Chat command should use LLM to converse as the agent

[DeveshParagiri]

Current Behavior

The extropy chat command is a stub that just dumps agent context on every message:

chat> hello
Agent `agent_01` context snapshot:
- Position: actively_oppose_ban
- Sentiment: 1.000
...

It doesn't actually have a conversation with the agent.

Expected Behavior

The chat command should use an LLM to roleplay AS the agent, grounded in their:

• Persona attributes
• Simulation state (position, sentiment, conviction)
• Memory traces / reasoning history
• Conversation history from simulation

Implementation

1. Build system prompt with agent persona + current state
2. Include memory traces as context
3. Send user messages to LLM (use simple_call_async)
4. Return responses as if the agent is speaking
5. Store conversation turns in chat_messages table (already exists)

Example

chat> Hey, what do you think about the book ban?
I think it's outrageous censorship. The school board made a heavy-handed 
decision without any transparent review process. I'm planning to attend 
the next board meeting to push back publicly.

chat> But don't you think some books aren't appropriate for kids?
Sure, some content isn't age-appropriate — I'm not naive about that. But 
a blanket ban isn't the answer. We need grade-level review and parental 
opt-out, not district-wide removals that override teachers and parents.

Files

• extropy/cli/commands/chat.py — replace _summarize_context with LLM call

[DeveshParagiri]

Fixed in PR #78 (feat(chat): LLM-backed agent chat + simpler chat targeting)