Skip to content

Fix agent stuck in waiting state after completion #343

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
satti-hari-krishna-reddy wants to merge 3 commits into
base: main
Choose a base branch
from
+44 −10
Open

Fix agent stuck in waiting state after completion #343

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
980b125
add http app
satti-hari-krishna-reddy Mar 12, 2026
d795ee3
fix agent stuck loading due to timeout
satti-hari-krishna-reddy Mar 12, 2026
eddf2c8
added auth injection for other internal apps as well
satti-hari-krishna-reddy Mar 12, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 11 additions & 2 deletions ai.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7464,6 +7464,11 @@ func HandleAiAgentExecutionStart(execution WorkflowExecution, startNode Action,

decidedApps += lowername + ", "
}

// Let's inject http.
if !strings.Contains(decidedApps, "http") {
decidedApps += "http, "
}
}

if len(decidedApps) > 0 {
Expand Down Expand Up @@ -8373,8 +8378,8 @@ You are the Action Execution Agent for the Shuffle platform. You receive tools (
}
}

if agentOutput.Status == "FINISHED" && agentOutput.CompletedAt > 0 && execution.Status == "EXECUTING" {
log.Printf("[INFO][%s] AI Agent action %s finished.", execution.ExecutionId, startNode.ID)
if agentOutput.Status == "FINISHED" && agentOutput.CompletedAt > 0 && execution.Status != "ABORTED" && execution.Status != "FAILURE" {
log.Printf("[INFO][%s] AI Agent action %s finished. Execution status: %s", execution.ExecutionId, startNode.ID, execution.Status)
for resultIndex, result := range execution.Results {
if result.Action.ID != startNode.ID {
continue
Expand Down Expand Up @@ -8424,6 +8429,10 @@ You are the Action Execution Agent for the Shuffle platform. You receive tools (
}
}
}

if createNextActions {
return startNode, nil
}

// 1. Map the response back
newResult, err := json.Marshal(resultMapping)
Expand Down
21 changes: 15 additions & 6 deletions db-connector.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1796,7 +1796,7 @@ func Fixexecution(ctx context.Context, workflowExecution WorkflowExecution) (Wor
finishedDecisions = append(finishedDecisions, decision.RunDetails.Id)
continue
} else if decision.RunDetails.Status == "FAILURE" {
//finishedDecisions = append(finishedDecisions, decision.RunDetails.Id)
finishedDecisions = append(finishedDecisions, decision.RunDetails.Id)
failedFound = true
continue
} else if decision.RunDetails.Status == "RUNNING" && decision.Action != "ask" {
Expand All @@ -1809,8 +1809,11 @@ func Fixexecution(ctx context.Context, workflowExecution WorkflowExecution) (Wor
mappedOutput.Decisions[decisionIndex].RunDetails.Status = "FAILURE"
mappedOutput.Decisions[decisionIndex].RunDetails.CompletedAt = time.Now().Unix()
mappedOutput.Decisions[decisionIndex].RunDetails.RawResponse += "\n[ERROR] Decision marked as FAILURE due to 5 minute timeout."
}

// Count this as finished + failed so recovery triggers in the same Fixexecution run
finishedDecisions = append(finishedDecisions, decision.RunDetails.Id)
failedFound = true
}
} else {
if decision.RunDetails.CompletedAt > 0 {
if debug {
Expand Down Expand Up @@ -1912,7 +1915,7 @@ func Fixexecution(ctx context.Context, workflowExecution WorkflowExecution) (Wor
go sendAgentActionSelfRequest("SUCCESS", workflowExecution, workflowExecution.Results[resultIndex])
}()
} else {
log.Printf("[INFO][%s] All decisions finished for agent action %s - but no finish action found, marking as WAITING.", workflowExecution.ExecutionId, action.ID)
log.Printf("[INFO][%s] All decisions finished for agent action %s - but no finish action found. Re-invoking agent to finalize (failedFound: %t).", workflowExecution.ExecutionId, action.ID, failedFound)

mappedOutput.Status = "RUNNING"
mappedOutput.CompletedAt = 0
Expand All @@ -1922,10 +1925,16 @@ func Fixexecution(ctx context.Context, workflowExecution WorkflowExecution) (Wor
workflowExecution.Status = "EXECUTING"
}

// To ensure the execution is actually updated
// Re-invoke the agent so the LLM can see the failure and produce a proper "finish" decision.

capturedExec := workflowExecution
capturedAction := action
go func() {
time.Sleep(1 * time.Second)
sendAgentActionSelfRequest("WAITING", workflowExecution, workflowExecution.Results[resultIndex])
time.Sleep(2 * time.Second)
_, err := HandleAiAgentExecutionStart(capturedExec, capturedAction, true)
if err != nil {
log.Printf("[ERROR][%s] Failed re-invoking agent after decisions completed for action %s: %s", capturedExec.ExecutionId, capturedAction.ID, err)
}
}()
}
} else if (result.Status == "" || result.Status == "WAITING") && mappedOutput.Status == "FINISHED" {
Expand Down
20 changes: 18 additions & 2 deletions shared.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17101,7 +17101,15 @@ func handleAgentDecisionStreamResult(workflowExecution WorkflowExecution, action
// Handle agent decisionmaking. Use the same
log.Printf("[INFO][%s] With the agent being finished, we are asking it whether it would like to do anything else", workflowExecution.ExecutionId)

returnAction, err := HandleAiAgentExecutionStart(workflowExecution, actionResult.Action, true)
var originalAction Action
if foundActionResultIndex >= 0 && foundActionResultIndex < len(workflowExecution.Results) {
Copy link
Copy Markdown
Member

@frikky frikky Mar 12, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What was the point of this? I'm always confused when we override actions, when things used to work without it

Copy link
Copy Markdown
Collaborator Author

@satti-hari-krishna-reddy satti-hari-krishna-reddy Mar 18, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

At line 8295 in ai, we are building actionResult, and inside of that we are setting the action's label name as fmt.Sprintf("Agent Decision %s", decision.RunDetails.Id). Then we call handleAgentDecisionStreamResult, and that one passes actionResult to HandleAiAgentExecutionStart. Now startNode is overwritten with an action with label, e.g., "Agent Decision RNnBU2aY". Due to this, the later nodes in workflow that are trying to refer to the AI agent node with its original label can't find it, resulting in being unable to get any data from the agent node at all.

originalAction = workflowExecution.Results[foundActionResultIndex].Action
} else {
// Fallback in case of an issue
originalAction = actionResult.Action
}

returnAction, err := HandleAiAgentExecutionStart(workflowExecution, originalAction, true)
if err != nil {
log.Printf("[ERROR][%s] Failed handling agent execution start: %s", workflowExecution.ExecutionId, err)
}
Expand Down Expand Up @@ -21407,8 +21415,16 @@ func PrepareSingleAction(ctx context.Context, user User, appId string, body []by
workflowExecution.OrgId = user.ActiveOrg.Id
}

formattedAppName := strings.ReplaceAll(strings.ToLower(app.Name), " ", "_")

isInternalShuffleApp := false
switch formattedAppName {
Copy link
Copy Markdown
Member

@frikky frikky Mar 12, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Something about just matching name seems off, but I'm not sure exactly why.

I wonder if there is a way to "steal" the URL field/apikey by formatting an app in a certain way.

Just need to think it over before merging 🤔

Copy link
Copy Markdown
Collaborator Author

@satti-hari-krishna-reddy satti-hari-krishna-reddy Mar 18, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yeah, I think having a special verified field for apps built by support users would be better. Given that we're already overriding the URL field when injecting the auth, if we compare additionally to this special field and both are true, we can trust it. What do you think?

case "shuffle_datastore", "shuffle_org_management", "shuffle_app_management", "shuffle_workflow_management":
isInternalShuffleApp = true
}

// Fallback to inject AI creds if the user don't have any
if len(workflowExecution.OrgId) > 0 && strings.ReplaceAll(strings.ToLower(app.Name), " ", "_") == "shuffle_datastore" && len(action.AuthenticationId) == 0 {
if len(workflowExecution.OrgId) > 0 && isInternalShuffleApp && len(action.AuthenticationId) == 0 {
backendUrl := os.Getenv("BASE_URL")
if len(os.Getenv("SHUFFLE_CLOUDRUN_URL")) > 0 && strings.Contains(os.Getenv("SHUFFLE_CLOUDRUN_URL"), "http") {
backendUrl = os.Getenv("SHUFFLE_CLOUDRUN_URL")
Expand Down
Loading