ETP 0.1 is an implementer draft. The repository does not provide a supported production release, a security certification, or a service-level agreement.

Security fixes before version 1.0 can change schemas, protocol behavior, storage, and compatibility.

Use GitHub private vulnerability reporting for this repository:

1. Open the repository Security tab.
2. Select Advisories.
3. Select Report a vulnerability.

If private reporting is not available, use the contact method on the maintainer's GitHub profile to request a private channel. Do not publish exploit details in an issue or discussion.

Include this information when it is safe to share:

• affected commit and platform;
• affected component and trust boundary;
• required access and preconditions;
• exact reproduction steps;
• observed and expected behavior;
• security impact;
• whether the test used credentials, external infrastructure, or private data;
• a minimal proof of concept.

Do not send live credentials, private keys, customer data, or unredacted logs.

Useful reports include:

• authorization without a valid current decision or grant;
• grant replay or more than one successful claim;
• target, argument, profile, pre-state, epoch, or audience substitution;
• a path that crosses a declared dispatch boundary without ETP mediation;
• a false not_dispatched outcome after external I/O can start;
• retry after an unresolved unknown outcome;
• conflicting receipts or reconciliation children;
• rollback that violates a documented protected-state assumption;
• parser ambiguity, acceptance of duplicate keys, or limit bypass;
• signature, role, audience, key, time, or revocation bypass;
• unintended disclosure of records, referenced documents, or keys.

The protocol does not make a compromised task authority, evaluator, issuer, executor, trust root, or host trustworthy. The protocol also does not provide an operating-system sandbox or exactly-once delivery to an arbitrary remote target. See THREAT_MODEL.md.

A report remains relevant when the implementation behaves more permissively than its specification, threat model, or declared boundary.

The maintainer will validate a report against the affected commit. The maintainer will coordinate disclosure after a fix or documented mitigation is available. Response times are best effort.

A security fix should add a regression test. It should also update the specification, threat model, or implementation status when the affected claim changes.