Only the latest release of the AI Native Dev Stack is supported. Security fixes are prepared for the current major-minor line and published as a patch release; older releases are not patched.
| Version | Supported |
|---|---|
| latest release (currently 2.2.x) | yes |
| any earlier release | no |
Use GitHub private vulnerability reporting on this repository: Security tab, Report a vulnerability. That channel keeps the report private until a fix is published.
Do not open a public issue for a suspected vulnerability. Do not include secret values, live credentials, tokens or personal data in a public issue, a pull request description, or a CI log: reports are public the moment they are filed. Share only what is necessary to reproduce, and reference the affected file and version.
If private reporting is unavailable to you, open a minimal public issue saying only that you have a security report and would like a private channel; leave every technical detail out of it.
- Acknowledgement on a best-effort basis. This is a single-maintainer project; no response-time SLA is promised.
- Confirmed findings are fixed on
dev, released as a patch onmain, and credited in the release notes unless you ask otherwise. - If a report is out of scope, the reasoning is stated in the same channel.
Read these before reporting; several properties are deliberate and documented:
docs/DISTRIBUTION-LIFECYCLE.md,docs/RELEASING.mdanddocs/adr/0009- what the lifecycle defends against, and what SHA-256 integrity does and does not prove. Release artifacts carry a GitHub build provenance attestation (gh attestation verify <file> -R Rwanbt/ai-native-dev-stack), which binds them to the workflow, repository and commit that produced them; they are still not signed by a human maintainer identity, and #24 tracks that distinction.stack/agents/anti-debt/docs/security-boundaries.md- the anti-debt agent sandboxing and egress boundaries.docs/knowledge/KNOWLEDGE-CONVERGENCE-MATRIX.md- Knowledge/Auto-Memory guarantees, including what is intentionally not shipped (canonical auto-promotion, K5 = STOP).docs/VERIFIED-WORK-PLANE.md- the Verified Work Plane trust model.
Out of scope: findings that amount to "the documentation says the trust model is limited" where it is; vulnerabilities requiring a compromised maintainer account or a compromised release source (that is the documented residual risk).