Security fixes are applied to the latest released minor version. Because the project is pre-1.0, we support only the most recent release line.
| Version | Supported |
|---|---|
| 0.1.x | ✅ |
| < 0.1 | ❌ |
Please report security issues privately. Do not open a public GitHub issue.
- Preferred: open a private advisory via GitHub Security Advisories.
- Alternatively, email hello@devopsaitoolkit.com with details and, if possible, a minimal reproduction.
We aim to acknowledge reports within 3 business days and to provide a remediation timeline after triage. Please give us a reasonable window to release a fix before any public disclosure.
devops-ai-toolkit is intentionally designed to be read-only, and this is a core security property rather than a side effect:
- It never executes commands. The tool analyzes the text you provide (logs,
YAML, Terraform plans, Kubernetes manifests, etc.). It does not run
apply,delete,restart, or any other mutating action against your infrastructure. "Diagnostic commands" surfaced by the knowledge base are suggestions for you to run, and are themselves read-only. - It never mutates the system it runs on, beyond writing its own output where you direct it.
- It does not transmit analyzed content anywhere by default. The offline, deterministic core runs entirely locally with no network access.
The only time analyzed content can leave your machine is when you explicitly enable an external AI provider (Anthropic, OpenAI, Gemini, or a self-hosted Ollama endpoint) for optional enrichment. In that case the content you submit is sent to the provider you configured, subject to that provider's terms. If you need a fully air-gapped guarantee, run the offline core only, or point enrichment at a self-hosted Ollama instance.
Because of this design, the most likely classes of vulnerability are input-handling issues (e.g., parsing untrusted input) and supply-chain issues in dependencies — these are exactly the reports we prioritize.