Malicious software (malware) with decentralized communication infrastructure, such as peer-to-peer botnets, is difficult to detect. In this paper, we describe a traffic-sanitization method for identifying malware-triggered outbound connections from a personal computer. Our solution correlates user activities with the content of outbound traffic. Our key observation is that user-initiated outbound traffic typically has corresponding human inputs, i.e., keystroke or mouse clicks. Our analysis on the causal relations between user inputs and packet payload enables the efficient enforcement of the inter-packet dependency at the application level. We formalize our approach within the framework of protocol-state machine. We define new application-level traffic-sanitization policies that enforce the inter-packet dependencies. The dependency is derived from the transitions among protocol states that involve both user actions and network events. We refer to our methodology as storytelling security. We demonstrate a concrete realization of our methodology in the context of peer-to-peer file-sharing application, describe its use in blocking traffic of P2P bots on a host. We implement and evaluate our prototype in Windows operating system in both online and offline deployment settings. Our experimental evaluation along with case studies of real-world P2P applications demonstrates the feasibility of verifying the inter-packet dependencies. Our deep packet inspection incurs overhead on the outbound network flow. Our solution can also be used as an offline collect-and-analyze tool.