Agent node (v3.1) loses upstream execution context when delegating to sub-agents or other agent workflows

[alejandrosnz]

Describe the problem

When using the AI Agent node version 3.1 (Latest), upstream execution context is lost when the Agent either:

• delegates to a sub-agent, or
• calls another workflow that contains an Agent

As a result, expressions referencing upstream nodes (e.g. a Telegram Trigger) stop working, even though the workflow structure remains visually unchanged.

This behavior did not occur with Agent node version 2.x, where upstream node expressions continued to work reliably.

---

Expected behavior

Expressions referencing upstream nodes (for example:

{{ $('Telegram Trigger').item.json.message.chat.id }}

) should behave consistently regardless of whether the Agent:

• answers directly,
• uses a sub-agent, or
• calls another workflow containing an Agent.

If this behavior is expected due to execution boundaries, the UI should clearly warn about it or document it explicitly.

---

Actual behavior

Workflow structure:

Telegram Trigger
 → Agent (v3.1, may use sub-agent OR call another workflow with an Agent)
 → Telegram Send Message

Observed behavior:

• If the Agent does NOT delegate, the expression works correctly.
• If the Agent uses a sub-agent, the expression evaluates to undefined.
• If the Agent calls another workflow that contains an Agent, the expression also evaluates to undefined.

This causes the Telegram Send node to fail due to a missing Chat ID.

No warning or error explains why the upstream node context is no longer available.

---

Why this is problematic

• The workflow is visually linear but behaves differently at runtime.
• The failure is non-deterministic, depending on LLM decisions.
• The same expression works or fails depending on Agent behavior.
• There is no UI indication that Agent v3 introduces execution boundaries.
• This represents a behavior change/regression compared to Agent v2.x.

From a user perspective, this breaks the expected mental model of how expressions and node context work in n8n.

---

Reproduction steps

1. Create a workflow:

   • Telegram Trigger
   • AI Agent node (version 3.1, Latest)
   • Telegram Send Message

2. Configure the Agent so it may:

   • use a sub-agent, or
   • call another workflow that contains an Agent

3. In the Telegram Send node, set Chat ID to:

   {{ $('Telegram Trigger').item.json.message.chat.id }}

4. Send a message to the bot:

   • Case A: Agent answers directly → message is sent successfully.
   • Case B: Agent delegates (sub-agent or workflow with Agent) → Chat ID is undefined and message fails.

---

Environment

• n8n version: 2.3.5
• AI Agent node version: 3.1 (Latest)
• Deployment: Docker
• OS: Linux (Docker host)
• Trigger: Telegram Trigger

---

Screenshots

The last node has the chat_id value when no sub-agent or workflow is called by the Main Agent:

The last node does not have the chat_id value when a sub-workflow was called:

The last node does not have the chat_id value when a sub-agent was called:

With AI Agent node version 2.x, the execution context appeared to be preserved when calling a workflow

[n8n-assistant]

Hey @alejandrosnz,

Thank you for reaching out! We’ve received your issue and are looking into it. To help us track this internally, we’ve created a Linear ticket with the reference: "GHC-6388".

We’ll keep you updated as we make progress, but if you have any additional details or context to share, feel free to add them here - it’s always helpful!

Thanks for your patience and for bringing this to our attention.

---

For INTERNAL use only: GHC-6388