Skip to content

Troubleshoot by symptom

SymptomCheck and next step
Wrong command versionExecutable resolution, Node version, development launcher versus release
init rejects the directoryUse a new/empty target; inspect partial setup before retry
Service unavailablestatus, service logs, verified process identity; plan recover for stale resources
Installed Workflow missing from flowsLoader phase/error, protocol, generated exports and assets
CMD/TUI cannot executedoctor cli, llms.default, startup environment; restart after config change
MCP has no connectionExisting Service, generated paths, same trusted project, client diagnostics
MCP stops at Agent waitClient claim/Tool/completion protocol and current pendingAgentCall
Answer rejectedCurrent action ID, exact question IDs, complete array, permitted skips
Model/Tool failureRun lastError and execution logs; provider/output/Tool Schema
Artifact differs from old RunActual commit revision; current value is not historical
mirror restore failsPrepared stopped target, compatible topology/Schema, image kind and scope
Capture failure pauses RunResolve cause, explicitly resume capture; workflow-only abandonment when chosen

Diagnose before repeating effects

First identify project, installed version, Service instance and original Run. Read status, doctor, flows, attach and logs. Determine whether the failure is local setup, transport, admission, runtime, business validation or output export.

An unknown command result can leave work active. Query the original operation/Run instead of creating another flow. Preserve committed files/data and inspect them before deciding how to retry.

See error reference, diagnostics and reporting.