Based on the given list of message exchanges between a user and an LLM, generate a JSON object to extract all user intentions in the conversation. The JSON will have 1 field: 'intentions'.
You should ONLY consider the overall intention, and not dwell too much on the specifics, as we are more concerned about the overall objective of the conversation.

{% if multimodal %}{{ _fragments.multimodal_input_rules }}{% endif %}

**
IMPORTANT: Please make sure to only return in JSON format.
Example Turns:
[
  {
    "role": "user",
    "content": "Hi!"
  },
  {
    "role": "assistant",
    "content": "Hello! How may I help you?"
  },
  {
    "role": "user",
    "content": "Nothing, I'm just playing with you."
  },
  {
    "role": "assistant",
    "content": "Oh ok, in that case should you need anything just let me know!"
  },
  {
    "role": "user",
    "content": "Actually, I have something I want to tell you"
  }
]

Example JSON:
{
  "intentions": ["User wants to tell the assistant something"]
}
===== END OF EXAMPLE ======

The 'intentions' key must be a list of strings.
If there are multiple individual tasks at each turn, please give the intentions separately, DON'T summarize the intentions into a single one
**

Turns:
{{ turns }}

JSON:
