Google open-sources Mantis toolkit for AI-assisted vulnerability reviews
Google has released Mantis, a stack-agnostic toolkit that adds security-review skills to AI coding agents. Mantis is not a standalone scanner: it provides slash commands that an existing agent can load and run under defined execution rules. It is intended for local and internal evaluation, while production deployment is not yet recommended. The toolkit is available under Apache 2.0 and can work with Gemini CLI, Antigravity CLI, Google ADK or comparable agent frameworks.
Its workflow covers the vulnerability lifecycle. Early skills examine version-control history, map directories, document architecture, identify trust boundaries and create a review plan. Other stages search code, remove duplicate or invalid findings and review suspected issues. Reproduction runs payloads in gVisor or a virtual machine with networking disabled. Mantis can then assemble exploit chains, apply and verify patches, re-attack the result, assign a risk score from 1 to 10 and generate a review report.
A supervisor command can coordinate the stages, while /mantis-advise uses earlier threat models, bug lineages and verified patch patterns before new code is written. Google says a hierarchical summary tree reduces token use by more than 85 percent and identifies sub-7 percent true-positive rates in naive AI code scanning as the problem Mantis addresses.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.