Before FileBelt 1.0, security fixes are provided for the latest released minor line. Older lines should upgrade to that line. The repository currently has no production release.

Use GitHub's private vulnerability reporting for this repository. Do not file a public issue or pull request containing exploitable details. Include affected versions, deployment assumptions, reproduction steps, impact, and any proposed mitigation. The project does not promise a contractual response SLA.

The maintainer will validate the report, coordinate a fix and advisory when appropriate, and credit reporters who request attribution. Security fixes must include regression coverage and upgrade or mitigation guidance.

Security review covers repository and release supply chain, identity and Virtual ACL, tenant separation, storage integrity, protocol adapters, browser rendering, integrations, secrets, Kubernetes isolation, and recovery behavior. The current core exposes an OIDC-authenticated browser/API, authenticated principal-to-principal shares, capability-limited payload streaming, PostgreSQL state, and optional Iggy notifications through a Kubernetes production topology and a Docker development/integration topology. See the current threat model and living engineering specifications.

Security-sensitive invariants include:

• every external identity resolves to a tenant-scoped internal principal;
• tenant administration grants no implicit access to user content;
• every content path enforces the common Virtual ACL and bounds stale access;
• the API has no payload mount and storage workers accept no browser session;
• PostgreSQL, not Iggy, is authoritative for metadata, policy, jobs, and audit;
• payload locators are UUIDs unrelated to user-controlled names; and
• build, runtime, and license evidence preserves the Apache/adapter boundary.

Docker profiles are for development, integration, and fault verification. Kubernetes 1.34-1.36 with Helm 4.2.4 is the supported production topology after the required acceptance gates pass. Production requires the shipped hardened chart, external PostgreSQL/OIDC/optional Iggy, an operator OIDC egress gateway, operator-managed Secrets, and an existing RWX POSIX claim. It makes no HA, online-backup, PITR, numeric RPO, or numeric RTO claim.

Operators must use a standards-compliant OIDC issuer, configure exact administrator issuer/subject pairs, provide trusted TLS and key material via secret files, use PostgreSQL 18, and use a POSIX storage filesystem that passes the startup fsync/rename/no-follow probes. Volume/provider encryption protects data at rest; FileBelt does not implement application-layer payload encryption.

OxiBelt is the public TLS edge. API and I/O backend ports require native mTLS with distinct exact OxiBelt client identities and remain isolated by NetworkPolicy. Never expose them directly or trust client identity headers. No FileBelt Pod receives general Internet egress, a Kubernetes API token, or an unnecessary payload mount. Apache Iggy is optional; only its digest-pinned helper may receive its documented SYS_NICE, memlock, and seccomp settings.

• Suspected session or identity compromise: locally suspend the user or revoke individual/all sessions, then review the append-only audit trail.
• Suspected capability-signing compromise: stop write admission, rotate the capability key generation, retain only the overlap required to drain the 60-second maximum capability lifetime, and reconcile operation state.
• Suspected session/share digest-key compromise: stop admission, rotate to a new generation, revoke affected credentials, and retain retiring material only for the documented validation window.
• Suspected payload corruption: quarantine rather than delete, stop affected reads, preserve evidence, identify every referencing version, and run an operator-directed scrub/recovery.
• Suspected database/storage inconsistency: quiesce writes, snapshot both planes, run read-only diagnostics, and follow the Kubernetes rollback runbook.

Do not include raw cookies, OIDC codes, CSRF values, share tokens, capabilities, signing/hash keys, database credentials, private payloads, or unredacted backup artifacts in a vulnerability report sent through a public channel.