AgentIsOK is pre-alpha design and interoperability work. The protocol, profiles, reference loop, and examples are incomplete, have not received an independent security assessment, and must not be used to authorize production actions.

There are currently no supported production versions.

Do not open a public issue for a vulnerability that could create exploitation risk.

Once the repository is public, use GitHub private vulnerability reporting from the repository Security tab. GitHub does not expose that reporting feature while a repository remains private.

Before changing the repository to public visibility, maintainers must:

1. Enable private vulnerability reporting.
2. Verify that a non-collaborator can open an advisory.
3. List any additional contact path in MAINTAINERS.md.

Until then, contact a repository maintainer privately through the Univeracity GitHub organization to establish a secure reporting channel before sending sensitive details.

Include when possible:

• affected document, schema, implementation, or commit;
• threat and likely impact;
• reproduction steps or a proof of concept;
• assumptions and deployment conditions;
• suggested mitigations; and
• whether the issue is already public or under active exploitation.

Do not include real credentials, personal data, or third-party confidential information.

Maintainers will attempt to:

1. Acknowledge receipt and establish a private coordination channel.
2. Assess whether the report affects the protocol, an implementation, or both.
3. Identify affected parties and safe mitigations.
4. Coordinate fixes and disclosure proportionate to exploitation risk.
5. Publish a security advisory and protocol clarification when appropriate.

Because the project is new and volunteer capacity may vary, this document does not promise a fixed response time. Reporters should state any disclosure deadline.

Non-embargoed design weaknesses belong in public issues and should update THREAT-MODEL.md, the protocol specification, profiles, and conformance vectors together.

Particularly useful reviews include:

• signature coverage and canonicalization;
• nonce replay and concurrent redemption;
• delegation attenuation and confused-deputy behavior;
• issuer and metadata substitution;
• parser differentials;
• privacy and correlation;
• policy obligation enforcement;
• fail-open and downgrade paths; and
• centralization or registry-capture risks.

The project supports good-faith security research that avoids privacy violations, service disruption, extortion, and unnecessary data access. A formal legal safe-harbor policy has not yet been adopted; researchers should obtain authorization before testing systems they do not own.