conductor-oss/conductor · warning · IllegalStateException

No pending HUMAN task found in execution {executionId}

Error message

No pending HUMAN task found in execution {executionId}

What it means

Thrown by AgentService.respond when no HUMAN-type task in IN_PROGRESS status is found in the execution. respond() is the human-in-the-loop (HITL) response endpoint — it searches the workflow's task list for a pending HUMAN task to complete with the human's output. If none exists, the execution is not currently waiting for human input. IllegalStateException maps to HTTP 409/422 depending on the controller mapping — it indicates a state precondition failure.

Source

Thrown at agentspan/src/main/java/org/conductoross/conductor/ai/agentspan/runtime/service/AgentService.java:1257

        return streamRegistry.register(executionId, lastEventId);
    }

    /** Respond to a pending HITL task in an agent execution. */
    public void respond(String executionId, Map<String, Object> output) {
        log.info("Responding to execution {}: {}", executionId, output);

        // Find the pending task (HUMAN type, IN_PROGRESS status)
        Workflow workflow = workflowService.getExecutionStatus(executionId, true);
        Task pendingTask = null;
        for (Task task : workflow.getTasks()) {
            if ("HUMAN".equals(task.getTaskType()) && task.getStatus() == Task.Status.IN_PROGRESS) {
                pendingTask = task;
                break;
            }
        }

        if (pendingTask == null) {
            throw new IllegalStateException(
                    "No pending HUMAN task found in execution " + executionId);
        }

        // Update the task with the human's response
        TaskResult taskResult = new TaskResult();
        taskResult.setTaskId(pendingTask.getTaskId());
        taskResult.setWorkflowInstanceId(executionId);
        taskResult.setStatus(TaskResult.Status.COMPLETED);
        Map<String, Object> outputData =
                new LinkedHashMap<>(
                        pendingTask.getOutputData() != null
                                ? pendingTask.getOutputData()
                                : Map.of());
        outputData.putAll(output);
        taskResult.setOutputData(outputData);
        taskService.updateTask(taskResult);
        log.info(
                "Completed HUMAN task {} in execution {}",

View on GitHub (pinned to cf7c3e4a8a)

Solutions

  1. Check getStatus() and verify there is a pending HUMAN task before calling respond().
  2. Guard against double-submit in the UI (disable the submit button after first click).
  3. Catch IllegalStateException and treat it as a no-op if the HITL task was already completed.
  4. Confirm the pending tool type via getStatus().pendingTool before responding.

Example fix

// before
agentService.respond(executionId, Map.of("answer", "yes"));

// after
AgentStatusResponse status = agentService.getStatus(executionId);
if (status.isWaiting() && "HUMAN".equals(status.getPendingToolType())) {
    agentService.respond(executionId, Map.of("answer", "yes"));
} else {
    log.info("No pending HITL task for {}, skipping respond", executionId);
}
Defensive patterns

Strategy: validation

Validate before calling

AgentStatusResponse status = agentService.getStatus(executionId);
if (!status.isWaiting()
    || !"HUMAN".equals(status.getPendingToolType())) {
    return ResponseEntity.status(409)
        .body("No pending HITL task in execution " + executionId);
}
agentService.respond(executionId, output);

Try / catch

try {
    agentService.respond(executionId, output);
} catch (IllegalStateException e) {
    if (e.getMessage().contains("No pending HUMAN task")) {
        // HITL task already completed or not yet scheduled
        log.info("No pending HITL task for {}", executionId);
        return ResponseEntity.noContent().build();
    }
    throw e;
}

Prevention

When it happens

Trigger: Calling respond() on an execution that has no active HITL task (the agent is mid-LLM-call or between tool calls); the HUMAN task was already completed by a prior respond call; calling respond on a terminal execution; the pending task is PULL_WORKFLOW_MESSAGES (not HUMAN) but the caller assumed HITL.

Common situations: Double-submit of a HITL response (user clicks twice); UI polls and submits before the HUMAN task is actually scheduled; the agent moved past the HITL point due to a timeout; caller confuses a PULL_WORKFLOW_MESSAGES wait with a HUMAN task wait.

Related errors

AI-assisted analysis of conductor-oss/conductor@cf7c3e4a8a (2026-08-14). Data as JSON: /api/errors/b1a3b2cdfc63a65a. Report an issue: GitHub.