stavros-dsh-redteamer

Most "AI pentest" setups are little more than prompt stacks and half-wired tools, with the model drifting off-script the moment it gets a chance. stavros-dsh-redteamer packages a complete red-team methodology, specialist roster, and code-level enforcement layer into a native DSH plugin: the Stavros orchestrator plans and delegates, 24 specialist subagents cover Web, network/AD, wireless, post-exploitation, cloud, and reporting, and all 78 zero-dependency tools are scope-guarded by scope.json—an empty file blocks everything, and any target not explicitly listed is refused at the engine level. Destructive or high-impact actions (persistence, lateral movement, DoS) require in-session confirmation for that specific step, and SSRF controls plus audit trails are hard-coded in the runtime, not left to the model's discretion.

The plugin ships no exploit code, no C2 framework, and no pre-canned attack payloads. It is a methodology-plus-enforcement layer: it decides what is authorized and keeps every action inside that boundary. The offensive capability comes from the operator's own tooling, invoked exclusively through the gated runner. Findings must carry reproducible evidence to count as verified; anything unconfirmed is labeled suspected explicitly.

It is built for engagement teams and penetration testers who hold written authorization and want a structured red-team pipeline: you supply the target scope and credentials, the Stavros crew drives stage-gated testing against a coverage matrix, and you approve every sensitive step before it executes. This is not a shortcut for "let the AI hack something"—it is an expert red-team crew working under your direct supervision.