves or the instance does not expose hasLiveTurnPendingEvidence — the fail-open
|
|
108
|
+
* remote/unknown case. Shared by every transcript-synth caller so the probe is resolved ONE
|
|
109
|
+
* way (UNIFY: same discriminator as the 3a48f660 native-event gate).
|
|
110
|
+
*/
|
|
111
|
+
export function resolveLiveTurnPendingEvidence(
|
|
112
|
+
components: { instanceManager?: { getInstance?: (sessionId: string) => unknown } } | undefined,
|
|
113
|
+
sessionId: string,
|
|
114
|
+
): (() => boolean) | undefined {
|
|
115
|
+
try {
|
|
116
|
+
const instance = components?.instanceManager?.getInstance?.(sessionId) as
|
|
117
|
+
{ hasLiveTurnPendingEvidence?: () => boolean } | undefined;
|
|
118
|
+
if (typeof instance?.hasLiveTurnPendingEvidence !== 'function') return undefined;
|
|
119
|
+
const probe = instance.hasLiveTurnPendingEvidence.bind(instance);
|
|
120
|
+
return () => probe();
|
|
121
|
+
} catch {
|
|
122
|
+
return undefined; // resolution itself failed → fail open
|
|
123
|
+
}
|
|
124
|
+
}
|
|
125
|
+
|
|
43
126
|
export function findRecentTerminalLedgerEvidence(args: {
|
|
44
127
|
meshId: string;
|
|
45
128
|
sessionId?: string;
|
|
@@ -229,6 +312,17 @@ export function reconcileDirectDispatchCompletionFromTranscript(args: {
|
|
|
229
312
|
* Direct-dispatch callers omit it → the gates apply exactly as before (no regression).
|
|
230
313
|
*/
|
|
231
314
|
preValidatedTranscriptEvidence?: boolean;
|
|
315
|
+
/**
|
|
316
|
+
* MID-TURN-CAUSAL-ADMISSION (rc.16): causal admission evidence for the unified guard —
|
|
317
|
+
* see TranscriptSynthCausalAdmission. Every transcript-synth caller (PHASE 4 reconcile,
|
|
318
|
+
* watchdog transcript propagation, MCP mesh_status poll) passes what it knows; the choke
|
|
319
|
+
* point evaluates it BEFORE writing the terminal ledger or queuing the coordinator
|
|
320
|
+
* completion. A veto returns { reconciled: false, reason } — callers treat that exactly
|
|
321
|
+
* like their other not-yet-proven outcomes (hold and re-evaluate next tick), so the held
|
|
322
|
+
* completion re-evaluates the causal evidence every pass and releases exactly once after
|
|
323
|
+
* the state clears (the terminal-ledger dedup above makes the release idempotent).
|
|
324
|
+
*/
|
|
325
|
+
causalAdmission?: TranscriptSynthCausalAdmission;
|
|
232
326
|
source?: string;
|
|
233
327
|
}): { reconciled: boolean; kind?: MeshLedgerKind; alreadyTerminal?: boolean; workerResult?: unknown; ledgerEntryId?: string; reason?: string } {
|
|
234
328
|
const finalSummary = readNonEmptyString(args.finalSummary);
|
|
@@ -251,6 +345,21 @@ export function reconcileDirectDispatchCompletionFromTranscript(args: {
|
|
|
251
345
|
return { reconciled: false, alreadyTerminal: true, reason: 'terminal_ledger_entry_exists' };
|
|
252
346
|
}
|
|
253
347
|
|
|
348
|
+
// MID-TURN-CAUSAL-ADMISSION (rc.16): the unified guard every gate-bypassing synth must
|
|
349
|
+
// pass. Ordered BEFORE any ledger write / queue so a vetoed synth leaves no trace other
|
|
350
|
+
// than the completion-gate trace (observability parity with the synth-fire trace below).
|
|
351
|
+
const admission = evaluateTranscriptSynthAdmission(args.causalAdmission);
|
|
352
|
+
if (!admission.admitted) {
|
|
353
|
+
LOG.info('MeshEvents', `Transcript synth vetoed for task ${args.taskId} session ${args.sessionId} (source ${args.source || 'direct_task_transcript_reconciliation'}): ${admission.reason} — holding the completion; it re-evaluates on the next reconcile pass`);
|
|
354
|
+
recordSynthCompletionGateTrace('synth-veto', {
|
|
355
|
+
producer: 'transcript_reconcile',
|
|
356
|
+
source: args.source || 'direct_task_transcript_reconciliation',
|
|
357
|
+
taskId: args.taskId,
|
|
358
|
+
reason: admission.reason,
|
|
359
|
+
});
|
|
360
|
+
return { reconciled: false, reason: admission.reason };
|
|
361
|
+
}
|
|
362
|
+
|
|
254
363
|
const nodeId = readNonEmptyString(args.nodeId) || dispatch?.nodeId;
|
|
255
364
|
const providerType = readNonEmptyString(args.providerType) || dispatch?.providerType || readNonEmptyString(dispatch?.payload.providerType);
|
|
256
365
|
const completedAt = args.completedAt || new Date().toISOString();
|
|
@@ -1331,7 +1331,12 @@ function isTerminalSessionStatus(status: string): boolean {
|
|
|
1331
1331
|
return ['stopped', 'failed', 'terminated', 'exited', 'closed'].includes(status);
|
|
1332
1332
|
}
|
|
1333
1333
|
|
|
1334
|
-
|
|
1334
|
+
// Exported (Fix B, rc.15 orchestration RCA) so the explicit mesh_send_task dispatch path
|
|
1335
|
+
// (med-family/cli-agent.ts) can reuse the SAME status/liveness probe to independently confirm a
|
|
1336
|
+
// pinned target session is genuinely ready before overriding the coarse worktreeBootstrap
|
|
1337
|
+
// 'running' defer for that one session. Never returns true for 'starting' / 'waiting_approval' /
|
|
1338
|
+
// 'generating' / any other non-idle status — those still refuse the override.
|
|
1339
|
+
export function isIdleSessionState(state: any): boolean {
|
|
1335
1340
|
const status = readNonEmptyString(state?.status).toLowerCase();
|
|
1336
1341
|
if (isTerminalSessionStatus(status)) return false;
|
|
1337
1342
|
return status === 'idle' || state?.activeChat?.status === 'waiting_input';
|
|
@@ -1387,6 +1392,19 @@ function nodeHasActiveMeshWork(components: DaemonComponents, meshId: string, nod
|
|
|
1387
1392
|
|
|
1388
1393
|
function isLaunchableNode(node: any): boolean {
|
|
1389
1394
|
if (!node || node.status === 'disabled' || node.status === 'removed') return false;
|
|
1395
|
+
// BOOTSTRAP-POLICY-CONSISTENCY (Fix B, rc.15 orchestration RCA): a worktree node whose
|
|
1396
|
+
// bootstrap is still 'running' must be excluded from auto-launch candidacy, not merely
|
|
1397
|
+
// deferred at claim time. Before this, isLaunchableNode passed a bootstrapping node through,
|
|
1398
|
+
// so maybeAutoLaunchOneQueueSession could spawn a session (launch_cli) on it — an ORPHAN,
|
|
1399
|
+
// since tryAssignQueueTask's own bootstrap gate (mirrors this same predicate) then defers the
|
|
1400
|
+
// claim onto that freshly-spawned session anyway, leaving it stranded with nothing assigned
|
|
1401
|
+
// while the node's real bootstrap-driven session comes up separately. Reusing
|
|
1402
|
+
// shouldDeferDispatchForBootstrap (rather than a raw status check) keeps this in agreement
|
|
1403
|
+
// with the claim gate's stale-running backstop: a 'running' stamp old enough and verified
|
|
1404
|
+
// git-clean is treated as silently complete and does not block launch.
|
|
1405
|
+
if (shouldDeferDispatchForBootstrap(node as { worktreeBootstrap?: { status?: string; startedAt?: string; updatedAt?: string; completedAt?: string }; workspace?: string })) {
|
|
1406
|
+
return false;
|
|
1407
|
+
}
|
|
1390
1408
|
// Delegate the health gate to the shared resolver so the auto-launch gate and the
|
|
1391
1409
|
// MAGI fan-out planner agree on exactly what "launchable health" means (online /
|
|
1392
1410
|
// unknown / absent pass; degraded / offline / dirty / wrong_branch are blocked).
|