Skip to content

Security: mya-ai/agentkit

SECURITY.md

Security Policy

Reporting a vulnerability

Please do not open a public issue for a security vulnerability.

Email security@heymya.ai — that always works. You can also report it privately through GitHub's Security Advisories; if that page 404s, the setting is not enabled yet, so use the email and please mention it. We will acknowledge within a few business days.

Please include what an attacker can do, the affected version or commit, and a reproduction if you have one.

Scope

agentkit is a library. It has no network listeners and no credential handling of its own — your provider adapter and your ActorStore own those. Relevant to this repo:

  • A way to bypass the tier gate so an irreversible tool call executes inside a loop instead of being captured as a proposal.
  • A way to make capability verification accept a call the agent never declared.
  • A way to make the single-writer guarantee admit two concurrent live instances for one unit, given a correct ActorStore.

Issues in your provider adapter, datastore, or prompt content are yours to fix, though we are glad to be told if the kit's interfaces made the mistake easy to write.

There aren't any published security advisories