SRE Agent is currently pre-1.0. Security fixes are provided for the latest release only.

Please do not open a public GitHub issue for security vulnerabilities.

Report vulnerabilities through GitHub private vulnerability reporting:

https://github.com/alparn/sre-agent/security/advisories/new

Include as much detail as possible:

• Affected version or commit
• Deployment mode (local, Docker Compose, Kubernetes)
• Steps to reproduce
• Expected and actual impact
• Relevant logs with secrets removed
• Suggested fix, if known

We aim to:

• Acknowledge reports within 48 hours
• Provide an initial assessment within 7 days
• Coordinate a fix and disclosure timeline with the reporter

SRE Agent is built around least privilege and explicit safety boundaries:

• Destructive production actions require human approval
• Secrets are loaded from environment variables, not source-controlled files
• Local vault data is kept outside Git by default
• LLM prompts should never include raw credentials
• Integrations should use scoped API keys and service accounts
• Container images run as a non-root user

Never commit:

• .env or .env.* files
• API keys or tokens
• Grafana, Slack, PagerDuty, GitHub, Linear, Anthropic, OpenAI, or Composio keys
• Vault incident data that contains customer or infrastructure details
• Logs, database dumps, or trace payloads from production systems

Use .env.example for documentation only.

If a secret is accidentally committed:

1. Rotate the key immediately at the provider
2. Remove it from Git history before publishing
3. Run a secret scanner before pushing
4. Document the incident privately with affected maintainers

• Use Kubernetes RBAC with least privilege
• Run the agent in a restricted namespace
• Use NetworkPolicies to limit egress
• Store secrets in a cloud secret manager or sealed Kubernetes Secret
• Disable anonymous Grafana access outside local demos
• Require approval for all act tools in production
• Enable dependency scanning and container image scanning in CI

The repository includes GitHub workflows for:

• CodeQL
• pip-audit
• Trivy filesystem and container scanning
• Secret scanning with TruffleHog

These scans reduce risk but do not replace manual security review.