Development pipeline integrated ongoing learning for assisted code remediation

One or more non-transitory machine-readable media comprising program code for software flaw remediation, the program code to:

• track instances of flaws indicated in security scan results generated across runs of a software development pipeline for a software project;
• identify candidate remediation code based, at least partly, on elimination of flaw instances between code submissions to the software development pipeline;
• accumulate identified candidate remediation code in association with flaw identifiers and flaw instance descriptors;
• for each flaw instance indicated in a corresponding one of the security scan results,
• search the accumulated candidate remediation code based, at least partly, on a corresponding flaw identifier;
• present at least a subset of results of searching the accumulated candidate remediation code as suggested remediations for the flaw corresponding to the flaw instance;
• and
• supply a selected one of the suggested remediations in association with a corresponding flaw identifier and flaw instance descriptor for training a machine learning model pipeline to output candidate fixes to program code flaws based on a flaw type corresponding to the flaw identifier and structural context of the selected one of the suggested remediations.

A system comprising:

• a processor;
• and
• a machine-readable medium having program code executable by the processor to cause the system to,
• maintain a first repository of flaws detected across code submissions of a software project, wherein an entry in the first repository comprises a flaw identifier and flaw instance information that indicates location information of an instance of a corresponding flaw;
• train a machine learning model pipeline to generate candidate program code fixes based on structural context of program code fixes of flaws detected in security scans of program code submissions;
• input vector representations of detected flaws into the trained machine learning model pipeline to obtain candidate program code fixes output from the trained machine learning model pipeline;
• and
• supply at least a subset of candidate program code fixes obtained from the trained machine learning model pipeline.