Responding to Sneaky Agents in Multi-agent Domains