crewly 1.20.250 → 1.20.251

package/dist/backend/backend/src/controllers/team/team.controller.js

@@ -3,7 +3,7 @@ import { isPersonId } from '../../services/people/people-directory.service.js';
 import { canMemberDelegate, getTeamLeadIds, isLeadRole, setTeamLead } from '../../utils/team.utils.js';
 import { v4 as uuidv4 } from 'uuid';
 import { ORCHESTRATOR_SESSION_NAME, ORCHESTRATOR_ROLE, RUNTIME_TYPES, NON_RECOVERABLE_ERROR_PATTERNS, CLAUDE_STARTUP_CONSTANTS, RUNTIME_STARTUP_CONSTANTS, } from '../../constants.js';
-import { CREWLY_CONSTANTS, AGENT_TIMEOUTS, AGENT_WAKE_ERROR_CODES, TEAM_LEAD_CONSTANTS, TEAM_PAUSE_CONSTANTS } from '../../constants.js';
+import { CREWLY_CONSTANTS, AGENT_TIMEOUTS, AGENT_WAKE_ERROR_CODES, TEAM_LEAD_CONSTANTS, TEAM_PAUSE_CONSTANTS, STALLED_AGENT_RECOVERY_CONSTANTS } from '../../constants.js';
 import { updateAgentHeartbeat } from '../../services/agent/agent-heartbeat.service.js';
 import { getSessionBackendSync, getSessionStatePersistence } from '../../services/session/index.js';
 import { removeCrewlyAgentFile } from '../../services/session/session-binding.js';
@@ -16,12 +16,13 @@ import { seedDeclaredCapabilities } from '../../services/memory/task-history-see
 import { validateAgentRole, InvalidAgentRoleError, } from '../../services/agent/agent-registration.service.js';
 import { SubAgentMessageQueue } from '../../services/messaging/sub-agent-message-queue.service.js';
 import { SUB_AGENT_QUEUE_CONSTANTS } from '../../constants.js';
+import { RESOURCE_MODE_CONSTANTS } from '../../services/agent/resource-mode.service.js';
 import { LoggerService } from '../../services/core/logger.service.js';
 import { getChatV2Service } from '../../services/chat-v2/chat-v2.singleton.js';
 import { OAuthReloginMonitorService } from '../../services/agent/oauth-relogin-monitor.service.js';
 import { normalizeMemberSkills } from '../../services/ai/prompt-builder.service.js';
 import { isSafeModelId, isSafeReasoningEffort } from '../../utils/runtime-model-flags.utils.js';
-import { evaluateColdLaunch, isDormantTeam, isPreAuthorizedSchedule, COMMITMENT_APPROVAL_LOOKBACK_MS, } from '../../services/orchestrator/commitment-approval-guard.js';
+import { evaluateColdLaunch, isDormantTeam, isHarnessParkedResume, isPreAuthorizedSchedule, COMMITMENT_APPROVAL_LOOKBACK_MS, } from '../../services/orchestrator/commitment-approval-guard.js';
 import { isOwnerDashboardRequest, readAgentSessionHeader } from '../../utils/agent-caller.utils.js';
 import { getCallerIdentity, isOwnerCaller, rejectNonOwner, sendOwnerAuthRequired } from '../../middleware/caller-identity.middleware.js';
 import { effectiveRuntimeType } from '../../services/runtime-fallback/effective-runtime.js';
@@ -692,6 +693,9 @@ async function _startTeamMemberCore(context, team, member, projectPath, options
             // Don't retry non-recoverable errors (e.g. missing CLI binary, or a
             // start-up blocked on the user such as Claude Code as root / never set up)
             if (createResult.errorCode === CLAUDE_STARTUP_CONSTANTS.BLOCKED_ERROR_CODE ||
+                // Waited for a slot behind the running-agent cap: retrying only waits
+                // again (CREW-304); the caller treats it as deferred.
+                createResult.errorCode === RESOURCE_MODE_CONSTANTS.START_DEFERRED_ERROR_CODE ||
                 (lastError && NON_RECOVERABLE_ERROR_PATTERNS.some(p => lastError.includes(p)))) {
                 logger.error('Non-recoverable error detected, skipping retries', { sessionName, lastError });
                 break;
@@ -1976,7 +1980,39 @@ export async function startTeamMember(req, res) {
                 workItemId: wakeWorkItemId,
             });
         }
-        if (!memberAlreadyActive && !ownerDashboardCall && isDormantTeam(team) && !scheduledPreAuthorized) {
+        // A team the harness parked (idle stop, slot freed, update) is resumed by
+        // the reconciler's wake for its own queued work, not cold-launched
+        // (CREW-303/304: CE sat 4h35m behind this gate on 2026-10-06).
+        let harnessParkedResume = false;
+        if (!memberAlreadyActive && !ownerDashboardCall && !scheduledPreAuthorized && wakeWorkItemId && isDormantTeam(team)) {
+            try {
+                const { TaskPoolService } = await import('../../services/task-pool/task-pool.service.js');
+                const wakeItem = await TaskPoolService.getInstance().findWorkItem(wakeWorkItemId);
+                harnessParkedResume = isHarnessParkedResume({
+                    team,
+                    sessionName: member.sessionName,
+                    workItem: wakeItem,
+                    now: Date.now(),
+                    parkReasons: STALLED_AGENT_RECOVERY_CONSTANTS.HARNESS_PARK_REASONS,
+                    windowMs: STALLED_AGENT_RECOVERY_CONSTANTS.PARKED_RESUME_WINDOW_MS,
+                });
+            }
+            catch (err) {
+                logger.warn('Commitment gate: could not read the wake WorkItem — treating as a cold launch', {
+                    workItemId: wakeWorkItemId,
+                    error: err instanceof Error ? err.message : String(err),
+                });
+            }
+            if (harnessParkedResume) {
+                logger.info('Commitment gate: resuming a team the harness parked (its own queued work)', {
+                    teamId,
+                    memberId,
+                    sessionName: member.sessionName,
+                    workItemId: wakeWorkItemId,
+                });
+            }
+        }
+        if (!memberAlreadyActive && !ownerDashboardCall && isDormantTeam(team) && !scheduledPreAuthorized && !harnessParkedResume) {
             let ownerMessages = [];
             let readOk = true;
             try {
@@ -2100,7 +2136,10 @@ export async function startTeamMember(req, res) {
             else {
                 res.status(500).json({
                     success: false,
-                    error: result.error || 'Failed to start team member'
+                    error: result.error || 'Failed to start team member',
+                    // Lets the reconciler tell a deferred start (RESOURCE_PRESSURE_CAP)
+                    // from a broken one (CREW-304).
+                    ...(result.errorCode ? { code: result.errorCode } : {}),
                 });
             }
         }