ADK reference campaign harness

Reference Google ADK implementation that configures and launches the 16-stage review graph with selectable sandboxes.

Parent repository GitHub

Type: workflow and CLI
Repository: google/mantis
Readiness: Use only with expert supervision in isolated, restricted environments; not production-ready by its own documentation
Keywords: 8

Location

Repository path

reference/README.md

Invocation

python3 scripts/configure.py --auto; python3 scripts/configure.py --test; ./run.sh PATH

Setup

Installation / activation

After reference/install.sh; launch against a target file or repository.

Keywords

ADKworkflowconfigurelaunchGCEgVisormicroVMstatic-only

Repository context

google/mantis

Apache-2.0 portable Agent Skills and a Python/Google ADK reference harness for staged defensive security review, covering architecture and threat modeling through vulnerability research, deduplication, validation, sandboxed reproduction, exploit chaining, patch verification, calibration, reflection, reporting, and secure-development advice; Claude Code is not explicitly documented, so applicability is inferred for skill-compatible coding agents rather than established as a native Claude extension.

Application securitySecure code reviewThreat modelingVulnerability researchStatic analysisDynamic reproductionExploit analysisSecurity patchingDevSecOpsHardware and firmware securityInfrastructure as Code securitySecure software development

Open full repository research