Strix - AI Penetration Testing Agent Cheatsheet
Strix is an open-source AI penetration testing tool that operates like a security researcher rather than a static scanner. Instead of pattern-matching source code, it dynamically runs the target application, interacts with it, probes for vulnerabilities, and validates each finding with a proof-of-concept exploit. That validation step is the differentiator: findings come with evidence they are actually exploitable, which cuts the false-positive noise that makes traditional scanner output expensive to triage.
Authorized testing only. Strix actively attacks the target and generates working exploits. Run it exclusively against systems you own or have explicit written permission to test. Unauthorized use is illegal.
Installation
| Method | Command |
|---|
| pip | pip install strix-agent |
| Docker | docker run -it strixagent/strix |
| From source | git clone https://github.com/usestrix/strix && pip install -e . |
| LLM key | export OPENAI_API_KEY=... (or another provider) |
| Verify | strix --version |
Core Usage
| Command | Description |
|---|
strix --target ./repo | Test a local codebase |
strix --target https://staging.example.com | Test a running application |
strix --target ./repo --instruction "focus on auth" | Steer the agent |
strix --help | Full option list |
How It Works
| Phase | What the agent does |
|---|
| Recon | Maps the app: routes, inputs, auth surfaces |
| Hypothesis | Reasons about likely vulnerability classes |
| Exploit | Crafts and sends real payloads |
| Validate | Confirms impact with a working PoC |
| Report | Documents finding + reproduction steps |
The validation phase is what separates Strix from a scanner: a finding only surfaces if the agent could actually trigger it.
Vulnerability Classes
| Class | Examples |
|---|
| Injection | SQLi, command injection, SSTI |
| XSS | Reflected, stored, DOM-based |
| Access control | IDOR, privilege escalation, broken authz |
| SSRF | Internal service access |
| Business logic | Workflow abuse a scanner cannot model |
| Auth | Session handling, token flaws |
Scope & Safety Controls
| Control | Purpose |
|---|
| Explicit target | Only the declared target is tested |
| Sandboxed runs | Execute the agent in an isolated container |
| Read-only modes | Limit destructive actions where supported |
| Rate limiting | Avoid overwhelming the target |
| Review PoCs | Always inspect generated exploits before reuse |
Run Strix against staging, not production, and confine network egress from the container.
Common Workflows
# Assess a staging deployment you own
strix --target https://staging.internal.example.com
# Focused review of an authentication module
strix --target ./app --instruction "audit the login and session handling"
# Run inside a container for isolation
docker run --rm -it -e OPENAI_API_KEY strixagent/strix \
--target https://staging.example.com
Reading the Output
| Field | Meaning |
|---|
| Finding | The vulnerability class and location |
| PoC | The concrete request/payload that triggered it |
| Impact | What an attacker gains |
| Reproduction | Steps to confirm manually |
Because each finding ships with a PoC, triage becomes verification rather than investigation.
| Aspect | Strix | SAST (Semgrep) | DAST (ZAP) |
|---|
| Method | AI agent, dynamic + validated | Static pattern match | Crawl + attack |
| False positives | Low (PoC-validated) | Higher | Medium |
| Business logic | Yes (reasons about intent) | No | Limited |
| Speed | Slower (agentic) | Fast | Medium |
| Best for | Deep validated assessment | Broad code coverage | App-wide DAST |
Complements rather than replaces: use Semgrep for breadth, ZAP for crawling DAST, and Strix for validated depth.
Resources
