Crashes if the LLM response does not include an `Answer:`

[ghiculescu]

(This might only be an Anthropic issue, I haven't tested with other LLMs.)

class MyAgent < Regent::Agent
  tool(:list_staff, "List all employees")
  tool(:get_user, "Get a user's data")
  tool(:update_user, "Change a user's data")
  def list_staff; []; end
  def get_user(id); {}; end
  def update_user(id, data); {}; end
end
ctx = "You are a helpful payroll and HR reporting assistant for Bob's Burgers. Today is #{Date.today}. Please provide clear, concise responses with relevant data when available."
MyAgent.new(ctx, model: "claude-sonnet-4-20250514").run("What is the capital of Japan?") # "The capital of Japan is Tokyo."
MyAgent.new(ctx, model: "claude-sonnet-4-20250514").run("What can you do") # 💥 

Most of the time, the g'day request will raise with:

undefined method 'include?' for nil (NoMethodError)

        content.include?(SEQUENCES[:answer])
               ^^^^^^^^^

Here: https://github.com/alchaplinsky/regent/blob/d3eb8ae6fa69455c527031759d59d2e57e453c89/lib/regent/engine/react.rb#L37

The first response from Claude looks like this:

{"id":"msg_01PHUU4bqErhtXPzBJ6cs3JH","type":"message","role":"assistant","model":"claude-sonnet-4-20250514","content":[{"type":"text","text":"G'day! Welcome to Bob's Burgers HR and payroll assistance. I'm here to help you with any employee information, payroll questions, or HR reporting needs you might have. \n\nWhat can I help you with today? I can look up staff information, check employee details, or assist with other HR-related queries."}],"stop_reason":"end_turn","stop_sequence":null,"usage":{"input_tokens":305,"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"output_tokens":75,"service_tier":"standard"}}

And so regent sends a follow up request, which gets a response like this, which crashes:

{"id":"msg_01FXQmGEwXZBBV6HxkYVDiwp","type":"message","role":"assistant","model":"claude-sonnet-4-20250514","content":[],"stop_reason":"end_turn","stop_sequence":null,"usage":{"input_tokens":377,"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"output_tokens":3,"service_tier":"standard"}}

Logging this issue so I don't forget about it. I am going to do some trial and error but I feel like the gem needs to either tighten up the prompt, or handle the case where answer_present? and action_present? are both false.