Locate the policy inside an agent

Locate the policy as the part of the agent that maps what it sees to what it does. Separate the policy from the environment, the reward rule, and any later training method.