Verdict. The right shape for a security pass before launch. Every finding comes with a working proof of concept, so it either reproduces or it doesn’t. That beats a static-analysis report full of maybes.

what it is

An open-source penetration-testing agent. It runs your application, finds vulnerabilities, and validates each one with a working exploit.

A team of recon, exploitation and post-exploitation agents works inside a Docker sandbox with a real offensive toolkit. That includes the Caido HTTP proxy, an automated browser for XSS, CSRF and auth bypass, an interactive shell, a Python exploit runtime, Nuclei and Playwright, plus SAST and DAST.

strix --target ./app-directory

The target can also be a URL, a GitHub repo, an OpenAPI file or a Postman collection. -n runs it headless for CI. strix view opens a local dashboard. Nine agent skills install with npx skills add usestrix/strix.

Results go to strix_runs/<run-name>. You get validated findings with CVSS scores and OWASP classes, optional remediation patches, and a non-zero exit code when it finds something, so it drops straight into CI.

what I found

Stars57,416 (6,220 forks), GitHub API, 2026-08-23
LicenseApache-2.0
Commits100 in three weeks, 68 contributors
Tests73 pytest files
Releasev1.5.3 (2026-08-10)
StackPython core, Go TUI, TypeScript viewer. PyPI strix-agent
  • It needs Docker running and an LLM API key. It can also sign in with a ChatGPT Plus or Pro subscription (strix auth login chatgpt).
  • Reasoning effort defaults to high. Use medium for quick scans and budget for it.
  • It’s open-core. The CLI is free, and a paid platform adds SSO, compliance reporting and self-hosted deployment.
  • Only one CI workflow, so test enforcement seems to lean on pre-commit hooks rather than PR gates.
  • strix view binds to 127.0.0.1 on a random port behind a tokened URL. It reads run files off disk and uploads nothing. You can steer a running scan by sending it new instructions mid-run.
  • The README says to only run it against systems you own or have written permission to test.