1.0.0rc1 is the only supported line while UCF Protocol is in release-candidate status. Security fixes are made on the default branch and included in the next release candidate. No older version is claimed to receive fixes.

Do not open a public issue for a vulnerability or attach private journal contents, credentials, or personal data to an issue. Use GitHub's private vulnerability-reporting feature for this repository if it is enabled. Otherwise, email support@samsarix.com with the subject UCF security report and include only the minimum reproduction needed. This mailbox is the private reporting channel unless a future release names a dedicated security address. The project does not promise a response-time SLA.

Please include the affected version, platform and Python version, impact, prerequisites, a minimal reproduction, and any suggested remediation. Give maintainers a reasonable opportunity to validate and fix the issue before public disclosure.

• confidentiality of user-supplied context and metadata;
• integrity and availability of observation history;
• correctness of metric, phase, score, and export data;
• integrity of built distributions and the CI process.

UCF runs with the invoking user's filesystem permissions. CLI arguments, environment variables, database/export paths, JSON files, stdin, metadata, and pre-existing SQLite contents are untrusted. A malicious local input producer may try to exhaust memory/disk, inject SQL, corrupt derived values, overwrite files, or trigger ambiguous parsing. A same-user process can directly read or modify the journal and is outside the protection the package can provide.

The supported runtime opens no network listener, sends no network traffic, loads no third-party plug-ins, and has no third-party dependencies. There are no server authentication, authorization, session, browser, or transport boundaries in this repository.

• Strict ucf/v1 parsing rejects unknown fields, non-finite/out-of-range metrics, invalid derived values, malformed timestamps, oversized input, and deeply nested or oversized metadata.
• SQLite statements bind values as parameters, schema constraints enforce metric ranges, duplicate event IDs fail closed, reads are bounded, and stored rows are revalidated.
• Export refuses replacement by default and uses a temporary file plus same-filesystem atomic replace.
• CI uses pinned action commits, least-privilege read permissions, strict lint/type checks, a branch coverage gate, distribution validation, and clean-wheel smoke installation.
• The product does not collect credentials or telemetry and does not silently transmit journal data.

The journal is not encrypted, signed, access-controlled beyond filesystem permissions, tamper-proof, or suitable as an audit log against a local adversary. Use appropriate OS permissions and disk encryption, keep tested backups, and choose retention and secure-deletion procedures. Do not record secrets, authentication tokens, regulated data, or unnecessary personal information in free-form fields.

SQLite is for local use, not a hostile multi-tenant or multi-host service. Export destinations are chosen by the operator; --force deliberately authorizes replacement. Installing a wheel still requires normal Python supply-chain controls: verify its source, provenance, and hash when those are available.

Security claims about hosted services under either the legacy Helix name or the current Samsarix brand, including APIs, dashboards, bots, data stores, cloud platforms, or other repositories, are out of scope. They are not part of this product or security boundary.