intent-spec is a proposed canonical intermediate representation for compiling operator intent into existing agent machinery.

Preferred: GitHub private vulnerability reporting. Open the Security tab on this repository and choose Report a vulnerability. That channel is private between you and the maintainer, requires no email, and nothing is posted publicly. Private reporting is enabled on this repository.

If you cannot use that channel, open a minimal public issue stating only that you have a security report and how to reach you. Please do not include exploit details, proof-of-concept code, or affected-version specifics in a public issue.

In scope: A representation that cannot round-trip the intent it claims to carry; a design that silently becomes a second router; and any ambiguity where two different downstream systems would read the same representation differently.

Out of scope / stated plainly: This is a specification proposal. It describes intended behaviour. It does not describe an implementation that has been adversarially tested at scale.

You will be credited in the fix or advisory unless you ask to remain anonymous.

There is no bug bounty, and no monetary reward is offered or implied. This is an independent research project maintained by one person. What it can offer is a fast, honest response and public credit.

• Our agent-systems threat posture and the method behind these reviews: see the adversarial-seat repository for the review method, and hermes-refuse for the fail-closed execution posture.