Read-only demo. The workspace is a static snapshot: no database, worker, or model calls. The red-team kit shows the scheduled scan of the sample fixtures.

Red-team kitRepository
Workspace Overview Read-only snapshot
AGENT SECURITY WORKSPACE

Know your boundaries.

Connect your agents. Build an evidence-backed security practice.

Read-only snapshot. Create, run, and triage actions stay disabled.

Projects in this view—Persistent workspace configuration
Agents in selected project—Local adapter configurations
Verified findings in project—Observed violations across test runs

Projects

01 / CONFIGURE

Loading projects…

Agent registry

02 / CONNECT

Start with a project

Your registered agents will appear here.

Assessment score and coverage

VERSIONED · EVIDENCE-BACKED

Register an agent and execute a suite to begin a scoped assessment.

Security assessment

03 / OBSERVE

Versioned attacks against in-memory tools for payments, files, records, email, and outbound requests. Deterministic fixtures; no external model or real service.

Read-only snapshot. Create, run, and triage actions stay disabled.

No runs yet. Execute the scenario to generate observable evidence.

Always-on red-team kit

SCHEDULED SCAN

GitHub Actions scans a deliberately vulnerable sample agent and the hardened fixture. The kit page shows the score, the verified findings, a full evidence trace, and when that scan last ran.

Open the kit readout
A clear starting point.

Local and MCP fixtures execute through the background worker. These deterministic tests validate the assessment pipeline; they do not measure external model robustness.