cisco-ai-defense/project-quarry is a rule engine that looks for threats and behavior patterns in agent conversations. It is a Python project published on GitHub and uses the Apache-2.0 license. The latest release is 0.1.0. The given material does not separately record which items changed in this release. So what can be confirmed here is not the addition of a particular new feature, but the purpose and operation of the tool as the repository describes it.
The tool inspects the messages of a conversation, tool calls and their results, and the order of events with rules that the user writes. Rules go through a syntax check and validation at load time, and rules that pass are compiled into a symbolic finite state automaton and applied to the conversation flow. A finite state automaton moves between defined states while checking the order of input. Rules that do not compile cannot be loaded. It can inspect text and regular expressions, structured fields, and whether a tool was used, and it can also express flows that span several turns. Results can include evidence, and they are printed on the command line in a human-readable format or as JSON and CSV. Limits on input size, compile time, run time, and model calls are also described.
However, the README states that Project Quarry is at the alpha stage and that the rule syntax may change. So the current rules and configuration cannot be assumed to stay the same in later versions. The purpose of checking multi-step conversation flows with explicit rules is clear, but until the syntax settles it is an early tool that requires accounting for the possibility of change.