Do not open a public GitHub issue for security vulnerabilities.
Please report security issues to the SkillOS security team via email. Include as much detail as possible: affected component, steps to reproduce, potential impact, and suggested fix if available.
You will receive a response within 48 hours. We will keep you updated on the remediation progress and credit you in the security advisory (unless you request anonymity).
| Version | Supported |
|---|---|
| 0.5.x | ✅ Active support |
| 0.4.x | ⏳ Limited support |
| < 0.4 | ❌ End of life |
SkillOS implements a defense-in-depth architecture for AI agent security:
| Layer | Feature | Description |
|---|---|---|
| L1 | Sandbox Isolation | Process-level execution isolation with resource limits, timeouts, and cleanup |
| L2 | Identity & Signing | Ed25519 challenge-response authentication with HMAC-SHA256 audit chain |
| L3 | Input Validation | Intent classification + risk assessment + keyword filtering |
| L4 | Human-in-the-Loop | High-risk operations require human gate approval before execution |
| L5 | Audit Logging | Append-only, HMAC-signed audit trail for all sensitive operations |
| L6 | Anti-Pattern Detection | NegMarket-based adversarial pattern recognition and warning |
| L7 | Secret Scanning | TruffleHog + pre-commit hooks prevent credential leakage |
| L8 | Dependency Auditing | pip-audit + Dependabot for CVE monitoring |
When deploying SkillOS in production:
- Always set
SKILLOS_API_KEYvia environment variable (never in code) - Use
--insecureflag only for local development - Enable all audit logging and monitor logs regularly
- Review pre-commit security hooks before bypassing
- Keep dependencies updated (
pip list --outdated) - Run
make securitybefore any release