retrieval-hub is a retrieval platform for AI agents running on OpenShift. It manages access to data sources on behalf of agents and users under a scoped, governed model. Taking vulnerability reports seriously is a first-class concern.

Please do not file public GitHub issues for security vulnerabilities.

Report vulnerabilities privately through GitHub's private vulnerability reporting:

• Go to the repository Security tab and select Report a vulnerability
• Or navigate directly to https://github.com/rdwj/retrieval-hub/security/advisories/new

If for any reason you cannot use GitHub's reporting flow, open a minimal public issue asking for a private contact channel (without disclosing details of the vulnerability).

When reporting, please include:

• A description of the issue and the affected component (retrieval_hub core library, retrieval-hub-mcp, retrieval-hub-auth, retrieval-hub-ui, or the deploy manifests)
• Steps to reproduce or a minimal proof of concept
• The version or commit SHA you observed the issue on
• The impact you believe the issue has (confidentiality, integrity, availability; scoped or cross-tenant)

You should receive an acknowledgement within a few business days. We will work with you to confirm the issue, assess impact, and coordinate a fix and disclosure timeline.

retrieval-hub is pre-1.0 and under active development. Fixes are applied to main; only the latest release of each published package receives security updates:

In scope:

• Authentication and authorization bypass
• Source access policy violations (cross-scope data access)
• Credential or token leakage
• Injection vulnerabilities in MCP tools, the auth server, or the UI BFF
• Provenance or content integrity bypass (forged signatures, tampered content hashes)
• Container or deployment misconfigurations that weaken FIPS or compliance posture

Out of scope:

• Findings that require compromising the underlying OpenShift cluster or PostgreSQL instance
• Denial-of-service via resource exhaustion against unauthenticated endpoints (rate limiting is a deployment concern)
• Issues only reproducible against example or scaffold code in scripts/ or sample datasets
• Best-practice recommendations without a demonstrated impact

We prefer coordinated disclosure. Once a fix is available, we will credit the reporter (unless anonymity is requested) in the release notes and any GitHub Security Advisory published for the issue.