An open research idea for AI-assisted adversarial simulation and defensive intelligence.

SynthRange is an open research idea and reference implementation for AI-assisted adversarial simulation ranges. It sketches how autonomous Red and Blue agents, target adapters, composable playbooks, and an evidence-driven Observer layer fit into one architecture for assessing systems you own—across web applications and, in the documented direction, Linux, Windows, and macOS—so that a run produces reproducible evidence, detections, and defensive intelligence.

The repository publishes the architecture, the versioned contracts, the components that exist today, and one sanitized illustrative run brief. It is offered for study, critique, and contribution rather than as a supported product or turnkey tool.

Architecture

SynthRange separates five responsibilities:

• Red performs adversarial work within a playbook and run scope.
• Blue investigates, detects, and develops defensive intelligence.
• Targets provide reproducible systems and applications to assess.
• Traffic Proxy records and optionally constrains web traffic.
• Observer collects artifacts and produces the neutral run record.

Red and Blue may use Hermes today, but the architecture is agent-framework independent.

Current status

Area	Status
Red Hermes agent	Available in the reference lab; public integration template planned
Blue Hermes agent	Available in the reference lab; autonomous finalization remains Prototype
Observer evidence helpers	Available; lifecycle and approval hardening available in source, lab rollout pending
Traffic Proxy addon	Available; mechanical pacing and budget policy available in source, lab rollout pending
Juice Shop target adapter	Available; machine-readable contract and Compose deployment
Initial composable playbooks	Available: shared evidence baseline, Red web discovery, Blue web triage
Observer UI	Prototype; source exists but is not deployed in the reference lab
Linux, Windows, and macOS targets	Planned
Whole-range automated deployment	Planned

Status labels mean Available, Prototype, Planned, or Vision. Planned architecture is documented openly without being presented as implemented.

There is no zero-to-first-run deployment path in this repository yet, and this preview does not promise one. That boundary is stated so readers can judge the material accurately; it is not a request for contributors to build an installer.

Start here

1. Overview
2. Architecture
3. Core concepts
4. Getting started
5. Playbooks
6. Targets
7. Contracts and schemas
8. Development
9. First real run lessons

Community

• Roadmap
• Contributing
• Governance
• Code of Conduct
• Security policy
• Support
• Public-readiness review
• Apache License 2.0

Repository map

components/     Observer and Traffic Proxy implementations
docs/           Architecture, concepts, decisions, and review notes
examples/       Sanitized illustrative contract documents
assets/         Canonical SVG diagrams and future approved media
integrations/   Agent-framework integrations, beginning with Hermes
playbooks/      Composable Red, Blue, and shared playbooks
schemas/        Versioned target-adapter and playbook contracts
targets/        Reproducible target adapters
templates/      Run and reporting templates
deploy/         Whole-range deployment material

See examples/runs/ for the sanitized run brief and its created-state document.