Skip to content

Security: Jake-Taylor02/AgentOS

Security

SECURITY.md

Security Policy

AgentOS includes components for secrets, permissions, agent isolation, runtime control, and observability. Security reports are important and welcome.

Reporting a Vulnerability

Please do not open a public issue for a security vulnerability.

Report privately to the project maintainer:

  • Maintainer: WAHIB EL KHADIRI
  • GitHub: WAHIB-EL-KHADIRI

Include:

  • affected component
  • reproduction steps
  • expected impact
  • suggested fix, if known

Scope

Security-sensitive areas include:

  • crates/vault
  • crates/kernel
  • crates/bus
  • crates/plugins
  • permission checks
  • agent isolation
  • secret handling
  • dashboard access control when implemented
  • Docker and Compose deployment defaults

Secure Defaults

AgentOS defaults to loopback runtime binding through RuntimeConfig, uses bounded protobuf request bodies on the bus, applies defensive HTTP response headers to built-in JSON/SSE endpoints, and runs the production container images as a non-root user.

The dashboard production image serves static assets with a restrictive Content Security Policy and related browser hardening headers. The Compose stack avoids publishing Redis/Postgres ports by default and requires an explicit AGENTOS_POSTGRES_PASSWORD before enabling the persistence profile.

Dependency And Build Checks

Run the repository check script before publishing or deploying changes:

bash scripts/check.sh

The script runs Rust formatting, workspace checks, Clippy with all targets, workspace tests, benches checks, demo smoke checks, dashboard build/lint/format checks/tests, TypeScript SDK build/tests, and Python SDK pytest/Ruff checks when the required tools are available.

Node package advisories should be checked with npm audit --audit-level=high in dashboard and crates/sdk/typescript. Rust advisory checks require installing cargo-audit or an equivalent policy tool in the local/CI environment.

Current Status

AgentOS has hardened local and container defaults, but production deployments must still add environment-specific controls such as authentication, authorization policy, TLS termination, secret injection, backup policy, and network segmentation. Treat the built-in APIs as trusted-network interfaces unless an authenticated gateway is placed in front of them.

There aren't any published security advisories