crewly 1.20.106 → 1.20.108

package/dist/cli/backend/src/services/agent/agent-registration.service.js

@@ -9,7 +9,6 @@ import { RuntimeServiceFactory } from './runtime-service.factory.js';
 import { registerInProcessRuntime, unregisterInProcessRuntime, } from './crewly-agent/in-process-runtime-registry.js';
 import { getCrewlyHomePath } from '../core/crewly-home.utils.js';
 import { CREWLY_CONSTANTS, ENV_CONSTANTS, AGENT_TIMEOUTS, ORCHESTRATOR_SESSION_NAME, ORCHESTRATOR_ROLE, RUNTIME_TYPES, SESSION_COMMAND_DELAYS, SESSION_RECREATION_CONSTANTS, EVENT_DELIVERY_CONSTANTS, TERMINAL_PATTERNS, GEMINI_SHELL_MODE_CONSTANTS, GEMINI_STUCK_CONNECTIVITY_PATTERN, GEMINI_ERROR_STATE_CONSTANTS, REGISTRATION_DELIVERY_CONSTANTS, ORC_CONVERSATION_CONSTANTS, } from '../../constants.js';
-import { WEB_CONSTANTS } from '../../../../config/constants.js';
 import { delay } from '../../utils/async.utils.js';
 import { buildRuntimeModelFlags } from '../../utils/runtime-model-flags.utils.js';
 import { stripToolCallMarkup } from '../../utils/tool-call-markup.utils.js';
@@ -21,6 +20,8 @@ import { RuntimeExitMonitorService } from './runtime-exit-monitor.service.js';
 import { ContextWindowMonitorService } from './context-window-monitor.service.js';
 import { OAuthReloginMonitorService } from './oauth-relogin-monitor.service.js';
 import { SubAgentMessageQueue } from '../messaging/sub-agent-message-queue.service.js';
+import { InFlightTurnTracker } from '../restart/in-flight-turn-tracker.service.js';
+import { RestartDrainService } from '../restart/restart-drain.service.js';
 import { AgentSuspendService } from './agent-suspend.service.js';
 import { PromptBuilderService, buildModuleConfigFromTeamMember, memberSkillsJson, } from '../ai/prompt-builder.service.js';
 import { PromptAssemblyService } from '../ai/prompt-modules/prompt-assembly.service.js';
@@ -29,7 +30,9 @@ import { stripAnsiCodes } from '../../utils/terminal-output.utils.js';
 import { isPromptLine, containsSpinnerOrWorkingIndicator, containsProcessingIndicator, containsBusyStatusBar, containsRewindMode, extractChatPrefix, stripTuiLineBorders, matchTuiPromptLine, } from '../../utils/terminal-string-ops.js';
 import { PtyActivityTrackerService } from './pty-activity-tracker.service.js';
 import { synthesizeSlackConversationId } from '../chat-v2/legacy-dto.utils.js';
+import { isRuntimeStartupBlockedError } from './runtime-startup-blocked.error.js';
 import { buildHandoverSummary, claudeTranscriptPath, conversationExists, lastTurnContextTokens, orcFreshContextTokens, planRuntimeSessionFlags, waitForCodexSessionId, } from './runtime-session-recovery.js';
+import { getLocalApiBaseUrl } from '../../utils/local-api-url.utils.js';
 /**
  * Whether a file exists (readable).
  *
@@ -146,9 +149,13 @@ export class AgentRegistrationService {
             this.cancelPendingRegistration(sessionName);
             this.unregisterTuiSession(sessionName);
             // Preserve queued messages for suspended agents (they'll be delivered on rehydrate)
-            if (!AgentSuspendService.getInstance().isSuspended(sessionName)) {
+            // and during shutdown: runtimes exit because the backend is stopping,
+            // and messages held back by the restart drain must survive it.
+            if (!AgentSuspendService.getInstance().isSuspended(sessionName) &&
+                !RestartDrainService.getInstance().isDeliveryPaused()) {
                 SubAgentMessageQueue.getInstance().clear(sessionName);
             }
+            InFlightTurnTracker.getInstance().markTurnComplete(sessionName, 'runtime exited');
         });
     }
     /**
@@ -1029,6 +1036,17 @@ export class AgentRegistrationService {
             }
         }
         catch (error) {
+            // Start-up blocked on the user (e.g. Claude Code as root, or never
+            // set up): full recreation would hit the same wall, so stop here and
+            // surface the actionable message instead of a generic timeout.
+            if (isRuntimeStartupBlockedError(error)) {
+                this.logger.error('Agent start-up blocked; not retrying', {
+                    sessionName,
+                    reason: error.reason,
+                    error: error.message,
+                });
+                return { success: false, error: error.message, errorCode: error.code };
+            }
             this.logger.warn('Step 1 (cleanup + reinit) failed', {
                 sessionName,
                 error: error instanceof Error ? error.message : String(error),
@@ -1048,6 +1066,14 @@ export class AgentRegistrationService {
                 }
             }
             catch (error) {
+                if (isRuntimeStartupBlockedError(error)) {
+                    this.logger.error('Agent start-up blocked; not retrying', {
+                        sessionName,
+                        reason: error.reason,
+                        error: error.message,
+                    });
+                    return { success: false, error: error.message, errorCode: error.code };
+                }
                 this.logger.warn('Step 2 (full recreation) failed', {
                     sessionName,
                     error: error instanceof Error ? error.message : String(error),
@@ -2601,7 +2627,8 @@ Loop until done, blocked, or explicitly reassigned:
         return {
             [ENV_CONSTANTS.CREWLY_SESSION_NAME]: sessionName,
             [ENV_CONSTANTS.CREWLY_ROLE]: role,
-            [ENV_CONSTANTS.CREWLY_API_URL]: `http://localhost:${WEB_CONSTANTS.PORTS.BACKEND}`,
+            // The port this instance actually runs on, not the default (#777).
+            [ENV_CONSTANTS.CREWLY_API_URL]: getLocalApiBaseUrl(),
             [ENV_CONSTANTS.CREWLY_PROJECT_PATH]: cwd,
             [ENV_CONSTANTS.CREWLY_INSTALL_DIR]: this.projectRoot,
         };
@@ -3100,7 +3127,7 @@ Loop until done, blocked, or explicitly reassigned:
             // Set environment variables for MCP connection
             await sessionHelper.setEnvironmentVariable(sessionName, ENV_CONSTANTS.CREWLY_SESSION_NAME, sessionName);
             await sessionHelper.setEnvironmentVariable(sessionName, ENV_CONSTANTS.CREWLY_ROLE, role);
-            await sessionHelper.setEnvironmentVariable(sessionName, ENV_CONSTANTS.CREWLY_API_URL, `http://localhost:${WEB_CONSTANTS.PORTS.BACKEND}`);
+            await sessionHelper.setEnvironmentVariable(sessionName, ENV_CONSTANTS.CREWLY_API_URL, getLocalApiBaseUrl());
             // #187: Set project path so memory skills can auto-inject it
             await sessionHelper.setEnvironmentVariable(sessionName, ENV_CONSTANTS.CREWLY_PROJECT_PATH, cwdToUse);
             // #222: Set install dir so agents can resolve skill paths regardless of CWD
@@ -3148,6 +3175,7 @@ Loop until done, blocked, or explicitly reassigned:
                     success: false,
                     sessionName,
                     error: initResult.error || 'Failed to initialize and register agent',
+                    ...(initResult.errorCode && { errorCode: initResult.errorCode }),
                 };
             }
             // Start context window monitoring for newly created non-orchestrator session
@@ -3325,6 +3353,22 @@ Loop until done, blocked, or explicitly reassigned:
                     error: 'Message is required and must be a string',
                 };
             }
+            // Safe restart: once shutdown has begun nothing new is written into an
+            // agent. Writing now would start a turn the drain then has to wait for
+            // (or cut off). The persistent queue carries it across the restart and
+            // register-self flushes it on the next boot.
+            if (RestartDrainService.getInstance().isDeliveryPaused()) {
+                SubAgentMessageQueue.getInstance().enqueue(sessionName, message);
+                this.logger.info('Shutdown in progress — message queued for delivery after restart', {
+                    sessionName,
+                    messageLength: message.length,
+                });
+                return {
+                    success: true,
+                    queued: true,
+                    message: '[RESTART_DRAIN] Message queued for delivery after the restart',
+                };
+            }
             // Auto-resolve runtime type if not provided.
             // CRITICAL: defaulting to CLAUDE_CODE is dangerous because
             // Ctrl+C cleanup (Claude Code behavior) triggers /quit on Gemini CLI.
@@ -3362,6 +3406,9 @@ Loop until done, blocked, or explicitly reassigned:
                     if (slackPrefixMatch[2])
                         slackMetadata.threadTs = slackPrefixMatch[2];
                 }
+                // In-process turns end exactly when handleMessage settles; the
+                // restart drain waits on this record until then.
+                const inFlight = InFlightTurnTracker.getInstance().recordDelivery(sessionName, message, 'in-process');
                 // Process message asynchronously — don't block the caller
                 crewlyRuntime.handleMessage(message, slackMetadata)
                     .then((result) => {
@@ -3475,6 +3522,9 @@ Loop until done, blocked, or explicitly reassigned:
                             error: statusError instanceof Error ? statusError.message : String(statusError),
                         });
                     }
+                })
+                    .finally(() => {
+                    InFlightTurnTracker.getInstance().completeMessage(sessionName, inFlight);
                 });
                 this.logger.info('Message dispatched to in-process Crewly Agent', {
                     sessionName, messageLength: message.length,
@@ -3531,7 +3581,14 @@ Loop until done, blocked, or explicitly reassigned:
             // issues (Tab+Enter recovery). Previously set to 1 which caused
             // Google Chat messages to fail permanently on first TUI hiccup.
             const maxDeliveryAttempts = runtimeType === RUNTIME_TYPES.GEMINI_CLI ? 2 : 3;
+            // Settle before writing: a resting agent's earlier deliveries belong to
+            // a finished turn. After the write the echo makes it look busy.
+            const turnTracker = InFlightTurnTracker.getInstance();
+            turnTracker.settle(sessionName);
             const delivered = await this.sendMessageWithRetry(sessionName, message, maxDeliveryAttempts, runtimeType);
+            if (delivered) {
+                turnTracker.recordDelivery(sessionName, message, 'pty');
+            }
             if (!delivered) {
                 // Check if the agent is actively processing (busy) — queue for later delivery.
                 // Use PTY idle time instead of regex patterns for robust cross-runtime detection.