Skip to main content

Execution errors

When a run fails, the timeline shows the step that stopped and the reason. This page translates the reasons.

The Retries? column says whether the platform tries again automatically, when the step has retries configured.

Identity and permission​

ReasonMeaningRetries?What to do
Service account not configuredThe organization has no service accountNoSee Service account
Service account disabledIt exists and is inactiveNoReactivate it
Divergent identityThe run was created under another identityNoTrigger it again
Authorization expiredThe authorization lapsed mid-workYesUsually resolves on its own
Authorization revoked or tampered withThe authorization is no longer validNoTrigger it again; if it persists, contact support
Capability outside the authorizationThe step asked for something the version did not declareNoDeclare the resource and publish again
Skill not permittedThe organization's policy does not grant itNoSee Allowed resources

AI steps​

ReasonMeaningRetries?What to do
Engine unavailableImaginne's engine did not answerYesUsually resolves on its own
Policy refusedThe engine refused the policy contextNoReview skills and resources
Required tool unavailableA required tool does not exist or did not completeNoSee Allowed resources
Tool failedA tool ended in failure, with no declared toleranceNoInvestigate the tool, or declare tolerance
Output does not match the declared formatA required field did not come backNoReduce the fields or make the objective more explicit
The AI asked a person for informationIn an automatic run there is nobody to answerNoComplete the input data
Work limit reachedIterations or answer size at the ceilingNoRaise the limit or reduce the scope
Partial resultWhat came back is not the whole answerNoReduce the task's scope
Stopped with no resultNo conclusion was reachedNoReview the objective
Model errorA model failureYesUsually resolves on its own
Context exhaustedThe content exceeded what fits in the contextYesReduce the data handed over
Unknown stop reasonThe engine returned something the platform does not recognizeNoContact support

That last row is a deliberate rule: a newer engine must not make an agent declare success out of ignorance.

External calls​

ReasonMeaningRetries?What to do
Invalid action inputThe parameters do not match what the action expectsNoFix the step's configuration
Action blockedPolicy or network protectionNoSee Calling an API
Step timeoutThe step exceeded its deadlineYesRaise the deadline or investigate the other side

Flow structure​

ReasonMeaningRetries?What to do
Unreadable definitionThe run's frozen content is invalidNoPublish again
Stuck flowNothing is ready and no Result was reachedNoReview the connections
Failure deciding a pathAn error evaluating a connection's conditionNoReview the condition
Failure assembling a valueAn expression could not be evaluatedNoReview the values used
Key conflict while combiningTwo sources would produce the same keyNoName the sources

Approvals and waiting​

ReasonMeaningWhat to do
Approval rejectedSomeone rejected itNothing — it is a legitimate outcome
Approval expiredThe deadline ran out with no decisionRaise the deadline or notify whoever decides

Agent chains​

ReasonMeaningWhat to do
Depth exceededThe chain is longer than the ceilingShorten the chain
Recursion detectedAn agent in the chain calls someone already in itRedesign the flow
Called agent not foundNo active version in that environmentActivate the agent in the same environment
Child run failedThe problem is in the called agentOpen the child run

Time​

ReasonMeaningWhat to do
Run expiredThe total deadline ran outRaise the timeout or reduce the work

What the messages never contain​

Every message goes through a cleanup before being stored: credentials, tokens, and database addresses are stripped. What remains is enough to diagnose, and nothing that should not circulate.

Next steps​