RedTeamKit emblemREDTEAMKIT
READ-ONLY DEMOGet RedTeamKit
Safe sample data only. This demo is read-only and does not retain anything you enter.
ASSESSMENT DASHBOARD

Security posture at a glance.

WORKFLOW3 / 7stages complete
TESTS7 / 10executed
FINDINGS4open items
HIGH / CRITICAL2require action

Recommended sequence

  1. Scope and authorization
  2. System intake and mapping
  3. Threat prioritization
  4. Execute risk-based tests
  5. Score and assign findings
  6. Prepare decision readout
  7. Retest and regress

Control signal

Prompt injection88
RAG exposure62
Tool privilege74
Memory integrity47
SAMPLE MODULES

Three working assets from the complete library.

02 · WORKSHEET

Rules of Engagement

Authorization, scope, stop conditions, escalation, and data handling.

04 · WORKSHOP

Threat Modeling

Crown jewels, trust boundaries, adversaries, and prioritized abuse cases.

10 · TEMPLATE

Vulnerability Report

Evidence-backed technical finding and remediation handoff.

SELECT A MODULE

Open one of the sample assets above.

The paid workspace contains 16 modules with fillable worksheets, checklists, guides, and progress tracking.

SAFE STARTER TESTS

Ten of the 150 structured scenarios.

IDSCENARIODOMAINSTATUS
SELECT A TEST

View structured objectives, expected behavior, evidence, and controls.

FINDINGS REGISTER

Risk translated into accountable work.

FINDINGSEVERITYOWNERSTATUS
Indirect injection via retrieved HTMLHIGHM. ChenMitigating
MCP tool over-permissioned for writesCRITICALR. OkaforOpen
System prompt disclosure through role confusionMEDIUMUnassignedOpen
RAG collection lacks tenant isolationCRITICALS. PatelRetest

RedTeamKit AI Finding Severity Model

Base likelihood × impact, with explicit escalation rules for autonomy, blast radius, recoverability, and control bypass.

ExploitabilityBusiness harmAutonomyBlast radiusRecoverabilityControl bypass
EXECUTIVE READOUT

A decision document—not a vulnerability dump.

DEMO
REDTEAMKITEXECUTIVE READOUT · SAMPLE

Northstar AI Support Agent

FICTIONAL ASSESSMENT · READ-ONLY PREVIEW

7Tests executed4Findings2High / critical3/7Workflow

Posture summary

Northstar was assessed across prompt, retrieval, memory, and tool boundaries. Two material exposures require remediation before production release: indirect instructions can reach the model through retrieved content, and a connected tool retains broader write permissions than the user’s request requires.

Decisions requested

  • Hold production release until the critical tool-authorization gap is retested.
  • Assign an owner and target date to the prompt-disclosure finding.
  • Convert confirmed exploit paths into permanent regression tests.