SLM-agent-1: add historic decisions + feedback for RLHF

[rcmohan]

1. Get the agent to learn from past decisions - reinforcement learning
2. Include a summary or raw log of the agent's past decisions,
3. Include the outcomes of those decisions, and any associated "award" (success metric) or "mistake" (error signal)
4. Embed all this directly in the SLM's prompt.