Troubleshooting¶
The agent selector is empty or shows an error¶
- The user needs the
administer flowdroppermission; the selector calls/api/flowdrop/ai-agents/autocomplete. - Check that
ai_agentsis enabled and the agent exists. - Rebuild caches with
drush cr. Agent definitions are cached by the AI Agents plugin manager.
The selector is a plain text field¶
The custom widget registers when the FlowDrop editor or playground fires its init event. If it does not appear:
- Check that
lib/build/agent-selector-field.jsexists. Rebuild it if you work from a git checkout that does not include it; see Editor field widgets. - Rebuild caches so the library dependency added by the module is picked up.
- Check the browser console for errors from
flowdrop_ui_components.
The preview has no system prompt¶
Expected for users without administer ai_agent. See Configuration.
The workflow stops at the agent node¶
The agent returned Needs Answers or Should Answer Question, and the workflow is waiting for a
reply. Answer the pending interrupt where the workflow runs. To avoid the question, give the agent
more detail in task_description or additional_context. See
Job status and questions.
success is false¶
The agent decided it cannot solve the task. Read error_message and response. Try another agent,
rephrase the task, or check that the agent has the tools the task needs.
The job fails with AI Agent "<id>" failed: ...¶
A runtime error happened while the agent ran. The text after the colon is the original exception
message. Common causes are a missing or misconfigured AI provider, an invalid llm_model value for
the provider, and errors raised by the agent's tools. Check the Drupal log for details.
tool_usage_limits has no effect¶
Limits apply to config-based agents only (agent_type is Config). Plugin-based agents ignore
them.