You have two competing chains: Chain 1 and chain 2. On chain 1, you spend a token on a good. On chain 2, you spend no token. You wait for chain 1 to win the race. You then stake on chain 2 to get your money back. If chain 2 then overtakes chain 1, you've executed a double-spend attack.

An equivalent attack wouldn't work on a PoW chain. If you do the equivalent of "staking" on chain 2, then you're computing hashes, which is costing real-life resources. In the PoS case, without slashing, staking on chain 2 is free. In fact, this is the rational move to make every time you spend a token; stake on competing chains to get your token back.