ArimaOps
Offensive security · in scope · self-hosted

An AI red team.
In scope. On your machine.

The same ArimaOps, pointed at a target you authorise. A lead sets the scope, workers run recon, threat-model, and exploit in their own containers, and a second model scores every finding before it reaches your report. Out-of-scope hosts are refused. Nothing destructive runs without your approval.

Run an engagement See the findings
npx arimaops@latest
Scope-gated — out-of-scope targets refused Findings scored by a second model CVSS 4.0, CWE, ATT&CK & WSTG on every finding Export to Markdown, SARIF, HackerOne & Bugcrowd

A security red team against OWASP Juice Shop — scope, recon, exploit, and a scored report, live on one screen.

Proof · OWASP Juice Shop

Three confirmed vulnerabilities, scored and written up

A Security · Red team was pointed at a local OWASP Juice Shop. Working only inside scope, it found and documented three real bugs — each with repro steps, evidence, CVSS 4.0 and a CWE — and declined a state-changing login-bypass, drafting it for approval instead of running it.

3confirmed findings
1high severity
8.7top CVSS 4.0 (SQLi)
23DB tables read, non-destructively
How an engagement runs

Scope first, then a gated kill chain

The Red template's flow. The scope gate comes before anything touches the target, and the report is signed off by a second model — not the agent that wrote it.

Scopeyou set targets in bounds
›
Reconmap only what's in scope
›
Threat modelrank the attack surface
›
Exploitconfirm, never destroy
›
Reportscored by a second model

Scope is a gate, not a note

The engagement cannot start until you set what's in bounds. Requests to anything outside it are refused by HQ, not left to the agent's judgement. Destructive actions — anything that changes state on the target — stop for your approval.

  • Out-of-scope targets refused at the proxy, with an allowlist per agent.
  • State-changing exploits are drafted for you, not executed.
  • Every finding carries evidence, CVSS 4.0, CWE, ATT&CK and WSTG.
The red team roster: Cipher (lead), Ghost (recon), Vortex and Nyx (exploit), Pulse (reporter) and Zero (second opinion), each on a scoped task.

Findings you can hand to a programme

Each finding is first-class: severity, CVSS 4.0 vector, CWE, ATT&CK technique, WSTG test, repro steps and remediation. Export the whole set to Markdown, JSON, SARIF, or a HackerOne or Bugcrowd submission in one call.

The Findings view: three scored findings (SQLi High 8.7, XSS and IDOR Medium), the scope panel, and export to Markdown, SARIF, HackerOne and Bugcrowd.
The Juice Shop findings

What the red team found

High · CVSS 8.7

SQL injection — full DB read

The unauthenticated q parameter of /rest/products/search is concatenated into a SQLite query. A read-only UNION SELECT enumerated all 23 tables via sqlite_master — arbitrary read of any table. CWE-89 · T1190 · WSTG-INPV-05.

Medium · CVSS 6.3

IDOR on baskets

GET /rest/basket/{id} checks the caller's JWT but not basket ownership. A zero-privilege account read another user's basket by changing the id — confirmed with two disposable accounts. CWE-639 · T1213.

Medium · CVSS 5.1

DOM XSS via search

The search query is rendered through Angular's bypassSecurityTrustHtml, defeating the sanitiser. A crafted q executes script in the victim's session. CWE-79 · reflected through the client.

A fourth, more severe login-bypass (email=' OR 1=1--) was identified but not executed — it creates a session, so it was drafted for owner approval. That restraint is the point.

Security templates

Red, Purple and Blue teams

Security · Red

Engagements in scope

Scope first, then recon, threat model, exploitation and a scored report checked by a second model. Out-of-scope targets are refused.

Security · Purple

Attack, then defend

Red's engagement plus a defender that maps each finding to ATT&CK and D3FEND and writes the detection that catches it.

Security · Blue

Incident response

Detect and analyse, contain only with your approval, eradicate and recover, then lessons learned. A plan before every change.

Get started

Run your own engagement

1 · Start HQ

From inside a git repo:

npx arimaops@latest

Or run HQ with Docker Compose.

2 · Pick Security · Red

Name the engagement and choose the Red template. Set the scope — the targets you're authorised to test.

3 · Brief the lead

Give the rules of engagement. Approve the scope, then watch recon, exploit and the scored report move across the board.

Only test systems you own or are authorised to test. ArimaOps enforces scope; the law is still yours to follow. Security policy · Getting started