agentred/web
Attack your apps before attackers do.
AI-powered penetration testing for web applications and REST & GraphQL APIs. Every material finding is verified before it reaches your team.
# replay order lookup as user_b
GET /api/v2/orders/1042 HTTP/1.1
Authorization: Bearer ••••user_b
HTTP/1.1 200 OK
{ "owner": "user_a", "total": 249.00 }
Vulnerability classes
What gets tested.
Agent Red tests the classes that lead to real compromise, across every route and endpoint in scope.
- 01Broken access control
Horizontal and vertical privilege checks across roles, tenants, and object IDs.
→ - 02Injection
SQL, command, and template injection through every parameter that reaches a sink.
→ - 03Cross-site scripting
Reflected, stored, and DOM-based XSS, confirmed in a real browser context.
→ - 04SSRF
Server-side requests pushed toward internal services and cloud metadata endpoints.
→ - 05Auth & sessions
Login, reset, token, and session flows tested for bypass and fixation.
→ - 06Business logic
Workflows abused out of order, replayed, or with values they never expected.
→ - 07Prompt injectionNew
LLM features attacked directly, then followed into the code and data around them.
→
Attack paths
Follow the path, not the checklist.
Single findings rarely tell the whole story. Agent Red chains them the way an attacker would.
- EntryPrompt injection
A crafted message steers the support assistant.
- PivotPath traversal
The assistant’s file tool reads outside its directory.
- EscalateSecrets exposure
Configuration files leak service credentials.
- ImpactStored XSS
Attacker content persists into other users’ sessions.
Illustrative chain built from the vulnerability classes Agent Red found in the ALAF case study.
Questions
Web & API testing, answered.
Do you test authenticated areas?
Yes. With credentials or test accounts you provide, Agent Red tests flows behind login, across the roles you put in scope.
Is testing intrusive?
Intrusive actions require approval. Your security team sets the rules of engagement and exclusions before any testing starts.
Do you test GraphQL?
Yes. REST and GraphQL APIs are both in scope, including schema analysis and the business logic behind each operation.
Can you test LLM features?
Yes. Prompt injection is tested as part of the wider application, so risk is followed into the code and data around the model.
One platform
Explore every surface.
Web & APIs
Application and API penetration testing across the paths attackers actually use.
Source code
AI-assisted security review that follows risk across code, dependencies, and CI/CD.
Cloud
Cloud posture and misconfiguration assessment across AWS, Google Cloud, and hosts reached over SSH.
- AWS
- Google Cloud
- SSH
Mobile
Static and dynamic mobile security analysis aligned to real application behavior.