Solution:
The set A={F2n:n=1,2,…}∪{−F2n+1:n=1,2,…}, where Fk is the kth Fibonacci number (F1=1, F2=1, Fk+2=Fk+1+Fk for k≥1) qualifies for an example. We then have F2n=F2n+2+(−F2n+1) and −F2n+1=(−F2n+3)+F2n+2 for all n≥1, so A is sum-full (and even with unique representations). On the other hand, we can never have
0=i=1∑sF2ni−j=1∑tF2nj+1
owing to the fact that Zeckendorf representations are known to be unique.
It remains to be shown that all nonzero values can be represented as sums of distinct numbers 1,−2,3,−5,8,−13,21,…. This may be done using a greedy algorithm: when representing n, the number largest in magnitude that is used is the element m=±Fk of A that is closest to 0 subject to having the same sign as n and ∣m∣≥∣n∣. That this algorithm terminates without using any member of A twice is a straightforward induction on k; the base case is k=2 (m=1) and the induction hypothesis is that for all n for which the above algorithm starts with ±Fℓ with ℓ≤k, it terminates without having used any member of A twice and without having used any ±Fj with j>ℓ.
Remark (James Aaronson and Adam P Goucher). Let n be a positive integer, and write u=2n; we claim that the set
{1,2,4,…,2n−1,−u,u+1,−(2u+1),3u+2,−(5u+3),8u+5,…}
is a sum-full zero-sum-free set. The proof is similar to that used for the standard examples.