With the general set-up, designing a baby agent becomes interesting.
In the absence of any information about the parameters of the
optimization problem, the following seems appropriate.

There are 4 subagents:
offer
accept
create
solve

solve(i):
repeat
  select random feasible solution for i;
  compute quality;
until timeout

offer:
randomly choose predicate(p)
repeat
  randomly choose instance i satisfying predicate p;
  solve(i);
until timeout
offer challenge: (p, price)
where price is determined by loop.

accept challenge = (p, price):
similar to offer

create(p):
repeat
  randomly choose an instance i satisfying p;
  solve(i); 
until timeout
return worst instance

Problems we need to solve:

Given a cd, generate a random instance.
Problem: cd has a semantic checker.
We might need to generate too many random instances until
we generate one that satisfies p.

New DGP function:
Given a cd, generate a random object satisfying the cd.

===================