#I have created a Workflow based Agent i am seeing consistent issue sub agent being skipped even tho

1 messages · Page 1 of 1 (latest)

frigid depotBOT
#
AI-generated Answer for m_adnan_

Thanks for flagging this. Based on ElevenLabs’ docs, here are the checks that most often explain “skipped” subagent nodes in Workflow-based Agents:

  • Confirm the graph actually routes through the subagent with a Forward edge. Edges control conversation flow; Forward edges move the conversation to the next node. If the entry/exit edges aren’t Forward (or aren’t connected as intended), the engine won’t traverse the subagent. (elevenlabs.io)

  • If you’re using LLM-based routing, make sure the condition is entered in the LLM Condition field, not just the Label. The Label is human-readable only and isn’t evaluated; the LLM Condition is what’s actually checked at runtime. (elevenlabs.io)

  • Remember that LLM Conditions are evaluated by an LLM in real time. If the phrasing is ambiguous, the edge may not trigger. To debug deterministically, simplify the condition wording and validate the path by running simulations or tests:

    • Run simulated conversations to inspect turns and verify the decision point is hit. (elevenlabs.io)
    • Use Agent Testing to create scenario tests around that edge (and tool-call tests if the subagent is expected to trigger tools). (elevenlabs.io)

  • If the step must execute a specific tool regardless of dialog, consider a Dispatch Tool node instead of relying on tool usage within a subagent. Tool nodes guarantee the tool call and let you branch on success/failure explicitly. (elevenlabs.io)

If you’ve verified the above and the subagent is still skipped, please share a minimal repro (workflow screenshot, the exact LLM Condition text, and a conversation ID) with ElevenLabs Support so they can inspect logs. (help.elevenlabs.io)

References:

  • Workflows: subagent nodes, edges, LLM Conditions, and tool nodes. (elevenlabs.io)
solar walrus
#

following is the condition to exit the subagent "You have verbally asked the user the CT contrast allergy screening question. The user has listened to the question and responded with a complete answer. If the user knows of any allergies to CT contrast, they have stated those allergies. If the user reported an allergy, they have also described the specific reaction they experienced."

even with this the agent is skipped

frigid depotBOT
#
AI-generated Answer for m_adnan_

Thanks for the detail. A few doc-backed checks that commonly explain “skipped” subagents in Workflows:

  • Make sure your condition text is in the LLM Condition field, not the Label. The Label is only a human-readable description and is not evaluated by the LLM. (elevenlabs.io)
  • Prefer an Expression edge for deterministic exits instead of a long natural-language LLM condition. Workflows support Expressions that can check dynamic variables (e.g., a flag like allergy_screening_complete or counts), which is more reliable than a compound English sentence. (elevenlabs.io)
  • If you must guarantee a specific action happens before evaluation (e.g., invoking a tool that records the screening result), place that action in a Tool node; tool nodes guarantee the tool is called, while subagents don’t. Then route based on the result. (elevenlabs.io)
  • Validate the edge with Agent Testing so you can reproduce and verify the transition fires under your expected transcript/variables before going live. (elevenlabs.io)

If none of the above resolves it, please share call/conversation IDs with support so we can inspect the run. (help.elevenlabs.io)

References:

  • Workflows (Edges, LLM Condition vs Label; Tool nodes) (elevenlabs.io)
  • Changelog: Workflow Expressions (deterministic conditions with variables) (elevenlabs.io)
  • Blog: Designing secure authentication flows (example of using workflow expressions and dynamic variables) (elevenlabs.io)
  • Agent Testing (elevenlabs.io)
  • Contact Support (help.elevenlabs.io)

Sources:
Workflows | ElevenLabs Documentation
September 29, 2025 | ElevenLabs Documentation
Workflows | ElevenLabs Documentation
Agent Testing | ElevenLabs Documentation
How can I contact support? – ElevenLabs
Workflows | ElevenLabs Documentation
September 29, 2025 | ElevenLabs Documentation
Designing secure caller identity authentication flows for voice agents
Agent Testing | ElevenLabs Documentation
How can I contact support? – ElevenLabs

solar walrus
#

the condition is in LLM condition not label