Introducing a new way to review your PRs.

Read the announcement

usestrix / strix

Open-source AI penetration testing tool to find and fix your app’s vulnerabilities.

View on GitHub

26 documents · Updated Nov 20, 2026

overview

Knowledge Base updated: 2026-11-19

Strix knowledge base

Strix is a security-scanning CLI that configures targets and runtimes, drives an LLM-backed testing agent, and produces findings and reports.

Its primary boundaries are the command-line experience, execution and runtime sessions, agent tools and model-context controls, result reporting, and local or cloud-managed integrations.

Critical per-scan state is coordinated by runner and session-management layers; configuration controls providers, runtimes, proxy/MCP connections, and cloud identity.

flowchart TD
  CLI[CLI commands] --> Setup[Scan setup]
  Setup --> Runner[Runner lifecycle]
  Runner --> Runtime[Runtime backends]
  Runner --> Session[Session manager]
  Runner --> Agent[Security agent]
  Agent --> Tools[Agent tools]
  Agent --> Context[Context management]
  Context --> Model[LLM providers]
  Tools --> Proxy[Proxy and Caido]
  Tools --> MCP[MCP servers]
  Agent --> Skills[Security skills]
  Agent --> Findings[Findings and reports]
  CLI --> Cloud[Cloud-managed scans]
  Runner --> Telemetry[Operational telemetry]
  Findings --> Output[Report exports]

Glossary

  • Runner — the scan-execution coordinator that manages lifecycle, inputs, and recovery.
  • Session manager — the component that scopes runtime resources to a scan and releases them afterward.
  • Skill — bundled or repository-level security-testing guidance available to an agent.
  • Caido — the proxy/runtime integration used for browser and traffic-assisted security work.
  • MCP — Model Context Protocol servers exposed to an agent as session-scoped tools.

Codebase-wide invariants

  • Scan resources must be isolated and released through session lifecycle handling (src/**).
  • Provider behavior is controlled by validated configuration rather than tool-local defaults (src/**).
  • Findings flow through report state and export paths before delivery (src/**).
  • Agent actions are mediated through the registered tool surface (src/**).
  • Long-running model interactions must account for available context and compaction (src/**).

Routing table