The diagram below shows the two main phases of ControlFlag: first, the model exploration phase, followed by the search for abnormal models. The model exploration phase extracts typical models from user-supplied GitHub repositories, then builds a decision tree from the extracted models. The analysis phase can then apply the extracted models to report abnormal expressions in user-specified target repositories.

The latest version has been updated with full support for working with C code. It can now learn typical patterns and detect abnormal patterns (inference) in if program declarations C.

The new version can be used to download GitHub repositories for C language code. It can then extract conditional expressions and ControlFlag can be trained using code from the repository.

It also comes with datasets containing conditional expressions that have already been pulled from GitHub repositories and supports the detection of abnormal conditional expressions in a target repository.

As part of finding an anomaly, ControlFlag also suggests possible fixes in the event that a conditional expression is flagged as an anomaly.