Cerefox is pre-1.0. Only the latest released minor version receives security fixes. Older releases — even recent ones — should be upgraded rather than patched in place.

Once Cerefox ships v1.0.0 this policy becomes more conservative — the matrix above will be revised at that time.

Please use GitHub's private vulnerability reporting rather than a public issue:

1. Go to https://github.com/fstamatelopoulos/cerefox/security/advisories/new (or: repo → Security tab → "Report a vulnerability").
2. Describe the issue, the affected version, and steps to reproduce.
3. Include any proof-of-concept or impact analysis you have.

Do not open a public issue, PR, or discussion for security reports. Public disclosure before a fix is shipped puts every Cerefox operator at risk — including non-technical users who installed the project to keep their own notes.

Cerefox is a single-user, self-hosted memory layer. The threat model assumes:

• You control the Supabase instance and its credentials.
• The legacy anon JWT used for MCP / Edge Function access is treated as a capability token: anyone with it can read and write your knowledge base.
• The Postgres database may contain personal information (notes, chats, research) — confidentiality of the data at rest is your responsibility (Supabase encryption, network controls, etc.).

In-scope security issues include:

• Authentication / authorization bypasses on Edge Functions or the local web UI.
• SQL injection, prompt injection, or command injection in any ingestion or search path.
• Secret leakage in logs, audit entries, or response bodies.
• Vulnerable dependencies that ship in a release.

Out-of-scope (please don't file as security):

• Self-XSS in the web UI when running with privileged access on localhost.
• Findings that require the attacker to already hold valid Supabase credentials.
• Best-practice suggestions without a concrete exploit path (open an issue with the "feature" template instead).

This is a hobby / open-source project maintained part-time. Acknowledgements should arrive within a few days; fixes ship as a patch release as soon as one is ready. If a vulnerability is severe enough to warrant a coordinated disclosure, we will work with you on a timeline.